This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
https://perl5.git.perl.org / perl5.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
RMG - Put all version-bump related information about INSTALL in one place
[perl5.git] / regexec.c
CommitLineData
a0d0e21e
LW		 1/* regexec.c
2 */
3
4/*
4ac71550
TC		 5 * One Ring to rule them all, One Ring to find them
6 &
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
81e983c1 83#include "inline_invlist.c"
1b0f46bf 84#include "unicode_constants.h"
81e983c1 85
e1cf74e3
CB		 86#ifdef DEBUGGING
87/* At least one required character in the target string is expressible only in
88 * UTF-8. */
89static const char* const non_utf8_target_but_utf8_required
90 = "Can't match, because target string needs to be in UTF-8\n";
91#endif
92
93#define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
94 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\
95 goto target; \
96} STMT_END
97
c74f6de9
KW		 98#define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
99
a687059c
LW		 100#ifndef STATIC
101#define STATIC static
102#endif
103
2d66f61e 104/* Valid only for non-utf8 strings: avoids the reginclass
7e2509c1
KW		 105 * call if there are no complications: i.e., if everything matchable is
106 * straight forward in the bitmap */
3db24e1e 107#define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,c+1,0) \
af364d03 108 : ANYOF_BITMAP_TEST(p,*(c)))
7d3e948e 109
c277df42
IZ		 110/*
111 * Forwards.
112 */
113
f2ed9b32 114#define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
ba44c216 115#define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b)
a0ed51b3 116
3dab1dad 117#define HOPc(pos,off) \
ba44c216 118 (char *)(reginfo->is_utf8_target \
220db18a 119 ? reghop3((U8*)pos, off, \
9d9163fb 120 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
3dab1dad 121 : (U8*)(pos + off))
557f47af 122
3dab1dad 123#define HOPBACKc(pos, off) \
ba44c216 124 (char*)(reginfo->is_utf8_target \
9d9163fb
DM		 125 ? reghopmaybe3((U8*)pos, -off, (U8*)(reginfo->strbeg)) \
126 : (pos - off >= reginfo->strbeg) \
8e11feef 127 ? (U8*)pos - off \
3dab1dad 128 : NULL)
efb30f32 129
ba44c216 130#define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
1aa99e6b 131#define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
1aa99e6b 132
557f47af
DM		 133/* lim must be +ve. Returns NULL on overshoot */
134#define HOPMAYBE3(pos,off,lim) \
135 (reginfo->is_utf8_target \
136 ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \
137 : ((U8*)pos + off <= lim) \
138 ? (U8*)pos + off \
139 : NULL)
140
8e9f2289
DM		 141/* like HOP3, but limits the result to <= lim even for the non-utf8 case.
142 * off must be >=0; args should be vars rather than expressions */
143#define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \
144 ? reghop3((U8*)(pos), off, (U8*)(lim)) \
145 : (U8*)((pos + off) > lim ? lim : (pos + off)))
146
2974eaec
DM		 147#define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \
148 ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \
149 : (U8*)(pos + off))
150#define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim))
7016d6eb
DM		 151
152#define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
153#define NEXTCHR_IS_EOS (nextchr < 0)
154
155#define SET_nextchr \
220db18a 156 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
7016d6eb
DM		 157
158#define SET_locinput(p) \
159 locinput = (p); \
160 SET_nextchr
161
162
2a16ac92 163#define LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) STMT_START { \
c7304fe2
KW		 164 if (!swash_ptr) { \
165 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
c7304fe2 166 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
2a16ac92 167 1, 0, invlist, &flags); \
c7304fe2
KW		 168 assert(swash_ptr); \
169 } \
170 } STMT_END
171
172/* If in debug mode, we test that a known character properly matches */
173#ifdef DEBUGGING
174# define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
175 property_name, \
2a16ac92 176 invlist, \
c7304fe2 177 utf8_char_in_property) \
2a16ac92 178 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist); \
c7304fe2
KW		 179 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
180#else
181# define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
182 property_name, \
2a16ac92 183 invlist, \
c7304fe2 184 utf8_char_in_property) \
2a16ac92 185 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist)
c7304fe2 186#endif
d1eb3177 187
c7304fe2
KW		 188#define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
189 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
2a16ac92
KW		 190 "", \
191 PL_XPosix_ptrs[_CC_WORDCHAR], \
df38da56 192 LATIN_CAPITAL_LETTER_SHARP_S_UTF8);
c7304fe2
KW		 193
194#define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
195 STMT_START { \
196 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \
197 "_X_regular_begin", \
2a16ac92 198 NULL, \
df38da56 199 LATIN_CAPITAL_LETTER_SHARP_S_UTF8); \
c7304fe2
KW		 200 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \
201 "_X_extend", \
2a16ac92 202 NULL, \
c7304fe2
KW		 203 COMBINING_GRAVE_ACCENT_UTF8); \
204 } STMT_END
d1eb3177 205
c7304fe2 206#define PLACEHOLDER /* Something for the preprocessor to grab onto */
3dab1dad
YO		 207/* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
208
5f80c4cf 209/* for use after a quantifier and before an EXACT-like node -- japhy */
c35dcbe2
YO		 210/* it would be nice to rework regcomp.sym to generate this stuff. sigh
211 *
212 * NOTE that *nothing* that affects backtracking should be in here, specifically
213 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
214 * node that is in between two EXACT like nodes when ascertaining what the required
215 * "follow" character is. This should probably be moved to regex compile time
216 * although it may be done at run time beause of the REF possibility - more
217 * investigation required. -- demerphq
218*/
baa60164
KW		 219#define JUMPABLE(rn) ( \
220 OP(rn) == OPEN || \
3e901dc0 221 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
baa60164
KW		 222 OP(rn) == EVAL || \
223 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
224 OP(rn) == PLUS || OP(rn) == MINMOD || \
225 OP(rn) == KEEPS || \
226 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
e2d8ce26 227)
ee9b8eae 228#define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
e2d8ce26 229
ee9b8eae
YO		 230#define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
231
232#if 0
233/* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
a4525e78 234 we don't need this definition. XXX These are now out-of-sync*/
ee9b8eae 235#define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
098b07d5 236#define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFA || OP(rn)==EXACTFA_NO_TRIE || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
ee9b8eae
YO		 237#define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
238
239#else
240/* ... so we use this as its faster. */
a4525e78
KW		 241#define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==EXACTL )
242#define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFLU8 || OP(rn)==EXACTFU_SS || OP(rn) == EXACTFA || OP(rn) == EXACTFA_NO_TRIE)
ee9b8eae
YO		 243#define IS_TEXTF(rn) ( OP(rn)==EXACTF )
244#define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
245
246#endif
e2d8ce26 247
a84d97b6
HS		 248/*
249 Search for mandatory following text node; for lookahead, the text must
250 follow but for lookbehind (rn->flags != 0) we skip to the next step.
251*/
baa60164 252#define FIND_NEXT_IMPT(rn) STMT_START { \
3dab1dad
YO		 253 while (JUMPABLE(rn)) { \
254 const OPCODE type = OP(rn); \
255 if (type == SUSPEND || PL_regkind[type] == CURLY) \
e2d8ce26 256 rn = NEXTOPER(NEXTOPER(rn)); \
3dab1dad 257 else if (type == PLUS) \
cca55fe3 258 rn = NEXTOPER(rn); \
3dab1dad 259 else if (type == IFMATCH) \
a84d97b6 260 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
e2d8ce26 261 else rn += NEXT_OFF(rn); \
3dab1dad 262 } \
5f80c4cf 263} STMT_END
74750237 264
22913b96
KW		 265/* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode.
266 * These are for the pre-composed Hangul syllables, which are all in a
267 * contiguous block and arranged there in such a way so as to facilitate
268 * alorithmic determination of their characteristics. As such, they don't need
269 * a swash, but can be determined by simple arithmetic. Almost all are
270 * GCB=LVT, but every 28th one is a GCB=LV */
271#define SBASE 0xAC00 /* Start of block */
272#define SCount 11172 /* Length of block */
273#define TCount 28
c476f425 274
006f26b2
DM		 275#define SLAB_FIRST(s) (&(s)->states[0])
276#define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
277
a75351a1 278static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
bf2039a9 279static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
bf2039a9 280static regmatch_state * S_push_slab(pTHX);
51371543 281
87c0511b 282#define REGCP_PAREN_ELEMS 3
f067efbf 283#define REGCP_OTHER_ELEMS 3
e0fa7e2b 284#define REGCP_FRAME_ELEMS 1
620d5b66
NC		 285/* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
286 * are needed for the regexp context stack bookkeeping. */
287
76e3520e 288STATIC CHECKPOINT
92da3157 289S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
a0d0e21e 290{
a3b680e6 291 const int retval = PL_savestack_ix;
92da3157
DM		 292 const int paren_elems_to_push =
293 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
e0fa7e2b
NC		 294 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
295 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
87c0511b 296 I32 p;
40a82448 297 GET_RE_DEBUG_FLAGS_DECL;
a0d0e21e 298
b93070ed
DM		 299 PERL_ARGS_ASSERT_REGCPPUSH;
300
e49a9654 301 if (paren_elems_to_push < 0)
e8a85d26
JH		 302 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
303 (int)paren_elems_to_push, (int)maxopenparen,
304 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
e49a9654 305
e0fa7e2b
NC		 306 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
307 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
5df417d0 308 " out of range (%lu-%ld)",
92da3157
DM		 309 total_elems,
310 (unsigned long)maxopenparen,
311 (long)parenfloor);
e0fa7e2b 312
620d5b66 313 SSGROW(total_elems + REGCP_FRAME_ELEMS);
7f69552c 314
495f47a5 315 DEBUG_BUFFERS_r(
92da3157 316 if ((int)maxopenparen > (int)parenfloor)
495f47a5
DM		 317 PerlIO_printf(Perl_debug_log,
318 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
319 PTR2UV(rex),
320 PTR2UV(rex->offs)
321 );
322 );
92da3157 323 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
b1ce53c5 324/* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
99a90e59
FC		 325 SSPUSHIV(rex->offs[p].end);
326 SSPUSHIV(rex->offs[p].start);
1ca2007e 327 SSPUSHINT(rex->offs[p].start_tmp);
e7707071 328 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
495f47a5
DM		 329 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
330 (UV)p,
331 (IV)rex->offs[p].start,
332 (IV)rex->offs[p].start_tmp,
333 (IV)rex->offs[p].end
40a82448 334 ));
a0d0e21e 335 }
b1ce53c5 336/* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
92da3157 337 SSPUSHINT(maxopenparen);
b93070ed
DM		 338 SSPUSHINT(rex->lastparen);
339 SSPUSHINT(rex->lastcloseparen);
e0fa7e2b 340 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
41123dfd 341
a0d0e21e
LW		 342 return retval;
343}
344
c277df42 345/* These are needed since we do not localize EVAL nodes: */
ab3bbdeb
YO		 346#define REGCP_SET(cp) \
347 DEBUG_STATE_r( \
ab3bbdeb 348 PerlIO_printf(Perl_debug_log, \
e4f74956 349 " Setting an EVAL scope, savestack=%"IVdf"\n", \
ab3bbdeb
YO		 350 (IV)PL_savestack_ix)); \
351 cp = PL_savestack_ix
c3464db5 352
ab3bbdeb 353#define REGCP_UNWIND(cp) \
e4f74956 354 DEBUG_STATE_r( \
ab3bbdeb 355 if (cp != PL_savestack_ix) \
e4f74956
YO		 356 PerlIO_printf(Perl_debug_log, \
357 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
ab3bbdeb
YO		 358 (IV)(cp), (IV)PL_savestack_ix)); \
359 regcpblow(cp)
c277df42 360
a8d1f4b4
DM		 361#define UNWIND_PAREN(lp, lcp) \
362 for (n = rex->lastparen; n > lp; n--) \
363 rex->offs[n].end = -1; \
364 rex->lastparen = n; \
365 rex->lastcloseparen = lcp;
366
367
f067efbf 368STATIC void
92da3157 369S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
a0d0e21e 370{
e0fa7e2b 371 UV i;
87c0511b 372 U32 paren;
a3621e74
YO		 373 GET_RE_DEBUG_FLAGS_DECL;
374
7918f24d
NC		 375 PERL_ARGS_ASSERT_REGCPPOP;
376
b1ce53c5 377 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
c6bf6a65 378 i = SSPOPUV;
e0fa7e2b
NC		 379 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
380 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
b93070ed
DM		 381 rex->lastcloseparen = SSPOPINT;
382 rex->lastparen = SSPOPINT;
92da3157 383 *maxopenparen_p = SSPOPINT;
b1ce53c5 384
620d5b66 385 i -= REGCP_OTHER_ELEMS;
b1ce53c5 386 /* Now restore the parentheses context. */
495f47a5
DM		 387 DEBUG_BUFFERS_r(
388 if (i || rex->lastparen + 1 <= rex->nparens)
389 PerlIO_printf(Perl_debug_log,
390 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
391 PTR2UV(rex),
392 PTR2UV(rex->offs)
393 );
394 );
92da3157 395 paren = *maxopenparen_p;
620d5b66 396 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
ea3daa5d 397 SSize_t tmps;
1ca2007e 398 rex->offs[paren].start_tmp = SSPOPINT;
99a90e59
FC		 399 rex->offs[paren].start = SSPOPIV;
400 tmps = SSPOPIV;
b93070ed
DM		 401 if (paren <= rex->lastparen)
402 rex->offs[paren].end = tmps;
495f47a5
DM		 403 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
404 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
405 (UV)paren,
406 (IV)rex->offs[paren].start,
407 (IV)rex->offs[paren].start_tmp,
408 (IV)rex->offs[paren].end,
409 (paren > rex->lastparen ? "(skipped)" : ""));
c277df42 410 );
87c0511b 411 paren--;
a0d0e21e 412 }
daf18116 413#if 1
dafc8851
JH		 414 /* It would seem that the similar code in regtry()
415 * already takes care of this, and in fact it is in
416 * a better location to since this code can #if 0-ed out
417 * but the code in regtry() is needed or otherwise tests
418 * requiring null fields (pat.t#187 and split.t#{13,14}
daf18116
JH		 419 * (as of patchlevel 7877) will fail. Then again,
420 * this code seems to be necessary or otherwise
225593e1
DM		 421 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
422 * --jhi updated by dapm */
b93070ed 423 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
92da3157 424 if (i > *maxopenparen_p)
b93070ed
DM		 425 rex->offs[i].start = -1;
426 rex->offs[i].end = -1;
495f47a5
DM		 427 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
428 " \\%"UVuf": %s ..-1 undeffing\n",
429 (UV)i,
92da3157 430 (i > *maxopenparen_p) ? "-1" : " "
495f47a5 431 ));
a0d0e21e 432 }
dafc8851 433#endif
a0d0e21e
LW		 434}
435
74088413
DM		 436/* restore the parens and associated vars at savestack position ix,
437 * but without popping the stack */
438
439STATIC void
92da3157 440S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
74088413
DM		 441{
442 I32 tmpix = PL_savestack_ix;
443 PL_savestack_ix = ix;
92da3157 444 regcppop(rex, maxopenparen_p);
74088413
DM		 445 PL_savestack_ix = tmpix;
446}
447
02db2b7b 448#define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
a0d0e21e 449
31c7f561
KW		 450STATIC bool
451S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
452{
453 /* Returns a boolean as to whether or not 'character' is a member of the
454 * Posix character class given by 'classnum' that should be equivalent to a
455 * value in the typedef '_char_class_number'.
456 *
457 * Ideally this could be replaced by a just an array of function pointers
458 * to the C library functions that implement the macros this calls.
459 * However, to compile, the precise function signatures are required, and
460 * these may vary from platform to to platform. To avoid having to figure
461 * out what those all are on each platform, I (khw) am using this method,
7aee35ff
KW		 462 * which adds an extra layer of function call overhead (unless the C
463 * optimizer strips it away). But we don't particularly care about
464 * performance with locales anyway. */
31c7f561
KW		 465
466 switch ((_char_class_number) classnum) {
15861f94 467 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
31c7f561 468 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
e8d596e0
KW		 469 case _CC_ENUM_ASCII: return isASCII_LC(character);
470 case _CC_ENUM_BLANK: return isBLANK_LC(character);
b0d691b2
KW		 471 case _CC_ENUM_CASED: return isLOWER_LC(character)
472 || isUPPER_LC(character);
e8d596e0 473 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
31c7f561
KW		 474 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
475 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
476 case _CC_ENUM_LOWER: return isLOWER_LC(character);
477 case _CC_ENUM_PRINT: return isPRINT_LC(character);
e8d596e0 478 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
31c7f561 479 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
e8d596e0 480 case _CC_ENUM_SPACE: return isSPACE_LC(character);
31c7f561
KW		 481 case _CC_ENUM_UPPER: return isUPPER_LC(character);
482 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
31c7f561 483 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
31c7f561
KW		 484 default: /* VERTSPACE should never occur in locales */
485 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
486 }
487
e5964223 488 NOT_REACHED; /* NOTREACHED */
31c7f561
KW		 489 return FALSE;
490}
491
3018b823
KW		 492STATIC bool
493S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
494{
495 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
496 * 'character' is a member of the Posix character class given by 'classnum'
497 * that should be equivalent to a value in the typedef
498 * '_char_class_number'.
499 *
500 * This just calls isFOO_lc on the code point for the character if it is in
2f306ab9 501 * the range 0-255. Outside that range, all characters use Unicode
3018b823
KW		 502 * rules, ignoring any locale. So use the Unicode function if this class
503 * requires a swash, and use the Unicode macro otherwise. */
504
505 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
506
507 if (UTF8_IS_INVARIANT(*character)) {
508 return isFOO_lc(classnum, *character);
509 }
510 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
511 return isFOO_lc(classnum,
94bb8c36 512 TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1)));
3018b823
KW		 513 }
514
613abc6d
KW		 515 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(character, character + UTF8SKIP(character));
516
3018b823
KW		 517 if (classnum < _FIRST_NON_SWASH_CC) {
518
519 /* Initialize the swash unless done already */
520 if (! PL_utf8_swash_ptrs[classnum]) {
521 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2a16ac92
KW		 522 PL_utf8_swash_ptrs[classnum] =
523 _core_swash_init("utf8",
524 "",
525 &PL_sv_undef, 1, 0,
526 PL_XPosix_ptrs[classnum], &flags);
3018b823
KW		 527 }
528
92a2046b
KW		 529 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
530 character,
531 TRUE /* is UTF */ ));
3018b823
KW		 532 }
533
534 switch ((_char_class_number) classnum) {
535 case _CC_ENUM_SPACE:
536 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
537
538 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
539 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
540 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
e1ee3960 541 default: break;
3018b823
KW		 542 }
543
e1ee3960 544 return FALSE; /* Things like CNTRL are always below 256 */
3018b823
KW		 545}
546
a687059c 547/*
e50aee73 548 * pregexec and friends
a687059c
LW		 549 */
550
76234dfb 551#ifndef PERL_IN_XSUB_RE
a687059c 552/*
c277df42 553 - pregexec - match a regexp against a string
a687059c 554 */
c277df42 555I32
5aaab254 556Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
ea3daa5d 557 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
8fd1a950
DM		 558/* stringarg: the point in the string at which to begin matching */
559/* strend: pointer to null at end of string */
560/* strbeg: real beginning of string */
561/* minend: end of match must be >= minend bytes after stringarg. */
562/* screamer: SV being matched: only used for utf8 flag, pos() etc; string
563 * itself is accessed via the pointers above */
564/* nosave: For optimizations. */
c277df42 565{
7918f24d
NC		 566 PERL_ARGS_ASSERT_PREGEXEC;
567
c277df42 568 return
9041c2e3 569 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
c277df42
IZ		 570 nosave ? 0 : REXEC_COPY_STR);
571}
76234dfb 572#endif
22e551b9 573
cad2e5aa 574
6eb5f6b9 575
1a4edc3c
DM		 576/* re_intuit_start():
577 *
578 * Based on some optimiser hints, try to find the earliest position in the
579 * string where the regex could match.
580 *
581 * rx: the regex to match against
582 * sv: the SV being matched: only used for utf8 flag; the string
583 * itself is accessed via the pointers below. Note that on
584 * something like an overloaded SV, SvPOK(sv) may be false
585 * and the string pointers may point to something unrelated to
586 * the SV itself.
587 * strbeg: real beginning of string
588 * strpos: the point in the string at which to begin matching
589 * strend: pointer to the byte following the last char of the string
590 * flags currently unused; set to 0
591 * data: currently unused; set to NULL
592 *
593 * The basic idea of re_intuit_start() is to use some known information
594 * about the pattern, namely:
595 *
596 * a) the longest known anchored substring (i.e. one that's at a
597 * constant offset from the beginning of the pattern; but not
598 * necessarily at a fixed offset from the beginning of the
599 * string);
600 * b) the longest floating substring (i.e. one that's not at a constant
601 * offset from the beginning of the pattern);
602 * c) Whether the pattern is anchored to the string; either
603 * an absolute anchor: /^../, or anchored to \n: /^.../m,
604 * or anchored to pos(): /\G/;
605 * d) A start class: a real or synthetic character class which
606 * represents which characters are legal at the start of the pattern;
607 *
608 * to either quickly reject the match, or to find the earliest position
609 * within the string at which the pattern might match, thus avoiding
610 * running the full NFA engine at those earlier locations, only to
611 * eventually fail and retry further along.
612 *
613 * Returns NULL if the pattern can't match, or returns the address within
614 * the string which is the earliest place the match could occur.
615 *
616 * The longest of the anchored and floating substrings is called 'check'
617 * and is checked first. The other is called 'other' and is checked
618 * second. The 'other' substring may not be present. For example,
619 *
620 * /(abc|xyz)ABC\d{0,3}DEFG/
621 *
622 * will have
623 *
624 * check substr (float) = "DEFG", offset 6..9 chars
625 * other substr (anchored) = "ABC", offset 3..3 chars
626 * stclass = [ax]
627 *
628 * Be aware that during the course of this function, sometimes 'anchored'
629 * refers to a substring being anchored relative to the start of the
630 * pattern, and sometimes to the pattern itself being anchored relative to
631 * the string. For example:
632 *
633 * /\dabc/: "abc" is anchored to the pattern;
634 * /^\dabc/: "abc" is anchored to the pattern and the string;
635 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
636 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
637 * but the pattern is anchored to the string.
52a21eb3
DM		 638 */
639
cad2e5aa 640char *
52a21eb3
DM		 641Perl_re_intuit_start(pTHX_
642 REGEXP * const rx,
643 SV *sv,
644 const char * const strbeg,
645 char *strpos,
646 char *strend,
647 const U32 flags,
648 re_scream_pos_data *data)
cad2e5aa 649{
8d919b0a 650 struct regexp *const prog = ReANY(rx);
6b071d16 651 SSize_t start_shift = prog->check_offset_min;
cad2e5aa 652 /* Should be nonnegative! */
ea3daa5d 653 SSize_t end_shift = 0;
0fc004dd
DM		 654 /* current lowest pos in string where the regex can start matching */
655 char *rx_origin = strpos;
eb578fdb 656 SV *check;
f2ed9b32 657 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
6480a6c4 658 U8 other_ix = 1 - prog->substrs->check_ix;
6ad5ffb3 659 bool ml_anch = 0;
8f4bf5fc 660 char *other_last = strpos;/* latest pos 'other' substr already checked to */
bd61b366 661 char *check_at = NULL; /* check substr found at this pos */
bbe252da 662 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
f8fc2ecf 663 RXi_GET_DECL(prog,progi);
02d5137b
DM		 664 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
665 regmatch_info *const reginfo = &reginfo_buf;
a3621e74
YO		 666 GET_RE_DEBUG_FLAGS_DECL;
667
7918f24d 668 PERL_ARGS_ASSERT_RE_INTUIT_START;
c33e64f0
FC		 669 PERL_UNUSED_ARG(flags);
670 PERL_UNUSED_ARG(data);
7918f24d 671
1dc475d0
DM		 672 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
673 "Intuit: trying to determine minimum start position...\n"));
674
fb9bbddb
DM		 675 /* for now, assume that all substr offsets are positive. If at some point
676 * in the future someone wants to do clever things with look-behind and
677 * -ve offsets, they'll need to fix up any code in this function
678 * which uses these offsets. See the thread beginning
679 * <20140113145929.GF27210@iabyn.com>
680 */
681 assert(prog->substrs->data[0].min_offset >= 0);
682 assert(prog->substrs->data[0].max_offset >= 0);
683 assert(prog->substrs->data[1].min_offset >= 0);
684 assert(prog->substrs->data[1].max_offset >= 0);
685 assert(prog->substrs->data[2].min_offset >= 0);
686 assert(prog->substrs->data[2].max_offset >= 0);
687
f7022b5a 688 /* for now, assume that if both present, that the floating substring
83f2232d 689 * doesn't start before the anchored substring.
f7022b5a
DM		 690 * If you break this assumption (e.g. doing better optimisations
691 * with lookahead/behind), then you'll need to audit the code in this
692 * function carefully first
693 */
694 assert(
695 ! ( (prog->anchored_utf8 || prog->anchored_substr)
696 && (prog->float_utf8 || prog->float_substr))
697 || (prog->float_min_offset >= prog->anchored_offset));
698
1a4edc3c
DM		 699 /* byte rather than char calculation for efficiency. It fails
700 * to quickly reject some cases that can't match, but will reject
701 * them later after doing full char arithmetic */
c344f387 702 if (prog->minlen > strend - strpos) {
a3621e74 703 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 704 " String too short...\n"));
cad2e5aa 705 goto fail;
2c2d71f5 706 }
d8da0584 707
ab4e48c1 708 RX_MATCH_UTF8_set(rx,utf8_target);
6c3fea77 709 reginfo->is_utf8_target = cBOOL(utf8_target);
bf2039a9 710 reginfo->info_aux = NULL;
9d9163fb 711 reginfo->strbeg = strbeg;
220db18a 712 reginfo->strend = strend;
aed7b151 713 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
02d5137b 714 reginfo->intuit = 1;
1cb48e53
DM		 715 /* not actually used within intuit, but zero for safety anyway */
716 reginfo->poscache_maxiter = 0;
02d5137b 717
f2ed9b32 718 if (utf8_target) {
33b8afdf
JH		 719 if (!prog->check_utf8 && prog->check_substr)
720 to_utf8_substr(prog);
721 check = prog->check_utf8;
722 } else {
7e0d5ad7
KW		 723 if (!prog->check_substr && prog->check_utf8) {
724 if (! to_byte_substr(prog)) {
6b54ddc5 725 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
7e0d5ad7
KW		 726 }
727 }
33b8afdf
JH		 728 check = prog->check_substr;
729 }
274cd312 730
1dc475d0
DM		 731 /* dump the various substring data */
732 DEBUG_OPTIMISE_MORE_r({
733 int i;
734 for (i=0; i<=2; i++) {
735 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
736 : prog->substrs->data[i].substr);
737 if (!sv)
738 continue;
739
740 PerlIO_printf(Perl_debug_log,
741 " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf
742 " useful=%"IVdf" utf8=%d [%s]\n",
743 i,
744 (IV)prog->substrs->data[i].min_offset,
745 (IV)prog->substrs->data[i].max_offset,
746 (IV)prog->substrs->data[i].end_shift,
747 BmUSEFUL(sv),
748 utf8_target ? 1 : 0,
749 SvPEEK(sv));
750 }
751 });
752
8e1490ee 753 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
9fc7410e
DM		 754
755 /* ml_anch: check after \n?
756 *
757 * A note about IMPLICIT: on an un-anchored pattern beginning
758 * with /.*.../, these flags will have been added by the
759 * compiler:
760 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
761 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
762 */
7d2d37f5
DM		 763 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
764 && !(prog->intflags & PREGf_IMPLICIT);
cad2e5aa 765
343c8a29 766 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
c889ccc8
DM		 767 /* we are only allowed to match at BOS or \G */
768
57fcbfa7 769 /* trivially reject if there's a BOS anchor and we're not at BOS.
7bb3b9eb
DM		 770 *
771 * Note that we don't try to do a similar quick reject for
772 * \G, since generally the caller will have calculated strpos
773 * based on pos() and gofs, so the string is already correctly
774 * anchored by definition; and handling the exceptions would
775 * be too fiddly (e.g. REXEC_IGNOREPOS).
57fcbfa7 776 */
7bb3b9eb 777 if ( strpos != strbeg
d3d47aac 778 && (prog->intflags & PREGf_ANCH_SBOL))
c889ccc8
DM		 779 {
780 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 781 " Not at start...\n"));
c889ccc8
DM		 782 goto fail;
783 }
784
a5d12a4b
DM		 785 /* in the presence of an anchor, the anchored (relative to the
786 * start of the regex) substr must also be anchored relative
66b7ec5c
DM		 787 * to strpos. So quickly reject if substr isn't found there.
788 * This works for \G too, because the caller will already have
789 * subtracted gofs from pos, and gofs is the offset from the
790 * \G to the start of the regex. For example, in /.abc\Gdef/,
791 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
792 * caller will have set strpos=pos()-4; we look for the substr
793 * at position pos()-4+1, which lines up with the "a" */
a5d12a4b 794
c889ccc8 795 if (prog->check_offset_min == prog->check_offset_max
d0d44648 796 && !(prog->intflags & PREGf_CANY_SEEN))
c889ccc8
DM		 797 {
798 /* Substring at constant offset from beg-of-str... */
d307bf57 799 SSize_t slen = SvCUR(check);
343c8a29 800 char *s = HOP3c(strpos, prog->check_offset_min, strend);
1de06328 801
1dc475d0
DM		 802 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
803 " Looking for check substr at fixed offset %"IVdf"...\n",
804 (IV)prog->check_offset_min));
805
7742aa66
DM		 806 if (SvTAIL(check)) {
807 /* In this case, the regex is anchored at the end too.
808 * Unless it's a multiline match, the lengths must match
809 * exactly, give or take a \n. NB: slen >= 1 since
810 * the last char of check is \n */
811 if (!multiline
812 && ( strend - s > slen
813 || strend - s < slen - 1
814 || (strend - s == slen && strend[-1] != '\n')))
c889ccc8
DM		 815 {
816 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 817 " String too long...\n"));
c889ccc8
DM		 818 goto fail_finish;
819 }
820 /* Now should match s[0..slen-2] */
821 slen--;
c889ccc8 822 }
d307bf57
DM		 823 if (slen && (*SvPVX_const(check) != *s
824 || (slen > 1 && memNE(SvPVX_const(check), s, slen))))
825 {
826 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 827 " String not equal...\n"));
d307bf57
DM		 828 goto fail_finish;
829 }
c889ccc8
DM		 830
831 check_at = s;
832 goto success_at_start;
cad2e5aa 833 }
cad2e5aa 834 }
cad2e5aa 835 }
0fc004dd 836
c0e0ec46 837 end_shift = prog->check_end_shift;
cad2e5aa 838
19188028 839#ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
0033605d 840 if (end_shift < 0)
1de06328 841 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
220fc49f 842 (IV)end_shift, RX_PRECOMP(prog));
2c2d71f5
JH		 843#endif
844
2c2d71f5 845 restart:
1de06328 846
66b7ec5c
DM		 847 /* This is the (re)entry point of the main loop in this function.
848 * The goal of this loop is to:
849 * 1) find the "check" substring in the region rx_origin..strend
850 * (adjusted by start_shift / end_shift). If not found, reject
851 * immediately.
852 * 2) If it exists, look for the "other" substr too if defined; for
853 * example, if the check substr maps to the anchored substr, then
854 * check the floating substr, and vice-versa. If not found, go
855 * back to (1) with rx_origin suitably incremented.
856 * 3) If we find an rx_origin position that doesn't contradict
857 * either of the substrings, then check the possible additional
858 * constraints on rx_origin of /^.../m or a known start class.
859 * If these fail, then depending on which constraints fail, jump
860 * back to here, or to various other re-entry points further along
861 * that skip some of the first steps.
862 * 4) If we pass all those tests, update the BmUSEFUL() count on the
863 * substring. If the start position was determined to be at the
864 * beginning of the string - so, not rejected, but not optimised,
865 * since we have to run regmatch from position 0 - decrement the
866 * BmUSEFUL() count. Otherwise increment it.
867 */
868
1a4edc3c
DM		 869
870 /* first, look for the 'check' substring */
871
1de06328 872 {
c33e64f0
FC		 873 U8* start_point;
874 U8* end_point;
c889ccc8 875
c889ccc8 876 DEBUG_OPTIMISE_MORE_r({
1dc475d0 877 PerlIO_printf(Perl_debug_log,
ae5d4331 878 " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf
1dc475d0 879 " Start shift: %"IVdf" End shift %"IVdf
4d281893 880 " Real end Shift: %"IVdf"\n",
197641b6 881 (IV)(rx_origin - strpos),
c889ccc8 882 (IV)prog->check_offset_min,
12fbc530
DM		 883 (IV)start_shift,
884 (IV)end_shift,
c889ccc8
DM		 885 (IV)prog->check_end_shift);
886 });
1de06328 887
0d331aaf 888 if (prog->intflags & PREGf_CANY_SEEN) {
0fc004dd 889 start_point= (U8*)(rx_origin + start_shift);
12fbc530 890 end_point= (U8*)(strend - end_shift);
557f47af
DM		 891 if (start_point > end_point)
892 goto fail_finish;
1de06328 893 } else {
557f47af
DM		 894 end_point = HOP3(strend, -end_shift, strbeg);
895 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
896 if (!start_point)
897 goto fail_finish;
1de06328 898 }
c889ccc8 899
557f47af 900
e0362b86 901 /* If the regex is absolutely anchored to either the start of the
d3d47aac 902 * string (SBOL) or to pos() (ANCH_GPOS), then
e0362b86
DM		 903 * check_offset_max represents an upper bound on the string where
904 * the substr could start. For the ANCH_GPOS case, we assume that
905 * the caller of intuit will have already set strpos to
906 * pos()-gofs, so in this case strpos + offset_max will still be
907 * an upper bound on the substr.
908 */
c19c836a
DM		 909 if (!ml_anch
910 && prog->intflags & PREGf_ANCH
e0362b86 911 && prog->check_offset_max != SSize_t_MAX)
c19c836a 912 {
1a08ba3a 913 SSize_t len = SvCUR(check) - !!SvTAIL(check);
e0362b86
DM		 914 const char * const anchor =
915 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
916
917 /* do a bytes rather than chars comparison. It's conservative;
918 * so it skips doing the HOP if the result can't possibly end
919 * up earlier than the old value of end_point.
920 */
921 if ((char*)end_point - anchor > prog->check_offset_max) {
922 end_point = HOP3lim((U8*)anchor,
923 prog->check_offset_max,
924 end_point -len)
925 + len;
926 }
d6ef1678
DM		 927 }
928
6bda09f9 929 DEBUG_OPTIMISE_MORE_r({
1dc475d0 930 PerlIO_printf(Perl_debug_log, " fbm_instr len=%d str=<%.*s>\n",
1de06328 931 (int)(end_point - start_point),
fc8cd66c 932 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
1de06328
YO		 933 start_point);
934 });
935
ae5d4331 936 check_at = fbm_instr( start_point, end_point,
7fba1cd6 937 check, multiline ? FBMrf_MULTILINE : 0);
c889ccc8 938
8fd34720
DM		 939 /* Update the count-of-usability, remove useless subpatterns,
940 unshift s. */
941
942 DEBUG_EXECUTE_r({
943 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
944 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
945 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s",
946 (check_at ? "Found" : "Did not find"),
947 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
948 ? "anchored" : "floating"),
949 quoted,
950 RE_SV_TAIL(check),
951 (check_at ? " at offset " : "...\n") );
952 });
2c2d71f5 953
8fd34720
DM		 954 if (!check_at)
955 goto fail_finish;
956 /* Finish the diagnostic message */
957 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(check_at - strpos)) );
2c2d71f5 958
8fd34720
DM		 959 /* set rx_origin to the minimum position where the regex could start
960 * matching, given the constraint of the just-matched check substring.
961 * But don't set it lower than previously.
962 */
fdc003fd 963
8fd34720
DM		 964 if (check_at - rx_origin > prog->check_offset_max)
965 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
966 }
fdc003fd
DM		 967
968
1a4edc3c 969 /* now look for the 'other' substring if defined */
2c2d71f5 970
6480a6c4
DM		 971 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
972 : prog->substrs->data[other_ix].substr)
1de06328 973 {
30944b6d 974 /* Take into account the "other" substring. */
6c3343a6
DM		 975 char *last, *last1;
976 char *s;
977 SV* must;
978 struct reg_substr_datum *other;
979
980 do_other_substr:
981 other = &prog->substrs->data[other_ix];
982
983 /* if "other" is anchored:
984 * we've previously found a floating substr starting at check_at.
985 * This means that the regex origin must lie somewhere
986 * between min (rx_origin): HOP3(check_at, -check_offset_max)
987 * and max: HOP3(check_at, -check_offset_min)
988 * (except that min will be >= strpos)
989 * So the fixed substr must lie somewhere between
990 * HOP3(min, anchored_offset)
991 * HOP3(max, anchored_offset) + SvCUR(substr)
992 */
993
994 /* if "other" is floating
995 * Calculate last1, the absolute latest point where the
996 * floating substr could start in the string, ignoring any
997 * constraints from the earlier fixed match. It is calculated
998 * as follows:
999 *
1000 * strend - prog->minlen (in chars) is the absolute latest
1001 * position within the string where the origin of the regex
1002 * could appear. The latest start point for the floating
1003 * substr is float_min_offset(*) on from the start of the
1004 * regex. last1 simply combines thee two offsets.
1005 *
1006 * (*) You might think the latest start point should be
1007 * float_max_offset from the regex origin, and technically
1008 * you'd be correct. However, consider
1009 * /a\d{2,4}bcd\w/
1010 * Here, float min, max are 3,5 and minlen is 7.
1011 * This can match either
1012 * /a\d\dbcd\w/
1013 * /a\d\d\dbcd\w/
1014 * /a\d\d\d\dbcd\w/
1015 * In the first case, the regex matches minlen chars; in the
1016 * second, minlen+1, in the third, minlen+2.
1017 * In the first case, the floating offset is 3 (which equals
1018 * float_min), in the second, 4, and in the third, 5 (which
1019 * equals float_max). In all cases, the floating string bcd
1020 * can never start more than 4 chars from the end of the
1021 * string, which equals minlen - float_min. As the substring
1022 * starts to match more than float_min from the start of the
1023 * regex, it makes the regex match more than minlen chars,
1024 * and the two cancel each other out. So we can always use
1025 * float_min - minlen, rather than float_max - minlen for the
1026 * latest position in the string.
1027 *
1028 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1029 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1030 */
1031
e7a14a9c 1032 assert(prog->minlen >= other->min_offset);
6c3343a6
DM		 1033 last1 = HOP3c(strend,
1034 other->min_offset - prog->minlen, strbeg);
1035
4d006249 1036 if (other_ix) {/* i.e. if (other-is-float) */
6c3343a6
DM		 1037 /* last is the latest point where the floating substr could
1038 * start, *given* any constraints from the earlier fixed
1039 * match. This constraint is that the floating string starts
1040 * <= float_max_offset chars from the regex origin (rx_origin).
1041 * If this value is less than last1, use it instead.
eb3831ce 1042 */
6c3343a6
DM		 1043 assert(rx_origin <= last1);
1044 last =
1045 /* this condition handles the offset==infinity case, and
1046 * is a short-cut otherwise. Although it's comparing a
1047 * byte offset to a char length, it does so in a safe way,
1048 * since 1 char always occupies 1 or more bytes,
1049 * so if a string range is (last1 - rx_origin) bytes,
1050 * it will be less than or equal to (last1 - rx_origin)
1051 * chars; meaning it errs towards doing the accurate HOP3
1052 * rather than just using last1 as a short-cut */
1053 (last1 - rx_origin) < other->max_offset
1054 ? last1
1055 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1056 }
1057 else {
1058 assert(strpos + start_shift <= check_at);
1059 last = HOP4c(check_at, other->min_offset - start_shift,
1060 strbeg, strend);
1061 }
ead917d0 1062
6c3343a6
DM		 1063 s = HOP3c(rx_origin, other->min_offset, strend);
1064 if (s < other_last) /* These positions already checked */
1065 s = other_last;
1066
1067 must = utf8_target ? other->utf8_substr : other->substr;
1068 assert(SvPOK(must));
1069 s = fbm_instr(
1070 (unsigned char*)s,
1071 (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0),
1072 must,
1073 multiline ? FBMrf_MULTILINE : 0
1074 );
1075 DEBUG_EXECUTE_r({
1076 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1077 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1078 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s",
1079 s ? "Found" : "Contradicts",
1080 other_ix ? "floating" : "anchored",
1081 quoted, RE_SV_TAIL(must));
1082 });
ead917d0 1083
ead917d0 1084
6c3343a6
DM		 1085 if (!s) {
1086 /* last1 is latest possible substr location. If we didn't
1087 * find it before there, we never will */
1088 if (last >= last1) {
1089 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1090 ", giving up...\n"));
1091 goto fail_finish;
ead917d0
DM		 1092 }
1093
6c3343a6
DM		 1094 /* try to find the check substr again at a later
1095 * position. Maybe next time we'll find the "other" substr
1096 * in range too */
1097 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1098 ", trying %s at offset %ld...\n",
1099 (other_ix ? "floating" : "anchored"),
197641b6 1100 (long)(HOP3c(check_at, 1, strend) - strpos)));
6c3343a6
DM		 1101
1102 other_last = HOP3c(last, 1, strend) /* highest failure */;
1103 rx_origin =
4d006249 1104 other_ix /* i.e. if other-is-float */
6c3343a6
DM		 1105 ? HOP3c(rx_origin, 1, strend)
1106 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1107 goto restart;
1108 }
1109 else {
1110 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
197641b6 1111 (long)(s - strpos)));
ead917d0 1112
4d006249 1113 if (other_ix) { /* if (other-is-float) */
6c3343a6
DM		 1114 /* other_last is set to s, not s+1, since its possible for
1115 * a floating substr to fail first time, then succeed
1116 * second time at the same floating position; e.g.:
1117 * "-AB--AABZ" =~ /\wAB\d*Z/
1118 * The first time round, anchored and float match at
1119 * "-(AB)--AAB(Z)" then fail on the initial \w character
1120 * class. Second time round, they match at "-AB--A(AB)(Z)".
1121 */
1122 other_last = s;
ead917d0
DM		 1123 }
1124 else {
6c3343a6
DM		 1125 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1126 other_last = HOP3c(s, 1, strend);
ead917d0 1127 }
6c3343a6 1128 }
cad2e5aa 1129 }
acba93e8
DM		 1130 else {
1131 DEBUG_OPTIMISE_MORE_r(
1132 PerlIO_printf(Perl_debug_log,
1133 " Check-only match: offset min:%"IVdf" max:%"IVdf
1c1c599d 1134 " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf
acba93e8
DM		 1135 " strend-strpos:%"IVdf"\n",
1136 (IV)prog->check_offset_min,
1137 (IV)prog->check_offset_max,
1c1c599d 1138 (IV)(check_at-strpos),
acba93e8 1139 (IV)(rx_origin-strpos),
1c1c599d 1140 (IV)(rx_origin-check_at),
acba93e8
DM		 1141 (IV)(strend-strpos)
1142 )
1143 );
1144 }
2c2d71f5 1145
acba93e8 1146 postprocess_substr_matches:
0991020e 1147
1a4edc3c 1148 /* handle the extra constraint of /^.../m if present */
e3c6feb0 1149
7d2d37f5 1150 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
4620cb61
DM		 1151 char *s;
1152
a62659bd
DM		 1153 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1154 " looking for /^/m anchor"));
d0880ea7
DM		 1155
1156 /* we have failed the constraint of a \n before rx_origin.
2e759faa
DM		 1157 * Find the next \n, if any, even if it's beyond the current
1158 * anchored and/or floating substrings. Whether we should be
1159 * scanning ahead for the next \n or the next substr is debatable.
1160 * On the one hand you'd expect rare substrings to appear less
1161 * often than \n's. On the other hand, searching for \n means
1162 * we're effectively flipping been check_substr and "\n" on each
1163 * iteration as the current "rarest" string candidate, which
1164 * means for example that we'll quickly reject the whole string if
1165 * hasn't got a \n, rather than trying every substr position
1166 * first
1167 */
d0880ea7 1168
4620cb61
DM		 1169 s = HOP3c(strend, - prog->minlen, strpos);
1170 if (s <= rx_origin ||
1171 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1172 {
d0880ea7
DM		 1173 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1174 " Did not find /%s^%s/m...\n",
1175 PL_colors[0], PL_colors[1]));
a62659bd
DM		 1176 goto fail_finish;
1177 }
d0880ea7 1178
4ada1233
DM		 1179 /* earliest possible origin is 1 char after the \n.
1180 * (since *rx_origin == '\n', it's safe to ++ here rather than
1181 * HOP(rx_origin, 1)) */
1182 rx_origin++;
d0880ea7 1183
f4f115de 1184 if (prog->substrs->check_ix == 0 /* check is anchored */
4ada1233 1185 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
f4f115de 1186 {
d0880ea7
DM		 1187 /* Position contradicts check-string; either because
1188 * check was anchored (and thus has no wiggle room),
4ada1233 1189 * or check was float and rx_origin is above the float range */
d0880ea7
DM		 1190 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1191 " Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
4ada1233 1192 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
d0880ea7
DM		 1193 goto restart;
1194 }
1195
1196 /* if we get here, the check substr must have been float,
2e759faa 1197 * is in range, and we may or may not have had an anchored
d0880ea7
DM		 1198 * "other" substr which still contradicts */
1199 assert(prog->substrs->check_ix); /* check is float */
1200
1201 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1202 /* whoops, the anchored "other" substr exists, so we still
1203 * contradict. On the other hand, the float "check" substr
1204 * didn't contradict, so just retry the anchored "other"
1205 * substr */
1206 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1207 " Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1208 PL_colors[0], PL_colors[1],
1209 (long)(rx_origin - strpos),
1210 (long)(rx_origin - strpos + prog->anchored_offset)));
1211 goto do_other_substr;
1212 }
1213
1214 /* success: we don't contradict the found floating substring
1215 * (and there's no anchored substr). */
d0880ea7
DM		 1216 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1217 " Found /%s^%s/m at offset %ld...\n",
1218 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
e3c6feb0
DM		 1219 }
1220 else {
2e759faa 1221 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
fe4f3442 1222 " (multiline anchor test skipped)\n"));
e3c6feb0
DM		 1223 }
1224
ffad1e6a 1225 success_at_start:
e3c6feb0 1226
cad2e5aa 1227
dd170ff5
DM		 1228 /* if we have a starting character class, then test that extra constraint.
1229 * (trie stclasses are too expensive to use here, we are better off to
1230 * leave it to regmatch itself) */
1231
f8fc2ecf 1232 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
f8fc2ecf 1233 const U8* const str = (U8*)STRING(progi->regstclass);
0991020e 1234
2c75e362 1235 /* XXX this value could be pre-computed */
f8fc2ecf 1236 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
2c75e362
DM		 1237 ? (reginfo->is_utf8_pat
1238 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1239 : STR_LEN(progi->regstclass))
66e933ab 1240 : 1);
1de06328 1241 char * endpos;
fa3bb21d 1242 char *s;
000dfd2d
DM		 1243 /* latest pos that a matching float substr constrains rx start to */
1244 char *rx_max_float = NULL;
1245
c75a3985
DM		 1246 /* if the current rx_origin is anchored, either by satisfying an
1247 * anchored substring constraint, or a /^.../m constraint, then we
1248 * can reject the current origin if the start class isn't found
1249 * at the current position. If we have a float-only match, then
1250 * rx_origin is constrained to a range; so look for the start class
1251 * in that range. if neither, then look for the start class in the
1252 * whole rest of the string */
1253
dd170ff5
DM		 1254 /* XXX DAPM it's not clear what the minlen test is for, and why
1255 * it's not used in the floating case. Nothing in the test suite
1256 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1257 * Here are some old comments, which may or may not be correct:
1258 *
1259 * minlen == 0 is possible if regstclass is \b or \B,
1260 * and the fixed substr is ''$.
1261 * Since minlen is already taken into account, rx_origin+1 is
1262 * before strend; accidentally, minlen >= 1 guaranties no false
1263 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1264 * 0) below assumes that regstclass does not come from lookahead...
1265 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1266 * This leaves EXACTF-ish only, which are dealt with in
1267 * find_byclass().
1268 */
1269
7d2d37f5 1270 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
fa3bb21d 1271 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
000dfd2d
DM		 1272 else if (prog->float_substr || prog->float_utf8) {
1273 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1274 endpos= HOP3c(rx_max_float, cl_l, strend);
1275 }
1de06328
YO		 1276 else
1277 endpos= strend;
1278
1dc475d0
DM		 1279 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1280 " looking for class: start_shift: %"IVdf" check_at: %"IVdf
c43b5520 1281 " rx_origin: %"IVdf" endpos: %"IVdf"\n",
1dc475d0 1282 (IV)start_shift, (IV)(check_at - strbeg),
c43b5520 1283 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
d8080198 1284
c43b5520 1285 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
f9176b44 1286 reginfo);
be778b1a 1287 if (!s) {
6eb5f6b9 1288 if (endpos == strend) {
a3621e74 1289 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1dc475d0 1290 " Could not match STCLASS...\n") );
6eb5f6b9
JH		 1291 goto fail;
1292 }
a3621e74 1293 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1dc475d0 1294 " This position contradicts STCLASS...\n") );
e0eb31e7
DM		 1295 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1296 && !(prog->intflags & PREGf_IMPLICIT))
653099ff 1297 goto fail;
9fed8d02 1298
6eb5f6b9 1299 /* Contradict one of substrings */
97136c8a
DM		 1300 if (prog->anchored_substr || prog->anchored_utf8) {
1301 if (prog->substrs->check_ix == 1) { /* check is float */
1302 /* Have both, check_string is floating */
1303 assert(rx_origin + start_shift <= check_at);
1304 if (rx_origin + start_shift != check_at) {
1305 /* not at latest position float substr could match:
c75a3985
DM		 1306 * Recheck anchored substring, but not floating.
1307 * The condition above is in bytes rather than
1308 * chars for efficiency. It's conservative, in
1309 * that it errs on the side of doing 'goto
1310 * do_other_substr', where a more accurate
1311 * char-based calculation will be done */
97136c8a
DM		 1312 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1313 " Looking for anchored substr starting at offset %ld...\n",
1314 (long)(other_last - strpos)) );
1315 goto do_other_substr;
3369914b 1316 }
3369914b
DM		 1317 }
1318 }
97136c8a 1319 else {
9fed8d02
DM		 1320 /* float-only */
1321
7d2d37f5 1322 if (ml_anch) {
c75a3985
DM		 1323 /* In the presence of ml_anch, we might be able to
1324 * find another \n without breaking the current float
1325 * constraint. */
1326
1327 /* strictly speaking this should be HOP3c(..., 1, ...),
1328 * but since we goto a block of code that's going to
1329 * search for the next \n if any, its safe here */
9fed8d02 1330 rx_origin++;
9fed8d02
DM		 1331 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1332 " Looking for /%s^%s/m starting at offset %ld...\n",
1333 PL_colors[0], PL_colors[1],
1334 (long)(rx_origin - strpos)) );
9fed8d02 1335 goto postprocess_substr_matches;
ab60c45a 1336 }
c75a3985
DM		 1337
1338 /* strictly speaking this can never be true; but might
1339 * be if we ever allow intuit without substrings */
1340 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
9fed8d02 1341 goto fail;
c75a3985 1342
000dfd2d 1343 rx_origin = rx_max_float;
9fed8d02
DM		 1344 }
1345
c75a3985
DM		 1346 /* at this point, any matching substrings have been
1347 * contradicted. Start again... */
1348
9fed8d02 1349 rx_origin = HOP3c(rx_origin, 1, strend);
557f47af
DM		 1350
1351 /* uses bytes rather than char calculations for efficiency.
1352 * It's conservative: it errs on the side of doing 'goto restart',
1353 * where there is code that does a proper char-based test */
9fed8d02 1354 if (rx_origin + start_shift + end_shift > strend) {
40268e5b 1355 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
9fed8d02
DM		 1356 " Could not match STCLASS...\n") );
1357 goto fail;
1358 }
1359 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1360 " Looking for %s substr starting at offset %ld...\n",
1361 (prog->substrs->check_ix ? "floating" : "anchored"),
1362 (long)(rx_origin + start_shift - strpos)) );
1363 goto restart;
6eb5f6b9 1364 }
9fed8d02 1365
c75a3985
DM		 1366 /* Success !!! */
1367
5f9c6575 1368 if (rx_origin != s) {
a3621e74 1369 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 1370 " By STCLASS: moving %ld --> %ld\n",
5f9c6575 1371 (long)(rx_origin - strpos), (long)(s - strpos))
b7953727
JH		 1372 );
1373 }
1374 else {
a3621e74 1375 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 1376 " Does not contradict STCLASS...\n");
b7953727
JH		 1377 );
1378 }
6eb5f6b9 1379 }
ffad1e6a
DM		 1380
1381 /* Decide whether using the substrings helped */
1382
1383 if (rx_origin != strpos) {
1384 /* Fixed substring is found far enough so that the match
1385 cannot start at strpos. */
1386
1387 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n"));
1388 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1389 }
1390 else {
70563e16
DM		 1391 /* The found rx_origin position does not prohibit matching at
1392 * strpos, so calling intuit didn't gain us anything. Decrement
1393 * the BmUSEFUL() count on the check substring, and if we reach
1394 * zero, free it. */
1395 if (!(prog->intflags & PREGf_NAUGHTY)
ffad1e6a
DM		 1396 && (utf8_target ? (
1397 prog->check_utf8 /* Could be deleted already */
1398 && --BmUSEFUL(prog->check_utf8) < 0
1399 && (prog->check_utf8 == prog->float_utf8)
1400 ) : (
1401 prog->check_substr /* Could be deleted already */
1402 && --BmUSEFUL(prog->check_substr) < 0
1403 && (prog->check_substr == prog->float_substr)
1404 )))
1405 {
1406 /* If flags & SOMETHING - do not do it many times on the same match */
1407 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n"));
1408 /* XXX Does the destruction order has to change with utf8_target? */
1409 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1410 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1411 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1412 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1413 check = NULL; /* abort */
ffad1e6a
DM		 1414 /* XXXX This is a remnant of the old implementation. It
1415 looks wasteful, since now INTUIT can use many
1416 other heuristics. */
1417 prog->extflags &= ~RXf_USE_INTUIT;
ffad1e6a
DM		 1418 }
1419 }
1420
1421 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1422 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1423 PL_colors[4], PL_colors[5], (long)(rx_origin - strpos)) );
1424
c765d6e0 1425 return rx_origin;
2c2d71f5
JH		 1426
1427 fail_finish: /* Substring not found */
33b8afdf 1428 if (prog->check_substr || prog->check_utf8) /* could be removed already */
f2ed9b32 1429 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
cad2e5aa 1430 fail:
a3621e74 1431 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
e4584336 1432 PL_colors[4], PL_colors[5]));
bd61b366 1433 return NULL;
cad2e5aa 1434}
9661b544 1435
70563e16 1436
a0a388a1 1437#define DECL_TRIE_TYPE(scan) \
e7fd4aa1 1438 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
a4525e78
KW		 1439 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold, \
1440 trie_utf8l, trie_flu8 } \
e7fd4aa1
KW		 1441 trie_type = ((scan->flags == EXACT) \
1442 ? (utf8_target ? trie_utf8 : trie_plain) \
a4525e78
KW		 1443 : (scan->flags == EXACTL) \
1444 ? (utf8_target ? trie_utf8l : trie_plain) \
1445 : (scan->flags == EXACTFA) \
1446 ? (utf8_target \
1447 ? trie_utf8_exactfa_fold \
1448 : trie_latin_utf8_exactfa_fold) \
1449 : (scan->flags == EXACTFLU8 \
1450 ? trie_flu8 \
1451 : (utf8_target \
1452 ? trie_utf8_fold \
1453 : trie_latin_utf8_fold)))
fab2782b 1454
fd3249ee 1455#define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
baa60164 1456STMT_START { \
fab2782b 1457 STRLEN skiplen; \
baa60164 1458 U8 flags = FOLD_FLAGS_FULL; \
fab2782b 1459 switch (trie_type) { \
a4525e78 1460 case trie_flu8: \
780fcc9f 1461 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
613abc6d
KW		 1462 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1463 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \
1464 } \
a4525e78 1465 goto do_trie_utf8_fold; \
31f05a37 1466 case trie_utf8_exactfa_fold: \
baa60164 1467 flags |= FOLD_FLAGS_NOMIX_ASCII; \
8e57b935 1468 /* FALLTHROUGH */ \
fab2782b 1469 case trie_utf8_fold: \
a4525e78 1470 do_trie_utf8_fold: \
fab2782b 1471 if ( foldlen>0 ) { \
c80e42f3 1472 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO		 1473 foldlen -= len; \
1474 uscan += len; \
1475 len=0; \
1476 } else { \
445bf929 1477 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
fab2782b
YO		 1478 len = UTF8SKIP(uc); \
1479 skiplen = UNISKIP( uvc ); \
1480 foldlen -= skiplen; \
1481 uscan = foldbuf + skiplen; \
1482 } \
1483 break; \
baa60164
KW		 1484 case trie_latin_utf8_exactfa_fold: \
1485 flags |= FOLD_FLAGS_NOMIX_ASCII; \
8e57b935 1486 /* FALLTHROUGH */ \
fab2782b
YO		 1487 case trie_latin_utf8_fold: \
1488 if ( foldlen>0 ) { \
c80e42f3 1489 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO		 1490 foldlen -= len; \
1491 uscan += len; \
1492 len=0; \
1493 } else { \
1494 len = 1; \
31f05a37 1495 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
fab2782b
YO		 1496 skiplen = UNISKIP( uvc ); \
1497 foldlen -= skiplen; \
1498 uscan = foldbuf + skiplen; \
1499 } \
1500 break; \
a4525e78 1501 case trie_utf8l: \
780fcc9f 1502 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
613abc6d
KW		 1503 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1504 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \
1505 } \
780fcc9f 1506 /* FALLTHROUGH */ \
fab2782b 1507 case trie_utf8: \
c80e42f3 1508 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO		 1509 break; \
1510 case trie_plain: \
1511 uvc = (UV)*uc; \
1512 len = 1; \
1513 } \
1514 if (uvc < 256) { \
1515 charid = trie->charmap[ uvc ]; \
1516 } \
1517 else { \
1518 charid = 0; \
1519 if (widecharmap) { \
1520 SV** const svpp = hv_fetch(widecharmap, \
1521 (char*)&uvc, sizeof(UV), 0); \
1522 if (svpp) \
1523 charid = (U16)SvIV(*svpp); \
1524 } \
1525 } \
4cadc6a9
YO		 1526} STMT_END
1527
ae7c5b9b
KW		 1528#define DUMP_EXEC_POS(li,s,doutf8) \
1529 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1530 startpos, doutf8)
1531
c84a03c5 1532#define REXEC_FBC_EXACTISH_SCAN(COND) \
4cadc6a9
YO		 1533STMT_START { \
1534 while (s <= e) { \
c84a03c5 1535 if ( (COND) \
fac1af77 1536 && (ln == 1 || folder(s, pat_string, ln)) \
02d5137b 1537 && (reginfo->intuit || regtry(reginfo, &s)) )\
4cadc6a9
YO		 1538 goto got_it; \
1539 s++; \
1540 } \
1541} STMT_END
1542
c84a03c5 1543#define REXEC_FBC_UTF8_SCAN(CODE) \
4cadc6a9 1544STMT_START { \
9a902117 1545 while (s < strend) { \
c84a03c5 1546 CODE \
9a902117 1547 s += UTF8SKIP(s); \
4cadc6a9
YO		 1548 } \
1549} STMT_END
1550
c84a03c5 1551#define REXEC_FBC_SCAN(CODE) \
4cadc6a9
YO		 1552STMT_START { \
1553 while (s < strend) { \
c84a03c5 1554 CODE \
4cadc6a9
YO		 1555 s++; \
1556 } \
1557} STMT_END
1558
05bd126c
KW		 1559#define REXEC_FBC_UTF8_CLASS_SCAN(COND) \
1560REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \
1561 if (COND) { \
1562 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1563 goto got_it; \
1564 else \
1565 tmp = doevery; \
1566 } \
1567 else \
1568 tmp = 1; \
4cadc6a9
YO		 1569)
1570
05bd126c
KW		 1571#define REXEC_FBC_CLASS_SCAN(COND) \
1572REXEC_FBC_SCAN( /* Loops while (s < strend) */ \
1573 if (COND) { \
1574 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1575 goto got_it; \
1576 else \
1577 tmp = doevery; \
1578 } \
1579 else \
1580 tmp = 1; \
4cadc6a9
YO		 1581)
1582
c84a03c5 1583#define REXEC_FBC_CSCAN(CONDUTF8,COND) \
baa60164 1584 if (utf8_target) { \
c84a03c5 1585 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \
e1d1eefb
YO		 1586 } \
1587 else { \
c84a03c5 1588 REXEC_FBC_CLASS_SCAN(COND); \
d981ef24 1589 }
05bd126c 1590
05bd126c
KW		 1591/* The three macros below are slightly different versions of the same logic.
1592 *
1593 * The first is for /a and /aa when the target string is UTF-8. This can only
1594 * match ascii, but it must advance based on UTF-8. The other two handle the
1595 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1596 * for the boundary (or non-boundary) between a word and non-word character.
1597 * The utf8 and non-utf8 cases have the same logic, but the details must be
1598 * different. Find the "wordness" of the character just prior to this one, and
1599 * compare it with the wordness of this one. If they differ, we have a
1600 * boundary. At the beginning of the string, pretend that the previous
1601 * character was a new-line.
1602 *
1603 * All these macros uncleanly have side-effects with each other and outside
1604 * variables. So far it's been too much trouble to clean-up
1605 *
1606 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1607 * a word character or not.
1608 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1609 * word/non-word
1610 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1611 *
1612 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1613 * are looking for a boundary or for a non-boundary. If we are looking for a
1614 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1615 * see if this tentative match actually works, and if so, to quit the loop
1616 * here. And vice-versa if we are looking for a non-boundary.
1617 *
1618 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1619 * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of
1620 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1621 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1622 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1623 * complement. But in that branch we complement tmp, meaning that at the
1624 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1625 * which means at the top of the loop in the next iteration, it is
1626 * TEST_NON_UTF8(s-1) */
b2f4e957 1627#define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
05bd126c
KW		 1628 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1629 tmp = TEST_NON_UTF8(tmp); \
1630 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1631 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1632 tmp = !tmp; \
1633 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1634 } \
1635 else { \
1636 IF_FAIL; \
1637 } \
1638 ); \
1639
1640/* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1641 * TEST_UTF8 is a macro that for the same input code points returns identically
1642 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
236d82fd 1643#define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
05bd126c
KW		 1644 if (s == reginfo->strbeg) { \
1645 tmp = '\n'; \
1646 } \
1647 else { /* Back-up to the start of the previous character */ \
1648 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1649 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
3db24e1e 1650 0, UTF8_ALLOW_DEFAULT); \
05bd126c
KW		 1651 } \
1652 tmp = TEST_UV(tmp); \
1653 LOAD_UTF8_CHARCLASS_ALNUM(); \
1654 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1655 if (tmp == ! (TEST_UTF8((U8 *) s))) { \
1656 tmp = !tmp; \
1657 IF_SUCCESS; \
1658 } \
1659 else { \
1660 IF_FAIL; \
1661 } \
1662 );
cfaf538b 1663
05bd126c
KW		 1664/* Like the above two macros. UTF8_CODE is the complete code for handling
1665 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1666 * macros below */
baa60164 1667#define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
63ac0dad 1668 if (utf8_target) { \
05bd126c 1669 UTF8_CODE \
63ac0dad
KW		 1670 } \
1671 else { /* Not utf8 */ \
9d9163fb 1672 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
63ac0dad 1673 tmp = TEST_NON_UTF8(tmp); \
05bd126c 1674 REXEC_FBC_SCAN( /* advances s while s < strend */ \
63ac0dad 1675 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
63ac0dad 1676 IF_SUCCESS; \
760cfa8e 1677 tmp = !tmp; \
63ac0dad
KW		 1678 } \
1679 else { \
1680 IF_FAIL; \
1681 } \
1682 ); \
1683 } \
c8519dc7
KW		 1684 /* Here, things have been set up by the previous code so that tmp is the \
1685 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1686 * utf8ness of the target). We also have to check if this matches against \
1687 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1688 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1689 * string */ \
1690 if (tmp == ! TEST_NON_UTF8('\n')) { \
1691 IF_SUCCESS; \
1692 } \
1693 else { \
1694 IF_FAIL; \
1695 }
63ac0dad 1696
ae7c5b9b
KW		 1697/* This is the macro to use when we want to see if something that looks like it
1698 * could match, actually does, and if so exits the loop */
1699#define REXEC_FBC_TRYIT \
1700 if ((reginfo->intuit || regtry(reginfo, &s))) \
1701 goto got_it
1702
1703/* The only difference between the BOUND and NBOUND cases is that
1704 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1705 * NBOUND. This is accomplished by passing it as either the if or else clause,
1706 * with the other one being empty (PLACEHOLDER is defined as empty).
1707 *
1708 * The TEST_FOO parameters are for operating on different forms of input, but
1709 * all should be ones that return identically for the same underlying code
1710 * points */
1711#define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1712 FBC_BOUND_COMMON( \
1713 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1714 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1715
1716#define FBC_BOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1717 FBC_BOUND_COMMON( \
1718 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1719 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1720
1721#define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1722 FBC_BOUND_COMMON( \
1723 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1724 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1725
1726#define FBC_NBOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1727 FBC_BOUND_COMMON( \
1728 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1729 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1730
05bd126c 1731
786e8c11 1732/* We know what class REx starts with. Try to find this position... */
02d5137b 1733/* if reginfo->intuit, its a dryrun */
786e8c11
YO		 1734/* annoyingly all the vars in this routine have different names from their counterparts
1735 in regmatch. /grrr */
3c3eec57 1736STATIC char *
07be1b83 1737S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
f9176b44 1738 const char *strend, regmatch_info *reginfo)
a687059c 1739{
73104a1b
KW		 1740 dVAR;
1741 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1742 char *pat_string; /* The pattern's exactish string */
1743 char *pat_end; /* ptr to end char of pat_string */
1744 re_fold_t folder; /* Function for computing non-utf8 folds */
1745 const U8 *fold_array; /* array for folding ords < 256 */
1746 STRLEN ln;
1747 STRLEN lnc;
73104a1b
KW		 1748 U8 c1;
1749 U8 c2;
1750 char *e;
1751 I32 tmp = 1; /* Scratch variable? */
ba44c216 1752 const bool utf8_target = reginfo->is_utf8_target;
73104a1b 1753 UV utf8_fold_flags = 0;
f9176b44 1754 const bool is_utf8_pat = reginfo->is_utf8_pat;
3018b823
KW		 1755 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1756 with a result inverts that result, as 0^1 =
1757 1 and 1^1 = 0 */
1758 _char_class_number classnum;
1759
73104a1b 1760 RXi_GET_DECL(prog,progi);
2f7f8cb1 1761
73104a1b 1762 PERL_ARGS_ASSERT_FIND_BYCLASS;
2f7f8cb1 1763
73104a1b
KW		 1764 /* We know what class it must start with. */
1765 switch (OP(c)) {
a4525e78 1766 case ANYOFL:
780fcc9f
KW		 1767 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1768 /* FALLTHROUGH */
73104a1b
KW		 1769 case ANYOF:
1770 if (utf8_target) {
1771 REXEC_FBC_UTF8_CLASS_SCAN(
3db24e1e 1772 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
73104a1b
KW		 1773 }
1774 else {
1775 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1776 }
1777 break;
1778 case CANY:
1779 REXEC_FBC_SCAN(
02d5137b 1780 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
73104a1b
KW		 1781 goto got_it;
1782 else
1783 tmp = doevery;
1784 );
1785 break;
1786
098b07d5
KW		 1787 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1788 assert(! is_utf8_pat);
924ba076 1789 /* FALLTHROUGH */
73104a1b 1790 case EXACTFA:
984e6dd1 1791 if (is_utf8_pat || utf8_target) {
73104a1b
KW		 1792 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1793 goto do_exactf_utf8;
1794 }
1795 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1796 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1797 goto do_exactf_non_utf8; /* isn't dealt with by these */
77a6d856 1798
2fdb7295
KW		 1799 case EXACTF: /* This node only generated for non-utf8 patterns */
1800 assert(! is_utf8_pat);
73104a1b 1801 if (utf8_target) {
73104a1b
KW		 1802 utf8_fold_flags = 0;
1803 goto do_exactf_utf8;
1804 }
1805 fold_array = PL_fold;
1806 folder = foldEQ;
1807 goto do_exactf_non_utf8;
1808
1809 case EXACTFL:
780fcc9f 1810 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
31f05a37 1811 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
cea315b6 1812 utf8_fold_flags = FOLDEQ_LOCALE;
73104a1b
KW		 1813 goto do_exactf_utf8;
1814 }
1815 fold_array = PL_fold_locale;
1816 folder = foldEQ_locale;
1817 goto do_exactf_non_utf8;
3c760661 1818
73104a1b 1819 case EXACTFU_SS:
984e6dd1 1820 if (is_utf8_pat) {
73104a1b
KW		 1821 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1822 }
1823 goto do_exactf_utf8;
16d951b7 1824
a4525e78
KW		 1825 case EXACTFLU8:
1826 if (! utf8_target) { /* All code points in this node require
1827 UTF-8 to express. */
1828 break;
1829 }
613abc6d
KW		 1830 utf8_fold_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
1831 | FOLDEQ_S2_FOLDS_SANE;
a4525e78
KW		 1832 goto do_exactf_utf8;
1833
73104a1b 1834 case EXACTFU:
984e6dd1
DM		 1835 if (is_utf8_pat || utf8_target) {
1836 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
73104a1b
KW		 1837 goto do_exactf_utf8;
1838 }
fac1af77 1839
73104a1b
KW		 1840 /* Any 'ss' in the pattern should have been replaced by regcomp,
1841 * so we don't have to worry here about this single special case
1842 * in the Latin1 range */
1843 fold_array = PL_fold_latin1;
1844 folder = foldEQ_latin1;
1845
924ba076 1846 /* FALLTHROUGH */
73104a1b 1847
c52b8b12 1848 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
73104a1b
KW		 1849 are no glitches with fold-length differences
1850 between the target string and pattern */
1851
1852 /* The idea in the non-utf8 EXACTF* cases is to first find the
1853 * first character of the EXACTF* node and then, if necessary,
1854 * case-insensitively compare the full text of the node. c1 is the
1855 * first character. c2 is its fold. This logic will not work for
1856 * Unicode semantics and the german sharp ss, which hence should
1857 * not be compiled into a node that gets here. */
1858 pat_string = STRING(c);
1859 ln = STR_LEN(c); /* length to match in octets/bytes */
1860
1861 /* We know that we have to match at least 'ln' bytes (which is the
1862 * same as characters, since not utf8). If we have to match 3
1863 * characters, and there are only 2 availabe, we know without
1864 * trying that it will fail; so don't start a match past the
1865 * required minimum number from the far end */
ea3daa5d 1866 e = HOP3c(strend, -((SSize_t)ln), s);
73104a1b 1867
02d5137b 1868 if (reginfo->intuit && e < s) {
73104a1b
KW		 1869 e = s; /* Due to minlen logic of intuit() */
1870 }
fac1af77 1871
73104a1b
KW		 1872 c1 = *pat_string;
1873 c2 = fold_array[c1];
1874 if (c1 == c2) { /* If char and fold are the same */
1875 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1876 }
1877 else {
1878 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1879 }
1880 break;
fac1af77 1881
c52b8b12
KW		 1882 do_exactf_utf8:
1883 {
73104a1b
KW		 1884 unsigned expansion;
1885
1886 /* If one of the operands is in utf8, we can't use the simpler folding
1887 * above, due to the fact that many different characters can have the
1888 * same fold, or portion of a fold, or different- length fold */
1889 pat_string = STRING(c);
1890 ln = STR_LEN(c); /* length to match in octets/bytes */
1891 pat_end = pat_string + ln;
984e6dd1 1892 lnc = is_utf8_pat /* length to match in characters */
73104a1b
KW		 1893 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1894 : ln;
1895
1896 /* We have 'lnc' characters to match in the pattern, but because of
1897 * multi-character folding, each character in the target can match
1898 * up to 3 characters (Unicode guarantees it will never exceed
1899 * this) if it is utf8-encoded; and up to 2 if not (based on the
1900 * fact that the Latin 1 folds are already determined, and the
1901 * only multi-char fold in that range is the sharp-s folding to
1902 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1903 * string character. Adjust lnc accordingly, rounding up, so that
1904 * if we need to match at least 4+1/3 chars, that really is 5. */
1905 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1906 lnc = (lnc + expansion - 1) / expansion;
1907
1908 /* As in the non-UTF8 case, if we have to match 3 characters, and
1909 * only 2 are left, it's guaranteed to fail, so don't start a
1910 * match that would require us to go beyond the end of the string
1911 */
ea3daa5d 1912 e = HOP3c(strend, -((SSize_t)lnc), s);
73104a1b 1913
02d5137b 1914 if (reginfo->intuit && e < s) {
73104a1b
KW		 1915 e = s; /* Due to minlen logic of intuit() */
1916 }
0658cdde 1917
73104a1b
KW		 1918 /* XXX Note that we could recalculate e to stop the loop earlier,
1919 * as the worst case expansion above will rarely be met, and as we
1920 * go along we would usually find that e moves further to the left.
1921 * This would happen only after we reached the point in the loop
1922 * where if there were no expansion we should fail. Unclear if
1923 * worth the expense */
1924
1925 while (s <= e) {
1926 char *my_strend= (char *)strend;
1927 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
984e6dd1 1928 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
02d5137b 1929 && (reginfo->intuit || regtry(reginfo, &s)) )
73104a1b
KW		 1930 {
1931 goto got_it;
1932 }
1933 s += (utf8_target) ? UTF8SKIP(s) : 1;
1934 }
1935 break;
1936 }
236d82fd 1937
73104a1b 1938 case BOUNDL:
780fcc9f 1939 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
236d82fd 1940 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
73104a1b
KW		 1941 break;
1942 case NBOUNDL:
780fcc9f 1943 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
236d82fd 1944 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
73104a1b
KW		 1945 break;
1946 case BOUND:
236d82fd 1947 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
73104a1b
KW		 1948 break;
1949 case BOUNDA:
236d82fd 1950 FBC_BOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
73104a1b
KW		 1951 break;
1952 case NBOUND:
236d82fd 1953 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
73104a1b
KW		 1954 break;
1955 case NBOUNDA:
236d82fd 1956 FBC_NBOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
73104a1b
KW		 1957 break;
1958 case BOUNDU:
236d82fd 1959 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
73104a1b
KW		 1960 break;
1961 case NBOUNDU:
236d82fd 1962 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
73104a1b 1963 break;
73104a1b
KW		 1964 case LNBREAK:
1965 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1966 is_LNBREAK_latin1_safe(s, strend)
1967 );
1968 break;
3018b823
KW		 1969
1970 /* The argument to all the POSIX node types is the class number to pass to
1971 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1972
1973 case NPOSIXL:
1974 to_complement = 1;
1975 /* FALLTHROUGH */
1976
1977 case POSIXL:
780fcc9f 1978 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3018b823
KW		 1979 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1980 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
73104a1b 1981 break;
3018b823
KW		 1982
1983 case NPOSIXD:
1984 to_complement = 1;
1985 /* FALLTHROUGH */
1986
1987 case POSIXD:
1988 if (utf8_target) {
1989 goto posix_utf8;
1990 }
1991 goto posixa;
1992
1993 case NPOSIXA:
1994 if (utf8_target) {
1995 /* The complement of something that matches only ASCII matches all
837226c8
KW		 1996 * non-ASCII, plus everything in ASCII that isn't in the class. */
1997 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
3018b823
KW		 1998 || ! _generic_isCC_A(*s, FLAGS(c)));
1999 break;
2000 }
2001
2002 to_complement = 1;
2003 /* FALLTHROUGH */
2004
73104a1b 2005 case POSIXA:
3018b823 2006 posixa:
73104a1b 2007 /* Don't need to worry about utf8, as it can match only a single
3018b823
KW		 2008 * byte invariant character. */
2009 REXEC_FBC_CLASS_SCAN(
2010 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
73104a1b 2011 break;
3018b823
KW		 2012
2013 case NPOSIXU:
2014 to_complement = 1;
2015 /* FALLTHROUGH */
2016
2017 case POSIXU:
2018 if (! utf8_target) {
2019 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
2020 FLAGS(c))));
2021 }
2022 else {
2023
c52b8b12 2024 posix_utf8:
3018b823
KW		 2025 classnum = (_char_class_number) FLAGS(c);
2026 if (classnum < _FIRST_NON_SWASH_CC) {
2027 while (s < strend) {
2028
2029 /* We avoid loading in the swash as long as possible, but
2030 * should we have to, we jump to a separate loop. This
2031 * extra 'if' statement is what keeps this code from being
2032 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
2033 if (UTF8_IS_ABOVE_LATIN1(*s)) {
2034 goto found_above_latin1;
2035 }
2036 if ((UTF8_IS_INVARIANT(*s)
2037 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
2038 classnum)))
2039 || (UTF8_IS_DOWNGRADEABLE_START(*s)
2040 && to_complement ^ cBOOL(
94bb8c36
KW		 2041 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
2042 *(s + 1)),
3018b823
KW		 2043 classnum))))
2044 {
02d5137b 2045 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
3018b823
KW		 2046 goto got_it;
2047 else {
2048 tmp = doevery;
2049 }
2050 }
2051 else {
2052 tmp = 1;
2053 }
2054 s += UTF8SKIP(s);
2055 }
2056 }
2057 else switch (classnum) { /* These classes are implemented as
2058 macros */
2059 case _CC_ENUM_SPACE: /* XXX would require separate code if we
2060 revert the change of \v matching this */
924ba076 2061 /* FALLTHROUGH */
3018b823
KW		 2062
2063 case _CC_ENUM_PSXSPC:
2064 REXEC_FBC_UTF8_CLASS_SCAN(
2065 to_complement ^ cBOOL(isSPACE_utf8(s)));
2066 break;
2067
2068 case _CC_ENUM_BLANK:
2069 REXEC_FBC_UTF8_CLASS_SCAN(
2070 to_complement ^ cBOOL(isBLANK_utf8(s)));
2071 break;
2072
2073 case _CC_ENUM_XDIGIT:
2074 REXEC_FBC_UTF8_CLASS_SCAN(
2075 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
2076 break;
2077
2078 case _CC_ENUM_VERTSPACE:
2079 REXEC_FBC_UTF8_CLASS_SCAN(
2080 to_complement ^ cBOOL(isVERTWS_utf8(s)));
2081 break;
2082
2083 case _CC_ENUM_CNTRL:
2084 REXEC_FBC_UTF8_CLASS_SCAN(
2085 to_complement ^ cBOOL(isCNTRL_utf8(s)));
2086 break;
2087
2088 default:
2089 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
e5964223 2090 NOT_REACHED; /* NOTREACHED */
3018b823
KW		 2091 }
2092 }
2093 break;
2094
2095 found_above_latin1: /* Here we have to load a swash to get the result
2096 for the current code point */
2097 if (! PL_utf8_swash_ptrs[classnum]) {
2098 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2099 PL_utf8_swash_ptrs[classnum] =
2a16ac92
KW		 2100 _core_swash_init("utf8",
2101 "",
2102 &PL_sv_undef, 1, 0,
2103 PL_XPosix_ptrs[classnum], &flags);
3018b823
KW		 2104 }
2105
2106 /* This is a copy of the loop above for swash classes, though using the
2107 * FBC macro instead of being expanded out. Since we've loaded the
2108 * swash, we don't have to check for that each time through the loop */
2109 REXEC_FBC_UTF8_CLASS_SCAN(
2110 to_complement ^ cBOOL(_generic_utf8(
2111 classnum,
2112 s,
2113 swash_fetch(PL_utf8_swash_ptrs[classnum],
2114 (U8 *) s, TRUE))));
73104a1b
KW		 2115 break;
2116
2117 case AHOCORASICKC:
2118 case AHOCORASICK:
2119 {
2120 DECL_TRIE_TYPE(c);
2121 /* what trie are we using right now */
2122 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2123 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2124 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2125
2126 const char *last_start = strend - trie->minlen;
6148ee25 2127#ifdef DEBUGGING
73104a1b 2128 const char *real_start = s;
6148ee25 2129#endif
73104a1b
KW		 2130 STRLEN maxlen = trie->maxlen;
2131 SV *sv_points;
2132 U8 **points; /* map of where we were in the input string
2133 when reading a given char. For ASCII this
2134 is unnecessary overhead as the relationship
2135 is always 1:1, but for Unicode, especially
2136 case folded Unicode this is not true. */
2137 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2138 U8 *bitmap=NULL;
2139
2140
2141 GET_RE_DEBUG_FLAGS_DECL;
2142
2143 /* We can't just allocate points here. We need to wrap it in
2144 * an SV so it gets freed properly if there is a croak while
2145 * running the match */
2146 ENTER;
2147 SAVETMPS;
2148 sv_points=newSV(maxlen * sizeof(U8 *));
2149 SvCUR_set(sv_points,
2150 maxlen * sizeof(U8 *));
2151 SvPOK_on(sv_points);
2152 sv_2mortal(sv_points);
2153 points=(U8**)SvPV_nolen(sv_points );
2154 if ( trie_type != trie_utf8_fold
2155 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2156 {
2157 if (trie->bitmap)
2158 bitmap=(U8*)trie->bitmap;
2159 else
2160 bitmap=(U8*)ANYOF_BITMAP(c);
2161 }
2162 /* this is the Aho-Corasick algorithm modified a touch
2163 to include special handling for long "unknown char" sequences.
2164 The basic idea being that we use AC as long as we are dealing
2165 with a possible matching char, when we encounter an unknown char
2166 (and we have not encountered an accepting state) we scan forward
2167 until we find a legal starting char.
2168 AC matching is basically that of trie matching, except that when
2169 we encounter a failing transition, we fall back to the current
2170 states "fail state", and try the current char again, a process
2171 we repeat until we reach the root state, state 1, or a legal
2172 transition. If we fail on the root state then we can either
2173 terminate if we have reached an accepting state previously, or
2174 restart the entire process from the beginning if we have not.
2175
2176 */
2177 while (s <= last_start) {
2178 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2179 U8 *uc = (U8*)s;
2180 U16 charid = 0;
2181 U32 base = 1;
2182 U32 state = 1;
2183 UV uvc = 0;
2184 STRLEN len = 0;
2185 STRLEN foldlen = 0;
2186 U8 *uscan = (U8*)NULL;
2187 U8 *leftmost = NULL;
2188#ifdef DEBUGGING
2189 U32 accepted_word= 0;
786e8c11 2190#endif
73104a1b
KW		 2191 U32 pointpos = 0;
2192
2193 while ( state && uc <= (U8*)strend ) {
2194 int failed=0;
2195 U32 word = aho->states[ state ].wordnum;
2196
2197 if( state==1 ) {
2198 if ( bitmap ) {
2199 DEBUG_TRIE_EXECUTE_r(
2200 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2201 dump_exec_pos( (char *)uc, c, strend, real_start,
2202 (char *)uc, utf8_target );
2203 PerlIO_printf( Perl_debug_log,
2204 " Scanning for legal start char...\n");
2205 }
2206 );
2207 if (utf8_target) {
2208 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2209 uc += UTF8SKIP(uc);
2210 }
2211 } else {
2212 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2213 uc++;
2214 }
786e8c11 2215 }
73104a1b 2216 s= (char *)uc;
07be1b83 2217 }
73104a1b
KW		 2218 if (uc >(U8*)last_start) break;
2219 }
2220
2221 if ( word ) {
2222 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2223 if (!leftmost || lpos < leftmost) {
2224 DEBUG_r(accepted_word=word);
2225 leftmost= lpos;
7016d6eb 2226 }
73104a1b 2227 if (base==0) break;
7016d6eb 2228
73104a1b
KW		 2229 }
2230 points[pointpos++ % maxlen]= uc;
2231 if (foldlen || uc < (U8*)strend) {
2232 REXEC_TRIE_READ_CHAR(trie_type, trie,
2233 widecharmap, uc,
2234 uscan, len, uvc, charid, foldlen,
2235 foldbuf, uniflags);
2236 DEBUG_TRIE_EXECUTE_r({
2237 dump_exec_pos( (char *)uc, c, strend,
2238 real_start, s, utf8_target);
2239 PerlIO_printf(Perl_debug_log,
2240 " Charid:%3u CP:%4"UVxf" ",
2241 charid, uvc);
2242 });
2243 }
2244 else {
2245 len = 0;
2246 charid = 0;
2247 }
07be1b83 2248
73104a1b
KW		 2249
2250 do {
6148ee25 2251#ifdef DEBUGGING
73104a1b 2252 word = aho->states[ state ].wordnum;
6148ee25 2253#endif
73104a1b
KW		 2254 base = aho->states[ state ].trans.base;
2255
2256 DEBUG_TRIE_EXECUTE_r({
2257 if (failed)
2258 dump_exec_pos( (char *)uc, c, strend, real_start,
2259 s, utf8_target );
2260 PerlIO_printf( Perl_debug_log,
2261 "%sState: %4"UVxf", word=%"UVxf,
2262 failed ? " Fail transition to " : "",
2263 (UV)state, (UV)word);
2264 });
2265 if ( base ) {
2266 U32 tmp;
2267 I32 offset;
2268 if (charid &&
2269 ( ((offset = base + charid
2270 - 1 - trie->uniquecharcount)) >= 0)
2271 && ((U32)offset < trie->lasttrans)
2272 && trie->trans[offset].check == state
2273 && (tmp=trie->trans[offset].next))
2274 {
2275 DEBUG_TRIE_EXECUTE_r(
2276 PerlIO_printf( Perl_debug_log," - legal\n"));
2277 state = tmp;
2278 break;
07be1b83
YO		 2279 }
2280 else {
786e8c11 2281 DEBUG_TRIE_EXECUTE_r(
73104a1b 2282 PerlIO_printf( Perl_debug_log," - fail\n"));
786e8c11 2283 failed = 1;
73104a1b 2284 state = aho->fail[state];
07be1b83 2285 }
07be1b83 2286 }
73104a1b
KW		 2287 else {
2288 /* we must be accepting here */
2289 DEBUG_TRIE_EXECUTE_r(
2290 PerlIO_printf( Perl_debug_log," - accepting\n"));
2291 failed = 1;
2292 break;
786e8c11 2293 }
73104a1b
KW		 2294 } while(state);
2295 uc += len;
2296 if (failed) {
2297 if (leftmost)
2298 break;
2299 if (!state) state = 1;
07be1b83 2300 }
73104a1b
KW		 2301 }
2302 if ( aho->states[ state ].wordnum ) {
2303 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2304 if (!leftmost || lpos < leftmost) {
2305 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2306 leftmost = lpos;
07be1b83
YO		 2307 }
2308 }
73104a1b
KW		 2309 if (leftmost) {
2310 s = (char*)leftmost;
2311 DEBUG_TRIE_EXECUTE_r({
2312 PerlIO_printf(
2313 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2314 (UV)accepted_word, (IV)(s - real_start)
2315 );
2316 });
02d5137b 2317 if (reginfo->intuit || regtry(reginfo, &s)) {
73104a1b
KW		 2318 FREETMPS;
2319 LEAVE;
2320 goto got_it;
2321 }
2322 s = HOPc(s,1);
2323 DEBUG_TRIE_EXECUTE_r({
2324 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2325 });
2326 } else {
2327 DEBUG_TRIE_EXECUTE_r(
2328 PerlIO_printf( Perl_debug_log,"No match.\n"));
2329 break;
2330 }
2331 }
2332 FREETMPS;
2333 LEAVE;
2334 }
2335 break;
2336 default:
2337 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
73104a1b
KW		 2338 }
2339 return 0;
2340 got_it:
2341 return s;
6eb5f6b9
JH		 2342}
2343
60165aa4
DM		 2344/* set RX_SAVED_COPY, RX_SUBBEG etc.
2345 * flags have same meanings as with regexec_flags() */
2346
749f4950
DM		 2347static void
2348S_reg_set_capture_string(pTHX_ REGEXP * const rx,
60165aa4
DM		 2349 char *strbeg,
2350 char *strend,
2351 SV *sv,
2352 U32 flags,
2353 bool utf8_target)
2354{
2355 struct regexp *const prog = ReANY(rx);
2356
60165aa4
DM		 2357 if (flags & REXEC_COPY_STR) {
2358#ifdef PERL_ANY_COW
2359 if (SvCANCOW(sv)) {
2360 if (DEBUG_C_TEST) {
2361 PerlIO_printf(Perl_debug_log,
2362 "Copy on write: regexp capture, type %d\n",
2363 (int) SvTYPE(sv));
2364 }
5411a0e5
DM		 2365 /* Create a new COW SV to share the match string and store
2366 * in saved_copy, unless the current COW SV in saved_copy
2367 * is valid and suitable for our purpose */
2368 if (( prog->saved_copy
2369 && SvIsCOW(prog->saved_copy)
2370 && SvPOKp(prog->saved_copy)
2371 && SvIsCOW(sv)
2372 && SvPOKp(sv)
2373 && SvPVX(sv) == SvPVX(prog->saved_copy)))
a76b0e90 2374 {
5411a0e5
DM		 2375 /* just reuse saved_copy SV */
2376 if (RXp_MATCH_COPIED(prog)) {
2377 Safefree(prog->subbeg);
2378 RXp_MATCH_COPIED_off(prog);
2379 }
2380 }
2381 else {
2382 /* create new COW SV to share string */
a76b0e90
DM		 2383 RX_MATCH_COPY_FREE(rx);
2384 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
a76b0e90 2385 }
5411a0e5
DM		 2386 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2387 assert (SvPOKp(prog->saved_copy));
60165aa4
DM		 2388 prog->sublen = strend - strbeg;
2389 prog->suboffset = 0;
2390 prog->subcoffset = 0;
2391 } else
2392#endif
2393 {
99a90e59
FC		 2394 SSize_t min = 0;
2395 SSize_t max = strend - strbeg;
ea3daa5d 2396 SSize_t sublen;
60165aa4
DM		 2397
2398 if ( (flags & REXEC_COPY_SKIP_POST)
e322109a 2399 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM		 2400 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2401 ) { /* don't copy $' part of string */
2402 U32 n = 0;
2403 max = -1;
2404 /* calculate the right-most part of the string covered
2405 * by a capture. Due to look-ahead, this may be to
2406 * the right of $&, so we have to scan all captures */
2407 while (n <= prog->lastparen) {
2408 if (prog->offs[n].end > max)
2409 max = prog->offs[n].end;
2410 n++;
2411 }
2412 if (max == -1)
2413 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2414 ? prog->offs[0].start
2415 : 0;
2416 assert(max >= 0 && max <= strend - strbeg);
2417 }
2418
2419 if ( (flags & REXEC_COPY_SKIP_PRE)
e322109a 2420 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM		 2421 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2422 ) { /* don't copy $` part of string */
2423 U32 n = 0;
2424 min = max;
2425 /* calculate the left-most part of the string covered
2426 * by a capture. Due to look-behind, this may be to
2427 * the left of $&, so we have to scan all captures */
2428 while (min && n <= prog->lastparen) {
2429 if ( prog->offs[n].start != -1
2430 && prog->offs[n].start < min)
2431 {
2432 min = prog->offs[n].start;
2433 }
2434 n++;
2435 }
2436 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2437 && min > prog->offs[0].end
2438 )
2439 min = prog->offs[0].end;
2440
2441 }
2442
2443 assert(min >= 0 && min <= max && min <= strend - strbeg);
2444 sublen = max - min;
2445
2446 if (RX_MATCH_COPIED(rx)) {
2447 if (sublen > prog->sublen)
2448 prog->subbeg =
2449 (char*)saferealloc(prog->subbeg, sublen+1);
2450 }
2451 else
2452 prog->subbeg = (char*)safemalloc(sublen+1);
2453 Copy(strbeg + min, prog->subbeg, sublen, char);
2454 prog->subbeg[sublen] = '\0';
2455 prog->suboffset = min;
2456 prog->sublen = sublen;
2457 RX_MATCH_COPIED_on(rx);
2458 }
2459 prog->subcoffset = prog->suboffset;
2460 if (prog->suboffset && utf8_target) {
2461 /* Convert byte offset to chars.
2462 * XXX ideally should only compute this if @-/@+
2463 * has been seen, a la PL_sawampersand ??? */
2464
2465 /* If there's a direct correspondence between the
2466 * string which we're matching and the original SV,
2467 * then we can use the utf8 len cache associated with
2468 * the SV. In particular, it means that under //g,
2469 * sv_pos_b2u() will use the previously cached
2470 * position to speed up working out the new length of
2471 * subcoffset, rather than counting from the start of
2472 * the string each time. This stops
2473 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2474 * from going quadratic */
2475 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
ea3daa5d
FC		 2476 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2477 SV_GMAGIC|SV_CONST_RETURN);
60165aa4
DM		 2478 else
2479 prog->subcoffset = utf8_length((U8*)strbeg,
2480 (U8*)(strbeg+prog->suboffset));
2481 }
2482 }
2483 else {
2484 RX_MATCH_COPY_FREE(rx);
2485 prog->subbeg = strbeg;
2486 prog->suboffset = 0;
2487 prog->subcoffset = 0;
2488 prog->sublen = strend - strbeg;
2489 }
2490}
2491
2492
2493
fae667d5 2494
6eb5f6b9
JH		 2495/*
2496 - regexec_flags - match a regexp against a string
2497 */
2498I32
5aaab254 2499Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
ea3daa5d 2500 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
8fd1a950
DM		 2501/* stringarg: the point in the string at which to begin matching */
2502/* strend: pointer to null at end of string */
2503/* strbeg: real beginning of string */
2504/* minend: end of match must be >= minend bytes after stringarg. */
2505/* sv: SV being matched: only used for utf8 flag, pos() etc; string
2506 * itself is accessed via the pointers above */
2507/* data: May be used for some additional optimizations.
d058ec57 2508 Currently unused. */
a340edde 2509/* flags: For optimizations. See REXEC_* in regexp.h */
8fd1a950 2510
6eb5f6b9 2511{
8d919b0a 2512 struct regexp *const prog = ReANY(rx);
5aaab254 2513 char *s;
eb578fdb 2514 regnode *c;
03c83e26 2515 char *startpos;
ea3daa5d
FC		 2516 SSize_t minlen; /* must match at least this many chars */
2517 SSize_t dontbother = 0; /* how many characters not to try at end */
f2ed9b32 2518 const bool utf8_target = cBOOL(DO_UTF8(sv));
2757e526 2519 I32 multiline;
f8fc2ecf 2520 RXi_GET_DECL(prog,progi);
02d5137b
DM		 2521 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2522 regmatch_info *const reginfo = &reginfo_buf;
e9105d30 2523 regexp_paren_pair *swap = NULL;
006f26b2 2524 I32 oldsave;
a3621e74
YO		 2525 GET_RE_DEBUG_FLAGS_DECL;
2526
7918f24d 2527 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
9d4ba2ae 2528 PERL_UNUSED_ARG(data);
6eb5f6b9
JH		 2529
2530 /* Be paranoid... */
03c83e26 2531 if (prog == NULL || stringarg == NULL) {
6eb5f6b9 2532 Perl_croak(aTHX_ "NULL regexp parameter");
6eb5f6b9
JH		 2533 }
2534
6c3fea77 2535 DEBUG_EXECUTE_r(
03c83e26 2536 debug_start_match(rx, utf8_target, stringarg, strend,
6c3fea77
DM		 2537 "Matching");
2538 );
8adc0f72 2539
b342a604
DM		 2540 startpos = stringarg;
2541
58430ea8 2542 if (prog->intflags & PREGf_GPOS_SEEN) {
d307c076
DM		 2543 MAGIC *mg;
2544
fef7148b
DM		 2545 /* set reginfo->ganch, the position where \G can match */
2546
2547 reginfo->ganch =
2548 (flags & REXEC_IGNOREPOS)
2549 ? stringarg /* use start pos rather than pos() */
2550 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
25fdce4a
FC		 2551 /* Defined pos(): */
2552 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
fef7148b
DM		 2553 : strbeg; /* pos() not defined; use start of string */
2554
2555 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
7b0eb0b8 2556 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg)));
fef7148b 2557
03c83e26
DM		 2558 /* in the presence of \G, we may need to start looking earlier in
2559 * the string than the suggested start point of stringarg:
0b2c2a84 2560 * if prog->gofs is set, then that's a known, fixed minimum
03c83e26
DM		 2561 * offset, such as
2562 * /..\G/: gofs = 2
2563 * /ab|c\G/: gofs = 1
2564 * or if the minimum offset isn't known, then we have to go back
2565 * to the start of the string, e.g. /w+\G/
2566 */
2bfbe302 2567
8e1490ee 2568 if (prog->intflags & PREGf_ANCH_GPOS) {
2bfbe302
DM		 2569 startpos = reginfo->ganch - prog->gofs;
2570 if (startpos <
2571 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2572 {
2573 DEBUG_r(PerlIO_printf(Perl_debug_log,
2574 "fail: ganch-gofs before earliest possible start\n"));
2575 return 0;
2576 }
2577 }
2578 else if (prog->gofs) {
b342a604
DM		 2579 if (startpos - prog->gofs < strbeg)
2580 startpos = strbeg;
2581 else
2582 startpos -= prog->gofs;
03c83e26 2583 }
58430ea8 2584 else if (prog->intflags & PREGf_GPOS_FLOAT)
b342a604 2585 startpos = strbeg;
03c83e26
DM		 2586 }
2587
2588 minlen = prog->minlen;
b342a604 2589 if ((startpos + minlen) > strend || startpos < strbeg) {
03c83e26
DM		 2590 DEBUG_r(PerlIO_printf(Perl_debug_log,
2591 "Regex match can't succeed, so not even tried\n"));
2592 return 0;
2593 }
2594
63a3746a
DM		 2595 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2596 * which will call destuctors to reset PL_regmatch_state, free higher
2597 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2598 * regmatch_info_aux_eval */
2599
2600 oldsave = PL_savestack_ix;
2601
dfa77d06
DM		 2602 s = startpos;
2603
e322109a 2604 if ((prog->extflags & RXf_USE_INTUIT)
7fadf4a7
DM		 2605 && !(flags & REXEC_CHECKED))
2606 {
dfa77d06 2607 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
7fadf4a7 2608 flags, NULL);
dfa77d06 2609 if (!s)
7fadf4a7
DM		 2610 return 0;
2611
e322109a 2612 if (prog->extflags & RXf_CHECK_ALL) {
7fadf4a7
DM		 2613 /* we can match based purely on the result of INTUIT.
2614 * Set up captures etc just for $& and $-[0]
2615 * (an intuit-only match wont have $1,$2,..) */
2616 assert(!prog->nparens);
d5e7783a
DM		 2617
2618 /* s/// doesn't like it if $& is earlier than where we asked it to
2619 * start searching (which can happen on something like /.\G/) */
2620 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2621 && (s < stringarg))
2622 {
2623 /* this should only be possible under \G */
58430ea8 2624 assert(prog->intflags & PREGf_GPOS_SEEN);
d5e7783a
DM		 2625 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2626 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2627 goto phooey;
2628 }
2629
7fadf4a7
DM		 2630 /* match via INTUIT shouldn't have any captures.
2631 * Let @-, @+, $^N know */
2632 prog->lastparen = prog->lastcloseparen = 0;
2633 RX_MATCH_UTF8_set(rx, utf8_target);
3ff69bd6
DM		 2634 prog->offs[0].start = s - strbeg;
2635 prog->offs[0].end = utf8_target
2636 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2637 : s - strbeg + prog->minlenret;
7fadf4a7 2638 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 2639 S_reg_set_capture_string(aTHX_ rx,
7fadf4a7
DM		 2640 strbeg, strend,
2641 sv, flags, utf8_target);
2642
7fadf4a7
DM		 2643 return 1;
2644 }
2645 }
2646
6c3fea77 2647 multiline = prog->extflags & RXf_PMf_MULTILINE;
1de06328 2648
dfa77d06 2649 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
a3621e74 2650 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
a72c7584
JH		 2651 "String too short [regexec_flags]...\n"));
2652 goto phooey;
1aa99e6b 2653 }
1de06328 2654
6eb5f6b9 2655 /* Check validity of program. */
f8fc2ecf 2656 if (UCHARAT(progi->program) != REG_MAGIC) {
6eb5f6b9
JH		 2657 Perl_croak(aTHX_ "corrupted regexp program");
2658 }
2659
1738e041 2660 RX_MATCH_TAINTED_off(rx);
ab4e48c1 2661 RX_MATCH_UTF8_set(rx, utf8_target);
1738e041 2662
6c3fea77
DM		 2663 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2664 reginfo->intuit = 0;
2665 reginfo->is_utf8_target = cBOOL(utf8_target);
02d5137b
DM		 2666 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2667 reginfo->warned = FALSE;
9d9163fb 2668 reginfo->strbeg = strbeg;
02d5137b 2669 reginfo->sv = sv;
1cb48e53 2670 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
220db18a 2671 reginfo->strend = strend;
6eb5f6b9 2672 /* see how far we have to get to not match where we matched before */
fe3974be 2673 reginfo->till = stringarg + minend;
6eb5f6b9 2674
60779a30 2675 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
82c23608
FC		 2676 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2677 S_cleanup_regmatch_info_aux has executed (registered by
2678 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2679 magic belonging to this SV.
2680 Not newSVsv, either, as it does not COW.
2681 */
2682 reginfo->sv = newSV(0);
4cba5ac0 2683 SvSetSV_nosteal(reginfo->sv, sv);
82c23608
FC		 2684 SAVEFREESV(reginfo->sv);
2685 }
2686
331b2dcc
DM		 2687 /* reserve next 2 or 3 slots in PL_regmatch_state:
2688 * slot N+0: may currently be in use: skip it
2689 * slot N+1: use for regmatch_info_aux struct
2690 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2691 * slot N+3: ready for use by regmatch()
2692 */
bf2039a9 2693
331b2dcc
DM		 2694 {
2695 regmatch_state *old_regmatch_state;
2696 regmatch_slab *old_regmatch_slab;
2697 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2698
2699 /* on first ever match, allocate first slab */
2700 if (!PL_regmatch_slab) {
2701 Newx(PL_regmatch_slab, 1, regmatch_slab);
2702 PL_regmatch_slab->prev = NULL;
2703 PL_regmatch_slab->next = NULL;
2704 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2705 }
bf2039a9 2706
331b2dcc
DM		 2707 old_regmatch_state = PL_regmatch_state;
2708 old_regmatch_slab = PL_regmatch_slab;
bf2039a9 2709
331b2dcc
DM		 2710 for (i=0; i <= max; i++) {
2711 if (i == 1)
2712 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2713 else if (i ==2)
2714 reginfo->info_aux_eval =
2715 reginfo->info_aux->info_aux_eval =
2716 &(PL_regmatch_state->u.info_aux_eval);
bf2039a9 2717
331b2dcc
DM		 2718 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2719 PL_regmatch_state = S_push_slab(aTHX);
2720 }
bf2039a9 2721
331b2dcc
DM		 2722 /* note initial PL_regmatch_state position; at end of match we'll
2723 * pop back to there and free any higher slabs */
bf2039a9 2724
331b2dcc
DM		 2725 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2726 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2ac8ff4b 2727 reginfo->info_aux->poscache = NULL;
bf2039a9 2728
331b2dcc 2729 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
bf2039a9 2730
331b2dcc
DM		 2731 if ((prog->extflags & RXf_EVAL_SEEN))
2732 S_setup_eval_state(aTHX_ reginfo);
2733 else
2734 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
bf2039a9 2735 }
d3aa529c 2736
6eb5f6b9 2737 /* If there is a "must appear" string, look for it. */
6eb5f6b9 2738
288b8c02 2739 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
e9105d30
GG		 2740 /* We have to be careful. If the previous successful match
2741 was from this regex we don't want a subsequent partially
2742 successful match to clobber the old results.
2743 So when we detect this possibility we add a swap buffer
d8da0584
KW		 2744 to the re, and switch the buffer each match. If we fail,
2745 we switch it back; otherwise we leave it swapped.
e9105d30
GG		 2746 */
2747 swap = prog->offs;
2748 /* do we need a save destructor here for eval dies? */
2749 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
495f47a5
DM		 2750 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2751 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2752 PTR2UV(prog),
2753 PTR2UV(swap),
2754 PTR2UV(prog->offs)
2755 ));
c74340f9 2756 }
6eb5f6b9
JH		 2757
2758 /* Simplest case: anchored match need be tried only once. */
d3d47aac 2759 /* [unless only anchor is MBOL - implying multiline is set] */
8e1490ee 2760 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
3542935d 2761 if (s == startpos && regtry(reginfo, &s))
6eb5f6b9 2762 goto got_it;
8e1490ee 2763 else if (multiline || (prog->intflags & (PREGf_IMPLICIT | PREGf_ANCH_MBOL))) /* XXXX SBOL? */
6eb5f6b9
JH		 2764 {
2765 char *end;
2766
2767 if (minlen)
2768 dontbother = minlen - 1;
1aa99e6b 2769 end = HOP3c(strend, -dontbother, strbeg) - 1;
6eb5f6b9 2770 /* for multiline we only have to try after newlines */
33b8afdf 2771 if (prog->check_substr || prog->check_utf8) {
92f3d482
YO		 2772 /* because of the goto we can not easily reuse the macros for bifurcating the
2773 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2774 if (utf8_target) {
2775 if (s == startpos)
2776 goto after_try_utf8;
2777 while (1) {
02d5137b 2778 if (regtry(reginfo, &s)) {
92f3d482
YO		 2779 goto got_it;
2780 }
2781 after_try_utf8:
2782 if (s > end) {
2783 goto phooey;
2784 }
2785 if (prog->extflags & RXf_USE_INTUIT) {
52a21eb3
DM		 2786 s = re_intuit_start(rx, sv, strbeg,
2787 s + UTF8SKIP(s), strend, flags, NULL);
92f3d482
YO		 2788 if (!s) {
2789 goto phooey;
2790 }
2791 }
2792 else {
2793 s += UTF8SKIP(s);
2794 }
2795 }
2796 } /* end search for check string in unicode */
2797 else {
2798 if (s == startpos) {
2799 goto after_try_latin;
2800 }
2801 while (1) {
02d5137b 2802 if (regtry(reginfo, &s)) {
92f3d482
YO		 2803 goto got_it;
2804 }
2805 after_try_latin:
2806 if (s > end) {
2807 goto phooey;
2808 }
2809 if (prog->extflags & RXf_USE_INTUIT) {
52a21eb3
DM		 2810 s = re_intuit_start(rx, sv, strbeg,
2811 s + 1, strend, flags, NULL);
92f3d482
YO		 2812 if (!s) {
2813 goto phooey;
2814 }
2815 }
2816 else {
2817 s++;
2818 }
2819 }
2820 } /* end search for check string in latin*/
2821 } /* end search for check string */
2822 else { /* search for newline */
2823 if (s > startpos) {
2824 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
6eb5f6b9 2825 s--;
92f3d482 2826 }
21eede78
YO		 2827 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2828 while (s <= end) { /* note it could be possible to match at the end of the string */
6eb5f6b9 2829 if (*s++ == '\n') { /* don't need PL_utf8skip here */
02d5137b 2830 if (regtry(reginfo, &s))
6eb5f6b9
JH		 2831 goto got_it;
2832 }
92f3d482
YO		 2833 }
2834 } /* end search for newline */
2835 } /* end anchored/multiline check string search */
6eb5f6b9 2836 goto phooey;
a8430a8b 2837 } else if (prog->intflags & PREGf_ANCH_GPOS)
f9f4320a 2838 {
a8430a8b
YO		 2839 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
2840 assert(prog->intflags & PREGf_GPOS_SEEN);
2bfbe302
DM		 2841 /* For anchored \G, the only position it can match from is
2842 * (ganch-gofs); we already set startpos to this above; if intuit
2843 * moved us on from there, we can't possibly succeed */
2844 assert(startpos == reginfo->ganch - prog->gofs);
2845 if (s == startpos && regtry(reginfo, &s))
6eb5f6b9
JH		 2846 goto got_it;
2847 goto phooey;
2848 }
2849
2850 /* Messy cases: unanchored match. */
bbe252da 2851 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
6eb5f6b9 2852 /* we have /x+whatever/ */
984e6dd1 2853 /* it must be a one character string (XXXX Except is_utf8_pat?) */
33b8afdf 2854 char ch;
bf93d4cc
GS		 2855#ifdef DEBUGGING
2856 int did_match = 0;
2857#endif
f2ed9b32 2858 if (utf8_target) {
7e0d5ad7
KW		 2859 if (! prog->anchored_utf8) {
2860 to_utf8_substr(prog);
2861 }
2862 ch = SvPVX_const(prog->anchored_utf8)[0];
4cadc6a9 2863 REXEC_FBC_SCAN(
6eb5f6b9 2864 if (*s == ch) {
a3621e74 2865 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 2866 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH		 2867 s += UTF8SKIP(s);
2868 while (s < strend && *s == ch)
2869 s += UTF8SKIP(s);
2870 }
4cadc6a9 2871 );
7e0d5ad7 2872
6eb5f6b9
JH		 2873 }
2874 else {
7e0d5ad7
KW		 2875 if (! prog->anchored_substr) {
2876 if (! to_byte_substr(prog)) {
6b54ddc5 2877 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW		 2878 }
2879 }
2880 ch = SvPVX_const(prog->anchored_substr)[0];
4cadc6a9 2881 REXEC_FBC_SCAN(
6eb5f6b9 2882 if (*s == ch) {
a3621e74 2883 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 2884 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH		 2885 s++;
2886 while (s < strend && *s == ch)
2887 s++;
2888 }
4cadc6a9 2889 );
6eb5f6b9 2890 }
a3621e74 2891 DEBUG_EXECUTE_r(if (!did_match)
bf93d4cc 2892 PerlIO_printf(Perl_debug_log,
b7953727
JH		 2893 "Did not find anchored character...\n")
2894 );
6eb5f6b9 2895 }
a0714e2c
SS		 2896 else if (prog->anchored_substr != NULL
2897 || prog->anchored_utf8 != NULL
2898 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
33b8afdf
JH		 2899 && prog->float_max_offset < strend - s)) {
2900 SV *must;
ea3daa5d
FC		 2901 SSize_t back_max;
2902 SSize_t back_min;
33b8afdf 2903 char *last;
6eb5f6b9 2904 char *last1; /* Last position checked before */
bf93d4cc
GS		 2905#ifdef DEBUGGING
2906 int did_match = 0;
2907#endif
33b8afdf 2908 if (prog->anchored_substr || prog->anchored_utf8) {
7e0d5ad7
KW		 2909 if (utf8_target) {
2910 if (! prog->anchored_utf8) {
2911 to_utf8_substr(prog);
2912 }
2913 must = prog->anchored_utf8;
2914 }
2915 else {
2916 if (! prog->anchored_substr) {
2917 if (! to_byte_substr(prog)) {
6b54ddc5 2918 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW		 2919 }
2920 }
2921 must = prog->anchored_substr;
2922 }
33b8afdf
JH		 2923 back_max = back_min = prog->anchored_offset;
2924 } else {
7e0d5ad7
KW		 2925 if (utf8_target) {
2926 if (! prog->float_utf8) {
2927 to_utf8_substr(prog);
2928 }
2929 must = prog->float_utf8;
2930 }
2931 else {
2932 if (! prog->float_substr) {
2933 if (! to_byte_substr(prog)) {
6b54ddc5 2934 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW		 2935 }
2936 }
2937 must = prog->float_substr;
2938 }
33b8afdf
JH		 2939 back_max = prog->float_max_offset;
2940 back_min = prog->float_min_offset;
2941 }
1de06328 2942
1de06328
YO		 2943 if (back_min<0) {
2944 last = strend;
2945 } else {
2946 last = HOP3c(strend, /* Cannot start after this */
ea3daa5d 2947 -(SSize_t)(CHR_SVLEN(must)
1de06328
YO		 2948 - (SvTAIL(must) != 0) + back_min), strbeg);
2949 }
9d9163fb 2950 if (s > reginfo->strbeg)
6eb5f6b9
JH		 2951 last1 = HOPc(s, -1);
2952 else
2953 last1 = s - 1; /* bogus */
2954
a0288114 2955 /* XXXX check_substr already used to find "s", can optimize if
6eb5f6b9 2956 check_substr==must. */
bf05793b 2957 dontbother = 0;
6eb5f6b9
JH		 2958 strend = HOPc(strend, -dontbother);
2959 while ( (s <= last) &&
e50d57d4 2960 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
9041c2e3 2961 (unsigned char*)strend, must,
c33e64f0 2962 multiline ? FBMrf_MULTILINE : 0)) ) {
a3621e74 2963 DEBUG_EXECUTE_r( did_match = 1 );
6eb5f6b9
JH		 2964 if (HOPc(s, -back_max) > last1) {
2965 last1 = HOPc(s, -back_min);
2966 s = HOPc(s, -back_max);
2967 }
2968 else {
9d9163fb
DM		 2969 char * const t = (last1 >= reginfo->strbeg)
2970 ? HOPc(last1, 1) : last1 + 1;
6eb5f6b9
JH		 2971
2972 last1 = HOPc(s, -back_min);
52657f30 2973 s = t;
6eb5f6b9 2974 }
f2ed9b32 2975 if (utf8_target) {
6eb5f6b9 2976 while (s <= last1) {
02d5137b 2977 if (regtry(reginfo, &s))
6eb5f6b9 2978 goto got_it;
7016d6eb
DM		 2979 if (s >= last1) {
2980 s++; /* to break out of outer loop */
2981 break;
2982 }
2983 s += UTF8SKIP(s);
6eb5f6b9
JH		 2984 }
2985 }
2986 else {
2987 while (s <= last1) {
02d5137b 2988 if (regtry(reginfo, &s))
6eb5f6b9
JH		 2989 goto got_it;
2990 s++;
2991 }
2992 }
2993 }
ab3bbdeb 2994 DEBUG_EXECUTE_r(if (!did_match) {
f2ed9b32 2995 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
ab3bbdeb
YO		 2996 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2997 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
33b8afdf 2998 ((must == prog->anchored_substr || must == prog->anchored_utf8)
bf93d4cc 2999 ? "anchored" : "floating"),
ab3bbdeb
YO		 3000 quoted, RE_SV_TAIL(must));
3001 });
6eb5f6b9
JH		 3002 goto phooey;
3003 }
f8fc2ecf 3004 else if ( (c = progi->regstclass) ) {
f14c76ed 3005 if (minlen) {
f8fc2ecf 3006 const OPCODE op = OP(progi->regstclass);
66e933ab 3007 /* don't bother with what can't match */
786e8c11 3008 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
f14c76ed
RGS		 3009 strend = HOPc(strend, -(minlen - 1));
3010 }
a3621e74 3011 DEBUG_EXECUTE_r({
be8e71aa 3012 SV * const prop = sv_newmortal();
8b9781c9 3013 regprop(prog, prop, c, reginfo, NULL);
0df25f3d 3014 {
f2ed9b32 3015 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb 3016 s,strend-s,60);
0df25f3d 3017 PerlIO_printf(Perl_debug_log,
1c8f8eb1 3018 "Matching stclass %.*s against %s (%d bytes)\n",
e4f74956 3019 (int)SvCUR(prop), SvPVX_const(prop),
ab3bbdeb 3020 quoted, (int)(strend - s));
0df25f3d 3021 }
ffc61ed2 3022 });
f9176b44 3023 if (find_byclass(prog, c, s, strend, reginfo))
6eb5f6b9 3024 goto got_it;
07be1b83 3025 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
d6a28714
JH		 3026 }
3027 else {
3028 dontbother = 0;
a0714e2c 3029 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
33b8afdf 3030 /* Trim the end. */
6af40bd7 3031 char *last= NULL;
33b8afdf 3032 SV* float_real;
c33e64f0
FC		 3033 STRLEN len;
3034 const char *little;
33b8afdf 3035
7e0d5ad7
KW		 3036 if (utf8_target) {
3037 if (! prog->float_utf8) {
3038 to_utf8_substr(prog);
3039 }
3040 float_real = prog->float_utf8;
3041 }
3042 else {
3043 if (! prog->float_substr) {
3044 if (! to_byte_substr(prog)) {
6b54ddc5 3045 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW		 3046 }
3047 }
3048 float_real = prog->float_substr;
3049 }
d6a28714 3050
c33e64f0
FC		 3051 little = SvPV_const(float_real, len);
3052 if (SvTAIL(float_real)) {
7f18ad16
KW		 3053 /* This means that float_real contains an artificial \n on
3054 * the end due to the presence of something like this:
3055 * /foo$/ where we can match both "foo" and "foo\n" at the
3056 * end of the string. So we have to compare the end of the
3057 * string first against the float_real without the \n and
3058 * then against the full float_real with the string. We
3059 * have to watch out for cases where the string might be
3060 * smaller than the float_real or the float_real without
3061 * the \n. */
1a13b075
YO		 3062 char *checkpos= strend - len;
3063 DEBUG_OPTIMISE_r(
3064 PerlIO_printf(Perl_debug_log,
3065 "%sChecking for float_real.%s\n",
3066 PL_colors[4], PL_colors[5]));
3067 if (checkpos + 1 < strbeg) {
7f18ad16
KW		 3068 /* can't match, even if we remove the trailing \n
3069 * string is too short to match */
1a13b075
YO		 3070 DEBUG_EXECUTE_r(
3071 PerlIO_printf(Perl_debug_log,
3072 "%sString shorter than required trailing substring, cannot match.%s\n",
3073 PL_colors[4], PL_colors[5]));
3074 goto phooey;
3075 } else if (memEQ(checkpos + 1, little, len - 1)) {
7f18ad16
KW		 3076 /* can match, the end of the string matches without the
3077 * "\n" */
1a13b075
YO		 3078 last = checkpos + 1;
3079 } else if (checkpos < strbeg) {
7f18ad16
KW		 3080 /* cant match, string is too short when the "\n" is
3081 * included */
1a13b075
YO		 3082 DEBUG_EXECUTE_r(
3083 PerlIO_printf(Perl_debug_log,
3084 "%sString does not contain required trailing substring, cannot match.%s\n",
3085 PL_colors[4], PL_colors[5]));
3086 goto phooey;
3087 } else if (!multiline) {
7f18ad16
KW		 3088 /* non multiline match, so compare with the "\n" at the
3089 * end of the string */
1a13b075
YO		 3090 if (memEQ(checkpos, little, len)) {
3091 last= checkpos;
3092 } else {
3093 DEBUG_EXECUTE_r(
3094 PerlIO_printf(Perl_debug_log,
3095 "%sString does not contain required trailing substring, cannot match.%s\n",
3096 PL_colors[4], PL_colors[5]));
3097 goto phooey;
3098 }
3099 } else {
7f18ad16
KW		 3100 /* multiline match, so we have to search for a place
3101 * where the full string is located */
d6a28714 3102 goto find_last;
1a13b075 3103 }
c33e64f0 3104 } else {
d6a28714 3105 find_last:
9041c2e3 3106 if (len)
d6a28714 3107 last = rninstr(s, strend, little, little + len);
b8c5462f 3108 else
a0288114 3109 last = strend; /* matching "$" */
b8c5462f 3110 }
6af40bd7 3111 if (!last) {
7f18ad16
KW		 3112 /* at one point this block contained a comment which was
3113 * probably incorrect, which said that this was a "should not
3114 * happen" case. Even if it was true when it was written I am
3115 * pretty sure it is not anymore, so I have removed the comment
3116 * and replaced it with this one. Yves */
6bda09f9
YO		 3117 DEBUG_EXECUTE_r(
3118 PerlIO_printf(Perl_debug_log,
b729e729
YO		 3119 "%sString does not contain required substring, cannot match.%s\n",
3120 PL_colors[4], PL_colors[5]
6af40bd7
YO		 3121 ));
3122 goto phooey;
bf93d4cc 3123 }
d6a28714
JH		 3124 dontbother = strend - last + prog->float_min_offset;
3125 }
3126 if (minlen && (dontbother < minlen))
3127 dontbother = minlen - 1;
3128 strend -= dontbother; /* this one's always in bytes! */
3129 /* We don't know much -- general case. */
f2ed9b32 3130 if (utf8_target) {
d6a28714 3131 for (;;) {
02d5137b 3132 if (regtry(reginfo, &s))
d6a28714
JH		 3133 goto got_it;
3134 if (s >= strend)
3135 break;
b8c5462f 3136 s += UTF8SKIP(s);
d6a28714
JH		 3137 };
3138 }
3139 else {
3140 do {
02d5137b 3141 if (regtry(reginfo, &s))
d6a28714
JH		 3142 goto got_it;
3143 } while (s++ < strend);
3144 }
3145 }
3146
3147 /* Failure. */
3148 goto phooey;
3149
7b52d656 3150 got_it:
d5e7783a
DM		 3151 /* s/// doesn't like it if $& is earlier than where we asked it to
3152 * start searching (which can happen on something like /.\G/) */
3153 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3154 && (prog->offs[0].start < stringarg - strbeg))
3155 {
3156 /* this should only be possible under \G */
58430ea8 3157 assert(prog->intflags & PREGf_GPOS_SEEN);
d5e7783a
DM		 3158 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3159 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3160 goto phooey;
3161 }
3162
495f47a5
DM		 3163 DEBUG_BUFFERS_r(
3164 if (swap)
3165 PerlIO_printf(Perl_debug_log,
3166 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
3167 PTR2UV(prog),
3168 PTR2UV(swap)
3169 );
3170 );
e9105d30 3171 Safefree(swap);
d6a28714 3172
bf2039a9
DM		 3173 /* clean up; this will trigger destructors that will free all slabs
3174 * above the current one, and cleanup the regmatch_info_aux
3175 * and regmatch_info_aux_eval sructs */
8adc0f72 3176
006f26b2
DM		 3177 LEAVE_SCOPE(oldsave);
3178
5daac39c
NC		 3179 if (RXp_PAREN_NAMES(prog))
3180 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
d6a28714
JH		 3181
3182 /* make sure $`, $&, $', and $digit will work later */
60165aa4 3183 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 3184 S_reg_set_capture_string(aTHX_ rx,
60165aa4
DM		 3185 strbeg, reginfo->strend,
3186 sv, flags, utf8_target);
9041c2e3 3187
d6a28714
JH		 3188 return 1;
3189
7b52d656 3190 phooey:
a3621e74 3191 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
e4584336 3192 PL_colors[4], PL_colors[5]));
8adc0f72 3193
bf2039a9
DM		 3194 /* clean up; this will trigger destructors that will free all slabs
3195 * above the current one, and cleanup the regmatch_info_aux
3196 * and regmatch_info_aux_eval sructs */
8adc0f72 3197
006f26b2
DM		 3198 LEAVE_SCOPE(oldsave);
3199
e9105d30 3200 if (swap) {
c74340f9 3201 /* we failed :-( roll it back */
495f47a5
DM		 3202 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
3203 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
3204 PTR2UV(prog),
3205 PTR2UV(prog->offs),
3206 PTR2UV(swap)
3207 ));
e9105d30
GG		 3208 Safefree(prog->offs);
3209 prog->offs = swap;
3210 }
d6a28714
JH		 3211 return 0;
3212}
3213
6bda09f9 3214
b3d298be 3215/* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
ec43f78b 3216 * Do inc before dec, in case old and new rex are the same */
baa60164 3217#define SET_reg_curpm(Re2) \
bf2039a9 3218 if (reginfo->info_aux_eval) { \
ec43f78b
DM		 3219 (void)ReREFCNT_inc(Re2); \
3220 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3221 PM_SETRE((PL_reg_curpm), (Re2)); \
3222 }
3223
3224
d6a28714
JH		 3225/*
3226 - regtry - try match at specific point
3227 */
3228STATIC I32 /* 0 failure, 1 success */
f73aaa43 3229S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
d6a28714 3230{
d6a28714 3231 CHECKPOINT lastcp;
288b8c02 3232 REGEXP *const rx = reginfo->prog;
8d919b0a 3233 regexp *const prog = ReANY(rx);
99a90e59 3234 SSize_t result;
f8fc2ecf 3235 RXi_GET_DECL(prog,progi);
a3621e74 3236 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC		 3237
3238 PERL_ARGS_ASSERT_REGTRY;
3239
24b23f37 3240 reginfo->cutpoint=NULL;
d6a28714 3241
9d9163fb 3242 prog->offs[0].start = *startposp - reginfo->strbeg;
d6a28714 3243 prog->lastparen = 0;
03994de8 3244 prog->lastcloseparen = 0;
d6a28714
JH		 3245
3246 /* XXXX What this code is doing here?!!! There should be no need
b93070ed 3247 to do this again and again, prog->lastparen should take care of
3dd2943c 3248 this! --ilya*/
dafc8851
JH		 3249
3250 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3251 * Actually, the code in regcppop() (which Ilya may be meaning by
b93070ed 3252 * prog->lastparen), is not needed at all by the test suite
225593e1
DM		 3253 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3254 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3255 * Meanwhile, this code *is* needed for the
daf18116
JH		 3256 * above-mentioned test suite tests to succeed. The common theme
3257 * on those tests seems to be returning null fields from matches.
225593e1 3258 * --jhi updated by dapm */
dafc8851 3259#if 1
d6a28714 3260 if (prog->nparens) {
b93070ed 3261 regexp_paren_pair *pp = prog->offs;
eb578fdb 3262 I32 i;
b93070ed 3263 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
f0ab9afb
NC		 3264 ++pp;
3265 pp->start = -1;
3266 pp->end = -1;
d6a28714
JH		 3267 }
3268 }
dafc8851 3269#endif
02db2b7b 3270 REGCP_SET(lastcp);
f73aaa43
DM		 3271 result = regmatch(reginfo, *startposp, progi->program + 1);
3272 if (result != -1) {
3273 prog->offs[0].end = result;
d6a28714
JH		 3274 return 1;
3275 }
24b23f37 3276 if (reginfo->cutpoint)
f73aaa43 3277 *startposp= reginfo->cutpoint;
02db2b7b 3278 REGCP_UNWIND(lastcp);
d6a28714
JH		 3279 return 0;
3280}
3281
02db2b7b 3282
8ba1375e
MJD		 3283#define sayYES goto yes
3284#define sayNO goto no
262b90c4 3285#define sayNO_SILENT goto no_silent
8ba1375e 3286
f9f4320a
YO		 3287/* we dont use STMT_START/END here because it leads to
3288 "unreachable code" warnings, which are bogus, but distracting. */
3289#define CACHEsayNO \
c476f425 3290 if (ST.cache_mask) \
2ac8ff4b 3291 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
f9f4320a 3292 sayNO
3298f257 3293
a3621e74 3294/* this is used to determine how far from the left messages like
265c4333
YO		 3295 'failed...' are printed. It should be set such that messages
3296 are inline with the regop output that created them.
a3621e74 3297*/
265c4333 3298#define REPORT_CODE_OFF 32
a3621e74
YO		 3299
3300
40a82448
DM		 3301#define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3302#define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
79a2a0e8
KW		 3303#define CHRTEST_NOT_A_CP_1 -999
3304#define CHRTEST_NOT_A_CP_2 -998
9e137952 3305
5d9a96ca
DM		 3306/* grab a new slab and return the first slot in it */
3307
3308STATIC regmatch_state *
3309S_push_slab(pTHX)
3310{
a35a87e7 3311#if PERL_VERSION < 9 && !defined(PERL_CORE)
54df2634
NC		 3312 dMY_CXT;
3313#endif
5d9a96ca
DM		 3314 regmatch_slab *s = PL_regmatch_slab->next;
3315 if (!s) {
3316 Newx(s, 1, regmatch_slab);
3317 s->prev = PL_regmatch_slab;
3318 s->next = NULL;
3319 PL_regmatch_slab->next = s;
3320 }
3321 PL_regmatch_slab = s;
86545054 3322 return SLAB_FIRST(s);
5d9a96ca 3323}
5b47454d 3324
95b24440 3325
40a82448
DM		 3326/* push a new state then goto it */
3327
4d5016e5
DM		 3328#define PUSH_STATE_GOTO(state, node, input) \
3329 pushinput = input; \
40a82448
DM		 3330 scan = node; \
3331 st->resume_state = state; \
3332 goto push_state;
3333
3334/* push a new state with success backtracking, then goto it */
3335
4d5016e5
DM		 3336#define PUSH_YES_STATE_GOTO(state, node, input) \
3337 pushinput = input; \
40a82448
DM		 3338 scan = node; \
3339 st->resume_state = state; \
3340 goto push_yes_state;
3341
aa283a38 3342
aa283a38 3343
4d5016e5 3344
d6a28714 3345/*
95b24440 3346
bf1f174e
DM		 3347regmatch() - main matching routine
3348
3349This is basically one big switch statement in a loop. We execute an op,
3350set 'next' to point the next op, and continue. If we come to a point which
3351we may need to backtrack to on failure such as (A|B|C), we push a
3352backtrack state onto the backtrack stack. On failure, we pop the top
3353state, and re-enter the loop at the state indicated. If there are no more
3354states to pop, we return failure.
3355
3356Sometimes we also need to backtrack on success; for example /A+/, where
3357after successfully matching one A, we need to go back and try to
3358match another one; similarly for lookahead assertions: if the assertion
3359completes successfully, we backtrack to the state just before the assertion
3360and then carry on. In these cases, the pushed state is marked as
3361'backtrack on success too'. This marking is in fact done by a chain of
3362pointers, each pointing to the previous 'yes' state. On success, we pop to
3363the nearest yes state, discarding any intermediate failure-only states.
3364Sometimes a yes state is pushed just to force some cleanup code to be
3365called at the end of a successful match or submatch; e.g. (??{$re}) uses
3366it to free the inner regex.
3367
3368Note that failure backtracking rewinds the cursor position, while
3369success backtracking leaves it alone.
3370
3371A pattern is complete when the END op is executed, while a subpattern
3372such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3373ops trigger the "pop to last yes state if any, otherwise return true"
3374behaviour.
3375
3376A common convention in this function is to use A and B to refer to the two
3377subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3378the subpattern to be matched possibly multiple times, while B is the entire
3379rest of the pattern. Variable and state names reflect this convention.
3380
3381The states in the main switch are the union of ops and failure/success of
3382substates associated with with that op. For example, IFMATCH is the op
3383that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3384'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3385successfully matched A and IFMATCH_A_fail is a state saying that we have
3386just failed to match A. Resume states always come in pairs. The backtrack
3387state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3388at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3389on success or failure.
3390
3391The struct that holds a backtracking state is actually a big union, with
3392one variant for each major type of op. The variable st points to the
3393top-most backtrack struct. To make the code clearer, within each
3394block of code we #define ST to alias the relevant union.
3395
3396Here's a concrete example of a (vastly oversimplified) IFMATCH
3397implementation:
3398
3399 switch (state) {
3400 ....
3401
3402#define ST st->u.ifmatch
3403
3404 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3405 ST.foo = ...; // some state we wish to save
95b24440 3406 ...
bf1f174e
DM		 3407 // push a yes backtrack state with a resume value of
3408 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3409 // first node of A:
4d5016e5 3410 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
bf1f174e
DM		 3411 // NOTREACHED
3412
3413 case IFMATCH_A: // we have successfully executed A; now continue with B
3414 next = B;
3415 bar = ST.foo; // do something with the preserved value
3416 break;
3417
3418 case IFMATCH_A_fail: // A failed, so the assertion failed
3419 ...; // do some housekeeping, then ...
3420 sayNO; // propagate the failure
3421
3422#undef ST
95b24440 3423
bf1f174e
DM		 3424 ...
3425 }
95b24440 3426
bf1f174e
DM		 3427For any old-timers reading this who are familiar with the old recursive
3428approach, the code above is equivalent to:
95b24440 3429
bf1f174e
DM		 3430 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3431 {
3432 int foo = ...
95b24440 3433 ...
bf1f174e
DM		 3434 if (regmatch(A)) {
3435 next = B;
3436 bar = foo;
3437 break;
95b24440 3438 }
bf1f174e
DM		 3439 ...; // do some housekeeping, then ...
3440 sayNO; // propagate the failure
95b24440 3441 }
bf1f174e
DM		 3442
3443The topmost backtrack state, pointed to by st, is usually free. If you
3444want to claim it, populate any ST.foo fields in it with values you wish to
3445save, then do one of
3446
4d5016e5
DM		 3447 PUSH_STATE_GOTO(resume_state, node, newinput);
3448 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
bf1f174e
DM		 3449
3450which sets that backtrack state's resume value to 'resume_state', pushes a
3451new free entry to the top of the backtrack stack, then goes to 'node'.
3452On backtracking, the free slot is popped, and the saved state becomes the
3453new free state. An ST.foo field in this new top state can be temporarily
3454accessed to retrieve values, but once the main loop is re-entered, it
3455becomes available for reuse.
3456
3457Note that the depth of the backtrack stack constantly increases during the
3458left-to-right execution of the pattern, rather than going up and down with
3459the pattern nesting. For example the stack is at its maximum at Z at the
3460end of the pattern, rather than at X in the following:
3461
3462 /(((X)+)+)+....(Y)+....Z/
3463
3464The only exceptions to this are lookahead/behind assertions and the cut,
3465(?>A), which pop all the backtrack states associated with A before
3466continuing.
3467
486ec47a 3468Backtrack state structs are allocated in slabs of about 4K in size.
bf1f174e
DM		 3469PL_regmatch_state and st always point to the currently active state,
3470and PL_regmatch_slab points to the slab currently containing
3471PL_regmatch_state. The first time regmatch() is called, the first slab is
3472allocated, and is never freed until interpreter destruction. When the slab
3473is full, a new one is allocated and chained to the end. At exit from
3474regmatch(), slabs allocated since entry are freed.
3475
3476*/
95b24440 3477
40a82448 3478
5bc10b2c 3479#define DEBUG_STATE_pp(pp) \
265c4333 3480 DEBUG_STATE_r({ \
baa60164 3481 DUMP_EXEC_POS(locinput, scan, utf8_target); \
5bc10b2c 3482 PerlIO_printf(Perl_debug_log, \
5d458dd8 3483 " %*s"pp" %s%s%s%s%s\n", \
5bc10b2c 3484 depth*2, "", \
baa60164 3485 PL_reg_name[st->resume_state], \
5d458dd8
YO		 3486 ((st==yes_state||st==mark_state) ? "[" : ""), \
3487 ((st==yes_state) ? "Y" : ""), \
3488 ((st==mark_state) ? "M" : ""), \
3489 ((st==yes_state||st==mark_state) ? "]" : "") \
3490 ); \
265c4333 3491 });
5bc10b2c 3492
40a82448 3493
3dab1dad 3494#define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
95b24440 3495
3df15adc 3496#ifdef DEBUGGING
5bc10b2c 3497
ab3bbdeb 3498STATIC void
f2ed9b32 3499S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
ab3bbdeb
YO		 3500 const char *start, const char *end, const char *blurb)
3501{
efd26800 3502 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
7918f24d
NC		 3503
3504 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3505
ab3bbdeb
YO		 3506 if (!PL_colorset)
3507 reginitcolors();
3508 {
3509 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
d2c6dc5e 3510 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
ab3bbdeb 3511
f2ed9b32 3512 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb
YO		 3513 start, end - start, 60);
3514
3515 PerlIO_printf(Perl_debug_log,
3516 "%s%s REx%s %s against %s\n",
3517 PL_colors[4], blurb, PL_colors[5], s0, s1);
3518
f2ed9b32 3519 if (utf8_target||utf8_pat)
1de06328
YO		 3520 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3521 utf8_pat ? "pattern" : "",
f2ed9b32
KW		 3522 utf8_pat && utf8_target ? " and " : "",
3523 utf8_target ? "string" : ""
ab3bbdeb
YO		 3524 );
3525 }
3526}
3df15adc
YO		 3527
3528STATIC void
786e8c11
YO		 3529S_dump_exec_pos(pTHX_ const char *locinput,
3530 const regnode *scan,
3531 const char *loc_regeol,
3532 const char *loc_bostr,
3533 const char *loc_reg_starttry,
f2ed9b32 3534 const bool utf8_target)
07be1b83 3535{
786e8c11 3536 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
07be1b83 3537 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
786e8c11 3538 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
07be1b83
YO		 3539 /* The part of the string before starttry has one color
3540 (pref0_len chars), between starttry and current
3541 position another one (pref_len - pref0_len chars),
3542 after the current position the third one.
3543 We assume that pref0_len <= pref_len, otherwise we
3544 decrease pref0_len. */
786e8c11
YO		 3545 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3546 ? (5 + taill) - l : locinput - loc_bostr;
07be1b83
YO		 3547 int pref0_len;
3548
7918f24d
NC		 3549 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3550
f2ed9b32 3551 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
07be1b83 3552 pref_len++;
786e8c11
YO		 3553 pref0_len = pref_len - (locinput - loc_reg_starttry);
3554 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3555 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3556 ? (5 + taill) - pref_len : loc_regeol - locinput);
f2ed9b32 3557 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
07be1b83
YO		 3558 l--;
3559 if (pref0_len < 0)
3560 pref0_len = 0;
3561 if (pref0_len > pref_len)
3562 pref0_len = pref_len;
3563 {
f2ed9b32 3564 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
0df25f3d 3565
ab3bbdeb 3566 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
1de06328 3567 (locinput - pref_len),pref0_len, 60, 4, 5);
0df25f3d 3568
ab3bbdeb 3569 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3df15adc 3570 (locinput - pref_len + pref0_len),
1de06328 3571 pref_len - pref0_len, 60, 2, 3);
0df25f3d 3572
ab3bbdeb 3573 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
1de06328 3574 locinput, loc_regeol - locinput, 10, 0, 1);
0df25f3d 3575
1de06328 3576 const STRLEN tlen=len0+len1+len2;
3df15adc 3577 PerlIO_printf(Perl_debug_log,
ab3bbdeb 3578 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
786e8c11 3579 (IV)(locinput - loc_bostr),
07be1b83 3580 len0, s0,
07be1b83 3581 len1, s1,
07be1b83 3582 (docolor ? "" : "> <"),
07be1b83 3583 len2, s2,
f9f4320a 3584 (int)(tlen > 19 ? 0 : 19 - tlen),
07be1b83
YO		 3585 "");
3586 }
3587}
3df15adc 3588
07be1b83
YO		 3589#endif
3590
0a4db386
YO		 3591/* reg_check_named_buff_matched()
3592 * Checks to see if a named buffer has matched. The data array of
3593 * buffer numbers corresponding to the buffer is expected to reside
3594 * in the regexp->data->data array in the slot stored in the ARG() of
3595 * node involved. Note that this routine doesn't actually care about the
3596 * name, that information is not preserved from compilation to execution.
3597 * Returns the index of the leftmost defined buffer with the given name
3598 * or 0 if non of the buffers matched.
3599 */
3600STATIC I32
dc3bf405 3601S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
7918f24d 3602{
0a4db386 3603 I32 n;
f8fc2ecf 3604 RXi_GET_DECL(rex,rexi);
ad64d0ec 3605 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
0a4db386 3606 I32 *nums=(I32*)SvPVX(sv_dat);
7918f24d
NC		 3607
3608 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3609
0a4db386 3610 for ( n=0; n<SvIVX(sv_dat); n++ ) {
b93070ed
DM		 3611 if ((I32)rex->lastparen >= nums[n] &&
3612 rex->offs[nums[n]].end != -1)
0a4db386
YO		 3613 {
3614 return nums[n];
3615 }
3616 }
3617 return 0;
3618}
3619
2f554ef7 3620
c74f6de9 3621static bool
984e6dd1 3622S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
aed7b151 3623 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
c74f6de9 3624{
79a2a0e8
KW		 3625 /* This function determines if there are one or two characters that match
3626 * the first character of the passed-in EXACTish node <text_node>, and if
3627 * so, returns them in the passed-in pointers.
c74f6de9 3628 *
79a2a0e8
KW		 3629 * If it determines that no possible character in the target string can
3630 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3631 * the first character in <text_node> requires UTF-8 to represent, and the
3632 * target string isn't in UTF-8.)
c74f6de9 3633 *
79a2a0e8
KW		 3634 * If there are more than two characters that could match the beginning of
3635 * <text_node>, or if more context is required to determine a match or not,
3636 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3637 *
3638 * The motiviation behind this function is to allow the caller to set up
3639 * tight loops for matching. If <text_node> is of type EXACT, there is
3640 * only one possible character that can match its first character, and so
3641 * the situation is quite simple. But things get much more complicated if
3642 * folding is involved. It may be that the first character of an EXACTFish
3643 * node doesn't participate in any possible fold, e.g., punctuation, so it
3644 * can be matched only by itself. The vast majority of characters that are
3645 * in folds match just two things, their lower and upper-case equivalents.
3646 * But not all are like that; some have multiple possible matches, or match
3647 * sequences of more than one character. This function sorts all that out.
3648 *
3649 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3650 * loop of trying to match A*, we know we can't exit where the thing
3651 * following it isn't a B. And something can't be a B unless it is the
3652 * beginning of B. By putting a quick test for that beginning in a tight
3653 * loop, we can rule out things that can't possibly be B without having to
3654 * break out of the loop, thus avoiding work. Similarly, if A is a single
3655 * character, we can make a tight loop matching A*, using the outputs of
3656 * this function.
3657 *
3658 * If the target string to match isn't in UTF-8, and there aren't
3659 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3660 * the one or two possible octets (which are characters in this situation)
3661 * that can match. In all cases, if there is only one character that can
3662 * match, *<c1p> and *<c2p> will be identical.
3663 *
3664 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3665 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3666 * can match the beginning of <text_node>. They should be declared with at
3667 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3668 * undefined what these contain.) If one or both of the buffers are
3669 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3670 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3671 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3672 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3673 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
c74f6de9 3674
ba44c216 3675 const bool utf8_target = reginfo->is_utf8_target;
79a2a0e8 3676
9cd990bf
JH		 3677 UV c1 = (UV)CHRTEST_NOT_A_CP_1;
3678 UV c2 = (UV)CHRTEST_NOT_A_CP_2;
79a2a0e8 3679 bool use_chrtest_void = FALSE;
aed7b151 3680 const bool is_utf8_pat = reginfo->is_utf8_pat;
79a2a0e8
KW		 3681
3682 /* Used when we have both utf8 input and utf8 output, to avoid converting
3683 * to/from code points */
3684 bool utf8_has_been_setup = FALSE;
3685
c74f6de9
KW		 3686 dVAR;
3687
b4291290 3688 U8 *pat = (U8*)STRING(text_node);
a6715020 3689 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
c74f6de9 3690
a4525e78 3691 if (OP(text_node) == EXACT || OP(text_node) == EXACTL) {
79a2a0e8
KW		 3692
3693 /* In an exact node, only one thing can be matched, that first
3694 * character. If both the pat and the target are UTF-8, we can just
3695 * copy the input to the output, avoiding finding the code point of
3696 * that character */
984e6dd1 3697 if (!is_utf8_pat) {
79a2a0e8
KW		 3698 c2 = c1 = *pat;
3699 }
3700 else if (utf8_target) {
3701 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3702 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3703 utf8_has_been_setup = TRUE;
3704 }
3705 else {
3706 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
c74f6de9 3707 }
79a2a0e8 3708 }
31f05a37
KW		 3709 else { /* an EXACTFish node */
3710 U8 *pat_end = pat + STR_LEN(text_node);
3711
3712 /* An EXACTFL node has at least some characters unfolded, because what
3713 * they match is not known until now. So, now is the time to fold
3714 * the first few of them, as many as are needed to determine 'c1' and
3715 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
3716 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
3717 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
3718 * need to fold as many characters as a single character can fold to,
3719 * so that later we can check if the first ones are such a multi-char
3720 * fold. But, in such a pattern only locale-problematic characters
3721 * aren't folded, so we can skip this completely if the first character
3722 * in the node isn't one of the tricky ones */
3723 if (OP(text_node) == EXACTFL) {
3724
3725 if (! is_utf8_pat) {
3726 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
3727 {
3728 folded[0] = folded[1] = 's';
3729 pat = folded;
3730 pat_end = folded + 2;
3731 }
3732 }
3733 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
3734 U8 *s = pat;
3735 U8 *d = folded;
3736 int i;
3737
3738 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
3739 if (isASCII(*s)) {
3740 *(d++) = (U8) toFOLD_LC(*s);
3741 s++;
3742 }
3743 else {
3744 STRLEN len;
3745 _to_utf8_fold_flags(s,
3746 d,
3747 &len,
3748 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE);
3749 d += len;
3750 s += UTF8SKIP(s);
3751 }
3752 }
3753
3754 pat = folded;
3755 pat_end = d;
3756 }
3757 }
3758
251b239f
KW		 3759 if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end))
3760 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end)))
baa60164
KW		 3761 {
3762 /* Multi-character folds require more context to sort out. Also
3763 * PL_utf8_foldclosures used below doesn't handle them, so have to
3764 * be handled outside this routine */
3765 use_chrtest_void = TRUE;
3766 }
3767 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3768 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
e8ea8356 3769 if (c1 > 255) {
baa60164
KW		 3770 /* Load the folds hash, if not already done */
3771 SV** listp;
3772 if (! PL_utf8_foldclosures) {
31667e6b 3773 _load_PL_utf8_foldclosures();
79a2a0e8 3774 }
79a2a0e8 3775
baa60164
KW		 3776 /* The fold closures data structure is a hash with the keys
3777 * being the UTF-8 of every character that is folded to, like
3778 * 'k', and the values each an array of all code points that
3779 * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ].
3780 * Multi-character folds are not included */
3781 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3782 (char *) pat,
3783 UTF8SKIP(pat),
3784 FALSE))))
3785 {
3786 /* Not found in the hash, therefore there are no folds
3787 * containing it, so there is only a single character that
3788 * could match */
3789 c2 = c1;
79a2a0e8 3790 }
baa60164
KW		 3791 else { /* Does participate in folds */
3792 AV* list = (AV*) *listp;
b9f2b683 3793 if (av_tindex(list) != 1) {
79a2a0e8 3794
baa60164
KW		 3795 /* If there aren't exactly two folds to this, it is
3796 * outside the scope of this function */
3797 use_chrtest_void = TRUE;
79a2a0e8 3798 }
baa60164
KW		 3799 else { /* There are two. Get them */
3800 SV** c_p = av_fetch(list, 0, FALSE);
3801 if (c_p == NULL) {
3802 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3803 }
3804 c1 = SvUV(*c_p);
3805
3806 c_p = av_fetch(list, 1, FALSE);
3807 if (c_p == NULL) {
3808 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3809 }
3810 c2 = SvUV(*c_p);
3811
3812 /* Folds that cross the 255/256 boundary are forbidden
3813 * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and
3814 * one is ASCIII. Since the pattern character is above
e8ea8356 3815 * 255, and its only other match is below 256, the only
baa60164
KW		 3816 * legal match will be to itself. We have thrown away
3817 * the original, so have to compute which is the one
e8ea8356 3818 * above 255. */
baa60164
KW		 3819 if ((c1 < 256) != (c2 < 256)) {
3820 if ((OP(text_node) == EXACTFL
3821 && ! IN_UTF8_CTYPE_LOCALE)
3822 || ((OP(text_node) == EXACTFA
3823 || OP(text_node) == EXACTFA_NO_TRIE)
3824 && (isASCII(c1) || isASCII(c2))))
3825 {
3826 if (c1 < 256) {
3827 c1 = c2;
3828 }
3829 else {
3830 c2 = c1;
3831 }
79a2a0e8
KW		 3832 }
3833 }
3834 }
3835 }
3836 }
e8ea8356 3837 else /* Here, c1 is <= 255 */
baa60164
KW		 3838 if (utf8_target
3839 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3840 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE))
3841 && ((OP(text_node) != EXACTFA
3842 && OP(text_node) != EXACTFA_NO_TRIE)
3843 || ! isASCII(c1)))
3844 {
3845 /* Here, there could be something above Latin1 in the target
3846 * which folds to this character in the pattern. All such
3847 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more
3848 * than two characters involved in their folds, so are outside
3849 * the scope of this function */
3850 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3851 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3852 }
3853 else {
3854 use_chrtest_void = TRUE;
3855 }
79a2a0e8 3856 }
baa60164
KW		 3857 else { /* Here nothing above Latin1 can fold to the pattern
3858 character */
3859 switch (OP(text_node)) {
c74f6de9 3860
baa60164
KW		 3861 case EXACTFL: /* /l rules */
3862 c2 = PL_fold_locale[c1];
3863 break;
c74f6de9 3864
baa60164
KW		 3865 case EXACTF: /* This node only generated for non-utf8
3866 patterns */
3867 assert(! is_utf8_pat);
3868 if (! utf8_target) { /* /d rules */
3869 c2 = PL_fold[c1];
3870 break;
3871 }
3872 /* FALLTHROUGH */
3873 /* /u rules for all these. This happens to work for
3874 * EXACTFA as nothing in Latin1 folds to ASCII */
3875 case EXACTFA_NO_TRIE: /* This node only generated for
3876 non-utf8 patterns */
3877 assert(! is_utf8_pat);
924ba076 3878 /* FALLTHROUGH */
baa60164
KW		 3879 case EXACTFA:
3880 case EXACTFU_SS:
3881 case EXACTFU:
3882 c2 = PL_fold_latin1[c1];
c74f6de9 3883 break;
c74f6de9 3884
baa60164
KW		 3885 default:
3886 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
e5964223 3887 NOT_REACHED; /* NOTREACHED */
baa60164 3888 }
c74f6de9
KW		 3889 }
3890 }
3891 }
79a2a0e8
KW		 3892
3893 /* Here have figured things out. Set up the returns */
3894 if (use_chrtest_void) {
3895 *c2p = *c1p = CHRTEST_VOID;
3896 }
3897 else if (utf8_target) {
3898 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3899 uvchr_to_utf8(c1_utf8, c1);
3900 uvchr_to_utf8(c2_utf8, c2);
c74f6de9 3901 }
c74f6de9 3902
79a2a0e8
KW		 3903 /* Invariants are stored in both the utf8 and byte outputs; Use
3904 * negative numbers otherwise for the byte ones. Make sure that the
3905 * byte ones are the same iff the utf8 ones are the same */
3906 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3907 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3908 ? *c2_utf8
3909 : (c1 == c2)
3910 ? CHRTEST_NOT_A_CP_1
3911 : CHRTEST_NOT_A_CP_2;
3912 }
3913 else if (c1 > 255) {
3914 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3915 can represent */
3916 return FALSE;
3917 }
c74f6de9 3918
79a2a0e8
KW		 3919 *c1p = *c2p = c2; /* c2 is the only representable value */
3920 }
3921 else { /* c1 is representable; see about c2 */
3922 *c1p = c1;
3923 *c2p = (c2 < 256) ? c2 : c1;
c74f6de9 3924 }
2f554ef7 3925
c74f6de9
KW		 3926 return TRUE;
3927}
2f554ef7 3928
f73aaa43 3929/* returns -1 on failure, $+[0] on success */
99a90e59 3930STATIC SSize_t
f73aaa43 3931S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
d6a28714 3932{
a35a87e7 3933#if PERL_VERSION < 9 && !defined(PERL_CORE)
54df2634
NC		 3934 dMY_CXT;
3935#endif
27da23d5 3936 dVAR;
ba44c216 3937 const bool utf8_target = reginfo->is_utf8_target;
4ad0818d 3938 const U32 uniflags = UTF8_ALLOW_DEFAULT;
288b8c02 3939 REGEXP *rex_sv = reginfo->prog;
8d919b0a 3940 regexp *rex = ReANY(rex_sv);
f8fc2ecf 3941 RXi_GET_DECL(rex,rexi);
5d9a96ca 3942 /* the current state. This is a cached copy of PL_regmatch_state */
eb578fdb 3943 regmatch_state *st;
5d9a96ca 3944 /* cache heavy used fields of st in registers */
eb578fdb
KW		 3945 regnode *scan;
3946 regnode *next;
3947 U32 n = 0; /* general value; init to avoid compiler warning */
ea3daa5d 3948 SSize_t ln = 0; /* len or last; init to avoid compiler warning */
d60de1d1 3949 char *locinput = startpos;
4d5016e5 3950 char *pushinput; /* where to continue after a PUSH */
eb578fdb 3951 I32 nextchr; /* is always set to UCHARAT(locinput) */
24d3c4a9 3952
b69b0499 3953 bool result = 0; /* return value of S_regmatch */
24d3c4a9 3954 int depth = 0; /* depth of backtrack stack */
4b196cd4
YO		 3955 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3956 const U32 max_nochange_depth =
3957 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3958 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
77cb431f
DM		 3959 regmatch_state *yes_state = NULL; /* state to pop to on success of
3960 subpattern */
e2e6a0f1
YO		 3961 /* mark_state piggy backs on the yes_state logic so that when we unwind
3962 the stack on success we can update the mark_state as we go */
3963 regmatch_state *mark_state = NULL; /* last mark state we have seen */
faec1544 3964 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
b8591aee 3965 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
40a82448 3966 U32 state_num;
5d458dd8
YO		 3967 bool no_final = 0; /* prevent failure from backtracking? */
3968 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
d60de1d1 3969 char *startpoint = locinput;
5d458dd8
YO		 3970 SV *popmark = NULL; /* are we looking for a mark? */
3971 SV *sv_commit = NULL; /* last mark name seen in failure */
3972 SV *sv_yes_mark = NULL; /* last mark name we have seen
486ec47a 3973 during a successful match */
5d458dd8
YO		 3974 U32 lastopen = 0; /* last open we saw */
3975 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
4a2b275c 3976 SV* const oreplsv = GvSVn(PL_replgv);
24d3c4a9
DM		 3977 /* these three flags are set by various ops to signal information to
3978 * the very next op. They have a useful lifetime of exactly one loop
3979 * iteration, and are not preserved or restored by state pushes/pops
3980 */
3981 bool sw = 0; /* the condition value in (?(cond)a|b) */
3982 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3983 int logical = 0; /* the following EVAL is:
3984 0: (?{...})
3985 1: (?(?{...})X|Y)
3986 2: (??{...})
3987 or the following IFMATCH/UNLESSM is:
3988 false: plain (?=foo)
3989 true: used as a condition: (?(?=foo))
3990 */
81ed78b2
DM		 3991 PAD* last_pad = NULL;
3992 dMULTICALL;
3993 I32 gimme = G_SCALAR;
3994 CV *caller_cv = NULL; /* who called us */
3995 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
74088413 3996 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
92da3157 3997 U32 maxopenparen = 0; /* max '(' index seen so far */
3018b823
KW		 3998 int to_complement; /* Invert the result? */
3999 _char_class_number classnum;
984e6dd1 4000 bool is_utf8_pat = reginfo->is_utf8_pat;
81ed78b2 4001
95b24440 4002#ifdef DEBUGGING
e68ec53f 4003 GET_RE_DEBUG_FLAGS_DECL;
d6a28714
JH		 4004#endif
4005
7e68f152
FC		 4006 /* protect against undef(*^R) */
4007 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv));
4008
81ed78b2
DM		 4009 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
4010 multicall_oldcatch = 0;
4011 multicall_cv = NULL;
4012 cx = NULL;
4f8dbb2d
JL		 4013 PERL_UNUSED_VAR(multicall_cop);
4014 PERL_UNUSED_VAR(newsp);
81ed78b2
DM		 4015
4016
7918f24d
NC		 4017 PERL_ARGS_ASSERT_REGMATCH;
4018
3b57cd43 4019 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
24b23f37 4020 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3b57cd43 4021 }));
5d9a96ca 4022
331b2dcc 4023 st = PL_regmatch_state;
5d9a96ca 4024
d6a28714 4025 /* Note that nextchr is a byte even in UTF */
7016d6eb 4026 SET_nextchr;
d6a28714
JH		 4027 scan = prog;
4028 while (scan != NULL) {
8ba1375e 4029
a3621e74 4030 DEBUG_EXECUTE_r( {
6136c704 4031 SV * const prop = sv_newmortal();
1de06328 4032 regnode *rnext=regnext(scan);
f2ed9b32 4033 DUMP_EXEC_POS( locinput, scan, utf8_target );
8b9781c9 4034 regprop(rex, prop, scan, reginfo, NULL);
07be1b83
YO		 4035
4036 PerlIO_printf(Perl_debug_log,
4037 "%3"IVdf":%*s%s(%"IVdf")\n",
f8fc2ecf 4038 (IV)(scan - rexi->program), depth*2, "",
07be1b83 4039 SvPVX_const(prop),
1de06328 4040 (PL_regkind[OP(scan)] == END || !rnext) ?
f8fc2ecf 4041 0 : (IV)(rnext - rexi->program));
2a782b5b 4042 });
d6a28714
JH		 4043
4044 next = scan + NEXT_OFF(scan);
4045 if (next == scan)
4046 next = NULL;
40a82448 4047 state_num = OP(scan);
d6a28714 4048
40a82448 4049 reenter_switch:
3018b823 4050 to_complement = 0;
34a81e2b 4051
7016d6eb 4052 SET_nextchr;
e6ca698c 4053 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
bf798dc4 4054
40a82448 4055 switch (state_num) {
d3d47aac 4056 case SBOL: /* /^../ and /\A../ */
9d9163fb 4057 if (locinput == reginfo->strbeg)
b8c5462f 4058 break;
d6a28714 4059 sayNO;
3c0563b9
DM		 4060
4061 case MBOL: /* /^../m */
9d9163fb 4062 if (locinput == reginfo->strbeg ||
7016d6eb 4063 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
d6a28714 4064 {
b8c5462f
JH		 4065 break;
4066 }
d6a28714 4067 sayNO;
3c0563b9 4068
3c0563b9 4069 case GPOS: /* \G */
3b0527fe 4070 if (locinput == reginfo->ganch)
d6a28714
JH		 4071 break;
4072 sayNO;
ee9b8eae 4073
3c0563b9 4074 case KEEPS: /* \K */
ee9b8eae 4075 /* update the startpoint */
b93070ed 4076 st->u.keeper.val = rex->offs[0].start;
9d9163fb 4077 rex->offs[0].start = locinput - reginfo->strbeg;
4d5016e5 4078 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
a74ff37d 4079 /* NOTREACHED */
e5964223 4080 NOT_REACHED;
a74ff37d 4081
ee9b8eae
YO		 4082 case KEEPS_next_fail:
4083 /* rollback the start point change */
b93070ed 4084 rex->offs[0].start = st->u.keeper.val;
ee9b8eae 4085 sayNO_SILENT;
a74ff37d 4086 /* NOTREACHED */
e5964223 4087 NOT_REACHED;
3c0563b9 4088
3c0563b9 4089 case MEOL: /* /..$/m */
7016d6eb 4090 if (!NEXTCHR_IS_EOS && nextchr != '\n')
b8c5462f 4091 sayNO;
b8c5462f 4092 break;
3c0563b9 4093
d3d47aac 4094 case SEOL: /* /..$/ */
7016d6eb 4095 if (!NEXTCHR_IS_EOS && nextchr != '\n')
b8c5462f 4096 sayNO;
220db18a 4097 if (reginfo->strend - locinput > 1)
b8c5462f 4098 sayNO;
b8c5462f 4099 break;
3c0563b9
DM		 4100
4101 case EOS: /* \z */
7016d6eb 4102 if (!NEXTCHR_IS_EOS)
b8c5462f 4103 sayNO;
d6a28714 4104 break;
3c0563b9
DM		 4105
4106 case SANY: /* /./s */
7016d6eb 4107 if (NEXTCHR_IS_EOS)
4633a7c4 4108 sayNO;
28b98f76 4109 goto increment_locinput;
3c0563b9
DM		 4110
4111 case CANY: /* \C */
7016d6eb 4112 if (NEXTCHR_IS_EOS)
f33976b4 4113 sayNO;
3640db6b 4114 locinput++;
a0d0e21e 4115 break;
3c0563b9
DM		 4116
4117 case REG_ANY: /* /./ */
7016d6eb 4118 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
1aa99e6b 4119 sayNO;
28b98f76
DM		 4120 goto increment_locinput;
4121
166ba7cd
DM		 4122
4123#undef ST
4124#define ST st->u.trie
3c0563b9 4125 case TRIEC: /* (ab|cd) with known charclass */
786e8c11
YO		 4126 /* In this case the charclass data is available inline so
4127 we can fail fast without a lot of extra overhead.
4128 */
7016d6eb 4129 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
fab2782b
YO		 4130 DEBUG_EXECUTE_r(
4131 PerlIO_printf(Perl_debug_log,
4132 "%*s %sfailed to match trie start class...%s\n",
4133 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4134 );
4135 sayNO_SILENT;
a74ff37d 4136 /* NOTREACHED */
e5964223 4137 NOT_REACHED;
786e8c11 4138 }
924ba076 4139 /* FALLTHROUGH */
3c0563b9 4140 case TRIE: /* (ab|cd) */
2e64971a
DM		 4141 /* the basic plan of execution of the trie is:
4142 * At the beginning, run though all the states, and
4143 * find the longest-matching word. Also remember the position
4144 * of the shortest matching word. For example, this pattern:
4145 * 1 2 3 4 5
4146 * ab|a|x|abcd|abc
4147 * when matched against the string "abcde", will generate
4148 * accept states for all words except 3, with the longest
895cc420 4149 * matching word being 4, and the shortest being 2 (with
2e64971a
DM		 4150 * the position being after char 1 of the string).
4151 *
4152 * Then for each matching word, in word order (i.e. 1,2,4,5),
4153 * we run the remainder of the pattern; on each try setting
4154 * the current position to the character following the word,
4155 * returning to try the next word on failure.
4156 *
4157 * We avoid having to build a list of words at runtime by
4158 * using a compile-time structure, wordinfo[].prev, which
4159 * gives, for each word, the previous accepting word (if any).
4160 * In the case above it would contain the mappings 1->2, 2->0,
4161 * 3->0, 4->5, 5->1. We can use this table to generate, from
4162 * the longest word (4 above), a list of all words, by
4163 * following the list of prev pointers; this gives us the
4164 * unordered list 4,5,1,2. Then given the current word we have
4165 * just tried, we can go through the list and find the
4166 * next-biggest word to try (so if we just failed on word 2,
4167 * the next in the list is 4).
4168 *
4169 * Since at runtime we don't record the matching position in
4170 * the string for each word, we have to work that out for
4171 * each word we're about to process. The wordinfo table holds
4172 * the character length of each word; given that we recorded
4173 * at the start: the position of the shortest word and its
4174 * length in chars, we just need to move the pointer the
4175 * difference between the two char lengths. Depending on
4176 * Unicode status and folding, that's cheap or expensive.
4177 *
4178 * This algorithm is optimised for the case where are only a
4179 * small number of accept states, i.e. 0,1, or maybe 2.
4180 * With lots of accepts states, and having to try all of them,
4181 * it becomes quadratic on number of accept states to find all
4182 * the next words.
4183 */
4184
3dab1dad 4185 {
07be1b83 4186 /* what type of TRIE am I? (utf8 makes this contextual) */
a0a388a1 4187 DECL_TRIE_TYPE(scan);
3dab1dad
YO		 4188
4189 /* what trie are we using right now */
be8e71aa 4190 reg_trie_data * const trie
f8fc2ecf 4191 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
85fbaab2 4192 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
3dab1dad 4193 U32 state = trie->startstate;
166ba7cd 4194
780fcc9f
KW		 4195 if (scan->flags == EXACTL || scan->flags == EXACTFLU8) {
4196 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
613abc6d
KW		 4197 if (utf8_target
4198 && UTF8_IS_ABOVE_LATIN1(nextchr)
4199 && scan->flags == EXACTL)
4200 {
4201 /* We only output for EXACTL, as we let the folder
4202 * output this message for EXACTFLU8 to avoid
4203 * duplication */
4204 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput,
4205 reginfo->strend);
4206 }
780fcc9f 4207 }
7016d6eb
DM		 4208 if ( trie->bitmap
4209 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
4210 {
3dab1dad
YO		 4211 if (trie->states[ state ].wordnum) {
4212 DEBUG_EXECUTE_r(
4213 PerlIO_printf(Perl_debug_log,
4214 "%*s %smatched empty string...%s\n",
5bc10b2c 4215 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3dab1dad 4216 );
20dbff7c
YO		 4217 if (!trie->jump)
4218 break;
3dab1dad
YO		 4219 } else {
4220 DEBUG_EXECUTE_r(
4221 PerlIO_printf(Perl_debug_log,
786e8c11 4222 "%*s %sfailed to match trie start class...%s\n",
5bc10b2c 4223 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
3dab1dad
YO		 4224 );
4225 sayNO_SILENT;
4226 }
4227 }
166ba7cd 4228
786e8c11
YO		 4229 {
4230 U8 *uc = ( U8* )locinput;
4231
4232 STRLEN len = 0;
4233 STRLEN foldlen = 0;
4234 U8 *uscan = (U8*)NULL;
786e8c11 4235 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2e64971a
DM		 4236 U32 charcount = 0; /* how many input chars we have matched */
4237 U32 accepted = 0; /* have we seen any accepting states? */
786e8c11 4238
786e8c11 4239 ST.jump = trie->jump;
786e8c11 4240 ST.me = scan;
2e64971a
DM		 4241 ST.firstpos = NULL;
4242 ST.longfold = FALSE; /* char longer if folded => it's harder */
4243 ST.nextword = 0;
4244
4245 /* fully traverse the TRIE; note the position of the
4246 shortest accept state and the wordnum of the longest
4247 accept state */
07be1b83 4248
220db18a 4249 while ( state && uc <= (U8*)(reginfo->strend) ) {
786e8c11 4250 U32 base = trie->states[ state ].trans.base;
f9f4320a 4251 UV uvc = 0;
acb909b4 4252 U16 charid = 0;
2e64971a
DM		 4253 U16 wordnum;
4254 wordnum = trie->states[ state ].wordnum;
4255
4256 if (wordnum) { /* it's an accept state */
4257 if (!accepted) {
4258 accepted = 1;
4259 /* record first match position */
4260 if (ST.longfold) {
4261 ST.firstpos = (U8*)locinput;
4262 ST.firstchars = 0;
5b47454d 4263 }
2e64971a
DM		 4264 else {
4265 ST.firstpos = uc;
4266 ST.firstchars = charcount;
4267 }
4268 }
4269 if (!ST.nextword || wordnum < ST.nextword)
4270 ST.nextword = wordnum;
4271 ST.topword = wordnum;
786e8c11 4272 }
a3621e74 4273
07be1b83 4274 DEBUG_TRIE_EXECUTE_r({
f2ed9b32 4275 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
a3621e74 4276 PerlIO_printf( Perl_debug_log,
2e64971a 4277 "%*s %sState: %4"UVxf" Accepted: %c ",
5bc10b2c 4278 2+depth * 2, "", PL_colors[4],
2e64971a 4279 (UV)state, (accepted ? 'Y' : 'N'));
07be1b83 4280 });
a3621e74 4281
2e64971a 4282 /* read a char and goto next state */
220db18a 4283 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
6dd2be57 4284 I32 offset;
55eed653
NC		 4285 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
4286 uscan, len, uvc, charid, foldlen,
4287 foldbuf, uniflags);
2e64971a
DM		 4288 charcount++;
4289 if (foldlen>0)
4290 ST.longfold = TRUE;
5b47454d 4291 if (charid &&
6dd2be57
DM		 4292 ( ((offset =
4293 base + charid - 1 - trie->uniquecharcount)) >= 0)
4294
4295 && ((U32)offset < trie->lasttrans)
4296 && trie->trans[offset].check == state)
5b47454d 4297 {
6dd2be57 4298 state = trie->trans[offset].next;
5b47454d
DM		 4299 }
4300 else {
4301 state = 0;
4302 }
4303 uc += len;
4304
4305 }
4306 else {
a3621e74
YO		 4307 state = 0;
4308 }
4309 DEBUG_TRIE_EXECUTE_r(
e4584336 4310 PerlIO_printf( Perl_debug_log,
786e8c11 4311 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
e4584336 4312 charid, uvc, (UV)state, PL_colors[5] );
a3621e74
YO		 4313 );
4314 }
2e64971a 4315 if (!accepted)
a3621e74 4316 sayNO;
a3621e74 4317
2e64971a
DM		 4318 /* calculate total number of accept states */
4319 {
4320 U16 w = ST.topword;
4321 accepted = 0;
4322 while (w) {
4323 w = trie->wordinfo[w].prev;
4324 accepted++;
4325 }
4326 ST.accepted = accepted;
4327 }
4328
166ba7cd
DM		 4329 DEBUG_EXECUTE_r(
4330 PerlIO_printf( Perl_debug_log,
4331 "%*s %sgot %"IVdf" possible matches%s\n",
5bc10b2c 4332 REPORT_CODE_OFF + depth * 2, "",
166ba7cd
DM		 4333 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
4334 );
2e64971a 4335 goto trie_first_try; /* jump into the fail handler */
786e8c11 4336 }}
a74ff37d 4337 /* NOTREACHED */
e5964223 4338 NOT_REACHED;
2e64971a
DM		 4339
4340 case TRIE_next_fail: /* we failed - try next alternative */
a059a757
DM		 4341 {
4342 U8 *uc;
fae667d5
YO		 4343 if ( ST.jump) {
4344 REGCP_UNWIND(ST.cp);
a8d1f4b4 4345 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
fae667d5 4346 }
2e64971a
DM		 4347 if (!--ST.accepted) {
4348 DEBUG_EXECUTE_r({
4349 PerlIO_printf( Perl_debug_log,
4350 "%*s %sTRIE failed...%s\n",
4351 REPORT_CODE_OFF+depth*2, "",
4352 PL_colors[4],
4353 PL_colors[5] );
4354 });
4355 sayNO_SILENT;
4356 }
4357 {
4358 /* Find next-highest word to process. Note that this code
4359 * is O(N^2) per trie run (O(N) per branch), so keep tight */
eb578fdb
KW		 4360 U16 min = 0;
4361 U16 word;
4362 U16 const nextword = ST.nextword;
4363 reg_trie_wordinfo * const wordinfo
2e64971a
DM		 4364 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
4365 for (word=ST.topword; word; word=wordinfo[word].prev) {
4366 if (word > nextword && (!min || word < min))
4367 min = word;
4368 }
4369 ST.nextword = min;
4370 }
4371
fae667d5 4372 trie_first_try:
5d458dd8
YO		 4373 if (do_cutgroup) {
4374 do_cutgroup = 0;
4375 no_final = 0;
4376 }
fae667d5
YO		 4377
4378 if ( ST.jump) {
b93070ed 4379 ST.lastparen = rex->lastparen;
f6033a9d 4380 ST.lastcloseparen = rex->lastcloseparen;
fae667d5 4381 REGCP_SET(ST.cp);
2e64971a 4382 }
a3621e74 4383
2e64971a 4384 /* find start char of end of current word */
166ba7cd 4385 {
2e64971a 4386 U32 chars; /* how many chars to skip */
2e64971a
DM		 4387 reg_trie_data * const trie
4388 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
4389
4390 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
4391 >= ST.firstchars);
4392 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
4393 - ST.firstchars;
a059a757 4394 uc = ST.firstpos;
2e64971a
DM		 4395
4396 if (ST.longfold) {
4397 /* the hard option - fold each char in turn and find
4398 * its folded length (which may be different */
4399 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
4400 STRLEN foldlen;
4401 STRLEN len;
d9a396a3 4402 UV uvc;
2e64971a
DM		 4403 U8 *uscan;
4404
4405 while (chars) {
f2ed9b32 4406 if (utf8_target) {
c80e42f3 4407 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
2e64971a
DM		 4408 uniflags);
4409 uc += len;
4410 }
4411 else {
4412 uvc = *uc;
4413 uc++;
4414 }
4415 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
4416 uscan = foldbuf;
4417 while (foldlen) {
4418 if (!--chars)
4419 break;
c80e42f3 4420 uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len,
2e64971a
DM		 4421 uniflags);
4422 uscan += len;
4423 foldlen -= len;
4424 }
4425 }
a3621e74 4426 }
2e64971a 4427 else {
f2ed9b32 4428 if (utf8_target)
2e64971a
DM		 4429 while (chars--)
4430 uc += UTF8SKIP(uc);
4431 else
4432 uc += chars;
4433 }
2e64971a 4434 }
166ba7cd 4435
6603fe3e
DM		 4436 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
4437 ? ST.jump[ST.nextword]
4438 : NEXT_OFF(ST.me));
166ba7cd 4439
2e64971a
DM		 4440 DEBUG_EXECUTE_r({
4441 PerlIO_printf( Perl_debug_log,
4442 "%*s %sTRIE matched word #%d, continuing%s\n",
4443 REPORT_CODE_OFF+depth*2, "",
4444 PL_colors[4],
4445 ST.nextword,
4446 PL_colors[5]
4447 );
4448 });
4449
4450 if (ST.accepted > 1 || has_cutgroup) {
a059a757 4451 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
a74ff37d 4452 /* NOTREACHED */
e5964223 4453 NOT_REACHED;
166ba7cd 4454 }
2e64971a
DM		 4455 /* only one choice left - just continue */
4456 DEBUG_EXECUTE_r({
4457 AV *const trie_words
4458 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
4459 SV ** const tmp = av_fetch( trie_words,
4460 ST.nextword-1, 0 );
4461 SV *sv= tmp ? sv_newmortal() : NULL;
4462
4463 PerlIO_printf( Perl_debug_log,
4464 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
4465 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
4466 ST.nextword,
4467 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
4468 PL_colors[0], PL_colors[1],
c89df6cf 4469 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
2e64971a
DM		 4470 )
4471 : "not compiled under -Dr",
4472 PL_colors[5] );
4473 });
4474
a059a757 4475 locinput = (char*)uc;
2e64971a 4476 continue; /* execute rest of RE */
a74ff37d 4477 /* NOTREACHED */
a059a757 4478 }
166ba7cd
DM		 4479#undef ST
4480
a4525e78 4481 case EXACTL: /* /abc/l */
780fcc9f 4482 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
613abc6d
KW		 4483
4484 /* Complete checking would involve going through every character
4485 * matched by the string to see if any is above latin1. But the
4486 * comparision otherwise might very well be a fast assembly
4487 * language routine, and I (khw) don't think slowing things down
4488 * just to check for this warning is worth it. So this just checks
4489 * the first character */
4490 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*locinput)) {
4491 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend);
4492 }
780fcc9f 4493 /* FALLTHROUGH */
3c0563b9 4494 case EXACT: { /* /abc/ */
95b24440 4495 char *s = STRING(scan);
24d3c4a9 4496 ln = STR_LEN(scan);
984e6dd1 4497 if (utf8_target != is_utf8_pat) {
bc517b45 4498 /* The target and the pattern have differing utf8ness. */
1aa99e6b 4499 char *l = locinput;
24d3c4a9 4500 const char * const e = s + ln;
a72c7584 4501
f2ed9b32 4502 if (utf8_target) {
e6a3850e
KW		 4503 /* The target is utf8, the pattern is not utf8.
4504 * Above-Latin1 code points can't match the pattern;
4505 * invariants match exactly, and the other Latin1 ones need
4506 * to be downgraded to a single byte in order to do the
4507 * comparison. (If we could be confident that the target
4508 * is not malformed, this could be refactored to have fewer
4509 * tests by just assuming that if the first bytes match, it
4510 * is an invariant, but there are tests in the test suite
4511 * dealing with (??{...}) which violate this) */
1aa99e6b 4512 while (s < e) {
220db18a
DM		 4513 if (l >= reginfo->strend
4514 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
4515 {
e6a3850e
KW		 4516 sayNO;
4517 }
4518 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4519 if (*l != *s) {
4520 sayNO;
4521 }
4522 l++;
4523 }
4524 else {
94bb8c36
KW		 4525 if (TWO_BYTE_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
4526 {
e6a3850e
KW		 4527 sayNO;
4528 }
4529 l += 2;
4530 }
4531 s++;
1aa99e6b 4532 }
5ff6fc6d
JH		 4533 }
4534 else {
4535 /* The target is not utf8, the pattern is utf8. */
1aa99e6b 4536 while (s < e) {
220db18a
DM		 4537 if (l >= reginfo->strend
4538 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
e6a3850e
KW		 4539 {
4540 sayNO;
4541 }
4542 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4543 if (*s != *l) {
4544 sayNO;
4545 }
4546 s++;
4547 }
4548 else {
94bb8c36
KW		 4549 if (TWO_BYTE_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
4550 {
e6a3850e
KW		 4551 sayNO;
4552 }
4553 s += 2;
4554 }
4555 l++;
1aa99e6b 4556 }
5ff6fc6d 4557 }
1aa99e6b 4558 locinput = l;
1aa99e6b 4559 }
5ac65bff
KW		 4560 else {
4561 /* The target and the pattern have the same utf8ness. */
4562 /* Inline the first character, for speed. */
220db18a 4563 if (reginfo->strend - locinput < ln
5ac65bff
KW		 4564 || UCHARAT(s) != nextchr
4565 || (ln > 1 && memNE(s, locinput, ln)))
4566 {
4567 sayNO;
4568 }
4569 locinput += ln;
4570 }
d6a28714 4571 break;
95b24440 4572 }
7016d6eb 4573
3c0563b9 4574 case EXACTFL: { /* /abc/il */
a932d541 4575 re_fold_t folder;
9a5a5549
KW		 4576 const U8 * fold_array;
4577 const char * s;
d513472c 4578 U32 fold_utf8_flags;
9a5a5549 4579
780fcc9f 4580 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
f67f9e53
KW		 4581 folder = foldEQ_locale;
4582 fold_array = PL_fold_locale;
cea315b6 4583 fold_utf8_flags = FOLDEQ_LOCALE;
9a5a5549
KW		 4584 goto do_exactf;
4585
a4525e78
KW		 4586 case EXACTFLU8: /* /abc/il; but all 'abc' are above 255, so
4587 is effectively /u; hence to match, target
4588 must be UTF-8. */
4589 if (! utf8_target) {
4590 sayNO;
4591 }
613abc6d
KW		 4592 fold_utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S1_ALREADY_FOLDED
4593 | FOLDEQ_S1_FOLDS_SANE;
ec5e0e1c
KW		 4594 folder = foldEQ_latin1;
4595 fold_array = PL_fold_latin1;
a4525e78
KW		 4596 goto do_exactf;
4597
3c0563b9 4598 case EXACTFU_SS: /* /\x{df}/iu */
3c0563b9 4599 case EXACTFU: /* /abc/iu */
9a5a5549
KW		 4600 folder = foldEQ_latin1;
4601 fold_array = PL_fold_latin1;
984e6dd1 4602 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
9a5a5549
KW		 4603 goto do_exactf;
4604
098b07d5
KW		 4605 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8
4606 patterns */
4607 assert(! is_utf8_pat);
924ba076 4608 /* FALLTHROUGH */
3c0563b9 4609 case EXACTFA: /* /abc/iaa */
2f7f8cb1
KW		 4610 folder = foldEQ_latin1;
4611 fold_array = PL_fold_latin1;
57014d77 4612 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
2f7f8cb1
KW		 4613 goto do_exactf;
4614
2fdb7295
KW		 4615 case EXACTF: /* /abc/i This node only generated for
4616 non-utf8 patterns */
4617 assert(! is_utf8_pat);
9a5a5549
KW		 4618 folder = foldEQ;
4619 fold_array = PL_fold;
62bf7766 4620 fold_utf8_flags = 0;
9a5a5549
KW		 4621
4622 do_exactf:
4623 s = STRING(scan);
24d3c4a9 4624 ln = STR_LEN(scan);
d6a28714 4625
31f05a37
KW		 4626 if (utf8_target
4627 || is_utf8_pat
4628 || state_num == EXACTFU_SS
4629 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
4630 {
3c760661
KW		 4631 /* Either target or the pattern are utf8, or has the issue where
4632 * the fold lengths may differ. */
be8e71aa 4633 const char * const l = locinput;
220db18a 4634 char *e = reginfo->strend;
bc517b45 4635
984e6dd1 4636 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
fa5b1667 4637 l, &e, 0, utf8_target, fold_utf8_flags))
c3e1d013
KW		 4638 {
4639 sayNO;
5486206c 4640 }
d07ddd77 4641 locinput = e;
d07ddd77 4642 break;
a0ed51b3 4643 }
d6a28714 4644
0a138b74 4645 /* Neither the target nor the pattern are utf8 */
1443c94c
DM		 4646 if (UCHARAT(s) != nextchr
4647 && !NEXTCHR_IS_EOS
4648 && UCHARAT(s) != fold_array[nextchr])
9a5a5549 4649 {
a0ed51b3 4650 sayNO;
9a5a5549 4651 }
220db18a 4652 if (reginfo->strend - locinput < ln)
b8c5462f 4653 sayNO;
9a5a5549 4654 if (ln > 1 && ! folder(s, locinput, ln))
4633a7c4 4655 sayNO;
24d3c4a9 4656 locinput += ln;
a0d0e21e 4657 break;
9a5a5549 4658 }
63ac0dad 4659
5c388b33 4660 /* XXX At that point regcomp.c would no longer * have to set the FLAGS fields of these */
3c0563b9 4661 case NBOUNDL: /* /\B/l */
5c388b33
KW		 4662 to_complement = 1;
4663 /* FALLTHROUGH */
4664
4665 case BOUNDL: /* /\b/l */
780fcc9f 4666 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5c388b33
KW		 4667 if (utf8_target) {
4668 if (locinput == reginfo->strbeg)
4669 ln = isWORDCHAR_LC('\n');
4670 else {
4671 ln = isWORDCHAR_LC_utf8(reghop3((U8*)locinput, -1,
4672 (U8*)(reginfo->strbeg)));
4673 }
4674 n = (NEXTCHR_IS_EOS)
4675 ? isWORDCHAR_LC('\n')
4676 : isWORDCHAR_LC_utf8((U8*)locinput);
4677 }
4678 else { /* Here the string isn't utf8 */
4679 ln = (locinput == reginfo->strbeg)
4680 ? isWORDCHAR_LC('\n')
4681 : isWORDCHAR_LC(UCHARAT(locinput - 1));
4682 n = (NEXTCHR_IS_EOS)
4683 ? isWORDCHAR_LC('\n')
4684 : isWORDCHAR_LC(nextchr);
4685 }
4686 if (to_complement ^ (ln == n)) {
4687 sayNO;
4688 }
4689 break;
4690
4691 case NBOUND: /* /\B/ */
4692 to_complement = 1;
780fcc9f 4693 /* FALLTHROUGH */
5c388b33 4694
3c0563b9 4695 case BOUND: /* /\b/ */
5c388b33
KW		 4696 if (utf8_target) {
4697 goto bound_utf8;
4698 }
4699 goto bound_ascii_match_only;
4700
4701 case NBOUNDA: /* /\B/a */
4702 to_complement = 1;
4703 /* FALLTHROUGH */
4704
3c0563b9 4705 case BOUNDA: /* /\b/a */
5c388b33 4706
c52b8b12 4707 bound_ascii_match_only:
5c388b33
KW		 4708 /* Here the string isn't utf8, or is utf8 and only ascii characters
4709 * are to match \w. In the latter case looking at the byte just
4710 * prior to the current one may be just the final byte of a
4711 * multi-byte character. This is ok. There are two cases:
4712 * 1) it is a single byte character, and then the test is doing
4713 * just what it's supposed to.
4714 * 2) it is a multi-byte character, in which case the final byte is
4715 * never mistakable for ASCII, and so the test will say it is
4716 * not a word character, which is the correct answer. */
4717 ln = (locinput == reginfo->strbeg)
4718 ? isWORDCHAR_A('\n')
4719 : isWORDCHAR_A(UCHARAT(locinput - 1));
4720 n = (NEXTCHR_IS_EOS)
4721 ? isWORDCHAR_A('\n')
4722 : isWORDCHAR_A(nextchr);
4723 if (to_complement ^ (ln == n)) {
4724 sayNO;
4725 }
4726 break;
4727
3c0563b9 4728 case NBOUNDU: /* /\B/u */
5c388b33
KW		 4729 to_complement = 1;
4730 /* FALLTHROUGH */
b2680017 4731
5c388b33
KW		 4732 case BOUNDU: /* /\b/u */
4733 if (utf8_target) {
4734
4735 bound_utf8:
4736 ln = (locinput == reginfo->strbeg)
4737 ? isWORDCHAR_L1('\n')
4738 : isWORDCHAR_utf8(reghop3((U8*)locinput, -1,
4739 (U8*)(reginfo->strbeg)));
4740 n = (NEXTCHR_IS_EOS)
4741 ? isWORDCHAR_L1('\n')
4742 : isWORDCHAR_utf8((U8*)locinput);
b2680017
YO		 4743 }
4744 else {
5c388b33
KW		 4745 ln = (locinput == reginfo->strbeg)
4746 ? isWORDCHAR_L1('\n')
4747 : isWORDCHAR_L1(UCHARAT(locinput - 1));
4748 n = (NEXTCHR_IS_EOS)
4749 ? isWORDCHAR_L1('\n')
4750 : isWORDCHAR_L1(nextchr);
cfaf538b 4751
b2680017 4752 }
5c388b33
KW		 4753
4754 if (to_complement ^ (ln == n)) {
4755 sayNO;
4756 }
b2680017 4757 break;
3c0563b9 4758
a4525e78 4759 case ANYOFL: /* /[abc]/l */
780fcc9f
KW		 4760 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4761 /* FALLTHROUGH */
a4525e78 4762 case ANYOF: /* /[abc]/ */
7016d6eb
DM		 4763 if (NEXTCHR_IS_EOS)
4764 sayNO;
e0193e47 4765 if (utf8_target) {
3db24e1e
KW		 4766 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend,
4767 utf8_target))
09b08e9b 4768 sayNO;
635cd5d4 4769 locinput += UTF8SKIP(locinput);
ffc61ed2
JH		 4770 }
4771 else {
20ed0b26 4772 if (!REGINCLASS(rex, scan, (U8*)locinput))
09b08e9b 4773 sayNO;
3640db6b 4774 locinput++;
e0f9d4a8 4775 }
b8c5462f 4776 break;
3c0563b9 4777
3018b823
KW		 4778 /* The argument (FLAGS) to all the POSIX node types is the class number
4779 * */
ee9a90b8 4780
3018b823
KW		 4781 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
4782 to_complement = 1;
4783 /* FALLTHROUGH */
4784
4785 case POSIXL: /* \w or [:punct:] etc. under /l */
780fcc9f 4786 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3018b823 4787 if (NEXTCHR_IS_EOS)
bedac28b 4788 sayNO;
bedac28b 4789
3018b823
KW		 4790 /* Use isFOO_lc() for characters within Latin1. (Note that
4791 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4792 * wouldn't be invariant) */
4793 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
eb4e9c04 4794 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
bedac28b
KW		 4795 sayNO;
4796 }
4797 }
3018b823
KW		 4798 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4799 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
94bb8c36 4800 (U8) TWO_BYTE_UTF8_TO_NATIVE(nextchr,
3018b823
KW		 4801 *(locinput + 1))))))
4802 {
bedac28b 4803 sayNO;
3018b823 4804 }
bedac28b 4805 }
3018b823 4806 else { /* Here, must be an above Latin-1 code point */
613abc6d 4807 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend);
01f55654 4808 goto utf8_posix_above_latin1;
bedac28b 4809 }
3018b823
KW		 4810
4811 /* Here, must be utf8 */
4812 locinput += UTF8SKIP(locinput);
bedac28b
KW		 4813 break;
4814
3018b823
KW		 4815 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
4816 to_complement = 1;
4817 /* FALLTHROUGH */
4818
4819 case POSIXD: /* \w or [:punct:] etc. under /d */
bedac28b 4820 if (utf8_target) {
3018b823 4821 goto utf8_posix;
bedac28b 4822 }
3018b823
KW		 4823 goto posixa;
4824
4825 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
bedac28b 4826
3018b823 4827 if (NEXTCHR_IS_EOS) {
bedac28b
KW		 4828 sayNO;
4829 }
bedac28b 4830
3018b823
KW		 4831 /* All UTF-8 variants match */
4832 if (! UTF8_IS_INVARIANT(nextchr)) {
4833 goto increment_locinput;
bedac28b 4834 }
ee9a90b8 4835
3018b823
KW		 4836 to_complement = 1;
4837 /* FALLTHROUGH */
4838
4839 case POSIXA: /* \w or [:punct:] etc. under /a */
4840
4841 posixa:
4842 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
4843 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
4844 * character is a single byte */
20d0b1e9 4845
3018b823
KW		 4846 if (NEXTCHR_IS_EOS
4847 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
4848 FLAGS(scan)))))
4849 {
0658cdde
KW		 4850 sayNO;
4851 }
3018b823
KW		 4852
4853 /* Here we are either not in utf8, or we matched a utf8-invariant,
4854 * so the next char is the next byte */
3640db6b 4855 locinput++;
0658cdde 4856 break;
3c0563b9 4857
3018b823
KW		 4858 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
4859 to_complement = 1;
4860 /* FALLTHROUGH */
4861
4862 case POSIXU: /* \w or [:punct:] etc. under /u */
4863 utf8_posix:
4864 if (NEXTCHR_IS_EOS) {
0658cdde
KW		 4865 sayNO;
4866 }
3018b823
KW		 4867
4868 /* Use _generic_isCC() for characters within Latin1. (Note that
4869 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4870 * wouldn't be invariant) */
4871 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4872 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
4873 FLAGS(scan)))))
4874 {
4875 sayNO;
4876 }
4877 locinput++;
4878 }
4879 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4880 if (! (to_complement
94bb8c36 4881 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(nextchr,
3018b823 4882 *(locinput + 1)),
94bb8c36 4883 FLAGS(scan)))))
3018b823
KW		 4884 {
4885 sayNO;
4886 }
4887 locinput += 2;
4888 }
4889 else { /* Handle above Latin-1 code points */
c52b8b12 4890 utf8_posix_above_latin1:
3018b823
KW		 4891 classnum = (_char_class_number) FLAGS(scan);
4892 if (classnum < _FIRST_NON_SWASH_CC) {
4893
4894 /* Here, uses a swash to find such code points. Load if if
4895 * not done already */
4896 if (! PL_utf8_swash_ptrs[classnum]) {
4897 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
4898 PL_utf8_swash_ptrs[classnum]
4899 = _core_swash_init("utf8",
2a16ac92
KW		 4900 "",
4901 &PL_sv_undef, 1, 0,
4902 PL_XPosix_ptrs[classnum], &flags);
3018b823
KW		 4903 }
4904 if (! (to_complement
4905 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
4906 (U8 *) locinput, TRUE))))
4907 {
4908 sayNO;
4909 }
4910 }
4911 else { /* Here, uses macros to find above Latin-1 code points */
4912 switch (classnum) {
4913 case _CC_ENUM_SPACE: /* XXX would require separate
4914 code if we revert the change
4915 of \v matching this */
4916 case _CC_ENUM_PSXSPC:
4917 if (! (to_complement
4918 ^ cBOOL(is_XPERLSPACE_high(locinput))))
4919 {
4920 sayNO;
4921 }
4922 break;
4923 case _CC_ENUM_BLANK:
4924 if (! (to_complement
4925 ^ cBOOL(is_HORIZWS_high(locinput))))
4926 {
4927 sayNO;
4928 }
4929 break;
4930 case _CC_ENUM_XDIGIT:
4931 if (! (to_complement
4932 ^ cBOOL(is_XDIGIT_high(locinput))))
4933 {
4934 sayNO;
4935 }
4936 break;
4937 case _CC_ENUM_VERTSPACE:
4938 if (! (to_complement
4939 ^ cBOOL(is_VERTWS_high(locinput))))
4940 {
4941 sayNO;
4942 }
4943 break;
4944 default: /* The rest, e.g. [:cntrl:], can't match
4945 above Latin1 */
4946 if (! to_complement) {
4947 sayNO;
4948 }
4949 break;
4950 }
4951 }
4952 locinput += UTF8SKIP(locinput);
4953 }
4954 break;
0658cdde 4955
37e2e78e
KW		 4956 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4957 a Unicode extended Grapheme Cluster */
4958 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4959 extended Grapheme Cluster is:
4960
7aee35ff
KW		 4961 CR LF
4962 | Prepend* Begin Extend*
4963 | .
37e2e78e 4964
7aee35ff
KW		 4965 Begin is: ( Special_Begin | ! Control )
4966 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4967 Extend is: ( Grapheme_Extend | Spacing_Mark )
4968 Control is: [ GCB_Control | CR | LF ]
4969 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
37e2e78e 4970
27d4fc33
KW		 4971 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4972 we can rewrite
4973
4974 Begin is ( Regular_Begin + Special Begin )
4975
4976 It turns out that 98.4% of all Unicode code points match
4977 Regular_Begin. Doing it this way eliminates a table match in
c101f46d 4978 the previous implementation for almost all Unicode code points.
27d4fc33 4979
37e2e78e
KW		 4980 There is a subtlety with Prepend* which showed up in testing.
4981 Note that the Begin, and only the Begin is required in:
4982 | Prepend* Begin Extend*
cc3b396d
KW		 4983 Also, Begin contains '! Control'. A Prepend must be a
4984 '! Control', which means it must also be a Begin. What it
4985 comes down to is that if we match Prepend* and then find no
4986 suitable Begin afterwards, that if we backtrack the last
4987 Prepend, that one will be a suitable Begin.
37e2e78e
KW		 4988 */
4989
7016d6eb 4990 if (NEXTCHR_IS_EOS)
a0ed51b3 4991 sayNO;
f2ed9b32 4992 if (! utf8_target) {
37e2e78e
KW		 4993
4994 /* Match either CR LF or '.', as all the other possibilities
4995 * require utf8 */
4996 locinput++; /* Match the . or CR */
cc3b396d
KW		 4997 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4998 match the LF */
220db18a 4999 && locinput < reginfo->strend
e699a1d5
KW		 5000 && UCHARAT(locinput) == '\n')
5001 {
5002 locinput++;
5003 }
37e2e78e
KW		 5004 }
5005 else {
5006
5007 /* Utf8: See if is ( CR LF ); already know that locinput <
220db18a
DM		 5008 * reginfo->strend, so locinput+1 is in bounds */
5009 if ( nextchr == '\r' && locinput+1 < reginfo->strend
f67f9e53 5010 && UCHARAT(locinput + 1) == '\n')
7016d6eb 5011 {
37e2e78e
KW		 5012 locinput += 2;
5013 }
5014 else {
45fdf108
KW		 5015 STRLEN len;
5016
37e2e78e
KW		 5017 /* In case have to backtrack to beginning, then match '.' */
5018 char *starting = locinput;
5019
5020 /* In case have to backtrack the last prepend */
e699a1d5 5021 char *previous_prepend = NULL;
37e2e78e
KW		 5022
5023 LOAD_UTF8_CHARCLASS_GCB();
5024
45fdf108 5025 /* Match (prepend)* */
220db18a 5026 while (locinput < reginfo->strend
45fdf108
KW		 5027 && (len = is_GCB_Prepend_utf8(locinput)))
5028 {
5029 previous_prepend = locinput;
5030 locinput += len;
a1853d78 5031 }
37e2e78e
KW		 5032
5033 /* As noted above, if we matched a prepend character, but
5034 * the next thing won't match, back off the last prepend we
5035 * matched, as it is guaranteed to match the begin */
5036 if (previous_prepend
220db18a 5037 && (locinput >= reginfo->strend
c101f46d
KW		 5038 || (! swash_fetch(PL_utf8_X_regular_begin,
5039 (U8*)locinput, utf8_target)
bff53399 5040 && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)))
c101f46d 5041 )
37e2e78e
KW		 5042 {
5043 locinput = previous_prepend;
5044 }
5045
220db18a 5046 /* Note that here we know reginfo->strend > locinput, as we
37e2e78e
KW		 5047 * tested that upon input to this switch case, and if we
5048 * moved locinput forward, we tested the result just above
5049 * and it either passed, or we backed off so that it will
5050 * now pass */
11dfcd49
KW		 5051 if (swash_fetch(PL_utf8_X_regular_begin,
5052 (U8*)locinput, utf8_target)) {
27d4fc33
KW		 5053 locinput += UTF8SKIP(locinput);
5054 }
bff53399 5055 else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) {
37e2e78e
KW		 5056
5057 /* Here did not match the required 'Begin' in the
5058 * second term. So just match the very first
5059 * character, the '.' of the final term of the regex */
5060 locinput = starting + UTF8SKIP(starting);
27d4fc33 5061 goto exit_utf8;
37e2e78e
KW		 5062 } else {
5063
11dfcd49
KW		 5064 /* Here is a special begin. It can be composed of
5065 * several individual characters. One possibility is
5066 * RI+ */
45fdf108
KW		 5067 if ((len = is_GCB_RI_utf8(locinput))) {
5068 locinput += len;
220db18a 5069 while (locinput < reginfo->strend
45fdf108 5070 && (len = is_GCB_RI_utf8(locinput)))
11dfcd49 5071 {
45fdf108 5072 locinput += len;
11dfcd49 5073 }
45fdf108
KW		 5074 } else if ((len = is_GCB_T_utf8(locinput))) {
5075 /* Another possibility is T+ */
5076 locinput += len;
220db18a 5077 while (locinput < reginfo->strend
45fdf108 5078 && (len = is_GCB_T_utf8(locinput)))
11dfcd49 5079 {
45fdf108 5080 locinput += len;
11dfcd49
KW		 5081 }
5082 } else {
5083
5084 /* Here, neither RI+ nor T+; must be some other
5085 * Hangul. That means it is one of the others: L,
5086 * LV, LVT or V, and matches:
5087 * L* (L | LVT T* | V * V* T* | LV V* T*) */
5088
5089 /* Match L* */
220db18a 5090 while (locinput < reginfo->strend
45fdf108 5091 && (len = is_GCB_L_utf8(locinput)))
11dfcd49 5092 {
45fdf108 5093 locinput += len;
11dfcd49 5094 }
37e2e78e 5095
11dfcd49
KW		 5096 /* Here, have exhausted L*. If the next character
5097 * is not an LV, LVT nor V, it means we had to have
5098 * at least one L, so matches L+ in the original
5099 * equation, we have a complete hangul syllable.
5100 * Are done. */
5101
220db18a 5102 if (locinput < reginfo->strend
45fdf108 5103 && is_GCB_LV_LVT_V_utf8(locinput))
11dfcd49 5104 {
11dfcd49 5105 /* Otherwise keep going. Must be LV, LVT or V.
7d43c479
KW		 5106 * See if LVT, by first ruling out V, then LV */
5107 if (! is_GCB_V_utf8(locinput)
5108 /* All but every TCount one is LV */
5109 && (valid_utf8_to_uvchr((U8 *) locinput,
5110 NULL)
5111 - SBASE)
5112 % TCount != 0)
5113 {
11dfcd49
KW		 5114 locinput += UTF8SKIP(locinput);
5115 } else {
5116
5117 /* Must be V or LV. Take it, then match
5118 * V* */
5119 locinput += UTF8SKIP(locinput);
220db18a 5120 while (locinput < reginfo->strend
45fdf108 5121 && (len = is_GCB_V_utf8(locinput)))
11dfcd49 5122 {
45fdf108 5123 locinput += len;
11dfcd49
KW		 5124 }
5125 }
37e2e78e 5126
11dfcd49 5127 /* And any of LV, LVT, or V can be followed
45fdf108 5128 * by T* */
220db18a 5129 while (locinput < reginfo->strend
45fdf108 5130 && (len = is_GCB_T_utf8(locinput)))
11dfcd49 5131 {
45fdf108 5132 locinput += len;
11dfcd49
KW		 5133 }
5134 }
cd94d768 5135 }
11dfcd49 5136 }
37e2e78e 5137
11dfcd49 5138 /* Match any extender */
220db18a 5139 while (locinput < reginfo->strend
11dfcd49
KW		 5140 && swash_fetch(PL_utf8_X_extend,
5141 (U8*)locinput, utf8_target))
5142 {
5143 locinput += UTF8SKIP(locinput);
5144 }
37e2e78e 5145 }
c52b8b12 5146 exit_utf8:
220db18a 5147 if (locinput > reginfo->strend) sayNO;
37e2e78e 5148 }
a0ed51b3 5149 break;
81714fb9 5150
3c0563b9 5151 case NREFFL: /* /\g{name}/il */
d7ef4b73
KW		 5152 { /* The capture buffer cases. The ones beginning with N for the
5153 named buffers just convert to the equivalent numbered and
5154 pretend they were called as the corresponding numbered buffer
5155 op. */
26ecd678
TC		 5156 /* don't initialize these in the declaration, it makes C++
5157 unhappy */
9d9163fb 5158 const char *s;
ff1157ca 5159 char type;
8368298a
TC		 5160 re_fold_t folder;
5161 const U8 *fold_array;
26ecd678 5162 UV utf8_fold_flags;
8368298a 5163
780fcc9f 5164 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
d7ef4b73
KW		 5165 folder = foldEQ_locale;
5166 fold_array = PL_fold_locale;
5167 type = REFFL;
cea315b6 5168 utf8_fold_flags = FOLDEQ_LOCALE;
d7ef4b73
KW		 5169 goto do_nref;
5170
3c0563b9 5171 case NREFFA: /* /\g{name}/iaa */
2f7f8cb1
KW		 5172 folder = foldEQ_latin1;
5173 fold_array = PL_fold_latin1;
5174 type = REFFA;
5175 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5176 goto do_nref;
5177
3c0563b9 5178 case NREFFU: /* /\g{name}/iu */
d7ef4b73
KW		 5179 folder = foldEQ_latin1;
5180 fold_array = PL_fold_latin1;
5181 type = REFFU;
d513472c 5182 utf8_fold_flags = 0;
d7ef4b73
KW		 5183 goto do_nref;
5184
3c0563b9 5185 case NREFF: /* /\g{name}/i */
d7ef4b73
KW		 5186 folder = foldEQ;
5187 fold_array = PL_fold;
5188 type = REFF;
d513472c 5189 utf8_fold_flags = 0;
d7ef4b73
KW		 5190 goto do_nref;
5191
3c0563b9 5192 case NREF: /* /\g{name}/ */
d7ef4b73 5193 type = REF;
83d7b90b
KW		 5194 folder = NULL;
5195 fold_array = NULL;
d513472c 5196 utf8_fold_flags = 0;
d7ef4b73
KW		 5197 do_nref:
5198
5199 /* For the named back references, find the corresponding buffer
5200 * number */
0a4db386
YO		 5201 n = reg_check_named_buff_matched(rex,scan);
5202
d7ef4b73 5203 if ( ! n ) {
81714fb9 5204 sayNO;
d7ef4b73
KW		 5205 }
5206 goto do_nref_ref_common;
5207
3c0563b9 5208 case REFFL: /* /\1/il */
780fcc9f 5209 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
d7ef4b73
KW		 5210 folder = foldEQ_locale;
5211 fold_array = PL_fold_locale;
cea315b6 5212 utf8_fold_flags = FOLDEQ_LOCALE;
d7ef4b73
KW		 5213 goto do_ref;
5214
3c0563b9 5215 case REFFA: /* /\1/iaa */
2f7f8cb1
KW		 5216 folder = foldEQ_latin1;
5217 fold_array = PL_fold_latin1;
5218 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5219 goto do_ref;
5220
3c0563b9 5221 case REFFU: /* /\1/iu */
d7ef4b73
KW		 5222 folder = foldEQ_latin1;
5223 fold_array = PL_fold_latin1;
d513472c 5224 utf8_fold_flags = 0;
d7ef4b73
KW		 5225 goto do_ref;
5226
3c0563b9 5227 case REFF: /* /\1/i */
d7ef4b73
KW		 5228 folder = foldEQ;
5229 fold_array = PL_fold;
d513472c 5230 utf8_fold_flags = 0;
83d7b90b 5231 goto do_ref;
d7ef4b73 5232
3c0563b9 5233 case REF: /* /\1/ */
83d7b90b
KW		 5234 folder = NULL;
5235 fold_array = NULL;
d513472c 5236 utf8_fold_flags = 0;
83d7b90b 5237
d7ef4b73 5238 do_ref:
81714fb9 5239 type = OP(scan);
d7ef4b73
KW		 5240 n = ARG(scan); /* which paren pair */
5241
5242 do_nref_ref_common:
b93070ed 5243 ln = rex->offs[n].start;
1cb48e53 5244 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
b93070ed 5245 if (rex->lastparen < n || ln == -1)
af3f8c16 5246 sayNO; /* Do not match unless seen CLOSEn. */
b93070ed 5247 if (ln == rex->offs[n].end)
a0d0e21e 5248 break;
a0ed51b3 5249
9d9163fb 5250 s = reginfo->strbeg + ln;
d7ef4b73 5251 if (type != REF /* REF can do byte comparison */
31f05a37
KW		 5252 && (utf8_target || type == REFFU || type == REFFL))
5253 {
220db18a 5254 char * limit = reginfo->strend;
d7ef4b73
KW		 5255
5256 /* This call case insensitively compares the entire buffer
5257 * at s, with the current input starting at locinput, but
220db18a
DM		 5258 * not going off the end given by reginfo->strend, and
5259 * returns in <limit> upon success, how much of the
5260 * current input was matched */
b93070ed 5261 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
d513472c 5262 locinput, &limit, 0, utf8_target, utf8_fold_flags))
d7ef4b73
KW		 5263 {
5264 sayNO;
a0ed51b3 5265 }
d7ef4b73 5266 locinput = limit;
a0ed51b3
LW		 5267 break;
5268 }
5269
d7ef4b73 5270 /* Not utf8: Inline the first character, for speed. */
7016d6eb
DM		 5271 if (!NEXTCHR_IS_EOS &&
5272 UCHARAT(s) != nextchr &&
81714fb9 5273 (type == REF ||
d7ef4b73 5274 UCHARAT(s) != fold_array[nextchr]))
4633a7c4 5275 sayNO;
b93070ed 5276 ln = rex->offs[n].end - ln;
220db18a 5277 if (locinput + ln > reginfo->strend)
4633a7c4 5278 sayNO;
81714fb9 5279 if (ln > 1 && (type == REF
24d3c4a9 5280 ? memNE(s, locinput, ln)
d7ef4b73 5281 : ! folder(s, locinput, ln)))
4633a7c4 5282 sayNO;
24d3c4a9 5283 locinput += ln;
a0d0e21e 5284 break;
81714fb9 5285 }
3c0563b9
DM		 5286
5287 case NOTHING: /* null op; e.g. the 'nothing' following
5288 * the '*' in m{(a+|b)*}' */
5289 break;
5290 case TAIL: /* placeholder while compiling (A|B|C) */
a0d0e21e 5291 break;
3c0563b9 5292
40a82448
DM		 5293#undef ST
5294#define ST st->u.eval
c277df42 5295 {
c277df42 5296 SV *ret;
d2f13c59 5297 REGEXP *re_sv;
6bda09f9 5298 regexp *re;
f8fc2ecf 5299 regexp_internal *rei;
1a147d38
YO		 5300 regnode *startpoint;
5301
3c0563b9 5302 case GOSTART: /* (?R) */
e7707071
YO		 5303 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
5304 if (cur_eval && cur_eval->locinput==locinput) {
24b23f37 5305 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
1a147d38 5306 Perl_croak(aTHX_ "Infinite recursion in regex");
4b196cd4 5307 if ( ++nochange_depth > max_nochange_depth )
1a147d38
YO		 5308 Perl_croak(aTHX_
5309 "Pattern subroutine nesting without pos change"
5310 " exceeded limit in regex");
6bda09f9
YO		 5311 } else {
5312 nochange_depth = 0;
1a147d38 5313 }
288b8c02 5314 re_sv = rex_sv;
6bda09f9 5315 re = rex;
f8fc2ecf 5316 rei = rexi;
1a147d38 5317 if (OP(scan)==GOSUB) {
6bda09f9
YO		 5318 startpoint = scan + ARG2L(scan);
5319 ST.close_paren = ARG(scan);
5320 } else {
f8fc2ecf 5321 startpoint = rei->program+1;
6bda09f9
YO		 5322 ST.close_paren = 0;
5323 }
d1b2014a
YO		 5324
5325 /* Save all the positions seen so far. */
5326 ST.cp = regcppush(rex, 0, maxopenparen);
5327 REGCP_SET(ST.lastcp);
5328
5329 /* and then jump to the code we share with EVAL */
6bda09f9 5330 goto eval_recurse_doit;
a74ff37d 5331 /* NOTREACHED */
3c0563b9 5332
6bda09f9
YO		 5333 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
5334 if (cur_eval && cur_eval->locinput==locinput) {
4b196cd4 5335 if ( ++nochange_depth > max_nochange_depth )
1a147d38 5336 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
6bda09f9
YO		 5337 } else {
5338 nochange_depth = 0;
5339 }
8e5e9ebe 5340 {
4aabdb9b 5341 /* execute the code in the {...} */
81ed78b2 5342
4aabdb9b 5343 dSP;
a6dc34f1 5344 IV before;
1f4d1a1e 5345 OP * const oop = PL_op;
4aabdb9b 5346 COP * const ocurcop = PL_curcop;
81ed78b2 5347 OP *nop;
81ed78b2 5348 CV *newcv;
91332126 5349
74088413 5350 /* save *all* paren positions */
92da3157 5351 regcppush(rex, 0, maxopenparen);
74088413
DM		 5352 REGCP_SET(runops_cp);
5353
81ed78b2
DM		 5354 if (!caller_cv)
5355 caller_cv = find_runcv(NULL);
5356
4aabdb9b 5357 n = ARG(scan);
81ed78b2 5358
b30fcab9 5359 if (rexi->data->what[n] == 'r') { /* code from an external qr */
8d919b0a 5360 newcv = (ReANY(
b30fcab9
DM		 5361 (REGEXP*)(rexi->data->data[n])
5362 ))->qr_anoncv
81ed78b2
DM		 5363 ;
5364 nop = (OP*)rexi->data->data[n+1];
b30fcab9
DM		 5365 }
5366 else if (rexi->data->what[n] == 'l') { /* literal code */
81ed78b2
DM		 5367 newcv = caller_cv;
5368 nop = (OP*)rexi->data->data[n];
5369 assert(CvDEPTH(newcv));
68e2671b
DM		 5370 }
5371 else {
d24ca0c5
DM		 5372 /* literal with own CV */
5373 assert(rexi->data->what[n] == 'L');
81ed78b2
DM		 5374 newcv = rex->qr_anoncv;
5375 nop = (OP*)rexi->data->data[n];
68e2671b 5376 }
81ed78b2 5377
0e458318
DM		 5378 /* normally if we're about to execute code from the same
5379 * CV that we used previously, we just use the existing
5380 * CX stack entry. However, its possible that in the
5381 * meantime we may have backtracked, popped from the save
5382 * stack, and undone the SAVECOMPPAD(s) associated with
5383 * PUSH_MULTICALL; in which case PL_comppad no longer
5384 * points to newcv's pad. */
5385 if (newcv != last_pushed_cv || PL_comppad != last_pad)
5386 {
b0065247
DM		 5387 U8 flags = (CXp_SUB_RE |
5388 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
0e458318 5389 if (last_pushed_cv) {
b0065247 5390 CHANGE_MULTICALL_FLAGS(newcv, flags);
0e458318
DM		 5391 }
5392 else {
b0065247 5393 PUSH_MULTICALL_FLAGS(newcv, flags);
0e458318
DM		 5394 }
5395 last_pushed_cv = newcv;
5396 }
c31ee3bb
DM		 5397 else {
5398 /* these assignments are just to silence compiler
5399 * warnings */
5400 multicall_cop = NULL;
5401 newsp = NULL;
5402 }
0e458318
DM		 5403 last_pad = PL_comppad;
5404
2e2e3f36
DM		 5405 /* the initial nextstate you would normally execute
5406 * at the start of an eval (which would cause error
5407 * messages to come from the eval), may be optimised
5408 * away from the execution path in the regex code blocks;
5409 * so manually set PL_curcop to it initially */
5410 {
81ed78b2 5411 OP *o = cUNOPx(nop)->op_first;
2e2e3f36
DM		 5412 assert(o->op_type == OP_NULL);
5413 if (o->op_targ == OP_SCOPE) {
5414 o = cUNOPo->op_first;
5415 }
5416 else {
5417 assert(o->op_targ == OP_LEAVE);
5418 o = cUNOPo->op_first;
5419 assert(o->op_type == OP_ENTER);
e6dae479 5420 o = OpSIBLING(o);
2e2e3f36
DM		 5421 }
5422
5423 if (o->op_type != OP_STUB) {
5424 assert( o->op_type == OP_NEXTSTATE
5425 || o->op_type == OP_DBSTATE
5426 || (o->op_type == OP_NULL
5427 && ( o->op_targ == OP_NEXTSTATE
5428 || o->op_targ == OP_DBSTATE
5429 )
5430 )
5431 );
5432 PL_curcop = (COP*)o;
5433 }
5434 }
81ed78b2 5435 nop = nop->op_next;
2e2e3f36 5436
24b23f37 5437 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
81ed78b2
DM		 5438 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
5439
8adc0f72 5440 rex->offs[0].end = locinput - reginfo->strbeg;
bf2039a9 5441 if (reginfo->info_aux_eval->pos_magic)
25fdce4a
FC		 5442 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
5443 reginfo->sv, reginfo->strbeg,
5444 locinput - reginfo->strbeg);
4aabdb9b 5445
2bf803e2
YO		 5446 if (sv_yes_mark) {
5447 SV *sv_mrk = get_sv("REGMARK", 1);
5448 sv_setsv(sv_mrk, sv_yes_mark);
5449 }
5450
81ed78b2
DM		 5451 /* we don't use MULTICALL here as we want to call the
5452 * first op of the block of interest, rather than the
5453 * first op of the sub */
a6dc34f1 5454 before = (IV)(SP-PL_stack_base);
81ed78b2 5455 PL_op = nop;
8e5e9ebe
RGS		 5456 CALLRUNOPS(aTHX); /* Scalar context. */
5457 SPAGAIN;
a6dc34f1 5458 if ((IV)(SP-PL_stack_base) == before)
075aa684 5459 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
8e5e9ebe
RGS		 5460 else {
5461 ret = POPs;
5462 PUTBACK;
5463 }
4aabdb9b 5464
e4bfbed3
DM		 5465 /* before restoring everything, evaluate the returned
5466 * value, so that 'uninit' warnings don't use the wrong
497d0a96
DM		 5467 * PL_op or pad. Also need to process any magic vars
5468 * (e.g. $1) *before* parentheses are restored */
e4bfbed3
DM		 5469
5470 PL_op = NULL;
5471
5e98dac2 5472 re_sv = NULL;
e4bfbed3
DM		 5473 if (logical == 0) /* (?{})/ */
5474 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
5475 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
5476 sw = cBOOL(SvTRUE(ret));
5477 logical = 0;
5478 }
5479 else { /* /(??{}) */
497d0a96
DM		 5480 /* if its overloaded, let the regex compiler handle
5481 * it; otherwise extract regex, or stringify */
237da807 5482 if (SvGMAGICAL(ret))
2685dc2d 5483 ret = sv_mortalcopy(ret);
497d0a96
DM		 5484 if (!SvAMAGIC(ret)) {
5485 SV *sv = ret;
5486 if (SvROK(sv))
5487 sv = SvRV(sv);
5488 if (SvTYPE(sv) == SVt_REGEXP)
5489 re_sv = (REGEXP*) sv;
63620942
FC		 5490 else if (SvSMAGICAL(ret)) {
5491 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
497d0a96
DM		 5492 if (mg)
5493 re_sv = (REGEXP *) mg->mg_obj;
5494 }
e4bfbed3 5495
2685dc2d 5496 /* force any undef warnings here */
237da807
FC		 5497 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
5498 ret = sv_mortalcopy(ret);
497d0a96
DM		 5499 (void) SvPV_force_nolen(ret);
5500 }
e4bfbed3
DM		 5501 }
5502
5503 }
5504
81ed78b2
DM		 5505 /* *** Note that at this point we don't restore
5506 * PL_comppad, (or pop the CxSUB) on the assumption it may
5507 * be used again soon. This is safe as long as nothing
5508 * in the regexp code uses the pad ! */
4aabdb9b 5509 PL_op = oop;
4aabdb9b 5510 PL_curcop = ocurcop;
92da3157 5511 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
f5df269c 5512 PL_curpm = PL_reg_curpm;
e4bfbed3
DM		 5513
5514 if (logical != 2)
4aabdb9b 5515 break;
8e5e9ebe 5516 }
e4bfbed3
DM		 5517
5518 /* only /(??{})/ from now on */
24d3c4a9 5519 logical = 0;
4aabdb9b 5520 {
4f639d21
DM		 5521 /* extract RE object from returned value; compiling if
5522 * necessary */
5c35adbb 5523
575c37f6
DM		 5524 if (re_sv) {
5525 re_sv = reg_temp_copy(NULL, re_sv);
288b8c02 5526 }
0f5d15d6 5527 else {
c737faaf 5528 U32 pm_flags = 0;
0f5d15d6 5529
9753d940
DM		 5530 if (SvUTF8(ret) && IN_BYTES) {
5531 /* In use 'bytes': make a copy of the octet
5532 * sequence, but without the flag on */
b9ad30b4
NC		 5533 STRLEN len;
5534 const char *const p = SvPV(ret, len);
5535 ret = newSVpvn_flags(p, len, SVs_TEMP);
5536 }
732caac7
DM		 5537 if (rex->intflags & PREGf_USE_RE_EVAL)
5538 pm_flags |= PMf_USE_RE_EVAL;
5539
5540 /* if we got here, it should be an engine which
5541 * supports compiling code blocks and stuff */
5542 assert(rex->engine && rex->engine->op_comp);
ec841a27 5543 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
575c37f6 5544 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
ec841a27 5545 rex->engine, NULL, NULL,
33be4c61 5546 /* copy /msixn etc to inner pattern */
13f27704 5547 ARG2L(scan),
ec841a27 5548 pm_flags);
732caac7 5549
9041c2e3 5550 if (!(SvFLAGS(ret)
237da807
FC		 5551 & (SVs_TEMP | SVs_GMG | SVf_ROK))
5552 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
a2794585
NC		 5553 /* This isn't a first class regexp. Instead, it's
5554 caching a regexp onto an existing, Perl visible
5555 scalar. */
575c37f6 5556 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
3ce3ed55 5557 }
0f5d15d6 5558 }
e1ff3a88 5559 SAVEFREESV(re_sv);
8d919b0a 5560 re = ReANY(re_sv);
4aabdb9b 5561 }
07bc277f 5562 RXp_MATCH_COPIED_off(re);
28d8d7f4
YO		 5563 re->subbeg = rex->subbeg;
5564 re->sublen = rex->sublen;
6502e081
DM		 5565 re->suboffset = rex->suboffset;
5566 re->subcoffset = rex->subcoffset;
d1b2014a
YO		 5567 re->lastparen = 0;
5568 re->lastcloseparen = 0;
f8fc2ecf 5569 rei = RXi_GET(re);
6bda09f9 5570 DEBUG_EXECUTE_r(
220db18a
DM		 5571 debug_start_match(re_sv, utf8_target, locinput,
5572 reginfo->strend, "Matching embedded");
6bda09f9 5573 );
f8fc2ecf 5574 startpoint = rei->program + 1;
1a147d38 5575 ST.close_paren = 0; /* only used for GOSUB */
d1b2014a
YO		 5576 /* Save all the seen positions so far. */
5577 ST.cp = regcppush(rex, 0, maxopenparen);
5578 REGCP_SET(ST.lastcp);
5579 /* and set maxopenparen to 0, since we are starting a "fresh" match */
5580 maxopenparen = 0;
5581 /* run the pattern returned from (??{...}) */
aa283a38 5582
c52b8b12 5583 eval_recurse_doit: /* Share code with GOSUB below this line
d1b2014a
YO		 5584 * At this point we expect the stack context to be
5585 * set up correctly */
4aabdb9b 5586
1cb95af7
DM		 5587 /* invalidate the S-L poscache. We're now executing a
5588 * different set of WHILEM ops (and their associated
5589 * indexes) against the same string, so the bits in the
5590 * cache are meaningless. Setting maxiter to zero forces
5591 * the cache to be invalidated and zeroed before reuse.
5592 * XXX This is too dramatic a measure. Ideally we should
5593 * save the old cache and restore when running the outer
5594 * pattern again */
1cb48e53 5595 reginfo->poscache_maxiter = 0;
4aabdb9b 5596
d1b2014a 5597 /* the new regexp might have a different is_utf8_pat than we do */
aed7b151 5598 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
faec1544 5599
288b8c02 5600 ST.prev_rex = rex_sv;
faec1544 5601 ST.prev_curlyx = cur_curlyx;
ec43f78b
DM		 5602 rex_sv = re_sv;
5603 SET_reg_curpm(rex_sv);
288b8c02 5604 rex = re;
f8fc2ecf 5605 rexi = rei;
faec1544 5606 cur_curlyx = NULL;
40a82448 5607 ST.B = next;
faec1544
DM		 5608 ST.prev_eval = cur_eval;
5609 cur_eval = st;
faec1544 5610 /* now continue from first node in postoned RE */
4d5016e5 5611 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
a74ff37d 5612 /* NOTREACHED */
e5964223 5613 NOT_REACHED;
c277df42 5614 }
40a82448 5615
faec1544
DM		 5616 case EVAL_AB: /* cleanup after a successful (??{A})B */
5617 /* note: this is called twice; first after popping B, then A */
ec43f78b 5618 rex_sv = ST.prev_rex;
aed7b151 5619 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
ec43f78b 5620 SET_reg_curpm(rex_sv);
8d919b0a 5621 rex = ReANY(rex_sv);
f8fc2ecf 5622 rexi = RXi_GET(rex);
4b22688e
YO		 5623 {
5624 /* preserve $^R across LEAVE's. See Bug 121070. */
5625 SV *save_sv= GvSV(PL_replgv);
5626 SvREFCNT_inc(save_sv);
5627 regcpblow(ST.cp); /* LEAVE in disguise */
5628 sv_setsv(GvSV(PL_replgv), save_sv);
5629 SvREFCNT_dec(save_sv);
5630 }
faec1544
DM		 5631 cur_eval = ST.prev_eval;
5632 cur_curlyx = ST.prev_curlyx;
34a81e2b 5633
1cb95af7 5634 /* Invalidate cache. See "invalidate" comment above. */
1cb48e53 5635 reginfo->poscache_maxiter = 0;
e7707071 5636 if ( nochange_depth )
4b196cd4 5637 nochange_depth--;
262b90c4 5638 sayYES;
40a82448 5639
40a82448 5640
faec1544
DM		 5641 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
5642 /* note: this is called twice; first after popping B, then A */
ec43f78b 5643 rex_sv = ST.prev_rex;
aed7b151 5644 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
ec43f78b 5645 SET_reg_curpm(rex_sv);
8d919b0a 5646 rex = ReANY(rex_sv);
f8fc2ecf 5647 rexi = RXi_GET(rex);
0357f1fd 5648
40a82448 5649 REGCP_UNWIND(ST.lastcp);
92da3157 5650 regcppop(rex, &maxopenparen);
faec1544
DM		 5651 cur_eval = ST.prev_eval;
5652 cur_curlyx = ST.prev_curlyx;
1cb95af7 5653 /* Invalidate cache. See "invalidate" comment above. */
1cb48e53 5654 reginfo->poscache_maxiter = 0;
e7707071 5655 if ( nochange_depth )
4b196cd4 5656 nochange_depth--;
40a82448 5657 sayNO_SILENT;
40a82448
DM		 5658#undef ST
5659
3c0563b9 5660 case OPEN: /* ( */
c277df42 5661 n = ARG(scan); /* which paren pair */
9d9163fb 5662 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
92da3157
DM		 5663 if (n > maxopenparen)
5664 maxopenparen = n;
495f47a5 5665 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
92da3157 5666 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n",
495f47a5
DM		 5667 PTR2UV(rex),
5668 PTR2UV(rex->offs),
5669 (UV)n,
5670 (IV)rex->offs[n].start_tmp,
92da3157 5671 (UV)maxopenparen
495f47a5 5672 ));
e2e6a0f1 5673 lastopen = n;
a0d0e21e 5674 break;
495f47a5
DM		 5675
5676/* XXX really need to log other places start/end are set too */
5677#define CLOSE_CAPTURE \
5678 rex->offs[n].start = rex->offs[n].start_tmp; \
9d9163fb 5679 rex->offs[n].end = locinput - reginfo->strbeg; \
495f47a5
DM		 5680 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
5681 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
5682 PTR2UV(rex), \
5683 PTR2UV(rex->offs), \
5684 (UV)n, \
5685 (IV)rex->offs[n].start, \
5686 (IV)rex->offs[n].end \
5687 ))
5688
3c0563b9 5689 case CLOSE: /* ) */
c277df42 5690 n = ARG(scan); /* which paren pair */
495f47a5 5691 CLOSE_CAPTURE;
b93070ed
DM		 5692 if (n > rex->lastparen)
5693 rex->lastparen = n;
5694 rex->lastcloseparen = n;
3b6647e0 5695 if (cur_eval && cur_eval->u.eval.close_paren == n) {
6bda09f9
YO		 5696 goto fake_end;
5697 }
a0d0e21e 5698 break;
3c0563b9
DM		 5699
5700 case ACCEPT: /* (*ACCEPT) */
e2e6a0f1
YO		 5701 if (ARG(scan)){
5702 regnode *cursor;
5703 for (cursor=scan;
5704 cursor && OP(cursor)!=END;
5705 cursor=regnext(cursor))
5706 {
5707 if ( OP(cursor)==CLOSE ){
5708 n = ARG(cursor);
5709 if ( n <= lastopen ) {
495f47a5 5710 CLOSE_CAPTURE;
b93070ed
DM		 5711 if (n > rex->lastparen)
5712 rex->lastparen = n;
5713 rex->lastcloseparen = n;
3b6647e0
RB		 5714 if ( n == ARG(scan) || (cur_eval &&
5715 cur_eval->u.eval.close_paren == n))
e2e6a0f1
YO		 5716 break;
5717 }
5718 }
5719 }
5720 }
5721 goto fake_end;
a74ff37d 5722 /* NOTREACHED */
3c0563b9
DM		 5723
5724 case GROUPP: /* (?(1)) */
c277df42 5725 n = ARG(scan); /* which paren pair */
b93070ed 5726 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
c277df42 5727 break;
3c0563b9
DM		 5728
5729 case NGROUPP: /* (?(<name>)) */
0a4db386 5730 /* reg_check_named_buff_matched returns 0 for no match */
f2338a2e 5731 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
0a4db386 5732 break;
3c0563b9
DM		 5733
5734 case INSUBP: /* (?(R)) */
0a4db386 5735 n = ARG(scan);
3b6647e0 5736 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
0a4db386 5737 break;
3c0563b9
DM		 5738
5739 case DEFINEP: /* (?(DEFINE)) */
0a4db386
YO		 5740 sw = 0;
5741 break;
3c0563b9
DM		 5742
5743 case IFTHEN: /* (?(cond)A|B) */
1cb48e53 5744 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
24d3c4a9 5745 if (sw)
c277df42
IZ		 5746 next = NEXTOPER(NEXTOPER(scan));
5747 else {
5748 next = scan + ARG(scan);
5749 if (OP(next) == IFTHEN) /* Fake one. */
5750 next = NEXTOPER(NEXTOPER(next));
5751 }
5752 break;
3c0563b9
DM		 5753
5754 case LOGICAL: /* modifier for EVAL and IFMATCH */
24d3c4a9 5755 logical = scan->flags;
c277df42 5756 break;
c476f425 5757
2ab05381 5758/*******************************************************************
2ab05381 5759
c476f425
DM		 5760The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5761pattern, where A and B are subpatterns. (For simple A, CURLYM or
5762STAR/PLUS/CURLY/CURLYN are used instead.)
2ab05381 5763
c476f425 5764A*B is compiled as <CURLYX><A><WHILEM><B>
2ab05381 5765
c476f425
DM		 5766On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5767state, which contains the current count, initialised to -1. It also sets
5768cur_curlyx to point to this state, with any previous value saved in the
5769state block.
2ab05381 5770
c476f425
DM		 5771CURLYX then jumps straight to the WHILEM op, rather than executing A,
5772since the pattern may possibly match zero times (i.e. it's a while {} loop
5773rather than a do {} while loop).
2ab05381 5774
c476f425
DM		 5775Each entry to WHILEM represents a successful match of A. The count in the
5776CURLYX block is incremented, another WHILEM state is pushed, and execution
5777passes to A or B depending on greediness and the current count.
2ab05381 5778
c476f425
DM		 5779For example, if matching against the string a1a2a3b (where the aN are
5780substrings that match /A/), then the match progresses as follows: (the
5781pushed states are interspersed with the bits of strings matched so far):
2ab05381 5782
c476f425
DM		 5783 <CURLYX cnt=-1>
5784 <CURLYX cnt=0><WHILEM>
5785 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5786 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5787 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5788 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
2ab05381 5789
c476f425
DM		 5790(Contrast this with something like CURLYM, which maintains only a single
5791backtrack state:
2ab05381 5792
c476f425
DM		 5793 <CURLYM cnt=0> a1
5794 a1 <CURLYM cnt=1> a2
5795 a1 a2 <CURLYM cnt=2> a3
5796 a1 a2 a3 <CURLYM cnt=3> b
5797)
2ab05381 5798
c476f425
DM		 5799Each WHILEM state block marks a point to backtrack to upon partial failure
5800of A or B, and also contains some minor state data related to that
5801iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5802overall state, such as the count, and pointers to the A and B ops.
2ab05381 5803
c476f425
DM		 5804This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5805must always point to the *current* CURLYX block, the rules are:
2ab05381 5806
c476f425
DM		 5807When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5808and set cur_curlyx to point the new block.
2ab05381 5809
c476f425
DM		 5810When popping the CURLYX block after a successful or unsuccessful match,
5811restore the previous cur_curlyx.
2ab05381 5812
c476f425
DM		 5813When WHILEM is about to execute B, save the current cur_curlyx, and set it
5814to the outer one saved in the CURLYX block.
2ab05381 5815
c476f425
DM		 5816When popping the WHILEM block after a successful or unsuccessful B match,
5817restore the previous cur_curlyx.
2ab05381 5818
c476f425
DM		 5819Here's an example for the pattern (AI* BI)*BO
5820I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
2ab05381 5821
c476f425
DM		 5822cur_
5823curlyx backtrack stack
5824------ ---------------
5825NULL
5826CO <CO prev=NULL> <WO>
5827CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5828CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5829NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
2ab05381 5830
c476f425
DM		 5831At this point the pattern succeeds, and we work back down the stack to
5832clean up, restoring as we go:
95b24440 5833
c476f425
DM		 5834CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5835CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5836CO <CO prev=NULL> <WO>
5837NULL
a0374537 5838
c476f425
DM		 5839*******************************************************************/
5840
5841#define ST st->u.curlyx
5842
5843 case CURLYX: /* start of /A*B/ (for complex A) */
5844 {
5845 /* No need to save/restore up to this paren */
5846 I32 parenfloor = scan->flags;
5847
5848 assert(next); /* keep Coverity happy */
5849 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5850 next += ARG(next);
5851
5852 /* XXXX Probably it is better to teach regpush to support
92da3157 5853 parenfloor > maxopenparen ... */
b93070ed
DM		 5854 if (parenfloor > (I32)rex->lastparen)
5855 parenfloor = rex->lastparen; /* Pessimization... */
c476f425
DM		 5856
5857 ST.prev_curlyx= cur_curlyx;
5858 cur_curlyx = st;
5859 ST.cp = PL_savestack_ix;
5860
5861 /* these fields contain the state of the current curly.
5862 * they are accessed by subsequent WHILEMs */
5863 ST.parenfloor = parenfloor;
d02d6d97 5864 ST.me = scan;
c476f425 5865 ST.B = next;
24d3c4a9
DM		 5866 ST.minmod = minmod;
5867 minmod = 0;
c476f425
DM		 5868 ST.count = -1; /* this will be updated by WHILEM */
5869 ST.lastloc = NULL; /* this will be updated by WHILEM */
5870
4d5016e5 5871 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
a74ff37d 5872 /* NOTREACHED */
e5964223 5873 NOT_REACHED;
c476f425 5874 }
a0d0e21e 5875
c476f425 5876 case CURLYX_end: /* just finished matching all of A*B */
c476f425
DM		 5877 cur_curlyx = ST.prev_curlyx;
5878 sayYES;
a74ff37d 5879 /* NOTREACHED */
e5964223 5880 NOT_REACHED;
a0d0e21e 5881
c476f425
DM		 5882 case CURLYX_end_fail: /* just failed to match all of A*B */
5883 regcpblow(ST.cp);
5884 cur_curlyx = ST.prev_curlyx;
5885 sayNO;
a74ff37d 5886 /* NOTREACHED */
e5964223 5887 NOT_REACHED;
4633a7c4 5888
a0d0e21e 5889
c476f425
DM		 5890#undef ST
5891#define ST st->u.whilem
5892
5893 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5894 {
5895 /* see the discussion above about CURLYX/WHILEM */
c476f425 5896 I32 n;
1a522d5d
JH		 5897 int min, max;
5898 regnode *A;
d02d6d97 5899
c476f425 5900 assert(cur_curlyx); /* keep Coverity happy */
1a522d5d
JH		 5901
5902 min = ARG1(cur_curlyx->u.curlyx.me);
5903 max = ARG2(cur_curlyx->u.curlyx.me);
5904 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
c476f425
DM		 5905 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5906 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5907 ST.cache_offset = 0;
5908 ST.cache_mask = 0;
5909
c476f425
DM		 5910
5911 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
d02d6d97
DM		 5912 "%*s whilem: matched %ld out of %d..%d\n",
5913 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
c476f425 5914 );
a0d0e21e 5915
c476f425 5916 /* First just match a string of min A's. */
a0d0e21e 5917
d02d6d97 5918 if (n < min) {
92da3157
DM		 5919 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5920 maxopenparen);
c476f425 5921 cur_curlyx->u.curlyx.lastloc = locinput;
92e82afa
YO		 5922 REGCP_SET(ST.lastcp);
5923
4d5016e5 5924 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
a74ff37d 5925 /* NOTREACHED */
e5964223 5926 NOT_REACHED;
c476f425
DM		 5927 }
5928
5929 /* If degenerate A matches "", assume A done. */
5930
5931 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5932 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5933 "%*s whilem: empty match detected, trying continuation...\n",
5934 REPORT_CODE_OFF+depth*2, "")
5935 );
5936 goto do_whilem_B_max;
5937 }
5938
1cb95af7
DM		 5939 /* super-linear cache processing.
5940 *
5941 * The idea here is that for certain types of CURLYX/WHILEM -
5942 * principally those whose upper bound is infinity (and
5943 * excluding regexes that have things like \1 and other very
5944 * non-regular expresssiony things), then if a pattern like
5945 * /....A*.../ fails and we backtrack to the WHILEM, then we
5946 * make a note that this particular WHILEM op was at string
5947 * position 47 (say) when the rest of pattern failed. Then, if
5948 * we ever find ourselves back at that WHILEM, and at string
5949 * position 47 again, we can just fail immediately rather than
5950 * running the rest of the pattern again.
5951 *
5952 * This is very handy when patterns start to go
5953 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
5954 * with a combinatorial explosion of backtracking.
5955 *
5956 * The cache is implemented as a bit array, with one bit per
5957 * string byte position per WHILEM op (up to 16) - so its
5958 * between 0.25 and 2x the string size.
5959 *
5960 * To avoid allocating a poscache buffer every time, we do an
5961 * initially countdown; only after we have executed a WHILEM
5962 * op (string-length x #WHILEMs) times do we allocate the
5963 * cache.
5964 *
5965 * The top 4 bits of scan->flags byte say how many different
5966 * relevant CURLLYX/WHILEM op pairs there are, while the
5967 * bottom 4-bits is the identifying index number of this
5968 * WHILEM.
5969 */
c476f425
DM		 5970
5971 if (scan->flags) {
a0d0e21e 5972
1cb48e53 5973 if (!reginfo->poscache_maxiter) {
c476f425
DM		 5974 /* start the countdown: Postpone detection until we
5975 * know the match is not *that* much linear. */
1cb48e53 5976 reginfo->poscache_maxiter
9d9163fb
DM		 5977 = (reginfo->strend - reginfo->strbeg + 1)
5978 * (scan->flags>>4);
66bf836d 5979 /* possible overflow for long strings and many CURLYX's */
1cb48e53
DM		 5980 if (reginfo->poscache_maxiter < 0)
5981 reginfo->poscache_maxiter = I32_MAX;
5982 reginfo->poscache_iter = reginfo->poscache_maxiter;
2c2d71f5 5983 }
c476f425 5984
1cb48e53 5985 if (reginfo->poscache_iter-- == 0) {
c476f425 5986 /* initialise cache */
ea3daa5d 5987 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
2ac8ff4b
DM		 5988 regmatch_info_aux *const aux = reginfo->info_aux;
5989 if (aux->poscache) {
ea3daa5d 5990 if ((SSize_t)reginfo->poscache_size < size) {
2ac8ff4b
DM		 5991 Renew(aux->poscache, size, char);
5992 reginfo->poscache_size = size;
2c2d71f5 5993 }
2ac8ff4b 5994 Zero(aux->poscache, size, char);
2c2d71f5
JH		 5995 }
5996 else {
2ac8ff4b
DM		 5997 reginfo->poscache_size = size;
5998 Newxz(aux->poscache, size, char);
2c2d71f5 5999 }
c476f425
DM		 6000 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
6001 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
6002 PL_colors[4], PL_colors[5])
6003 );
2c2d71f5 6004 }
c476f425 6005
1cb48e53 6006 if (reginfo->poscache_iter < 0) {
c476f425 6007 /* have we already failed at this position? */
ea3daa5d 6008 SSize_t offset, mask;
338e600a
DM		 6009
6010 reginfo->poscache_iter = -1; /* stop eventual underflow */
c476f425 6011 offset = (scan->flags & 0xf) - 1
9d9163fb
DM		 6012 + (locinput - reginfo->strbeg)
6013 * (scan->flags>>4);
c476f425
DM		 6014 mask = 1 << (offset % 8);
6015 offset /= 8;
2ac8ff4b 6016 if (reginfo->info_aux->poscache[offset] & mask) {
c476f425
DM		 6017 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
6018 "%*s whilem: (cache) already tried at this position...\n",
6019 REPORT_CODE_OFF+depth*2, "")
2c2d71f5 6020 );
3298f257 6021 sayNO; /* cache records failure */
2c2d71f5 6022 }
c476f425
DM		 6023 ST.cache_offset = offset;
6024 ST.cache_mask = mask;
2c2d71f5 6025 }
c476f425 6026 }
2c2d71f5 6027
c476f425 6028 /* Prefer B over A for minimal matching. */
a687059c 6029
c476f425
DM		 6030 if (cur_curlyx->u.curlyx.minmod) {
6031 ST.save_curlyx = cur_curlyx;
6032 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
92da3157
DM		 6033 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
6034 maxopenparen);
c476f425 6035 REGCP_SET(ST.lastcp);
4d5016e5
DM		 6036 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
6037 locinput);
a74ff37d 6038 /* NOTREACHED */
e5964223 6039 NOT_REACHED;
c476f425 6040 }
a0d0e21e 6041
c476f425
DM		 6042 /* Prefer A over B for maximal matching. */
6043
d02d6d97 6044 if (n < max) { /* More greed allowed? */
92da3157
DM		 6045 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
6046 maxopenparen);
c476f425
DM		 6047 cur_curlyx->u.curlyx.lastloc = locinput;
6048 REGCP_SET(ST.lastcp);
4d5016e5 6049 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
a74ff37d 6050 /* NOTREACHED */
e5964223 6051 NOT_REACHED;
c476f425
DM		 6052 }
6053 goto do_whilem_B_max;
6054 }
a74ff37d 6055 /* NOTREACHED */
e5964223 6056 NOT_REACHED;
c476f425
DM		 6057
6058 case WHILEM_B_min: /* just matched B in a minimal match */
6059 case WHILEM_B_max: /* just matched B in a maximal match */
6060 cur_curlyx = ST.save_curlyx;
6061 sayYES;
a74ff37d 6062 /* NOTREACHED */
e5964223 6063 NOT_REACHED;
c476f425
DM		 6064
6065 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
6066 cur_curlyx = ST.save_curlyx;
6067 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
6068 cur_curlyx->u.curlyx.count--;
6069 CACHEsayNO;
a74ff37d 6070 /* NOTREACHED */
e5964223 6071 NOT_REACHED;
c476f425
DM		 6072
6073 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
924ba076 6074 /* FALLTHROUGH */
c476f425 6075 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
92e82afa 6076 REGCP_UNWIND(ST.lastcp);
92da3157 6077 regcppop(rex, &maxopenparen);
c476f425
DM		 6078 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
6079 cur_curlyx->u.curlyx.count--;
6080 CACHEsayNO;
a74ff37d 6081 /* NOTREACHED */
e5964223 6082 NOT_REACHED;
c476f425
DM		 6083
6084 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
6085 REGCP_UNWIND(ST.lastcp);
92da3157 6086 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
c476f425
DM		 6087 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6088 "%*s whilem: failed, trying continuation...\n",
6089 REPORT_CODE_OFF+depth*2, "")
6090 );
6091 do_whilem_B_max:
6092 if (cur_curlyx->u.curlyx.count >= REG_INFTY
6093 && ckWARN(WARN_REGEXP)
39819bd9 6094 && !reginfo->warned)
c476f425 6095 {
39819bd9 6096 reginfo->warned = TRUE;
dcbac5bb
FC		 6097 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
6098 "Complex regular subexpression recursion limit (%d) "
6099 "exceeded",
c476f425
DM		 6100 REG_INFTY - 1);
6101 }
6102
6103 /* now try B */
6104 ST.save_curlyx = cur_curlyx;
6105 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
4d5016e5
DM		 6106 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
6107 locinput);
a74ff37d 6108 /* NOTREACHED */
e5964223 6109 NOT_REACHED;
c476f425
DM		 6110
6111 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
6112 cur_curlyx = ST.save_curlyx;
6113 REGCP_UNWIND(ST.lastcp);
92da3157 6114 regcppop(rex, &maxopenparen);
c476f425 6115
d02d6d97 6116 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
c476f425
DM		 6117 /* Maximum greed exceeded */
6118 if (cur_curlyx->u.curlyx.count >= REG_INFTY
6119 && ckWARN(WARN_REGEXP)
39819bd9 6120 && !reginfo->warned)
c476f425 6121 {
39819bd9 6122 reginfo->warned = TRUE;
c476f425 6123 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
dcbac5bb
FC		 6124 "Complex regular subexpression recursion "
6125 "limit (%d) exceeded",
c476f425 6126 REG_INFTY - 1);
a0d0e21e 6127 }
c476f425 6128 cur_curlyx->u.curlyx.count--;
3ab3c9b4 6129 CACHEsayNO;
a0d0e21e 6130 }
c476f425
DM		 6131
6132 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6133 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
6134 );
6135 /* Try grabbing another A and see if it helps. */
c476f425 6136 cur_curlyx->u.curlyx.lastloc = locinput;
92da3157
DM		 6137 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
6138 maxopenparen);
c476f425 6139 REGCP_SET(ST.lastcp);
d02d6d97 6140 PUSH_STATE_GOTO(WHILEM_A_min,
4d5016e5
DM		 6141 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
6142 locinput);
a74ff37d 6143 /* NOTREACHED */
e5964223 6144 NOT_REACHED;
40a82448
DM		 6145
6146#undef ST
6147#define ST st->u.branch
6148
6149 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
c277df42
IZ		 6150 next = scan + ARG(scan);
6151 if (next == scan)
6152 next = NULL;
40a82448 6153 scan = NEXTOPER(scan);
924ba076 6154 /* FALLTHROUGH */
c277df42 6155
40a82448
DM		 6156 case BRANCH: /* /(...|A|...)/ */
6157 scan = NEXTOPER(scan); /* scan now points to inner node */
b93070ed 6158 ST.lastparen = rex->lastparen;
f6033a9d 6159 ST.lastcloseparen = rex->lastcloseparen;
40a82448
DM		 6160 ST.next_branch = next;
6161 REGCP_SET(ST.cp);
02db2b7b 6162
40a82448 6163 /* Now go into the branch */
5d458dd8 6164 if (has_cutgroup) {
4d5016e5 6165 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
5d458dd8 6166 } else {
4d5016e5 6167 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
5d458dd8 6168 }
a74ff37d 6169 /* NOTREACHED */
e5964223 6170 NOT_REACHED;
3c0563b9
DM		 6171
6172 case CUTGROUP: /* /(*THEN)/ */
5d458dd8 6173 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
ad64d0ec 6174 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4d5016e5 6175 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
a74ff37d 6176 /* NOTREACHED */
e5964223 6177 NOT_REACHED;
3c0563b9 6178
5d458dd8
YO		 6179 case CUTGROUP_next_fail:
6180 do_cutgroup = 1;
6181 no_final = 1;
6182 if (st->u.mark.mark_name)
6183 sv_commit = st->u.mark.mark_name;
6184 sayNO;
a74ff37d 6185 /* NOTREACHED */
e5964223 6186 NOT_REACHED;
3c0563b9 6187
5d458dd8
YO		 6188 case BRANCH_next:
6189 sayYES;
a74ff37d 6190 /* NOTREACHED */
e5964223 6191 NOT_REACHED;
3c0563b9 6192
40a82448 6193 case BRANCH_next_fail: /* that branch failed; try the next, if any */
5d458dd8
YO		 6194 if (do_cutgroup) {
6195 do_cutgroup = 0;
6196 no_final = 0;
6197 }
40a82448 6198 REGCP_UNWIND(ST.cp);
a8d1f4b4 6199 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
40a82448
DM		 6200 scan = ST.next_branch;
6201 /* no more branches? */
5d458dd8
YO		 6202 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
6203 DEBUG_EXECUTE_r({
6204 PerlIO_printf( Perl_debug_log,
6205 "%*s %sBRANCH failed...%s\n",
6206 REPORT_CODE_OFF+depth*2, "",
6207 PL_colors[4],
6208 PL_colors[5] );
6209 });
6210 sayNO_SILENT;
6211 }
40a82448 6212 continue; /* execute next BRANCH[J] op */
a74ff37d 6213 /* NOTREACHED */
40a82448 6214
3c0563b9 6215 case MINMOD: /* next op will be non-greedy, e.g. A*? */
24d3c4a9 6216 minmod = 1;
a0d0e21e 6217 break;
40a82448
DM		 6218
6219#undef ST
6220#define ST st->u.curlym
6221
6222 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
6223
6224 /* This is an optimisation of CURLYX that enables us to push
84d2fa14 6225 * only a single backtracking state, no matter how many matches
40a82448
DM		 6226 * there are in {m,n}. It relies on the pattern being constant
6227 * length, with no parens to influence future backrefs
6228 */
6229
6230 ST.me = scan;
dc45a647 6231 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
40a82448 6232
f6033a9d
DM		 6233 ST.lastparen = rex->lastparen;
6234 ST.lastcloseparen = rex->lastcloseparen;
6235
40a82448
DM		 6236 /* if paren positive, emulate an OPEN/CLOSE around A */
6237 if (ST.me->flags) {
3b6647e0 6238 U32 paren = ST.me->flags;
92da3157
DM		 6239 if (paren > maxopenparen)
6240 maxopenparen = paren;
c277df42 6241 scan += NEXT_OFF(scan); /* Skip former OPEN. */
6407bf3b 6242 }
40a82448
DM		 6243 ST.A = scan;
6244 ST.B = next;
6245 ST.alen = 0;
6246 ST.count = 0;
24d3c4a9
DM		 6247 ST.minmod = minmod;
6248 minmod = 0;
40a82448
DM		 6249 ST.c1 = CHRTEST_UNINIT;
6250 REGCP_SET(ST.cp);
6407bf3b 6251
40a82448
DM		 6252 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
6253 goto curlym_do_B;
6254
6255 curlym_do_A: /* execute the A in /A{m,n}B/ */
4d5016e5 6256 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
a74ff37d 6257 /* NOTREACHED */
e5964223 6258 NOT_REACHED;
5f80c4cf 6259
40a82448 6260 case CURLYM_A: /* we've just matched an A */
40a82448
DM		 6261 ST.count++;
6262 /* after first match, determine A's length: u.curlym.alen */
6263 if (ST.count == 1) {
ba44c216 6264 if (reginfo->is_utf8_target) {
c07e9d7b
DM		 6265 char *s = st->locinput;
6266 while (s < locinput) {
40a82448
DM		 6267 ST.alen++;
6268 s += UTF8SKIP(s);
6269 }
6270 }
6271 else {
c07e9d7b 6272 ST.alen = locinput - st->locinput;
40a82448
DM		 6273 }
6274 if (ST.alen == 0)
6275 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
6276 }
0cadcf80
DM		 6277 DEBUG_EXECUTE_r(
6278 PerlIO_printf(Perl_debug_log,
40a82448 6279 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
5bc10b2c 6280 (int)(REPORT_CODE_OFF+(depth*2)), "",
40a82448 6281 (IV) ST.count, (IV)ST.alen)
0cadcf80
DM		 6282 );
6283
0a4db386 6284 if (cur_eval && cur_eval->u.eval.close_paren &&
24b23f37 6285 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
0a4db386
YO		 6286 goto fake_end;
6287
c966426a
DM		 6288 {
6289 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
6290 if ( max == REG_INFTY || ST.count < max )
6291 goto curlym_do_A; /* try to match another A */
6292 }
40a82448 6293 goto curlym_do_B; /* try to match B */
5f80c4cf 6294
40a82448
DM		 6295 case CURLYM_A_fail: /* just failed to match an A */
6296 REGCP_UNWIND(ST.cp);
0a4db386
YO		 6297
6298 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
6299 || (cur_eval && cur_eval->u.eval.close_paren &&
24b23f37 6300 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
40a82448 6301 sayNO;
0cadcf80 6302
40a82448 6303 curlym_do_B: /* execute the B in /A{m,n}B/ */
40a82448
DM		 6304 if (ST.c1 == CHRTEST_UNINIT) {
6305 /* calculate c1 and c2 for possible match of 1st char
6306 * following curly */
6307 ST.c1 = ST.c2 = CHRTEST_VOID;
d20a21f4 6308 assert(ST.B);
40a82448
DM		 6309 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
6310 regnode *text_node = ST.B;
6311 if (! HAS_TEXT(text_node))
6312 FIND_NEXT_IMPT(text_node);
ee9b8eae
YO		 6313 /* this used to be
6314
6315 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
6316
6317 But the former is redundant in light of the latter.
6318
6319 if this changes back then the macro for
6320 IS_TEXT and friends need to change.
6321 */
c74f6de9 6322 if (PL_regkind[OP(text_node)] == EXACT) {
79a2a0e8 6323 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
984e6dd1 6324 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
aed7b151 6325 reginfo))
c74f6de9
KW		 6326 {
6327 sayNO;
6328 }
c277df42 6329 }
c277df42 6330 }
40a82448
DM		 6331 }
6332
6333 DEBUG_EXECUTE_r(
6334 PerlIO_printf(Perl_debug_log,
6335 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
5bc10b2c 6336 (int)(REPORT_CODE_OFF+(depth*2)),
40a82448
DM		 6337 "", (IV)ST.count)
6338 );
c74f6de9 6339 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
79a2a0e8
KW		 6340 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
6341 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6342 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6343 {
6344 /* simulate B failing */
6345 DEBUG_OPTIMISE_r(
6346 PerlIO_printf(Perl_debug_log,
33daa3a5 6347 "%*s CURLYM Fast bail next target=0x%"UVXf" c1=0x%"UVXf" c2=0x%"UVXf"\n",
79a2a0e8
KW		 6348 (int)(REPORT_CODE_OFF+(depth*2)),"",
6349 valid_utf8_to_uvchr((U8 *) locinput, NULL),
6350 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
6351 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
6352 );
6353 state_num = CURLYM_B_fail;
6354 goto reenter_switch;
6355 }
6356 }
6357 else if (nextchr != ST.c1 && nextchr != ST.c2) {
5400f398
KW		 6358 /* simulate B failing */
6359 DEBUG_OPTIMISE_r(
6360 PerlIO_printf(Perl_debug_log,
33daa3a5 6361 "%*s CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
5400f398 6362 (int)(REPORT_CODE_OFF+(depth*2)),"",
79a2a0e8
KW		 6363 (int) nextchr, ST.c1, ST.c2)
6364 );
5400f398
KW		 6365 state_num = CURLYM_B_fail;
6366 goto reenter_switch;
6367 }
c74f6de9 6368 }
40a82448
DM		 6369
6370 if (ST.me->flags) {
f6033a9d 6371 /* emulate CLOSE: mark current A as captured */
40a82448
DM		 6372 I32 paren = ST.me->flags;
6373 if (ST.count) {
b93070ed 6374 rex->offs[paren].start
9d9163fb
DM		 6375 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
6376 rex->offs[paren].end = locinput - reginfo->strbeg;
f6033a9d
DM		 6377 if ((U32)paren > rex->lastparen)
6378 rex->lastparen = paren;
6379 rex->lastcloseparen = paren;
c277df42 6380 }
40a82448 6381 else
b93070ed 6382 rex->offs[paren].end = -1;
0a4db386 6383 if (cur_eval && cur_eval->u.eval.close_paren &&
24b23f37 6384 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
0a4db386
YO		 6385 {
6386 if (ST.count)
6387 goto fake_end;
6388 else
6389 sayNO;
6390 }
c277df42 6391 }
0a4db386 6392
4d5016e5 6393 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
a74ff37d 6394 /* NOTREACHED */
e5964223 6395 NOT_REACHED;
40a82448
DM		 6396
6397 case CURLYM_B_fail: /* just failed to match a B */
6398 REGCP_UNWIND(ST.cp);
a8d1f4b4 6399 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
40a82448 6400 if (ST.minmod) {
84d2fa14
HS		 6401 I32 max = ARG2(ST.me);
6402 if (max != REG_INFTY && ST.count == max)
40a82448
DM		 6403 sayNO;
6404 goto curlym_do_A; /* try to match a further A */
6405 }
6406 /* backtrack one A */
6407 if (ST.count == ARG1(ST.me) /* min */)
6408 sayNO;
6409 ST.count--;
7016d6eb 6410 SET_locinput(HOPc(locinput, -ST.alen));
40a82448
DM		 6411 goto curlym_do_B; /* try to match B */
6412
c255a977
DM		 6413#undef ST
6414#define ST st->u.curly
40a82448 6415
c255a977
DM		 6416#define CURLY_SETPAREN(paren, success) \
6417 if (paren) { \
6418 if (success) { \
9d9163fb
DM		 6419 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
6420 rex->offs[paren].end = locinput - reginfo->strbeg; \
f6033a9d
DM		 6421 if (paren > rex->lastparen) \
6422 rex->lastparen = paren; \
b93070ed 6423 rex->lastcloseparen = paren; \
c255a977 6424 } \
f6033a9d 6425 else { \
b93070ed 6426 rex->offs[paren].end = -1; \
f6033a9d
DM		 6427 rex->lastparen = ST.lastparen; \
6428 rex->lastcloseparen = ST.lastcloseparen; \
6429 } \
c255a977
DM		 6430 }
6431
b40a2c17 6432 case STAR: /* /A*B/ where A is width 1 char */
c255a977
DM		 6433 ST.paren = 0;
6434 ST.min = 0;
6435 ST.max = REG_INFTY;
a0d0e21e
LW		 6436 scan = NEXTOPER(scan);
6437 goto repeat;
3c0563b9 6438
b40a2c17 6439 case PLUS: /* /A+B/ where A is width 1 char */
c255a977
DM		 6440 ST.paren = 0;
6441 ST.min = 1;
6442 ST.max = REG_INFTY;
c277df42 6443 scan = NEXTOPER(scan);
c255a977 6444 goto repeat;
3c0563b9 6445
b40a2c17 6446 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
5400f398
KW		 6447 ST.paren = scan->flags; /* Which paren to set */
6448 ST.lastparen = rex->lastparen;
f6033a9d 6449 ST.lastcloseparen = rex->lastcloseparen;
92da3157
DM		 6450 if (ST.paren > maxopenparen)
6451 maxopenparen = ST.paren;
c255a977
DM		 6452 ST.min = ARG1(scan); /* min to match */
6453 ST.max = ARG2(scan); /* max to match */
0a4db386 6454 if (cur_eval && cur_eval->u.eval.close_paren &&
86413ec0 6455 cur_eval->u.eval.close_paren == (U32)ST.paren) {
0a4db386
YO		 6456 ST.min=1;
6457 ST.max=1;
6458 }
c255a977
DM		 6459 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
6460 goto repeat;
3c0563b9 6461
b40a2c17 6462 case CURLY: /* /A{m,n}B/ where A is width 1 char */
c255a977
DM		 6463 ST.paren = 0;
6464 ST.min = ARG1(scan); /* min to match */
6465 ST.max = ARG2(scan); /* max to match */
6466 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
c277df42 6467 repeat:
a0d0e21e
LW		 6468 /*
6469 * Lookahead to avoid useless match attempts
6470 * when we know what character comes next.
c255a977 6471 *
5f80c4cf
JP		 6472 * Used to only do .*x and .*?x, but now it allows
6473 * for )'s, ('s and (?{ ... })'s to be in the way
6474 * of the quantifier and the EXACT-like node. -- japhy
6475 */
6476
eb5c1be8 6477 assert(ST.min <= ST.max);
3337dfe3
KW		 6478 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
6479 ST.c1 = ST.c2 = CHRTEST_VOID;
6480 }
6481 else {
5f80c4cf
JP		 6482 regnode *text_node = next;
6483
3dab1dad
YO		 6484 if (! HAS_TEXT(text_node))
6485 FIND_NEXT_IMPT(text_node);
5f80c4cf 6486
9e137952 6487 if (! HAS_TEXT(text_node))
c255a977 6488 ST.c1 = ST.c2 = CHRTEST_VOID;
5f80c4cf 6489 else {
ee9b8eae 6490 if ( PL_regkind[OP(text_node)] != EXACT ) {
c255a977 6491 ST.c1 = ST.c2 = CHRTEST_VOID;
cca55fe3 6492 }
c74f6de9 6493 else {
ee9b8eae
YO		 6494
6495 /* Currently we only get here when
6496
6497 PL_rekind[OP(text_node)] == EXACT
6498
6499 if this changes back then the macro for IS_TEXT and
6500 friends need to change. */
79a2a0e8 6501 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
984e6dd1 6502 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
aed7b151 6503 reginfo))
c74f6de9
KW		 6504 {
6505 sayNO;
6506 }
6507 }
1aa99e6b 6508 }
bbce6d69 6509 }
c255a977
DM		 6510
6511 ST.A = scan;
6512 ST.B = next;
24d3c4a9 6513 if (minmod) {
eb72505d 6514 char *li = locinput;
24d3c4a9 6515 minmod = 0;
984e6dd1 6516 if (ST.min &&
f9176b44 6517 regrepeat(rex, &li, ST.A, reginfo, ST.min, depth)
984e6dd1 6518 < ST.min)
4633a7c4 6519 sayNO;
7016d6eb 6520 SET_locinput(li);
c255a977 6521 ST.count = ST.min;
c255a977
DM		 6522 REGCP_SET(ST.cp);
6523 if (ST.c1 == CHRTEST_VOID)
6524 goto curly_try_B_min;
6525
6526 ST.oldloc = locinput;
6527
6528 /* set ST.maxpos to the furthest point along the
6529 * string that could possibly match */
6530 if (ST.max == REG_INFTY) {
220db18a 6531 ST.maxpos = reginfo->strend - 1;
f2ed9b32 6532 if (utf8_target)
c255a977
DM		 6533 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
6534 ST.maxpos--;
6535 }
f2ed9b32 6536 else if (utf8_target) {
c255a977
DM		 6537 int m = ST.max - ST.min;
6538 for (ST.maxpos = locinput;
220db18a 6539 m >0 && ST.maxpos < reginfo->strend; m--)
c255a977
DM		 6540 ST.maxpos += UTF8SKIP(ST.maxpos);
6541 }
6542 else {
6543 ST.maxpos = locinput + ST.max - ST.min;
220db18a
DM		 6544 if (ST.maxpos >= reginfo->strend)
6545 ST.maxpos = reginfo->strend - 1;
c255a977
DM		 6546 }
6547 goto curly_try_B_min_known;
6548
6549 }
6550 else {
eb72505d
DM		 6551 /* avoid taking address of locinput, so it can remain
6552 * a register var */
6553 char *li = locinput;
f9176b44 6554 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth);
c255a977
DM		 6555 if (ST.count < ST.min)
6556 sayNO;
7016d6eb 6557 SET_locinput(li);
c255a977
DM		 6558 if ((ST.count > ST.min)
6559 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
6560 {
6561 /* A{m,n} must come at the end of the string, there's
6562 * no point in backing off ... */
6563 ST.min = ST.count;
6564 /* ...except that $ and \Z can match before *and* after
6565 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
6566 We may back off by one in this case. */
eb72505d 6567 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
c255a977
DM		 6568 ST.min--;
6569 }
6570 REGCP_SET(ST.cp);
6571 goto curly_try_B_max;
6572 }
a74ff37d 6573 /* NOTREACHED */
e5964223 6574 NOT_REACHED;
c255a977
DM		 6575
6576 case CURLY_B_min_known_fail:
6577 /* failed to find B in a non-greedy match where c1,c2 valid */
c255a977 6578
c255a977 6579 REGCP_UNWIND(ST.cp);
a8d1f4b4
DM		 6580 if (ST.paren) {
6581 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6582 }
c255a977
DM		 6583 /* Couldn't or didn't -- move forward. */
6584 ST.oldloc = locinput;
f2ed9b32 6585 if (utf8_target)
c255a977
DM		 6586 locinput += UTF8SKIP(locinput);
6587 else
6588 locinput++;
6589 ST.count++;
6590 curly_try_B_min_known:
6591 /* find the next place where 'B' could work, then call B */
6592 {
6593 int n;
f2ed9b32 6594 if (utf8_target) {
c255a977
DM		 6595 n = (ST.oldloc == locinput) ? 0 : 1;
6596 if (ST.c1 == ST.c2) {
c255a977 6597 /* set n to utf8_distance(oldloc, locinput) */
79a2a0e8
KW		 6598 while (locinput <= ST.maxpos
6599 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
6600 {
6601 locinput += UTF8SKIP(locinput);
c255a977
DM		 6602 n++;
6603 }
1aa99e6b
IH		 6604 }
6605 else {
c255a977 6606 /* set n to utf8_distance(oldloc, locinput) */
79a2a0e8
KW		 6607 while (locinput <= ST.maxpos
6608 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6609 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6610 {
6611 locinput += UTF8SKIP(locinput);
c255a977 6612 n++;
1aa99e6b 6613 }
0fe9bf95
IZ		 6614 }
6615 }
5400f398 6616 else { /* Not utf8_target */
c255a977
DM		 6617 if (ST.c1 == ST.c2) {
6618 while (locinput <= ST.maxpos &&
6619 UCHARAT(locinput) != ST.c1)
6620 locinput++;
bbce6d69 6621 }
c255a977
DM		 6622 else {
6623 while (locinput <= ST.maxpos
6624 && UCHARAT(locinput) != ST.c1
6625 && UCHARAT(locinput) != ST.c2)
6626 locinput++;
a0ed51b3 6627 }
c255a977
DM		 6628 n = locinput - ST.oldloc;
6629 }
6630 if (locinput > ST.maxpos)
6631 sayNO;
c255a977 6632 if (n) {
eb72505d
DM		 6633 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
6634 * at b; check that everything between oldloc and
6635 * locinput matches */
6636 char *li = ST.oldloc;
c255a977 6637 ST.count += n;
f9176b44 6638 if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n)
4633a7c4 6639 sayNO;
eb72505d 6640 assert(n == REG_INFTY || locinput == li);
a0d0e21e 6641 }
c255a977 6642 CURLY_SETPAREN(ST.paren, ST.count);
0a4db386 6643 if (cur_eval && cur_eval->u.eval.close_paren &&
86413ec0 6644 cur_eval->u.eval.close_paren == (U32)ST.paren) {
0a4db386
YO		 6645 goto fake_end;
6646 }
4d5016e5 6647 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
a0d0e21e 6648 }
a74ff37d 6649 /* NOTREACHED */
e5964223 6650 NOT_REACHED;
c255a977
DM		 6651
6652 case CURLY_B_min_fail:
6653 /* failed to find B in a non-greedy match where c1,c2 invalid */
c255a977
DM		 6654
6655 REGCP_UNWIND(ST.cp);
a8d1f4b4
DM		 6656 if (ST.paren) {
6657 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6658 }
c255a977 6659 /* failed -- move forward one */
f73aaa43 6660 {
eb72505d 6661 char *li = locinput;
f9176b44 6662 if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) {
f73aaa43
DM		 6663 sayNO;
6664 }
eb72505d 6665 locinput = li;
f73aaa43
DM		 6666 }
6667 {
c255a977 6668 ST.count++;
c255a977
DM		 6669 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
6670 ST.count > 0)) /* count overflow ? */
15272685 6671 {
c255a977
DM		 6672 curly_try_B_min:
6673 CURLY_SETPAREN(ST.paren, ST.count);
0a4db386 6674 if (cur_eval && cur_eval->u.eval.close_paren &&
86413ec0 6675 cur_eval->u.eval.close_paren == (U32)ST.paren) {
0a4db386
YO		 6676 goto fake_end;
6677 }
4d5016e5 6678 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
a0d0e21e
LW		 6679 }
6680 }
c74f6de9 6681 sayNO;
a74ff37d 6682 /* NOTREACHED */
e5964223 6683 NOT_REACHED;
c255a977 6684
c52b8b12 6685 curly_try_B_max:
c255a977 6686 /* a successful greedy match: now try to match B */
40d049e4 6687 if (cur_eval && cur_eval->u.eval.close_paren &&
86413ec0 6688 cur_eval->u.eval.close_paren == (U32)ST.paren) {
40d049e4
YO		 6689 goto fake_end;
6690 }
c255a977 6691 {
220db18a 6692 bool could_match = locinput < reginfo->strend;
79a2a0e8 6693
c255a977 6694 /* If it could work, try it. */
79a2a0e8
KW		 6695 if (ST.c1 != CHRTEST_VOID && could_match) {
6696 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
6697 {
6698 could_match = memEQ(locinput,
6699 ST.c1_utf8,
6700 UTF8SKIP(locinput))
6701 || memEQ(locinput,
6702 ST.c2_utf8,
6703 UTF8SKIP(locinput));
6704 }
6705 else {
6706 could_match = UCHARAT(locinput) == ST.c1
6707 || UCHARAT(locinput) == ST.c2;
6708 }
6709 }
6710 if (ST.c1 == CHRTEST_VOID || could_match) {
c255a977 6711 CURLY_SETPAREN(ST.paren, ST.count);
4d5016e5 6712 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
a74ff37d 6713 /* NOTREACHED */
e5964223 6714 NOT_REACHED;
c255a977
DM		 6715 }
6716 }
924ba076 6717 /* FALLTHROUGH */
3c0563b9 6718
c255a977
DM		 6719 case CURLY_B_max_fail:
6720 /* failed to find B in a greedy match */
c255a977
DM		 6721
6722 REGCP_UNWIND(ST.cp);
a8d1f4b4
DM		 6723 if (ST.paren) {
6724 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6725 }
c255a977
DM		 6726 /* back up. */
6727 if (--ST.count < ST.min)
6728 sayNO;
eb72505d 6729 locinput = HOPc(locinput, -1);
c255a977
DM		 6730 goto curly_try_B_max;
6731
6732#undef ST
6733
3c0563b9 6734 case END: /* last op of main pattern */
c52b8b12 6735 fake_end:
faec1544
DM		 6736 if (cur_eval) {
6737 /* we've just finished A in /(??{A})B/; now continue with B */
faec1544 6738
288b8c02 6739 st->u.eval.prev_rex = rex_sv; /* inner */
92da3157
DM		 6740
6741 /* Save *all* the positions. */
6742 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
ec43f78b 6743 rex_sv = cur_eval->u.eval.prev_rex;
aed7b151 6744 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
ec43f78b 6745 SET_reg_curpm(rex_sv);
8d919b0a 6746 rex = ReANY(rex_sv);
f8fc2ecf 6747 rexi = RXi_GET(rex);
faec1544 6748 cur_curlyx = cur_eval->u.eval.prev_curlyx;
34a81e2b 6749
faec1544 6750 REGCP_SET(st->u.eval.lastcp);
faec1544
DM		 6751
6752 /* Restore parens of the outer rex without popping the
6753 * savestack */
92da3157
DM		 6754 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp,
6755 &maxopenparen);
faec1544
DM		 6756
6757 st->u.eval.prev_eval = cur_eval;
6758 cur_eval = cur_eval->u.eval.prev_eval;
6759 DEBUG_EXECUTE_r(
2a49f0f5
JH		 6760 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
6761 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
e7707071
YO		 6762 if ( nochange_depth )
6763 nochange_depth--;
6764
4d5016e5
DM		 6765 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
6766 locinput); /* match B */
faec1544
DM		 6767 }
6768
3b0527fe 6769 if (locinput < reginfo->till) {
a3621e74 6770 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
7821416a
IZ		 6771 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6772 PL_colors[4],
6d59b646
DM		 6773 (long)(locinput - startpos),
6774 (long)(reginfo->till - startpos),
7821416a 6775 PL_colors[5]));
58e23c8d 6776
262b90c4 6777 sayNO_SILENT; /* Cannot match: too short. */
7821416a 6778 }
262b90c4 6779 sayYES; /* Success! */
dad79028
DM		 6780
6781 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6782 DEBUG_EXECUTE_r(
6783 PerlIO_printf(Perl_debug_log,
6784 "%*s %ssubpattern success...%s\n",
5bc10b2c 6785 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
262b90c4 6786 sayYES; /* Success! */
dad79028 6787
40a82448
DM		 6788#undef ST
6789#define ST st->u.ifmatch
6790
37f53970
DM		 6791 {
6792 char *newstart;
6793
40a82448
DM		 6794 case SUSPEND: /* (?>A) */
6795 ST.wanted = 1;
37f53970 6796 newstart = locinput;
9041c2e3 6797 goto do_ifmatch;
dad79028 6798
40a82448
DM		 6799 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6800 ST.wanted = 0;
dad79028
DM		 6801 goto ifmatch_trivial_fail_test;
6802
40a82448
DM		 6803 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6804 ST.wanted = 1;
dad79028 6805 ifmatch_trivial_fail_test:
a0ed51b3 6806 if (scan->flags) {
52657f30 6807 char * const s = HOPBACKc(locinput, scan->flags);
dad79028
DM		 6808 if (!s) {
6809 /* trivial fail */
24d3c4a9
DM		 6810 if (logical) {
6811 logical = 0;
f2338a2e 6812 sw = 1 - cBOOL(ST.wanted);
dad79028 6813 }
40a82448 6814 else if (ST.wanted)
dad79028
DM		 6815 sayNO;
6816 next = scan + ARG(scan);
6817 if (next == scan)
6818 next = NULL;
6819 break;
6820 }
37f53970 6821 newstart = s;
a0ed51b3
LW		 6822 }
6823 else
37f53970 6824 newstart = locinput;
a0ed51b3 6825
c277df42 6826 do_ifmatch:
40a82448 6827 ST.me = scan;
24d3c4a9 6828 ST.logical = logical;
24d786f4
YO		 6829 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6830
40a82448 6831 /* execute body of (?...A) */
37f53970 6832 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
a74ff37d 6833 /* NOTREACHED */
e5964223 6834 NOT_REACHED;
37f53970 6835 }
40a82448
DM		 6836
6837 case IFMATCH_A_fail: /* body of (?...A) failed */
6838 ST.wanted = !ST.wanted;
924ba076 6839 /* FALLTHROUGH */
40a82448
DM		 6840
6841 case IFMATCH_A: /* body of (?...A) succeeded */
24d3c4a9 6842 if (ST.logical) {
f2338a2e 6843 sw = cBOOL(ST.wanted);
40a82448
DM		 6844 }
6845 else if (!ST.wanted)
6846 sayNO;
6847
37f53970
DM		 6848 if (OP(ST.me) != SUSPEND) {
6849 /* restore old position except for (?>...) */
6850 locinput = st->locinput;
40a82448
DM		 6851 }
6852 scan = ST.me + ARG(ST.me);
6853 if (scan == ST.me)
6854 scan = NULL;
6855 continue; /* execute B */
6856
6857#undef ST
dad79028 6858
3c0563b9
DM		 6859 case LONGJMP: /* alternative with many branches compiles to
6860 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
c277df42
IZ		 6861 next = scan + ARG(scan);
6862 if (next == scan)
6863 next = NULL;
a0d0e21e 6864 break;
3c0563b9
DM		 6865
6866 case COMMIT: /* (*COMMIT) */
220db18a 6867 reginfo->cutpoint = reginfo->strend;
e2e6a0f1 6868 /* FALLTHROUGH */
3c0563b9
DM		 6869
6870 case PRUNE: /* (*PRUNE) */
e2e6a0f1 6871 if (!scan->flags)
ad64d0ec 6872 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4d5016e5 6873 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
a74ff37d 6874 /* NOTREACHED */
e5964223 6875 NOT_REACHED;
3c0563b9 6876
54612592
YO		 6877 case COMMIT_next_fail:
6878 no_final = 1;
6879 /* FALLTHROUGH */
3c0563b9
DM		 6880
6881 case OPFAIL: /* (*FAIL) */
7f69552c 6882 sayNO;
a74ff37d 6883 /* NOTREACHED */
e5964223 6884 NOT_REACHED;
e2e6a0f1
YO		 6885
6886#define ST st->u.mark
3c0563b9 6887 case MARKPOINT: /* (*MARK:foo) */
e2e6a0f1 6888 ST.prev_mark = mark_state;
5d458dd8 6889 ST.mark_name = sv_commit = sv_yes_mark
ad64d0ec 6890 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
e2e6a0f1 6891 mark_state = st;
4d5016e5
DM		 6892 ST.mark_loc = locinput;
6893 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
a74ff37d 6894 /* NOTREACHED */
e5964223 6895 NOT_REACHED;
3c0563b9 6896
e2e6a0f1
YO		 6897 case MARKPOINT_next:
6898 mark_state = ST.prev_mark;
6899 sayYES;
a74ff37d 6900 /* NOTREACHED */
e5964223 6901 NOT_REACHED;
3c0563b9 6902
e2e6a0f1 6903 case MARKPOINT_next_fail:
5d458dd8 6904 if (popmark && sv_eq(ST.mark_name,popmark))
e2e6a0f1
YO		 6905 {
6906 if (ST.mark_loc > startpoint)
6907 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6908 popmark = NULL; /* we found our mark */
6909 sv_commit = ST.mark_name;
6910
6911 DEBUG_EXECUTE_r({
5d458dd8 6912 PerlIO_printf(Perl_debug_log,
e2e6a0f1
YO		 6913 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6914 REPORT_CODE_OFF+depth*2, "",
be2597df 6915 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
e2e6a0f1
YO		 6916 });
6917 }
6918 mark_state = ST.prev_mark;
5d458dd8
YO		 6919 sv_yes_mark = mark_state ?
6920 mark_state->u.mark.mark_name : NULL;
e2e6a0f1 6921 sayNO;
a74ff37d 6922 /* NOTREACHED */
e5964223 6923 NOT_REACHED;
3c0563b9
DM		 6924
6925 case SKIP: /* (*SKIP) */
5d458dd8 6926 if (scan->flags) {
2bf803e2 6927 /* (*SKIP) : if we fail we cut here*/
5d458dd8 6928 ST.mark_name = NULL;
e2e6a0f1 6929 ST.mark_loc = locinput;
4d5016e5 6930 PUSH_STATE_GOTO(SKIP_next,next, locinput);
5d458dd8 6931 } else {
2bf803e2 6932 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
5d458dd8
YO		 6933 otherwise do nothing. Meaning we need to scan
6934 */
6935 regmatch_state *cur = mark_state;
ad64d0ec 6936 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
5d458dd8
YO		 6937
6938 while (cur) {
6939 if ( sv_eq( cur->u.mark.mark_name,
6940 find ) )
6941 {
6942 ST.mark_name = find;
4d5016e5 6943 PUSH_STATE_GOTO( SKIP_next, next, locinput);
5d458dd8
YO		 6944 }
6945 cur = cur->u.mark.prev_mark;
6946 }
e2e6a0f1 6947 }
2bf803e2 6948 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
5d458dd8 6949 break;
3c0563b9 6950
5d458dd8
YO		 6951 case SKIP_next_fail:
6952 if (ST.mark_name) {
6953 /* (*CUT:NAME) - Set up to search for the name as we
6954 collapse the stack*/
6955 popmark = ST.mark_name;
6956 } else {
6957 /* (*CUT) - No name, we cut here.*/
e2e6a0f1
YO		 6958 if (ST.mark_loc > startpoint)
6959 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
5d458dd8
YO		 6960 /* but we set sv_commit to latest mark_name if there
6961 is one so they can test to see how things lead to this
6962 cut */
6963 if (mark_state)
6964 sv_commit=mark_state->u.mark.mark_name;
6965 }
e2e6a0f1
YO		 6966 no_final = 1;
6967 sayNO;
a74ff37d 6968 /* NOTREACHED */
e5964223 6969 NOT_REACHED;
e2e6a0f1 6970#undef ST
3c0563b9
DM		 6971
6972 case LNBREAK: /* \R */
220db18a 6973 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
e1d1eefb 6974 locinput += n;
e1d1eefb
YO		 6975 } else
6976 sayNO;
6977 break;
6978
a0d0e21e 6979 default:
b900a521 6980 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
d7d93a81 6981 PTR2UV(scan), OP(scan));
cea2e8a9 6982 Perl_croak(aTHX_ "regexp memory corruption");
28b98f76
DM		 6983
6984 /* this is a point to jump to in order to increment
6985 * locinput by one character */
c52b8b12 6986 increment_locinput:
e6ca698c 6987 assert(!NEXTCHR_IS_EOS);
28b98f76
DM		 6988 if (utf8_target) {
6989 locinput += PL_utf8skip[nextchr];
7016d6eb 6990 /* locinput is allowed to go 1 char off the end, but not 2+ */
220db18a 6991 if (locinput > reginfo->strend)
28b98f76 6992 sayNO;
28b98f76
DM		 6993 }
6994 else
3640db6b 6995 locinput++;
28b98f76 6996 break;
5d458dd8
YO		 6997
6998 } /* end switch */
95b24440 6999
5d458dd8
YO		 7000 /* switch break jumps here */
7001 scan = next; /* prepare to execute the next op and ... */
7002 continue; /* ... jump back to the top, reusing st */
a74ff37d 7003 /* NOTREACHED */
95b24440 7004
40a82448
DM		 7005 push_yes_state:
7006 /* push a state that backtracks on success */
7007 st->u.yes.prev_yes_state = yes_state;
7008 yes_state = st;
924ba076 7009 /* FALLTHROUGH */
40a82448
DM		 7010 push_state:
7011 /* push a new regex state, then continue at scan */
7012 {
7013 regmatch_state *newst;
7014
24b23f37
YO		 7015 DEBUG_STACK_r({
7016 regmatch_state *cur = st;
7017 regmatch_state *curyes = yes_state;
7018 int curd = depth;
7019 regmatch_slab *slab = PL_regmatch_slab;
7020 for (;curd > -1;cur--,curd--) {
7021 if (cur < SLAB_FIRST(slab)) {
7022 slab = slab->prev;
7023 cur = SLAB_LAST(slab);
7024 }
7025 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
7026 REPORT_CODE_OFF + 2 + depth * 2,"",
13d6edb4 7027 curd, PL_reg_name[cur->resume_state],
24b23f37
YO		 7028 (curyes == cur) ? "yes" : ""
7029 );
7030 if (curyes == cur)
7031 curyes = cur->u.yes.prev_yes_state;
7032 }
7033 } else
7034 DEBUG_STATE_pp("push")
7035 );
40a82448 7036 depth++;
40a82448
DM		 7037 st->locinput = locinput;
7038 newst = st+1;
7039 if (newst > SLAB_LAST(PL_regmatch_slab))
7040 newst = S_push_slab(aTHX);
7041 PL_regmatch_state = newst;
786e8c11 7042
4d5016e5 7043 locinput = pushinput;
40a82448
DM		 7044 st = newst;
7045 continue;
a74ff37d 7046 /* NOTREACHED */
40a82448 7047 }
a0d0e21e 7048 }
a687059c 7049
a0d0e21e
LW		 7050 /*
7051 * We get here only if there's trouble -- normally "case END" is
7052 * the terminating point.
7053 */
cea2e8a9 7054 Perl_croak(aTHX_ "corrupted regexp pointers");
a74ff37d 7055 /* NOTREACHED */
4633a7c4 7056 sayNO;
e5964223 7057 NOT_REACHED;
4633a7c4 7058
7b52d656 7059 yes:
77cb431f
DM		 7060 if (yes_state) {
7061 /* we have successfully completed a subexpression, but we must now
7062 * pop to the state marked by yes_state and continue from there */
77cb431f 7063 assert(st != yes_state);
5bc10b2c
DM		 7064#ifdef DEBUGGING
7065 while (st != yes_state) {
7066 st--;
7067 if (st < SLAB_FIRST(PL_regmatch_slab)) {
7068 PL_regmatch_slab = PL_regmatch_slab->prev;
7069 st = SLAB_LAST(PL_regmatch_slab);
7070 }
e2e6a0f1 7071 DEBUG_STATE_r({
54612592
YO		 7072 if (no_final) {
7073 DEBUG_STATE_pp("pop (no final)");
7074 } else {
7075 DEBUG_STATE_pp("pop (yes)");
7076 }
e2e6a0f1 7077 });
5bc10b2c
DM		 7078 depth--;
7079 }
7080#else
77cb431f
DM		 7081 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
7082 || yes_state > SLAB_LAST(PL_regmatch_slab))
7083 {
7084 /* not in this slab, pop slab */
7085 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
7086 PL_regmatch_slab = PL_regmatch_slab->prev;
7087 st = SLAB_LAST(PL_regmatch_slab);
7088 }
7089 depth -= (st - yes_state);
5bc10b2c 7090#endif
77cb431f
DM		 7091 st = yes_state;
7092 yes_state = st->u.yes.prev_yes_state;
7093 PL_regmatch_state = st;
24b23f37 7094
3640db6b 7095 if (no_final)
5d458dd8 7096 locinput= st->locinput;
54612592 7097 state_num = st->resume_state + no_final;
24d3c4a9 7098 goto reenter_switch;
77cb431f
DM		 7099 }
7100
a3621e74 7101 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
e4584336 7102 PL_colors[4], PL_colors[5]));
02db2b7b 7103
bf2039a9 7104 if (reginfo->info_aux_eval) {
19b95bf0
DM		 7105 /* each successfully executed (?{...}) block does the equivalent of
7106 * local $^R = do {...}
7107 * When popping the save stack, all these locals would be undone;
7108 * bypass this by setting the outermost saved $^R to the latest
7109 * value */
4b22688e
YO		 7110 /* I dont know if this is needed or works properly now.
7111 * see code related to PL_replgv elsewhere in this file.
7112 * Yves
7113 */
19b95bf0
DM		 7114 if (oreplsv != GvSV(PL_replgv))
7115 sv_setsv(oreplsv, GvSV(PL_replgv));
7116 }
95b24440 7117 result = 1;
aa283a38 7118 goto final_exit;
4633a7c4 7119
7b52d656 7120 no:
a3621e74 7121 DEBUG_EXECUTE_r(
7821416a 7122 PerlIO_printf(Perl_debug_log,
786e8c11 7123 "%*s %sfailed...%s\n",
5bc10b2c 7124 REPORT_CODE_OFF+depth*2, "",
786e8c11 7125 PL_colors[4], PL_colors[5])
7821416a 7126 );
aa283a38 7127
7b52d656 7128 no_silent:
54612592
YO		 7129 if (no_final) {
7130 if (yes_state) {
7131 goto yes;
7132 } else {
7133 goto final_exit;
7134 }
7135 }
aa283a38
DM		 7136 if (depth) {
7137 /* there's a previous state to backtrack to */
40a82448
DM		 7138 st--;
7139 if (st < SLAB_FIRST(PL_regmatch_slab)) {
7140 PL_regmatch_slab = PL_regmatch_slab->prev;
7141 st = SLAB_LAST(PL_regmatch_slab);
7142 }
7143 PL_regmatch_state = st;
40a82448 7144 locinput= st->locinput;
40a82448 7145
5bc10b2c
DM		 7146 DEBUG_STATE_pp("pop");
7147 depth--;
262b90c4
DM		 7148 if (yes_state == st)
7149 yes_state = st->u.yes.prev_yes_state;
5bc10b2c 7150
24d3c4a9
DM		 7151 state_num = st->resume_state + 1; /* failure = success + 1 */
7152 goto reenter_switch;
95b24440 7153 }
24d3c4a9 7154 result = 0;
aa283a38 7155
262b90c4 7156 final_exit:
bbe252da 7157 if (rex->intflags & PREGf_VERBARG_SEEN) {
5d458dd8
YO		 7158 SV *sv_err = get_sv("REGERROR", 1);
7159 SV *sv_mrk = get_sv("REGMARK", 1);
7160 if (result) {
e2e6a0f1 7161 sv_commit = &PL_sv_no;
5d458dd8
YO		 7162 if (!sv_yes_mark)
7163 sv_yes_mark = &PL_sv_yes;
7164 } else {
7165 if (!sv_commit)
7166 sv_commit = &PL_sv_yes;
7167 sv_yes_mark = &PL_sv_no;
7168 }
316ebaf2
JH		 7169 assert(sv_err);
7170 assert(sv_mrk);
5d458dd8
YO		 7171 sv_setsv(sv_err, sv_commit);
7172 sv_setsv(sv_mrk, sv_yes_mark);
e2e6a0f1 7173 }
19b95bf0 7174
81ed78b2
DM		 7175
7176 if (last_pushed_cv) {
7177 dSP;
7178 POP_MULTICALL;
4f8dbb2d 7179 PERL_UNUSED_VAR(SP);
81ed78b2
DM		 7180 }
7181
9d9163fb
DM		 7182 assert(!result || locinput - reginfo->strbeg >= 0);
7183 return result ? locinput - reginfo->strbeg : -1;
a687059c
LW		 7184}
7185
7186/*
7187 - regrepeat - repeatedly match something simple, report how many
d60de1d1 7188 *
e64f369d
KW		 7189 * What 'simple' means is a node which can be the operand of a quantifier like
7190 * '+', or {1,3}
7191 *
d60de1d1
DM		 7192 * startposp - pointer a pointer to the start position. This is updated
7193 * to point to the byte following the highest successful
7194 * match.
7195 * p - the regnode to be repeatedly matched against.
220db18a 7196 * reginfo - struct holding match state, such as strend
4063ade8 7197 * max - maximum number of things to match.
d60de1d1 7198 * depth - (for debugging) backtracking depth.
a687059c 7199 */
76e3520e 7200STATIC I32
272d35c9 7201S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
f9176b44 7202 regmatch_info *const reginfo, I32 max, int depth)
a687059c 7203{
4063ade8 7204 char *scan; /* Pointer to current position in target string */
eb578fdb 7205 I32 c;
220db18a 7206 char *loceol = reginfo->strend; /* local version */
4063ade8 7207 I32 hardcount = 0; /* How many matches so far */
ba44c216 7208 bool utf8_target = reginfo->is_utf8_target;
3018b823 7209 int to_complement = 0; /* Invert the result? */
d513472c 7210 UV utf8_flags;
3018b823 7211 _char_class_number classnum;
4f55667c
SP		 7212#ifndef DEBUGGING
7213 PERL_UNUSED_ARG(depth);
7214#endif
a0d0e21e 7215
7918f24d
NC		 7216 PERL_ARGS_ASSERT_REGREPEAT;
7217
f73aaa43 7218 scan = *startposp;
faf11cac
HS		 7219 if (max == REG_INFTY)
7220 max = I32_MAX;
dfb8f192 7221 else if (! utf8_target && loceol - scan > max)
7f596f4c 7222 loceol = scan + max;
4063ade8
KW		 7223
7224 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
7225 * to the maximum of how far we should go in it (leaving it set to the real
7226 * end, if the maximum permissible would take us beyond that). This allows
7227 * us to make the loop exit condition that we haven't gone past <loceol> to
7228 * also mean that we haven't exceeded the max permissible count, saving a
7229 * test each time through the loop. But it assumes that the OP matches a
7230 * single byte, which is true for most of the OPs below when applied to a
7231 * non-UTF-8 target. Those relatively few OPs that don't have this
7232 * characteristic will have to compensate.
7233 *
7234 * There is no adjustment for UTF-8 targets, as the number of bytes per
7235 * character varies. OPs will have to test both that the count is less
7236 * than the max permissible (using <hardcount> to keep track), and that we
7237 * are still within the bounds of the string (using <loceol>. A few OPs
7238 * match a single byte no matter what the encoding. They can omit the max
7239 * test if, for the UTF-8 case, they do the adjustment that was skipped
7240 * above.
7241 *
7242 * Thus, the code above sets things up for the common case; and exceptional
7243 * cases need extra work; the common case is to make sure <scan> doesn't
7244 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
7245 * count doesn't exceed the maximum permissible */
7246
a0d0e21e 7247 switch (OP(p)) {
22c35a8c 7248 case REG_ANY:
f2ed9b32 7249 if (utf8_target) {
1aa99e6b 7250 while (scan < loceol && hardcount < max && *scan != '\n') {
ffc61ed2
JH		 7251 scan += UTF8SKIP(scan);
7252 hardcount++;
7253 }
7254 } else {
7255 while (scan < loceol && *scan != '\n')
7256 scan++;
a0ed51b3
LW		 7257 }
7258 break;
ffc61ed2 7259 case SANY:
f2ed9b32 7260 if (utf8_target) {
a0804c9e 7261 while (scan < loceol && hardcount < max) {
def8e4ea
JH		 7262 scan += UTF8SKIP(scan);
7263 hardcount++;
7264 }
7265 }
7266 else
7267 scan = loceol;
a0ed51b3 7268 break;
4063ade8 7269 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
9597860a 7270 if (utf8_target && loceol - scan > max) {
4063ade8
KW		 7271
7272 /* <loceol> hadn't been adjusted in the UTF-8 case */
7273 scan += max;
7274 }
7275 else {
7276 scan = loceol;
7277 }
f33976b4 7278 break;
a4525e78 7279 case EXACTL:
780fcc9f 7280 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
613abc6d
KW		 7281 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*scan)) {
7282 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(scan, loceol);
7283 }
780fcc9f 7284 /* FALLTHROUGH */
59d32103 7285 case EXACT:
f9176b44 7286 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
613a425d 7287
59d32103 7288 c = (U8)*STRING(p);
59d32103 7289
5e4a1da1
KW		 7290 /* Can use a simple loop if the pattern char to match on is invariant
7291 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
7292 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
7293 * true iff it doesn't matter if the argument is in UTF-8 or not */
f9176b44 7294 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
e9369824 7295 if (utf8_target && loceol - scan > max) {
4063ade8
KW		 7296 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
7297 * since here, to match at all, 1 char == 1 byte */
7298 loceol = scan + max;
7299 }
59d32103
KW		 7300 while (scan < loceol && UCHARAT(scan) == c) {
7301 scan++;
7302 }
7303 }
f9176b44 7304 else if (reginfo->is_utf8_pat) {
5e4a1da1
KW		 7305 if (utf8_target) {
7306 STRLEN scan_char_len;
5e4a1da1 7307
4063ade8 7308 /* When both target and pattern are UTF-8, we have to do
5e4a1da1
KW		 7309 * string EQ */
7310 while (hardcount < max
9a902117
KW		 7311 && scan < loceol
7312 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
5e4a1da1
KW		 7313 && memEQ(scan, STRING(p), scan_char_len))
7314 {
4200a00c 7315 scan += scan_char_len;
5e4a1da1
KW		 7316 hardcount++;
7317 }
7318 }
7319 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
b40a2c17 7320
5e4a1da1
KW		 7321 /* Target isn't utf8; convert the character in the UTF-8
7322 * pattern to non-UTF8, and do a simple loop */
94bb8c36 7323 c = TWO_BYTE_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
5e4a1da1
KW		 7324 while (scan < loceol && UCHARAT(scan) == c) {
7325 scan++;
7326 }
7327 } /* else pattern char is above Latin1, can't possibly match the
7328 non-UTF-8 target */
b40a2c17 7329 }
5e4a1da1 7330 else {
59d32103 7331
5e4a1da1
KW		 7332 /* Here, the string must be utf8; pattern isn't, and <c> is
7333 * different in utf8 than not, so can't compare them directly.
7334 * Outside the loop, find the two utf8 bytes that represent c, and
7335 * then look for those in sequence in the utf8 string */
59d32103
KW		 7336 U8 high = UTF8_TWO_BYTE_HI(c);
7337 U8 low = UTF8_TWO_BYTE_LO(c);
59d32103
KW		 7338
7339 while (hardcount < max
7340 && scan + 1 < loceol
7341 && UCHARAT(scan) == high
7342 && UCHARAT(scan + 1) == low)
7343 {
7344 scan += 2;
7345 hardcount++;
7346 }
7347 }
7348 break;
5e4a1da1 7349
098b07d5
KW		 7350 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
7351 assert(! reginfo->is_utf8_pat);
924ba076 7352 /* FALLTHROUGH */
2f7f8cb1 7353 case EXACTFA:
098b07d5 7354 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
2f7f8cb1
KW		 7355 goto do_exactf;
7356
d4e0b827 7357 case EXACTFL:
780fcc9f 7358 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
cea315b6 7359 utf8_flags = FOLDEQ_LOCALE;
17580e7a
KW		 7360 goto do_exactf;
7361
2fdb7295
KW		 7362 case EXACTF: /* This node only generated for non-utf8 patterns */
7363 assert(! reginfo->is_utf8_pat);
098b07d5
KW		 7364 utf8_flags = 0;
7365 goto do_exactf;
62bf7766 7366
a4525e78
KW		 7367 case EXACTFLU8:
7368 if (! utf8_target) {
7369 break;
7370 }
613abc6d
KW		 7371 utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
7372 | FOLDEQ_S2_FOLDS_SANE;
a4525e78
KW		 7373 goto do_exactf;
7374
3c760661 7375 case EXACTFU_SS:
9a5a5549 7376 case EXACTFU:
f9176b44 7377 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
59d32103 7378
c52b8b12 7379 do_exactf: {
613a425d
KW		 7380 int c1, c2;
7381 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
d4e0b827 7382
f9176b44 7383 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
613a425d 7384
984e6dd1 7385 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
aed7b151 7386 reginfo))
984e6dd1 7387 {
613a425d 7388 if (c1 == CHRTEST_VOID) {
49b95fad 7389 /* Use full Unicode fold matching */
220db18a 7390 char *tmpeol = reginfo->strend;
f9176b44 7391 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
49b95fad
KW		 7392 while (hardcount < max
7393 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
7394 STRING(p), NULL, pat_len,
f9176b44 7395 reginfo->is_utf8_pat, utf8_flags))
49b95fad
KW		 7396 {
7397 scan = tmpeol;
220db18a 7398 tmpeol = reginfo->strend;
49b95fad
KW		 7399 hardcount++;
7400 }
613a425d
KW		 7401 }
7402 else if (utf8_target) {
7403 if (c1 == c2) {
4063ade8
KW		 7404 while (scan < loceol
7405 && hardcount < max
613a425d
KW		 7406 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
7407 {
7408 scan += UTF8SKIP(scan);
7409 hardcount++;
7410 }
7411 }
7412 else {
4063ade8
KW		 7413 while (scan < loceol
7414 && hardcount < max
613a425d
KW		 7415 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
7416 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
7417 {
7418 scan += UTF8SKIP(scan);
7419 hardcount++;
7420 }
7421 }
7422 }
7423 else if (c1 == c2) {
7424 while (scan < loceol && UCHARAT(scan) == c1) {
7425 scan++;
7426 }
7427 }
7428 else {
7429 while (scan < loceol &&
7430 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
7431 {
7432 scan++;
7433 }
7434 }
634c83a2 7435 }
bbce6d69 7436 break;
613a425d 7437 }
a4525e78 7438 case ANYOFL:
780fcc9f
KW		 7439 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
7440 /* FALLTHROUGH */
a0d0e21e 7441 case ANYOF:
e0193e47 7442 if (utf8_target) {
4e8910e0 7443 while (hardcount < max
9a902117 7444 && scan < loceol
3db24e1e 7445 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target))
4e8910e0 7446 {
9a902117 7447 scan += UTF8SKIP(scan);
ffc61ed2
JH		 7448 hardcount++;
7449 }
7450 } else {
32fc9b6a 7451 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
ffc61ed2
JH		 7452 scan++;
7453 }
a0d0e21e 7454 break;
4063ade8 7455
3018b823 7456 /* The argument (FLAGS) to all the POSIX node types is the class number */
980866de 7457
3018b823
KW		 7458 case NPOSIXL:
7459 to_complement = 1;
7460 /* FALLTHROUGH */
980866de 7461
3018b823 7462 case POSIXL:
780fcc9f 7463 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3018b823
KW		 7464 if (! utf8_target) {
7465 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
7466 *scan)))
a12cf05f 7467 {
3018b823
KW		 7468 scan++;
7469 }
7470 } else {
7471 while (hardcount < max && scan < loceol
7472 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
7473 (U8 *) scan)))
7474 {
7475 scan += UTF8SKIP(scan);
ffc61ed2
JH		 7476 hardcount++;
7477 }
a0ed51b3
LW		 7478 }
7479 break;
0658cdde 7480
3018b823
KW		 7481 case POSIXD:
7482 if (utf8_target) {
7483 goto utf8_posix;
7484 }
7485 /* FALLTHROUGH */
7486
0658cdde 7487 case POSIXA:
0430522f 7488 if (utf8_target && loceol - scan > max) {
4063ade8 7489
7aee35ff
KW		 7490 /* We didn't adjust <loceol> at the beginning of this routine
7491 * because is UTF-8, but it is actually ok to do so, since here, to
7492 * match, 1 char == 1 byte. */
4063ade8
KW		 7493 loceol = scan + max;
7494 }
7495 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
0658cdde
KW		 7496 scan++;
7497 }
7498 break;
980866de 7499
3018b823
KW		 7500 case NPOSIXD:
7501 if (utf8_target) {
7502 to_complement = 1;
7503 goto utf8_posix;
7504 }
924ba076 7505 /* FALLTHROUGH */
980866de 7506
3018b823
KW		 7507 case NPOSIXA:
7508 if (! utf8_target) {
7509 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
a12cf05f
KW		 7510 scan++;
7511 }
4063ade8 7512 }
3018b823 7513 else {
980866de 7514
3018b823 7515 /* The complement of something that matches only ASCII matches all
837226c8 7516 * non-ASCII, plus everything in ASCII that isn't in the class. */
bedac28b 7517 while (hardcount < max && scan < loceol
837226c8 7518 && (! isASCII_utf8(scan)
3018b823 7519 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
a12cf05f 7520 {
3018b823 7521 scan += UTF8SKIP(scan);
ffc61ed2
JH		 7522 hardcount++;
7523 }
3018b823
KW		 7524 }
7525 break;
980866de 7526
3018b823
KW		 7527 case NPOSIXU:
7528 to_complement = 1;
7529 /* FALLTHROUGH */
7530
7531 case POSIXU:
7532 if (! utf8_target) {
7533 while (scan < loceol && to_complement
7534 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
4063ade8 7535 {
3018b823
KW		 7536 scan++;
7537 }
cfaf538b
KW		 7538 }
7539 else {
c52b8b12 7540 utf8_posix:
3018b823
KW		 7541 classnum = (_char_class_number) FLAGS(p);
7542 if (classnum < _FIRST_NON_SWASH_CC) {
7543
7544 /* Here, a swash is needed for above-Latin1 code points.
7545 * Process as many Latin1 code points using the built-in rules.
7546 * Go to another loop to finish processing upon encountering
7547 * the first Latin1 code point. We could do that in this loop
7548 * as well, but the other way saves having to test if the swash
7549 * has been loaded every time through the loop: extra space to
7550 * save a test. */
7551 while (hardcount < max && scan < loceol) {
7552 if (UTF8_IS_INVARIANT(*scan)) {
7553 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
7554 classnum))))
7555 {
7556 break;
7557 }
7558 scan++;
7559 }
7560 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
7561 if (! (to_complement
94bb8c36
KW		 7562 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*scan,
7563 *(scan + 1)),
3018b823
KW		 7564 classnum))))
7565 {
7566 break;
7567 }
7568 scan += 2;
7569 }
7570 else {
7571 goto found_above_latin1;
7572 }
7573
7574 hardcount++;
7575 }
7576 }
7577 else {
7578 /* For these character classes, the knowledge of how to handle
7579 * every code point is compiled in to Perl via a macro. This
7580 * code is written for making the loops as tight as possible.
7581 * It could be refactored to save space instead */
7582 switch (classnum) {
7583 case _CC_ENUM_SPACE: /* XXX would require separate code
7584 if we revert the change of \v
7585 matching this */
924ba076 7586 /* FALLTHROUGH */
3018b823
KW		 7587 case _CC_ENUM_PSXSPC:
7588 while (hardcount < max
7589 && scan < loceol
7590 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
7591 {
7592 scan += UTF8SKIP(scan);
7593 hardcount++;
7594 }
7595 break;
7596 case _CC_ENUM_BLANK:
7597 while (hardcount < max
7598 && scan < loceol
7599 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
7600 {
7601 scan += UTF8SKIP(scan);
7602 hardcount++;
7603 }
7604 break;
7605 case _CC_ENUM_XDIGIT:
7606 while (hardcount < max
7607 && scan < loceol
7608 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
7609 {
7610 scan += UTF8SKIP(scan);
7611 hardcount++;
7612 }
7613 break;
7614 case _CC_ENUM_VERTSPACE:
7615 while (hardcount < max
7616 && scan < loceol
7617 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
7618 {
7619 scan += UTF8SKIP(scan);
7620 hardcount++;
7621 }
7622 break;
7623 case _CC_ENUM_CNTRL:
7624 while (hardcount < max
7625 && scan < loceol
7626 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
7627 {
7628 scan += UTF8SKIP(scan);
7629 hardcount++;
7630 }
7631 break;
7632 default:
7633 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
7634 }
7635 }
a0ed51b3 7636 }
3018b823 7637 break;
4063ade8 7638
3018b823
KW		 7639 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
7640
7641 /* Load the swash if not already present */
7642 if (! PL_utf8_swash_ptrs[classnum]) {
7643 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7644 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
2a16ac92
KW		 7645 "utf8",
7646 "",
7647 &PL_sv_undef, 1, 0,
7648 PL_XPosix_ptrs[classnum], &flags);
4063ade8 7649 }
3018b823
KW		 7650
7651 while (hardcount < max && scan < loceol
7652 && to_complement ^ cBOOL(_generic_utf8(
7653 classnum,
7654 scan,
7655 swash_fetch(PL_utf8_swash_ptrs[classnum],
7656 (U8 *) scan,
7657 TRUE))))
7658 {
7659 scan += UTF8SKIP(scan);
7660 hardcount++;
7661 }
7662 break;
7663
e1d1eefb 7664 case LNBREAK:
e64f369d
KW		 7665 if (utf8_target) {
7666 while (hardcount < max && scan < loceol &&
7667 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
7668 scan += c;
7669 hardcount++;
7670 }
7671 } else {
7672 /* LNBREAK can match one or two latin chars, which is ok, but we
7673 * have to use hardcount in this situation, and throw away the
7674 * adjustment to <loceol> done before the switch statement */
220db18a 7675 loceol = reginfo->strend;
e64f369d
KW		 7676 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
7677 scan+=c;
7678 hardcount++;
7679 }
7680 }
7681 break;
e1d1eefb 7682
780fcc9f
KW		 7683 case BOUNDL:
7684 case NBOUNDL:
7685 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
7686 /* FALLTHROUGH */
584b1f02
KW		 7687 case BOUND:
7688 case BOUNDA:
584b1f02
KW		 7689 case BOUNDU:
7690 case EOS:
7691 case GPOS:
7692 case KEEPS:
7693 case NBOUND:
7694 case NBOUNDA:
584b1f02
KW		 7695 case NBOUNDU:
7696 case OPFAIL:
7697 case SBOL:
7698 case SEOL:
7699 /* These are all 0 width, so match right here or not at all. */
7700 break;
7701
7702 default:
7703 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
a74ff37d 7704 /* NOTREACHED */
e5964223 7705 NOT_REACHED;
584b1f02 7706
a0d0e21e 7707 }
a687059c 7708
a0ed51b3
LW		 7709 if (hardcount)
7710 c = hardcount;
7711 else
f73aaa43
DM		 7712 c = scan - *startposp;
7713 *startposp = scan;
a687059c 7714
a3621e74 7715 DEBUG_r({
e68ec53f 7716 GET_RE_DEBUG_FLAGS_DECL;
be8e71aa 7717 DEBUG_EXECUTE_r({
e68ec53f 7718 SV * const prop = sv_newmortal();
8b9781c9 7719 regprop(prog, prop, p, reginfo, NULL);
e68ec53f 7720 PerlIO_printf(Perl_debug_log,
be8e71aa 7721 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
e2e6a0f1 7722 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
a3621e74 7723 });
be8e71aa 7724 });
9041c2e3 7725
a0d0e21e 7726 return(c);
a687059c
LW		 7727}
7728
c277df42 7729
be8e71aa 7730#if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
c277df42 7731/*
6c6525b8 7732- regclass_swash - prepare the utf8 swash. Wraps the shared core version to
e0193e47
KW		 7733create a copy so that changes the caller makes won't change the shared one.
7734If <altsvp> is non-null, will return NULL in it, for back-compat.
6c6525b8 7735 */
ffc61ed2 7736SV *
5aaab254 7737Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
ffc61ed2 7738{
6c6525b8 7739 PERL_ARGS_ASSERT_REGCLASS_SWASH;
e0193e47
KW		 7740
7741 if (altsvp) {
7742 *altsvp = NULL;
7743 }
7744
ef9bc832 7745 return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL, NULL));
6c6525b8 7746}
6c6525b8 7747
3e63bed3 7748#endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */
ffc61ed2
JH		 7749
7750/*
ba7b4546 7751 - reginclass - determine if a character falls into a character class
832705d4 7752
a4525e78 7753 n is the ANYOF-type regnode
6698fab5 7754 p is the target string
3db24e1e 7755 p_end points to one byte beyond the end of the target string
6698fab5 7756 utf8_target tells whether p is in UTF-8.
832705d4 7757
635cd5d4 7758 Returns true if matched; false otherwise.
eba1359e 7759
d5788240
KW		 7760 Note that this can be a synthetic start class, a combination of various
7761 nodes, so things you think might be mutually exclusive, such as locale,
7762 aren't. It can match both locale and non-locale
7763
bbce6d69 7764 */
7765
76e3520e 7766STATIC bool
3db24e1e 7767S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
bbce6d69 7768{
27da23d5 7769 dVAR;
a3b680e6 7770 const char flags = ANYOF_FLAGS(n);
bbce6d69 7771 bool match = FALSE;
cc07378b 7772 UV c = *p;
1aa99e6b 7773
7918f24d
NC		 7774 PERL_ARGS_ASSERT_REGINCLASS;
7775
afd2eb18
KW		 7776 /* If c is not already the code point, get it. Note that
7777 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
7778 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
635cd5d4 7779 STRLEN c_len = 0;
3db24e1e 7780 c = utf8n_to_uvchr(p, p_end - p, &c_len,
6182169b
KW		 7781 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7782 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7783 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7784 * UTF8_ALLOW_FFFF */
f7ab54c6 7785 if (c_len == (STRLEN)-1)
e8a70c6f 7786 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
613abc6d
KW		 7787 if (c > 255 && OP(n) == ANYOFL && ! is_ANYOF_SYNTHETIC(n)) {
7788 _CHECK_AND_OUTPUT_WIDE_LOCALE_CP_MSG(c);
7789 }
19f67299 7790 }
4b3cda86 7791
7cdde544 7792 /* If this character is potentially in the bitmap, check it */
dcb20b36 7793 if (c < NUM_ANYOF_CODE_POINTS) {
ffc61ed2
JH		 7794 if (ANYOF_BITMAP_TEST(n, c))
7795 match = TRUE;
93e92956
KW		 7796 else if ((flags & ANYOF_MATCHES_ALL_NON_UTF8_NON_ASCII)
7797 && ! utf8_target
7798 && ! isASCII(c))
11454c59
KW		 7799 {
7800 match = TRUE;
7801 }
1462525b 7802 else if (flags & ANYOF_LOCALE_FLAGS) {
6942fd9a 7803 if ((flags & ANYOF_LOC_FOLD)
e0a1ff7a 7804 && c < 256
6942fd9a
KW		 7805 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
7806 {
7807 match = TRUE;
b99851e1 7808 }
e0a1ff7a
KW		 7809 else if (ANYOF_POSIXL_TEST_ANY_SET(n)
7810 && c < 256
7811 ) {
31c7f561
KW		 7812
7813 /* The data structure is arranged so bits 0, 2, 4, ... are set
7814 * if the class includes the Posix character class given by
7815 * bit/2; and 1, 3, 5, ... are set if the class includes the
7816 * complemented Posix class given by int(bit/2). So we loop
7817 * through the bits, each time changing whether we complement
7818 * the result or not. Suppose for the sake of illustration
7819 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
7820 * is set, it means there is a match for this ANYOF node if the
7821 * character is in the class given by the expression (0 / 2 = 0
7822 * = \w). If it is in that class, isFOO_lc() will return 1,
7823 * and since 'to_complement' is 0, the result will stay TRUE,
7824 * and we exit the loop. Suppose instead that bit 0 is 0, but
7825 * bit 1 is 1. That means there is a match if the character
7826 * matches \W. We won't bother to call isFOO_lc() on bit 0,
7827 * but will on bit 1. On the second iteration 'to_complement'
7828 * will be 1, so the exclusive or will reverse things, so we
7829 * are testing for \W. On the third iteration, 'to_complement'
7830 * will be 0, and we would be testing for \s; the fourth
b0d691b2
KW		 7831 * iteration would test for \S, etc.
7832 *
7833 * Note that this code assumes that all the classes are closed
7834 * under folding. For example, if a character matches \w, then
7835 * its fold does too; and vice versa. This should be true for
7836 * any well-behaved locale for all the currently defined Posix
7837 * classes, except for :lower: and :upper:, which are handled
7838 * by the pseudo-class :cased: which matches if either of the
7839 * other two does. To get rid of this assumption, an outer
7840 * loop could be used below to iterate over both the source
7841 * character, and its fold (if different) */
31c7f561
KW		 7842
7843 int count = 0;
7844 int to_complement = 0;
522b3c1e 7845
31c7f561 7846 while (count < ANYOF_MAX) {
8efd3f97 7847 if (ANYOF_POSIXL_TEST(n, count)
31c7f561
KW		 7848 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
7849 {
7850 match = TRUE;
7851 break;
7852 }
7853 count++;
7854 to_complement ^= 1;
7855 }
ffc61ed2 7856 }
a0ed51b3 7857 }
a0ed51b3
LW		 7858 }
7859
31f05a37 7860
7cdde544 7861 /* If the bitmap didn't (or couldn't) match, and something outside the
3b04b210 7862 * bitmap could match, try that. */
ef87b810 7863 if (!match) {
93e92956
KW		 7864 if (c >= NUM_ANYOF_CODE_POINTS
7865 && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP))
7866 {
7867 match = TRUE; /* Everything above the bitmap matches */
e051a21d 7868 }
93e92956
KW		 7869 else if ((flags & ANYOF_HAS_NONBITMAP_NON_UTF8_MATCHES)
7870 || (utf8_target && (flags & ANYOF_HAS_UTF8_NONBITMAP_MATCHES))
1ee208c4
KW		 7871 || ((flags & ANYOF_LOC_FOLD)
7872 && IN_UTF8_CTYPE_LOCALE
93e92956 7873 && ARG(n) != ANYOF_ONLY_HAS_BITMAP))
3b04b210 7874 {
1ee208c4
KW		 7875 SV* only_utf8_locale = NULL;
7876 SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0,
ef9bc832 7877 &only_utf8_locale, NULL);
7cdde544 7878 if (sw) {
893ef8be 7879 U8 utf8_buffer[2];
7cdde544
KW		 7880 U8 * utf8_p;
7881 if (utf8_target) {
7882 utf8_p = (U8 *) p;
e0193e47 7883 } else { /* Convert to utf8 */
893ef8be
KW		 7884 utf8_p = utf8_buffer;
7885 append_utf8_from_native_byte(*p, &utf8_p);
7886 utf8_p = utf8_buffer;
7cdde544 7887 }
f56b6394 7888
e0193e47 7889 if (swash_fetch(sw, utf8_p, TRUE)) {
7cdde544 7890 match = TRUE;
e0193e47 7891 }
7cdde544 7892 }
1ee208c4
KW		 7893 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
7894 match = _invlist_contains_cp(only_utf8_locale, c);
7895 }
7cdde544 7896 }
5073ffbd
KW		 7897
7898 if (UNICODE_IS_SUPER(c)
ae986089 7899 && (flags & ANYOF_WARN_SUPER)
5073ffbd
KW		 7900 && ckWARN_d(WARN_NON_UNICODE))
7901 {
7902 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2d88a86a 7903 "Matched non-Unicode code point 0x%04"UVXf" against Unicode property; may not be portable", c);
5073ffbd 7904 }
7cdde544
KW		 7905 }
7906
5dbb0c08
KW		 7907#if ANYOF_INVERT != 1
7908 /* Depending on compiler optimization cBOOL takes time, so if don't have to
7909 * use it, don't */
7910# error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
7911#endif
7912
f0fdc1c9 7913 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
5dbb0c08 7914 return (flags & ANYOF_INVERT) ^ match;
a0ed51b3 7915}
161b471a 7916
dfe13c55 7917STATIC U8 *
ea3daa5d 7918S_reghop3(U8 *s, SSize_t off, const U8* lim)
9041c2e3 7919{
6af86488
KW		 7920 /* return the position 'off' UTF-8 characters away from 's', forward if
7921 * 'off' >= 0, backwards if negative. But don't go outside of position
7922 * 'lim', which better be < s if off < 0 */
7923
7918f24d
NC		 7924 PERL_ARGS_ASSERT_REGHOP3;
7925
a0ed51b3 7926 if (off >= 0) {
1aa99e6b 7927 while (off-- && s < lim) {
ffc61ed2 7928 /* XXX could check well-formedness here */
a0ed51b3 7929 s += UTF8SKIP(s);
ffc61ed2 7930 }
a0ed51b3
LW		 7931 }
7932 else {
1de06328
YO		 7933 while (off++ && s > lim) {
7934 s--;
7935 if (UTF8_IS_CONTINUED(*s)) {
7936 while (s > lim && UTF8_IS_CONTINUATION(*s))
7937 s--;
a0ed51b3 7938 }
1de06328 7939 /* XXX could check well-formedness here */
a0ed51b3
LW		 7940 }
7941 }
7942 return s;
7943}
161b471a 7944
dfe13c55 7945STATIC U8 *
ea3daa5d 7946S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
1de06328 7947{
7918f24d
NC		 7948 PERL_ARGS_ASSERT_REGHOP4;
7949
1de06328
YO		 7950 if (off >= 0) {
7951 while (off-- && s < rlim) {
7952 /* XXX could check well-formedness here */
7953 s += UTF8SKIP(s);
7954 }
7955 }
7956 else {
7957 while (off++ && s > llim) {
7958 s--;
7959 if (UTF8_IS_CONTINUED(*s)) {
7960 while (s > llim && UTF8_IS_CONTINUATION(*s))
7961 s--;
7962 }
7963 /* XXX could check well-formedness here */
7964 }
7965 }
7966 return s;
7967}
1de06328 7968
557f47af
DM		 7969/* like reghop3, but returns NULL on overrun, rather than returning last
7970 * char pos */
7971
1de06328 7972STATIC U8 *
ea3daa5d 7973S_reghopmaybe3(U8* s, SSize_t off, const U8* lim)
a0ed51b3 7974{
7918f24d
NC		 7975 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7976
a0ed51b3 7977 if (off >= 0) {
1aa99e6b 7978 while (off-- && s < lim) {
ffc61ed2 7979 /* XXX could check well-formedness here */
a0ed51b3 7980 s += UTF8SKIP(s);
ffc61ed2 7981 }
a0ed51b3 7982 if (off >= 0)
3dab1dad 7983 return NULL;
a0ed51b3
LW		 7984 }
7985 else {
1de06328
YO		 7986 while (off++ && s > lim) {
7987 s--;
7988 if (UTF8_IS_CONTINUED(*s)) {
7989 while (s > lim && UTF8_IS_CONTINUATION(*s))
7990 s--;
a0ed51b3 7991 }
1de06328 7992 /* XXX could check well-formedness here */
a0ed51b3
LW		 7993 }
7994 if (off <= 0)
3dab1dad 7995 return NULL;
a0ed51b3
LW		 7996 }
7997 return s;
7998}
51371543 7999
a75351a1
DM		 8000
8001/* when executing a regex that may have (?{}), extra stuff needs setting
8002 up that will be visible to the called code, even before the current
8003 match has finished. In particular:
8004
8005 * $_ is localised to the SV currently being matched;
8006 * pos($_) is created if necessary, ready to be updated on each call-out
8007 to code;
8008 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
8009 isn't set until the current pattern is successfully finished), so that
8010 $1 etc of the match-so-far can be seen;
8011 * save the old values of subbeg etc of the current regex, and set then
8012 to the current string (again, this is normally only done at the end
8013 of execution)
a75351a1
DM		 8014*/
8015
8016static void
8017S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
8018{
8019 MAGIC *mg;
8020 regexp *const rex = ReANY(reginfo->prog);
bf2039a9 8021 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
8adc0f72 8022
8adc0f72 8023 eval_state->rex = rex;
a75351a1 8024
a75351a1
DM		 8025 if (reginfo->sv) {
8026 /* Make $_ available to executed code. */
8027 if (reginfo->sv != DEFSV) {
8028 SAVE_DEFSV;
8029 DEFSV_set(reginfo->sv);
8030 }
8031
96c2a8ff 8032 if (!(mg = mg_find_mglob(reginfo->sv))) {
a75351a1 8033 /* prepare for quick setting of pos */
96c2a8ff 8034 mg = sv_magicext_mglob(reginfo->sv);
a75351a1
DM		 8035 mg->mg_len = -1;
8036 }
8adc0f72
DM		 8037 eval_state->pos_magic = mg;
8038 eval_state->pos = mg->mg_len;
25fdce4a 8039 eval_state->pos_flags = mg->mg_flags;
a75351a1 8040 }
8adc0f72
DM		 8041 else
8042 eval_state->pos_magic = NULL;
8043
a75351a1 8044 if (!PL_reg_curpm) {
f65e70f5
DM		 8045 /* PL_reg_curpm is a fake PMOP that we can attach the current
8046 * regex to and point PL_curpm at, so that $1 et al are visible
8047 * within a /(?{})/. It's just allocated once per interpreter the
8048 * first time its needed */
a75351a1
DM		 8049 Newxz(PL_reg_curpm, 1, PMOP);
8050#ifdef USE_ITHREADS
8051 {
8052 SV* const repointer = &PL_sv_undef;
8053 /* this regexp is also owned by the new PL_reg_curpm, which
8054 will try to free it. */
8055 av_push(PL_regex_padav, repointer);
b9f2b683 8056 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
a75351a1
DM		 8057 PL_regex_pad = AvARRAY(PL_regex_padav);
8058 }
8059#endif
8060 }
8061 SET_reg_curpm(reginfo->prog);
8adc0f72 8062 eval_state->curpm = PL_curpm;
a75351a1
DM		 8063 PL_curpm = PL_reg_curpm;
8064 if (RXp_MATCH_COPIED(rex)) {
8065 /* Here is a serious problem: we cannot rewrite subbeg,
8066 since it may be needed if this match fails. Thus
8067 $` inside (?{}) could fail... */
8adc0f72
DM		 8068 eval_state->subbeg = rex->subbeg;
8069 eval_state->sublen = rex->sublen;
8070 eval_state->suboffset = rex->suboffset;
a8ee055f 8071 eval_state->subcoffset = rex->subcoffset;
a75351a1 8072#ifdef PERL_ANY_COW
8adc0f72 8073 eval_state->saved_copy = rex->saved_copy;
a75351a1
DM		 8074#endif
8075 RXp_MATCH_COPIED_off(rex);
8076 }
8077 else
8adc0f72 8078 eval_state->subbeg = NULL;
a75351a1
DM		 8079 rex->subbeg = (char *)reginfo->strbeg;
8080 rex->suboffset = 0;
8081 rex->subcoffset = 0;
8082 rex->sublen = reginfo->strend - reginfo->strbeg;
8083}
8084
bf2039a9
DM		 8085
8086/* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
a75351a1 8087
51371543 8088static void
bf2039a9 8089S_cleanup_regmatch_info_aux(pTHX_ void *arg)
51371543 8090{
bf2039a9
DM		 8091 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
8092 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
331b2dcc 8093 regmatch_slab *s;
bf2039a9 8094
2ac8ff4b
DM		 8095 Safefree(aux->poscache);
8096
331b2dcc 8097 if (eval_state) {
bf2039a9 8098
331b2dcc 8099 /* undo the effects of S_setup_eval_state() */
bf2039a9 8100
331b2dcc
DM		 8101 if (eval_state->subbeg) {
8102 regexp * const rex = eval_state->rex;
8103 rex->subbeg = eval_state->subbeg;
8104 rex->sublen = eval_state->sublen;
8105 rex->suboffset = eval_state->suboffset;
8106 rex->subcoffset = eval_state->subcoffset;
db2c6cb3 8107#ifdef PERL_ANY_COW
331b2dcc 8108 rex->saved_copy = eval_state->saved_copy;
ed252734 8109#endif
331b2dcc
DM		 8110 RXp_MATCH_COPIED_on(rex);
8111 }
8112 if (eval_state->pos_magic)
25fdce4a 8113 {
331b2dcc 8114 eval_state->pos_magic->mg_len = eval_state->pos;
25fdce4a
FC		 8115 eval_state->pos_magic->mg_flags =
8116 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
8117 | (eval_state->pos_flags & MGf_BYTES);
8118 }
331b2dcc
DM		 8119
8120 PL_curpm = eval_state->curpm;
8adc0f72 8121 }
bf2039a9 8122
331b2dcc
DM		 8123 PL_regmatch_state = aux->old_regmatch_state;
8124 PL_regmatch_slab = aux->old_regmatch_slab;
8125
8126 /* free all slabs above current one - this must be the last action
8127 * of this function, as aux and eval_state are allocated within
8128 * slabs and may be freed here */
8129
8130 s = PL_regmatch_slab->next;
8131 if (s) {
8132 PL_regmatch_slab->next = NULL;
8133 while (s) {
8134 regmatch_slab * const osl = s;
8135 s = s->next;
8136 Safefree(osl);
8137 }
8138 }
51371543 8139}
33b8afdf 8140
8adc0f72 8141
33b8afdf 8142STATIC void
5aaab254 8143S_to_utf8_substr(pTHX_ regexp *prog)
33b8afdf 8144{
7e0d5ad7
KW		 8145 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
8146 * on the converted value */
8147
a1cac82e 8148 int i = 1;
7918f24d
NC		 8149
8150 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
8151
a1cac82e
NC		 8152 do {
8153 if (prog->substrs->data[i].substr
8154 && !prog->substrs->data[i].utf8_substr) {
8155 SV* const sv = newSVsv(prog->substrs->data[i].substr);
8156 prog->substrs->data[i].utf8_substr = sv;
8157 sv_utf8_upgrade(sv);
610460f9 8158 if (SvVALID(prog->substrs->data[i].substr)) {
cffe132d 8159 if (SvTAIL(prog->substrs->data[i].substr)) {
610460f9
NC		 8160 /* Trim the trailing \n that fbm_compile added last
8161 time. */
8162 SvCUR_set(sv, SvCUR(sv) - 1);
8163 /* Whilst this makes the SV technically "invalid" (as its
8164 buffer is no longer followed by "\0") when fbm_compile()
8165 adds the "\n" back, a "\0" is restored. */
cffe132d
NC		 8166 fbm_compile(sv, FBMcf_TAIL);
8167 } else
8168 fbm_compile(sv, 0);
610460f9 8169 }
a1cac82e
NC		 8170 if (prog->substrs->data[i].substr == prog->check_substr)
8171 prog->check_utf8 = sv;
8172 }
8173 } while (i--);
33b8afdf
JH		 8174}
8175
7e0d5ad7 8176STATIC bool
5aaab254 8177S_to_byte_substr(pTHX_ regexp *prog)
33b8afdf 8178{
7e0d5ad7
KW		 8179 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
8180 * on the converted value; returns FALSE if can't be converted. */
8181
a1cac82e 8182 int i = 1;
7918f24d
NC		 8183
8184 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
8185
a1cac82e
NC		 8186 do {
8187 if (prog->substrs->data[i].utf8_substr
8188 && !prog->substrs->data[i].substr) {
8189 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
7e0d5ad7
KW		 8190 if (! sv_utf8_downgrade(sv, TRUE)) {
8191 return FALSE;
8192 }
5400f398
KW		 8193 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
8194 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
8195 /* Trim the trailing \n that fbm_compile added last
8196 time. */
8197 SvCUR_set(sv, SvCUR(sv) - 1);
8198 fbm_compile(sv, FBMcf_TAIL);
8199 } else
8200 fbm_compile(sv, 0);
8201 }
a1cac82e
NC		 8202 prog->substrs->data[i].substr = sv;
8203 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
8204 prog->check_substr = sv;
33b8afdf 8205 }
a1cac82e 8206 } while (i--);
7e0d5ad7
KW		 8207
8208 return TRUE;
33b8afdf 8209}
66610fdd
RGS		 8210
8211/*
8212 * Local variables:
8213 * c-indentation-style: bsd
8214 * c-basic-offset: 4
14d04a33 8215 * indent-tabs-mode: nil
66610fdd
RGS		 8216 * End:
8217 *
14d04a33 8218 * ex: set ts=8 sts=4 sw=4 et:
37442d52 8219 */
Unnamed repository; edit this file 'description' to name the repository.