This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
For character case changing, create macros and use
[perl5.git] / regexec.c
CommitLineData
a0d0e21e
LW
1/* regexec.c
2 */
3
4/*
f65819ce
CO
5 * One Ring to rule them all, One Ring to find them
6 *
4ac71550
TC
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
a0d0e21e
LW
10 */
11
61296642
DM
12/* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
166f8a29 14 * a regular expression.
e4a054ea
DM
15 *
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
166f8a29
DM
20 */
21
a687059c
LW
22/* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
24 */
25
26/* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
29 */
30
e50aee73
AD
31/* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
34*/
35
b9d5759e 36#ifdef PERL_EXT_RE_BUILD
54df2634 37#include "re_top.h"
9041c2e3 38#endif
56953603 39
a687059c 40/*
e50aee73 41 * pregcomp and pregexec -- regsub and regerror are not used in perl
a687059c
LW
42 *
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
45 *
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
49 *
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
52 * from defects in it.
53 *
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
56 *
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
59 *
60 **** Alterations to Henry's code are...
61 ****
4bb101f2 62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
1129b882
NC
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
a687059c 65 ****
9ef589d8
LW
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
a687059c
LW
68 *
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
72 */
73#include "EXTERN.h"
864dbfa3 74#define PERL_IN_REGEXEC_C
a687059c 75#include "perl.h"
0f5d15d6 76
54df2634
NC
77#ifdef PERL_IN_XSUB_RE
78# include "re_comp.h"
79#else
80# include "regcomp.h"
81#endif
a687059c 82
b992490d 83#include "invlist_inline.h"
1b0f46bf 84#include "unicode_constants.h"
81e983c1 85
bbac6b20
KW
86#define B_ON_NON_UTF8_LOCALE_IS_WRONG \
87 "Use of \\b{} or \\B{} for non-UTF-8 locale is wrong. Assuming a UTF-8 locale"
88
a0bd1a30
KW
89static const char utf8_locale_required[] =
90 "Use of (?[ ]) for non-UTF-8 locale is wrong. Assuming a UTF-8 locale";
91
e1cf74e3
CB
92#ifdef DEBUGGING
93/* At least one required character in the target string is expressible only in
94 * UTF-8. */
95static const char* const non_utf8_target_but_utf8_required
96 = "Can't match, because target string needs to be in UTF-8\n";
97#endif
98
7b031478 99#define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
6ad9a8ab 100 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%s", non_utf8_target_but_utf8_required));\
7b031478 101 goto target; \
e1cf74e3
CB
102} STMT_END
103
c74f6de9
KW
104#define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
105
a687059c
LW
106#ifndef STATIC
107#define STATIC static
108#endif
109
451c6e0b
KW
110/* Valid only if 'c', the character being looke-up, is an invariant under
111 * UTF-8: it avoids the reginclass call if there are no complications: i.e., if
112 * everything matchable is straight forward in the bitmap */
113#define REGINCLASS(prog,p,c,u) (ANYOF_FLAGS(p) \
114 ? reginclass(prog,p,c,c+1,u) \
115 : ANYOF_BITMAP_TEST(p,*(c)))
7d3e948e 116
c277df42
IZ
117/*
118 * Forwards.
119 */
120
f2ed9b32 121#define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
ba44c216 122#define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b)
a0ed51b3 123
3dab1dad 124#define HOPc(pos,off) \
ba44c216 125 (char *)(reginfo->is_utf8_target \
220db18a 126 ? reghop3((U8*)pos, off, \
9d9163fb 127 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
3dab1dad 128 : (U8*)(pos + off))
557f47af 129
3dab1dad 130#define HOPBACKc(pos, off) \
ba44c216 131 (char*)(reginfo->is_utf8_target \
c708944d 132 ? reghopmaybe3((U8*)pos, (SSize_t)0-off, (U8*)(reginfo->strbeg)) \
9d9163fb 133 : (pos - off >= reginfo->strbeg) \
8e11feef 134 ? (U8*)pos - off \
3dab1dad 135 : NULL)
efb30f32 136
ba44c216 137#define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
1aa99e6b 138#define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
1aa99e6b 139
557f47af
DM
140/* lim must be +ve. Returns NULL on overshoot */
141#define HOPMAYBE3(pos,off,lim) \
142 (reginfo->is_utf8_target \
143 ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \
144 : ((U8*)pos + off <= lim) \
145 ? (U8*)pos + off \
146 : NULL)
147
8e9f2289
DM
148/* like HOP3, but limits the result to <= lim even for the non-utf8 case.
149 * off must be >=0; args should be vars rather than expressions */
150#define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \
151 ? reghop3((U8*)(pos), off, (U8*)(lim)) \
152 : (U8*)((pos + off) > lim ? lim : (pos + off)))
153
2974eaec
DM
154#define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \
155 ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \
156 : (U8*)(pos + off))
157#define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim))
7016d6eb
DM
158
159#define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
160#define NEXTCHR_IS_EOS (nextchr < 0)
161
162#define SET_nextchr \
220db18a 163 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
7016d6eb
DM
164
165#define SET_locinput(p) \
166 locinput = (p); \
167 SET_nextchr
168
169
2a16ac92 170#define LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) STMT_START { \
c7304fe2
KW
171 if (!swash_ptr) { \
172 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
c7304fe2 173 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
2a16ac92 174 1, 0, invlist, &flags); \
c7304fe2
KW
175 assert(swash_ptr); \
176 } \
177 } STMT_END
178
179/* If in debug mode, we test that a known character properly matches */
180#ifdef DEBUGGING
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
KW
186 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
187#else
188# define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
189 property_name, \
2a16ac92 190 invlist, \
c7304fe2 191 utf8_char_in_property) \
2a16ac92 192 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist)
c7304fe2 193#endif
d1eb3177 194
c7304fe2
KW
195#define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
196 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
2a16ac92
KW
197 "", \
198 PL_XPosix_ptrs[_CC_WORDCHAR], \
0766489e 199 LATIN_SMALL_LIGATURE_LONG_S_T_UTF8);
c7304fe2 200
c7304fe2 201#define PLACEHOLDER /* Something for the preprocessor to grab onto */
3dab1dad
YO
202/* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
203
5f80c4cf 204/* for use after a quantifier and before an EXACT-like node -- japhy */
c35dcbe2
YO
205/* it would be nice to rework regcomp.sym to generate this stuff. sigh
206 *
207 * NOTE that *nothing* that affects backtracking should be in here, specifically
208 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
209 * node that is in between two EXACT like nodes when ascertaining what the required
210 * "follow" character is. This should probably be moved to regex compile time
211 * although it may be done at run time beause of the REF possibility - more
212 * investigation required. -- demerphq
213*/
baa60164
KW
214#define JUMPABLE(rn) ( \
215 OP(rn) == OPEN || \
24be3102
YO
216 (OP(rn) == CLOSE && \
217 !EVAL_CLOSE_PAREN_IS(cur_eval,ARG(rn)) ) || \
baa60164
KW
218 OP(rn) == EVAL || \
219 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
220 OP(rn) == PLUS || OP(rn) == MINMOD || \
221 OP(rn) == KEEPS || \
222 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
e2d8ce26 223)
ee9b8eae 224#define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
e2d8ce26 225
ee9b8eae
YO
226#define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
227
228#if 0
229/* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
a4525e78 230 we don't need this definition. XXX These are now out-of-sync*/
ee9b8eae 231#define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
098b07d5 232#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
233#define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
234
235#else
236/* ... so we use this as its faster. */
a4525e78
KW
237#define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==EXACTL )
238#define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFLU8 || OP(rn)==EXACTFU_SS || OP(rn) == EXACTFA || OP(rn) == EXACTFA_NO_TRIE)
ee9b8eae
YO
239#define IS_TEXTF(rn) ( OP(rn)==EXACTF )
240#define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
241
242#endif
e2d8ce26 243
a84d97b6
HS
244/*
245 Search for mandatory following text node; for lookahead, the text must
246 follow but for lookbehind (rn->flags != 0) we skip to the next step.
247*/
baa60164 248#define FIND_NEXT_IMPT(rn) STMT_START { \
3dab1dad
YO
249 while (JUMPABLE(rn)) { \
250 const OPCODE type = OP(rn); \
251 if (type == SUSPEND || PL_regkind[type] == CURLY) \
e2d8ce26 252 rn = NEXTOPER(NEXTOPER(rn)); \
3dab1dad 253 else if (type == PLUS) \
cca55fe3 254 rn = NEXTOPER(rn); \
3dab1dad 255 else if (type == IFMATCH) \
a84d97b6 256 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
e2d8ce26 257 else rn += NEXT_OFF(rn); \
3dab1dad 258 } \
5f80c4cf 259} STMT_END
74750237 260
006f26b2
DM
261#define SLAB_FIRST(s) (&(s)->states[0])
262#define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
263
a75351a1 264static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
bf2039a9 265static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
bf2039a9 266static regmatch_state * S_push_slab(pTHX);
51371543 267
87c0511b 268#define REGCP_PAREN_ELEMS 3
f067efbf 269#define REGCP_OTHER_ELEMS 3
e0fa7e2b 270#define REGCP_FRAME_ELEMS 1
620d5b66
NC
271/* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
272 * are needed for the regexp context stack bookkeeping. */
273
76e3520e 274STATIC CHECKPOINT
21553840 275S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen _pDEPTH)
a0d0e21e 276{
a3b680e6 277 const int retval = PL_savestack_ix;
92da3157
DM
278 const int paren_elems_to_push =
279 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
e0fa7e2b
NC
280 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
281 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
87c0511b 282 I32 p;
40a82448 283 GET_RE_DEBUG_FLAGS_DECL;
a0d0e21e 284
b93070ed
DM
285 PERL_ARGS_ASSERT_REGCPPUSH;
286
e49a9654 287 if (paren_elems_to_push < 0)
e8a85d26
JH
288 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
289 (int)paren_elems_to_push, (int)maxopenparen,
290 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
e49a9654 291
e0fa7e2b 292 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
147e3846 293 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %" UVuf
5df417d0 294 " out of range (%lu-%ld)",
92da3157
DM
295 total_elems,
296 (unsigned long)maxopenparen,
297 (long)parenfloor);
e0fa7e2b 298
620d5b66 299 SSGROW(total_elems + REGCP_FRAME_ELEMS);
7f69552c 300
495f47a5 301 DEBUG_BUFFERS_r(
92da3157 302 if ((int)maxopenparen > (int)parenfloor)
2b1a3689 303 Perl_re_exec_indentf( aTHX_
147e3846 304 "rex=0x%" UVxf " offs=0x%" UVxf ": saving capture indices:\n",
2b1a3689
YO
305 depth,
306 PTR2UV(rex),
495f47a5
DM
307 PTR2UV(rex->offs)
308 );
309 );
92da3157 310 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
b1ce53c5 311/* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
99a90e59
FC
312 SSPUSHIV(rex->offs[p].end);
313 SSPUSHIV(rex->offs[p].start);
1ca2007e 314 SSPUSHINT(rex->offs[p].start_tmp);
2b1a3689 315 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
147e3846 316 " \\%" UVuf ": %" IVdf "(%" IVdf ")..%" IVdf "\n",
2b1a3689
YO
317 depth,
318 (UV)p,
495f47a5
DM
319 (IV)rex->offs[p].start,
320 (IV)rex->offs[p].start_tmp,
321 (IV)rex->offs[p].end
40a82448 322 ));
a0d0e21e 323 }
b1ce53c5 324/* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
92da3157 325 SSPUSHINT(maxopenparen);
b93070ed
DM
326 SSPUSHINT(rex->lastparen);
327 SSPUSHINT(rex->lastcloseparen);
e0fa7e2b 328 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
41123dfd 329
a0d0e21e
LW
330 return retval;
331}
332
c277df42 333/* These are needed since we do not localize EVAL nodes: */
ab3bbdeb
YO
334#define REGCP_SET(cp) \
335 DEBUG_STATE_r( \
147e3846
KW
336 Perl_re_exec_indentf( aTHX_ \
337 "Setting an EVAL scope, savestack=%" IVdf ",\n", \
cb41e5d6
YO
338 depth, (IV)PL_savestack_ix \
339 ) \
340 ); \
ab3bbdeb 341 cp = PL_savestack_ix
c3464db5 342
ab3bbdeb 343#define REGCP_UNWIND(cp) \
e4f74956 344 DEBUG_STATE_r( \
cb41e5d6 345 if (cp != PL_savestack_ix) \
147e3846
KW
346 Perl_re_exec_indentf( aTHX_ \
347 "Clearing an EVAL scope, savestack=%" \
348 IVdf "..%" IVdf "\n", \
cb41e5d6
YO
349 depth, (IV)(cp), (IV)PL_savestack_ix \
350 ) \
351 ); \
ab3bbdeb 352 regcpblow(cp)
c277df42 353
a8d1f4b4
DM
354#define UNWIND_PAREN(lp, lcp) \
355 for (n = rex->lastparen; n > lp; n--) \
356 rex->offs[n].end = -1; \
357 rex->lastparen = n; \
358 rex->lastcloseparen = lcp;
359
360
f067efbf 361STATIC void
21553840 362S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p _pDEPTH)
a0d0e21e 363{
e0fa7e2b 364 UV i;
87c0511b 365 U32 paren;
a3621e74
YO
366 GET_RE_DEBUG_FLAGS_DECL;
367
7918f24d
NC
368 PERL_ARGS_ASSERT_REGCPPOP;
369
b1ce53c5 370 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
c6bf6a65 371 i = SSPOPUV;
e0fa7e2b
NC
372 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
373 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
b93070ed
DM
374 rex->lastcloseparen = SSPOPINT;
375 rex->lastparen = SSPOPINT;
92da3157 376 *maxopenparen_p = SSPOPINT;
b1ce53c5 377
620d5b66 378 i -= REGCP_OTHER_ELEMS;
b1ce53c5 379 /* Now restore the parentheses context. */
495f47a5
DM
380 DEBUG_BUFFERS_r(
381 if (i || rex->lastparen + 1 <= rex->nparens)
2b1a3689 382 Perl_re_exec_indentf( aTHX_
147e3846 383 "rex=0x%" UVxf " offs=0x%" UVxf ": restoring capture indices to:\n",
2b1a3689
YO
384 depth,
385 PTR2UV(rex),
495f47a5
DM
386 PTR2UV(rex->offs)
387 );
388 );
92da3157 389 paren = *maxopenparen_p;
620d5b66 390 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
ea3daa5d 391 SSize_t tmps;
1ca2007e 392 rex->offs[paren].start_tmp = SSPOPINT;
99a90e59
FC
393 rex->offs[paren].start = SSPOPIV;
394 tmps = SSPOPIV;
b93070ed
DM
395 if (paren <= rex->lastparen)
396 rex->offs[paren].end = tmps;
2b1a3689 397 DEBUG_BUFFERS_r( Perl_re_exec_indentf( aTHX_
147e3846 398 " \\%" UVuf ": %" IVdf "(%" IVdf ")..%" IVdf "%s\n",
2b1a3689
YO
399 depth,
400 (UV)paren,
495f47a5
DM
401 (IV)rex->offs[paren].start,
402 (IV)rex->offs[paren].start_tmp,
403 (IV)rex->offs[paren].end,
404 (paren > rex->lastparen ? "(skipped)" : ""));
c277df42 405 );
87c0511b 406 paren--;
a0d0e21e 407 }
daf18116 408#if 1
dafc8851
JH
409 /* It would seem that the similar code in regtry()
410 * already takes care of this, and in fact it is in
411 * a better location to since this code can #if 0-ed out
412 * but the code in regtry() is needed or otherwise tests
413 * requiring null fields (pat.t#187 and split.t#{13,14}
daf18116
JH
414 * (as of patchlevel 7877) will fail. Then again,
415 * this code seems to be necessary or otherwise
225593e1
DM
416 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
417 * --jhi updated by dapm */
b93070ed 418 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
92da3157 419 if (i > *maxopenparen_p)
b93070ed
DM
420 rex->offs[i].start = -1;
421 rex->offs[i].end = -1;
2b1a3689 422 DEBUG_BUFFERS_r( Perl_re_exec_indentf( aTHX_
147e3846 423 " \\%" UVuf ": %s ..-1 undeffing\n",
2b1a3689
YO
424 depth,
425 (UV)i,
92da3157 426 (i > *maxopenparen_p) ? "-1" : " "
495f47a5 427 ));
a0d0e21e 428 }
dafc8851 429#endif
a0d0e21e
LW
430}
431
74088413
DM
432/* restore the parens and associated vars at savestack position ix,
433 * but without popping the stack */
434
435STATIC void
21553840 436S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p _pDEPTH)
74088413
DM
437{
438 I32 tmpix = PL_savestack_ix;
85882954
YO
439 PERL_ARGS_ASSERT_REGCP_RESTORE;
440
74088413 441 PL_savestack_ix = ix;
21553840 442 regcppop(rex, maxopenparen_p);
74088413
DM
443 PL_savestack_ix = tmpix;
444}
445
02db2b7b 446#define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
a0d0e21e 447
24e16d7b
KW
448bool
449Perl_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
31c7f561
KW
450{
451 /* Returns a boolean as to whether or not 'character' is a member of the
452 * Posix character class given by 'classnum' that should be equivalent to a
453 * value in the typedef '_char_class_number'.
454 *
455 * Ideally this could be replaced by a just an array of function pointers
456 * to the C library functions that implement the macros this calls.
457 * However, to compile, the precise function signatures are required, and
458 * these may vary from platform to to platform. To avoid having to figure
459 * out what those all are on each platform, I (khw) am using this method,
7aee35ff
KW
460 * which adds an extra layer of function call overhead (unless the C
461 * optimizer strips it away). But we don't particularly care about
462 * performance with locales anyway. */
31c7f561
KW
463
464 switch ((_char_class_number) classnum) {
15861f94 465 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
31c7f561 466 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
e8d596e0
KW
467 case _CC_ENUM_ASCII: return isASCII_LC(character);
468 case _CC_ENUM_BLANK: return isBLANK_LC(character);
cee69f79 469 case _CC_ENUM_CASED: return isLOWER_LC(character)
b0d691b2 470 || isUPPER_LC(character);
e8d596e0 471 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
31c7f561
KW
472 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
473 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
474 case _CC_ENUM_LOWER: return isLOWER_LC(character);
475 case _CC_ENUM_PRINT: return isPRINT_LC(character);
476 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
e8d596e0 477 case _CC_ENUM_SPACE: return isSPACE_LC(character);
31c7f561
KW
478 case _CC_ENUM_UPPER: return isUPPER_LC(character);
479 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
31c7f561 480 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
31c7f561
KW
481 default: /* VERTSPACE should never occur in locales */
482 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
483 }
484
e5964223 485 NOT_REACHED; /* NOTREACHED */
31c7f561
KW
486 return FALSE;
487}
488
3018b823
KW
489STATIC bool
490S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
491{
492 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
493 * 'character' is a member of the Posix character class given by 'classnum'
494 * that should be equivalent to a value in the typedef
495 * '_char_class_number'.
496 *
497 * This just calls isFOO_lc on the code point for the character if it is in
2f306ab9 498 * the range 0-255. Outside that range, all characters use Unicode
3018b823
KW
499 * rules, ignoring any locale. So use the Unicode function if this class
500 * requires a swash, and use the Unicode macro otherwise. */
501
502 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
503
504 if (UTF8_IS_INVARIANT(*character)) {
505 return isFOO_lc(classnum, *character);
506 }
507 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
508 return isFOO_lc(classnum,
a62b247b 509 EIGHT_BIT_UTF8_TO_NATIVE(*character, *(character + 1)));
3018b823
KW
510 }
511
613abc6d
KW
512 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(character, character + UTF8SKIP(character));
513
3018b823
KW
514 if (classnum < _FIRST_NON_SWASH_CC) {
515
516 /* Initialize the swash unless done already */
517 if (! PL_utf8_swash_ptrs[classnum]) {
518 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2a16ac92
KW
519 PL_utf8_swash_ptrs[classnum] =
520 _core_swash_init("utf8",
521 "",
522 &PL_sv_undef, 1, 0,
523 PL_XPosix_ptrs[classnum], &flags);
3018b823
KW
524 }
525
92a2046b
KW
526 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
527 character,
528 TRUE /* is UTF */ ));
3018b823
KW
529 }
530
531 switch ((_char_class_number) classnum) {
779cf272 532 case _CC_ENUM_SPACE: return is_XPERLSPACE_high(character);
3018b823
KW
533 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
534 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
535 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
e1ee3960 536 default: break;
3018b823
KW
537 }
538
e1ee3960 539 return FALSE; /* Things like CNTRL are always below 256 */
3018b823
KW
540}
541
a687059c 542/*
e50aee73 543 * pregexec and friends
a687059c
LW
544 */
545
76234dfb 546#ifndef PERL_IN_XSUB_RE
a687059c 547/*
c277df42 548 - pregexec - match a regexp against a string
a687059c 549 */
c277df42 550I32
5aaab254 551Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
ea3daa5d 552 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
8fd1a950
DM
553/* stringarg: the point in the string at which to begin matching */
554/* strend: pointer to null at end of string */
555/* strbeg: real beginning of string */
556/* minend: end of match must be >= minend bytes after stringarg. */
557/* screamer: SV being matched: only used for utf8 flag, pos() etc; string
558 * itself is accessed via the pointers above */
559/* nosave: For optimizations. */
c277df42 560{
7918f24d
NC
561 PERL_ARGS_ASSERT_PREGEXEC;
562
c277df42 563 return
9041c2e3 564 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
c277df42
IZ
565 nosave ? 0 : REXEC_COPY_STR);
566}
76234dfb 567#endif
22e551b9 568
cad2e5aa 569
6eb5f6b9 570
1a4edc3c
DM
571/* re_intuit_start():
572 *
573 * Based on some optimiser hints, try to find the earliest position in the
574 * string where the regex could match.
575 *
576 * rx: the regex to match against
577 * sv: the SV being matched: only used for utf8 flag; the string
578 * itself is accessed via the pointers below. Note that on
579 * something like an overloaded SV, SvPOK(sv) may be false
580 * and the string pointers may point to something unrelated to
581 * the SV itself.
582 * strbeg: real beginning of string
583 * strpos: the point in the string at which to begin matching
584 * strend: pointer to the byte following the last char of the string
585 * flags currently unused; set to 0
586 * data: currently unused; set to NULL
587 *
588 * The basic idea of re_intuit_start() is to use some known information
589 * about the pattern, namely:
590 *
591 * a) the longest known anchored substring (i.e. one that's at a
592 * constant offset from the beginning of the pattern; but not
593 * necessarily at a fixed offset from the beginning of the
594 * string);
595 * b) the longest floating substring (i.e. one that's not at a constant
596 * offset from the beginning of the pattern);
597 * c) Whether the pattern is anchored to the string; either
598 * an absolute anchor: /^../, or anchored to \n: /^.../m,
599 * or anchored to pos(): /\G/;
600 * d) A start class: a real or synthetic character class which
601 * represents which characters are legal at the start of the pattern;
602 *
603 * to either quickly reject the match, or to find the earliest position
604 * within the string at which the pattern might match, thus avoiding
605 * running the full NFA engine at those earlier locations, only to
606 * eventually fail and retry further along.
607 *
608 * Returns NULL if the pattern can't match, or returns the address within
609 * the string which is the earliest place the match could occur.
610 *
611 * The longest of the anchored and floating substrings is called 'check'
612 * and is checked first. The other is called 'other' and is checked
613 * second. The 'other' substring may not be present. For example,
614 *
615 * /(abc|xyz)ABC\d{0,3}DEFG/
616 *
617 * will have
618 *
619 * check substr (float) = "DEFG", offset 6..9 chars
620 * other substr (anchored) = "ABC", offset 3..3 chars
621 * stclass = [ax]
622 *
623 * Be aware that during the course of this function, sometimes 'anchored'
624 * refers to a substring being anchored relative to the start of the
625 * pattern, and sometimes to the pattern itself being anchored relative to
626 * the string. For example:
627 *
628 * /\dabc/: "abc" is anchored to the pattern;
629 * /^\dabc/: "abc" is anchored to the pattern and the string;
630 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
631 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
632 * but the pattern is anchored to the string.
52a21eb3
DM
633 */
634
cad2e5aa 635char *
52a21eb3
DM
636Perl_re_intuit_start(pTHX_
637 REGEXP * const rx,
638 SV *sv,
639 const char * const strbeg,
640 char *strpos,
641 char *strend,
642 const U32 flags,
643 re_scream_pos_data *data)
cad2e5aa 644{
8d919b0a 645 struct regexp *const prog = ReANY(rx);
6b071d16 646 SSize_t start_shift = prog->check_offset_min;
cad2e5aa 647 /* Should be nonnegative! */
ea3daa5d 648 SSize_t end_shift = 0;
0fc004dd
DM
649 /* current lowest pos in string where the regex can start matching */
650 char *rx_origin = strpos;
eb578fdb 651 SV *check;
f2ed9b32 652 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
6480a6c4 653 U8 other_ix = 1 - prog->substrs->check_ix;
6ad5ffb3 654 bool ml_anch = 0;
8f4bf5fc 655 char *other_last = strpos;/* latest pos 'other' substr already checked to */
bd61b366 656 char *check_at = NULL; /* check substr found at this pos */
bbe252da 657 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
f8fc2ecf 658 RXi_GET_DECL(prog,progi);
02d5137b
DM
659 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
660 regmatch_info *const reginfo = &reginfo_buf;
a3621e74
YO
661 GET_RE_DEBUG_FLAGS_DECL;
662
7918f24d 663 PERL_ARGS_ASSERT_RE_INTUIT_START;
c33e64f0
FC
664 PERL_UNUSED_ARG(flags);
665 PERL_UNUSED_ARG(data);
7918f24d 666
6ad9a8ab 667 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0
DM
668 "Intuit: trying to determine minimum start position...\n"));
669
fb9bbddb 670 /* for now, assume that all substr offsets are positive. If at some point
f67a5002 671 * in the future someone wants to do clever things with lookbehind and
fb9bbddb
DM
672 * -ve offsets, they'll need to fix up any code in this function
673 * which uses these offsets. See the thread beginning
674 * <20140113145929.GF27210@iabyn.com>
675 */
676 assert(prog->substrs->data[0].min_offset >= 0);
677 assert(prog->substrs->data[0].max_offset >= 0);
678 assert(prog->substrs->data[1].min_offset >= 0);
679 assert(prog->substrs->data[1].max_offset >= 0);
680 assert(prog->substrs->data[2].min_offset >= 0);
681 assert(prog->substrs->data[2].max_offset >= 0);
682
f7022b5a 683 /* for now, assume that if both present, that the floating substring
83f2232d 684 * doesn't start before the anchored substring.
f7022b5a
DM
685 * If you break this assumption (e.g. doing better optimisations
686 * with lookahead/behind), then you'll need to audit the code in this
687 * function carefully first
688 */
689 assert(
690 ! ( (prog->anchored_utf8 || prog->anchored_substr)
691 && (prog->float_utf8 || prog->float_substr))
692 || (prog->float_min_offset >= prog->anchored_offset));
693
1a4edc3c
DM
694 /* byte rather than char calculation for efficiency. It fails
695 * to quickly reject some cases that can't match, but will reject
696 * them later after doing full char arithmetic */
c344f387 697 if (prog->minlen > strend - strpos) {
6ad9a8ab 698 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0 699 " String too short...\n"));
cad2e5aa 700 goto fail;
2c2d71f5 701 }
d8da0584 702
ab4e48c1 703 RX_MATCH_UTF8_set(rx,utf8_target);
6c3fea77 704 reginfo->is_utf8_target = cBOOL(utf8_target);
bf2039a9 705 reginfo->info_aux = NULL;
9d9163fb 706 reginfo->strbeg = strbeg;
220db18a 707 reginfo->strend = strend;
aed7b151 708 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
02d5137b 709 reginfo->intuit = 1;
1cb48e53
DM
710 /* not actually used within intuit, but zero for safety anyway */
711 reginfo->poscache_maxiter = 0;
02d5137b 712
f2ed9b32 713 if (utf8_target) {
2814f4b3
HS
714 if ((!prog->anchored_utf8 && prog->anchored_substr)
715 || (!prog->float_utf8 && prog->float_substr))
33b8afdf
JH
716 to_utf8_substr(prog);
717 check = prog->check_utf8;
718 } else {
7e0d5ad7
KW
719 if (!prog->check_substr && prog->check_utf8) {
720 if (! to_byte_substr(prog)) {
6b54ddc5 721 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
7e0d5ad7
KW
722 }
723 }
33b8afdf
JH
724 check = prog->check_substr;
725 }
274cd312 726
1dc475d0
DM
727 /* dump the various substring data */
728 DEBUG_OPTIMISE_MORE_r({
729 int i;
730 for (i=0; i<=2; i++) {
731 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
732 : prog->substrs->data[i].substr);
733 if (!sv)
734 continue;
735
6ad9a8ab 736 Perl_re_printf( aTHX_
147e3846
KW
737 " substrs[%d]: min=%" IVdf " max=%" IVdf " end shift=%" IVdf
738 " useful=%" IVdf " utf8=%d [%s]\n",
1dc475d0
DM
739 i,
740 (IV)prog->substrs->data[i].min_offset,
741 (IV)prog->substrs->data[i].max_offset,
742 (IV)prog->substrs->data[i].end_shift,
743 BmUSEFUL(sv),
744 utf8_target ? 1 : 0,
745 SvPEEK(sv));
746 }
747 });
748
8e1490ee 749 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
9fc7410e
DM
750
751 /* ml_anch: check after \n?
752 *
0fa70a06 753 * A note about PREGf_IMPLICIT: on an un-anchored pattern beginning
9fc7410e
DM
754 * with /.*.../, these flags will have been added by the
755 * compiler:
756 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
757 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
758 */
7d2d37f5
DM
759 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
760 && !(prog->intflags & PREGf_IMPLICIT);
cad2e5aa 761
343c8a29 762 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
c889ccc8
DM
763 /* we are only allowed to match at BOS or \G */
764
57fcbfa7 765 /* trivially reject if there's a BOS anchor and we're not at BOS.
7bb3b9eb
DM
766 *
767 * Note that we don't try to do a similar quick reject for
768 * \G, since generally the caller will have calculated strpos
769 * based on pos() and gofs, so the string is already correctly
770 * anchored by definition; and handling the exceptions would
771 * be too fiddly (e.g. REXEC_IGNOREPOS).
57fcbfa7 772 */
7bb3b9eb 773 if ( strpos != strbeg
d3d47aac 774 && (prog->intflags & PREGf_ANCH_SBOL))
c889ccc8 775 {
6ad9a8ab 776 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0 777 " Not at start...\n"));
c889ccc8
DM
778 goto fail;
779 }
780
a5d12a4b
DM
781 /* in the presence of an anchor, the anchored (relative to the
782 * start of the regex) substr must also be anchored relative
66b7ec5c
DM
783 * to strpos. So quickly reject if substr isn't found there.
784 * This works for \G too, because the caller will already have
785 * subtracted gofs from pos, and gofs is the offset from the
786 * \G to the start of the regex. For example, in /.abc\Gdef/,
787 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
788 * caller will have set strpos=pos()-4; we look for the substr
789 * at position pos()-4+1, which lines up with the "a" */
a5d12a4b 790
33c28ab2 791 if (prog->check_offset_min == prog->check_offset_max) {
c889ccc8 792 /* Substring at constant offset from beg-of-str... */
d307bf57 793 SSize_t slen = SvCUR(check);
343c8a29 794 char *s = HOP3c(strpos, prog->check_offset_min, strend);
1de06328 795
6ad9a8ab 796 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 797 " Looking for check substr at fixed offset %" IVdf "...\n",
1dc475d0
DM
798 (IV)prog->check_offset_min));
799
7742aa66
DM
800 if (SvTAIL(check)) {
801 /* In this case, the regex is anchored at the end too.
802 * Unless it's a multiline match, the lengths must match
803 * exactly, give or take a \n. NB: slen >= 1 since
804 * the last char of check is \n */
805 if (!multiline
806 && ( strend - s > slen
807 || strend - s < slen - 1
808 || (strend - s == slen && strend[-1] != '\n')))
c889ccc8 809 {
6ad9a8ab 810 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0 811 " String too long...\n"));
c889ccc8
DM
812 goto fail_finish;
813 }
814 /* Now should match s[0..slen-2] */
815 slen--;
c889ccc8 816 }
26fb2318
TC
817 if (slen && (strend - s < slen
818 || *SvPVX_const(check) != *s
819 || (slen > 1 && (memNE(SvPVX_const(check), s, slen)))))
d307bf57 820 {
6ad9a8ab 821 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0 822 " String not equal...\n"));
d307bf57
DM
823 goto fail_finish;
824 }
c889ccc8
DM
825
826 check_at = s;
827 goto success_at_start;
cad2e5aa 828 }
cad2e5aa 829 }
cad2e5aa 830 }
0fc004dd 831
c0e0ec46 832 end_shift = prog->check_end_shift;
cad2e5aa 833
19188028 834#ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
0033605d 835 if (end_shift < 0)
147e3846 836 Perl_croak(aTHX_ "panic: end_shift: %" IVdf " pattern:\n%s\n ",
220fc49f 837 (IV)end_shift, RX_PRECOMP(prog));
2c2d71f5
JH
838#endif
839
2c2d71f5 840 restart:
1de06328 841
66b7ec5c
DM
842 /* This is the (re)entry point of the main loop in this function.
843 * The goal of this loop is to:
844 * 1) find the "check" substring in the region rx_origin..strend
845 * (adjusted by start_shift / end_shift). If not found, reject
846 * immediately.
847 * 2) If it exists, look for the "other" substr too if defined; for
848 * example, if the check substr maps to the anchored substr, then
849 * check the floating substr, and vice-versa. If not found, go
850 * back to (1) with rx_origin suitably incremented.
851 * 3) If we find an rx_origin position that doesn't contradict
852 * either of the substrings, then check the possible additional
853 * constraints on rx_origin of /^.../m or a known start class.
854 * If these fail, then depending on which constraints fail, jump
855 * back to here, or to various other re-entry points further along
856 * that skip some of the first steps.
857 * 4) If we pass all those tests, update the BmUSEFUL() count on the
858 * substring. If the start position was determined to be at the
859 * beginning of the string - so, not rejected, but not optimised,
860 * since we have to run regmatch from position 0 - decrement the
861 * BmUSEFUL() count. Otherwise increment it.
862 */
863
1a4edc3c
DM
864
865 /* first, look for the 'check' substring */
866
1de06328 867 {
c33e64f0
FC
868 U8* start_point;
869 U8* end_point;
c889ccc8 870
c889ccc8 871 DEBUG_OPTIMISE_MORE_r({
6ad9a8ab 872 Perl_re_printf( aTHX_
147e3846
KW
873 " At restart: rx_origin=%" IVdf " Check offset min: %" IVdf
874 " Start shift: %" IVdf " End shift %" IVdf
875 " Real end Shift: %" IVdf "\n",
675e93ee 876 (IV)(rx_origin - strbeg),
c889ccc8 877 (IV)prog->check_offset_min,
12fbc530
DM
878 (IV)start_shift,
879 (IV)end_shift,
c889ccc8
DM
880 (IV)prog->check_end_shift);
881 });
1de06328 882
33c28ab2
DM
883 end_point = HOP3(strend, -end_shift, strbeg);
884 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
885 if (!start_point)
886 goto fail_finish;
c889ccc8 887
557f47af 888
e0362b86 889 /* If the regex is absolutely anchored to either the start of the
d3d47aac 890 * string (SBOL) or to pos() (ANCH_GPOS), then
e0362b86
DM
891 * check_offset_max represents an upper bound on the string where
892 * the substr could start. For the ANCH_GPOS case, we assume that
893 * the caller of intuit will have already set strpos to
894 * pos()-gofs, so in this case strpos + offset_max will still be
895 * an upper bound on the substr.
896 */
c19c836a
DM
897 if (!ml_anch
898 && prog->intflags & PREGf_ANCH
e0362b86 899 && prog->check_offset_max != SSize_t_MAX)
c19c836a 900 {
1a08ba3a 901 SSize_t len = SvCUR(check) - !!SvTAIL(check);
e0362b86
DM
902 const char * const anchor =
903 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
904
905 /* do a bytes rather than chars comparison. It's conservative;
906 * so it skips doing the HOP if the result can't possibly end
907 * up earlier than the old value of end_point.
908 */
909 if ((char*)end_point - anchor > prog->check_offset_max) {
910 end_point = HOP3lim((U8*)anchor,
911 prog->check_offset_max,
912 end_point -len)
913 + len;
914 }
d6ef1678
DM
915 }
916
ae5d4331 917 check_at = fbm_instr( start_point, end_point,
7fba1cd6 918 check, multiline ? FBMrf_MULTILINE : 0);
c889ccc8 919
6ad9a8ab 920 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 921 " doing 'check' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
675e93ee
DM
922 (IV)((char*)start_point - strbeg),
923 (IV)((char*)end_point - strbeg),
924 (IV)(check_at ? check_at - strbeg : -1)
925 ));
926
8fd34720
DM
927 /* Update the count-of-usability, remove useless subpatterns,
928 unshift s. */
929
930 DEBUG_EXECUTE_r({
931 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
932 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
6ad9a8ab 933 Perl_re_printf( aTHX_ " %s %s substr %s%s%s",
8fd34720
DM
934 (check_at ? "Found" : "Did not find"),
935 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
936 ? "anchored" : "floating"),
937 quoted,
938 RE_SV_TAIL(check),
939 (check_at ? " at offset " : "...\n") );
940 });
2c2d71f5 941
8fd34720
DM
942 if (!check_at)
943 goto fail_finish;
8fd34720
DM
944 /* set rx_origin to the minimum position where the regex could start
945 * matching, given the constraint of the just-matched check substring.
946 * But don't set it lower than previously.
947 */
fdc003fd 948
8fd34720
DM
949 if (check_at - rx_origin > prog->check_offset_max)
950 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
675e93ee 951 /* Finish the diagnostic message */
6ad9a8ab 952 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 953 "%ld (rx_origin now %" IVdf ")...\n",
675e93ee
DM
954 (long)(check_at - strbeg),
955 (IV)(rx_origin - strbeg)
956 ));
8fd34720 957 }
fdc003fd
DM
958
959
1a4edc3c 960 /* now look for the 'other' substring if defined */
2c2d71f5 961
6480a6c4
DM
962 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
963 : prog->substrs->data[other_ix].substr)
1de06328 964 {
30944b6d 965 /* Take into account the "other" substring. */
6c3343a6
DM
966 char *last, *last1;
967 char *s;
968 SV* must;
969 struct reg_substr_datum *other;
970
971 do_other_substr:
972 other = &prog->substrs->data[other_ix];
973
974 /* if "other" is anchored:
975 * we've previously found a floating substr starting at check_at.
976 * This means that the regex origin must lie somewhere
977 * between min (rx_origin): HOP3(check_at, -check_offset_max)
978 * and max: HOP3(check_at, -check_offset_min)
979 * (except that min will be >= strpos)
980 * So the fixed substr must lie somewhere between
981 * HOP3(min, anchored_offset)
982 * HOP3(max, anchored_offset) + SvCUR(substr)
983 */
984
985 /* if "other" is floating
986 * Calculate last1, the absolute latest point where the
987 * floating substr could start in the string, ignoring any
988 * constraints from the earlier fixed match. It is calculated
989 * as follows:
990 *
991 * strend - prog->minlen (in chars) is the absolute latest
992 * position within the string where the origin of the regex
993 * could appear. The latest start point for the floating
994 * substr is float_min_offset(*) on from the start of the
995 * regex. last1 simply combines thee two offsets.
996 *
997 * (*) You might think the latest start point should be
998 * float_max_offset from the regex origin, and technically
999 * you'd be correct. However, consider
1000 * /a\d{2,4}bcd\w/
1001 * Here, float min, max are 3,5 and minlen is 7.
1002 * This can match either
1003 * /a\d\dbcd\w/
1004 * /a\d\d\dbcd\w/
1005 * /a\d\d\d\dbcd\w/
1006 * In the first case, the regex matches minlen chars; in the
1007 * second, minlen+1, in the third, minlen+2.
1008 * In the first case, the floating offset is 3 (which equals
1009 * float_min), in the second, 4, and in the third, 5 (which
1010 * equals float_max). In all cases, the floating string bcd
1011 * can never start more than 4 chars from the end of the
1012 * string, which equals minlen - float_min. As the substring
1013 * starts to match more than float_min from the start of the
1014 * regex, it makes the regex match more than minlen chars,
1015 * and the two cancel each other out. So we can always use
1016 * float_min - minlen, rather than float_max - minlen for the
1017 * latest position in the string.
1018 *
1019 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1020 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1021 */
1022
e7a14a9c 1023 assert(prog->minlen >= other->min_offset);
6c3343a6
DM
1024 last1 = HOP3c(strend,
1025 other->min_offset - prog->minlen, strbeg);
1026
4d006249 1027 if (other_ix) {/* i.e. if (other-is-float) */
6c3343a6
DM
1028 /* last is the latest point where the floating substr could
1029 * start, *given* any constraints from the earlier fixed
1030 * match. This constraint is that the floating string starts
1031 * <= float_max_offset chars from the regex origin (rx_origin).
1032 * If this value is less than last1, use it instead.
eb3831ce 1033 */
6c3343a6
DM
1034 assert(rx_origin <= last1);
1035 last =
1036 /* this condition handles the offset==infinity case, and
1037 * is a short-cut otherwise. Although it's comparing a
1038 * byte offset to a char length, it does so in a safe way,
1039 * since 1 char always occupies 1 or more bytes,
1040 * so if a string range is (last1 - rx_origin) bytes,
1041 * it will be less than or equal to (last1 - rx_origin)
1042 * chars; meaning it errs towards doing the accurate HOP3
1043 * rather than just using last1 as a short-cut */
1044 (last1 - rx_origin) < other->max_offset
1045 ? last1
1046 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1047 }
1048 else {
1049 assert(strpos + start_shift <= check_at);
1050 last = HOP4c(check_at, other->min_offset - start_shift,
1051 strbeg, strend);
1052 }
ead917d0 1053
6c3343a6
DM
1054 s = HOP3c(rx_origin, other->min_offset, strend);
1055 if (s < other_last) /* These positions already checked */
1056 s = other_last;
1057
1058 must = utf8_target ? other->utf8_substr : other->substr;
1059 assert(SvPOK(must));
675e93ee
DM
1060 {
1061 char *from = s;
1062 char *to = last + SvCUR(must) - (SvTAIL(must)!=0);
1063
71a9d105
DM
1064 if (to > strend)
1065 to = strend;
88203927
DM
1066 if (from > to) {
1067 s = NULL;
6ad9a8ab 1068 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 1069 " skipping 'other' fbm scan: %" IVdf " > %" IVdf "\n",
88203927
DM
1070 (IV)(from - strbeg),
1071 (IV)(to - strbeg)
1072 ));
1073 }
1074 else {
1075 s = fbm_instr(
1076 (unsigned char*)from,
1077 (unsigned char*)to,
1078 must,
1079 multiline ? FBMrf_MULTILINE : 0
1080 );
6ad9a8ab 1081 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 1082 " doing 'other' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
88203927
DM
1083 (IV)(from - strbeg),
1084 (IV)(to - strbeg),
1085 (IV)(s ? s - strbeg : -1)
1086 ));
1087 }
675e93ee
DM
1088 }
1089
6c3343a6
DM
1090 DEBUG_EXECUTE_r({
1091 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1092 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
6ad9a8ab 1093 Perl_re_printf( aTHX_ " %s %s substr %s%s",
6c3343a6
DM
1094 s ? "Found" : "Contradicts",
1095 other_ix ? "floating" : "anchored",
1096 quoted, RE_SV_TAIL(must));
1097 });
ead917d0 1098
ead917d0 1099
6c3343a6
DM
1100 if (!s) {
1101 /* last1 is latest possible substr location. If we didn't
1102 * find it before there, we never will */
1103 if (last >= last1) {
6ad9a8ab 1104 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
675e93ee 1105 "; giving up...\n"));
6c3343a6 1106 goto fail_finish;
ead917d0
DM
1107 }
1108
6c3343a6
DM
1109 /* try to find the check substr again at a later
1110 * position. Maybe next time we'll find the "other" substr
1111 * in range too */
6c3343a6
DM
1112 other_last = HOP3c(last, 1, strend) /* highest failure */;
1113 rx_origin =
4d006249 1114 other_ix /* i.e. if other-is-float */
6c3343a6
DM
1115 ? HOP3c(rx_origin, 1, strend)
1116 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
6ad9a8ab 1117 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 1118 "; about to retry %s at offset %ld (rx_origin now %" IVdf ")...\n",
675e93ee
DM
1119 (other_ix ? "floating" : "anchored"),
1120 (long)(HOP3c(check_at, 1, strend) - strbeg),
1121 (IV)(rx_origin - strbeg)
1122 ));
6c3343a6
DM
1123 goto restart;
1124 }
1125 else {
4d006249 1126 if (other_ix) { /* if (other-is-float) */
6c3343a6
DM
1127 /* other_last is set to s, not s+1, since its possible for
1128 * a floating substr to fail first time, then succeed
1129 * second time at the same floating position; e.g.:
1130 * "-AB--AABZ" =~ /\wAB\d*Z/
1131 * The first time round, anchored and float match at
1132 * "-(AB)--AAB(Z)" then fail on the initial \w character
1133 * class. Second time round, they match at "-AB--A(AB)(Z)".
1134 */
1135 other_last = s;
ead917d0
DM
1136 }
1137 else {
6c3343a6
DM
1138 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1139 other_last = HOP3c(s, 1, strend);
ead917d0 1140 }
6ad9a8ab 1141 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 1142 " at offset %ld (rx_origin now %" IVdf ")...\n",
675e93ee
DM
1143 (long)(s - strbeg),
1144 (IV)(rx_origin - strbeg)
1145 ));
1146
6c3343a6 1147 }
cad2e5aa 1148 }
acba93e8
DM
1149 else {
1150 DEBUG_OPTIMISE_MORE_r(
6ad9a8ab 1151 Perl_re_printf( aTHX_
147e3846
KW
1152 " Check-only match: offset min:%" IVdf " max:%" IVdf
1153 " check_at:%" IVdf " rx_origin:%" IVdf " rx_origin-check_at:%" IVdf
1154 " strend:%" IVdf "\n",
acba93e8
DM
1155 (IV)prog->check_offset_min,
1156 (IV)prog->check_offset_max,
675e93ee
DM
1157 (IV)(check_at-strbeg),
1158 (IV)(rx_origin-strbeg),
1c1c599d 1159 (IV)(rx_origin-check_at),
675e93ee 1160 (IV)(strend-strbeg)
acba93e8
DM
1161 )
1162 );
1163 }
2c2d71f5 1164
acba93e8 1165 postprocess_substr_matches:
0991020e 1166
1a4edc3c 1167 /* handle the extra constraint of /^.../m if present */
e3c6feb0 1168
7d2d37f5 1169 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
4620cb61
DM
1170 char *s;
1171
6ad9a8ab 1172 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
a62659bd 1173 " looking for /^/m anchor"));
d0880ea7
DM
1174
1175 /* we have failed the constraint of a \n before rx_origin.
2e759faa
DM
1176 * Find the next \n, if any, even if it's beyond the current
1177 * anchored and/or floating substrings. Whether we should be
1178 * scanning ahead for the next \n or the next substr is debatable.
1179 * On the one hand you'd expect rare substrings to appear less
1180 * often than \n's. On the other hand, searching for \n means
675e93ee 1181 * we're effectively flipping between check_substr and "\n" on each
2e759faa
DM
1182 * iteration as the current "rarest" string candidate, which
1183 * means for example that we'll quickly reject the whole string if
1184 * hasn't got a \n, rather than trying every substr position
1185 * first
1186 */
d0880ea7 1187
4620cb61
DM
1188 s = HOP3c(strend, - prog->minlen, strpos);
1189 if (s <= rx_origin ||
1190 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1191 {
6ad9a8ab 1192 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
d0880ea7
DM
1193 " Did not find /%s^%s/m...\n",
1194 PL_colors[0], PL_colors[1]));
a62659bd
DM
1195 goto fail_finish;
1196 }
d0880ea7 1197
4ada1233
DM
1198 /* earliest possible origin is 1 char after the \n.
1199 * (since *rx_origin == '\n', it's safe to ++ here rather than
1200 * HOP(rx_origin, 1)) */
1201 rx_origin++;
d0880ea7 1202
f4f115de 1203 if (prog->substrs->check_ix == 0 /* check is anchored */
4ada1233 1204 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
f4f115de 1205 {
d0880ea7
DM
1206 /* Position contradicts check-string; either because
1207 * check was anchored (and thus has no wiggle room),
4ada1233 1208 * or check was float and rx_origin is above the float range */
6ad9a8ab 1209 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
675e93ee
DM
1210 " Found /%s^%s/m, about to restart lookup for check-string with rx_origin %ld...\n",
1211 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
d0880ea7
DM
1212 goto restart;
1213 }
1214
1215 /* if we get here, the check substr must have been float,
2e759faa 1216 * is in range, and we may or may not have had an anchored
d0880ea7
DM
1217 * "other" substr which still contradicts */
1218 assert(prog->substrs->check_ix); /* check is float */
1219
1220 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1221 /* whoops, the anchored "other" substr exists, so we still
1222 * contradict. On the other hand, the float "check" substr
1223 * didn't contradict, so just retry the anchored "other"
1224 * substr */
6ad9a8ab 1225 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846 1226 " Found /%s^%s/m, rescanning for anchored from offset %" IVdf " (rx_origin now %" IVdf ")...\n",
d0880ea7 1227 PL_colors[0], PL_colors[1],
73e8ff00
DM
1228 (IV)(rx_origin - strbeg + prog->anchored_offset),
1229 (IV)(rx_origin - strbeg)
675e93ee 1230 ));
d0880ea7
DM
1231 goto do_other_substr;
1232 }
1233
1234 /* success: we don't contradict the found floating substring
1235 * (and there's no anchored substr). */
6ad9a8ab 1236 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
675e93ee
DM
1237 " Found /%s^%s/m with rx_origin %ld...\n",
1238 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
e3c6feb0
DM
1239 }
1240 else {
6ad9a8ab 1241 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
fe4f3442 1242 " (multiline anchor test skipped)\n"));
e3c6feb0
DM
1243 }
1244
ffad1e6a 1245 success_at_start:
e3c6feb0 1246
cad2e5aa 1247
dd170ff5
DM
1248 /* if we have a starting character class, then test that extra constraint.
1249 * (trie stclasses are too expensive to use here, we are better off to
1250 * leave it to regmatch itself) */
1251
f8fc2ecf 1252 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
f8fc2ecf 1253 const U8* const str = (U8*)STRING(progi->regstclass);
0991020e 1254
2c75e362 1255 /* XXX this value could be pre-computed */
f8fc2ecf 1256 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
2c75e362
DM
1257 ? (reginfo->is_utf8_pat
1258 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1259 : STR_LEN(progi->regstclass))
66e933ab 1260 : 1);
1de06328 1261 char * endpos;
fa3bb21d 1262 char *s;
000dfd2d
DM
1263 /* latest pos that a matching float substr constrains rx start to */
1264 char *rx_max_float = NULL;
1265
c75a3985
DM
1266 /* if the current rx_origin is anchored, either by satisfying an
1267 * anchored substring constraint, or a /^.../m constraint, then we
1268 * can reject the current origin if the start class isn't found
1269 * at the current position. If we have a float-only match, then
1270 * rx_origin is constrained to a range; so look for the start class
1271 * in that range. if neither, then look for the start class in the
1272 * whole rest of the string */
1273
dd170ff5
DM
1274 /* XXX DAPM it's not clear what the minlen test is for, and why
1275 * it's not used in the floating case. Nothing in the test suite
1276 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1277 * Here are some old comments, which may or may not be correct:
1278 *
1279 * minlen == 0 is possible if regstclass is \b or \B,
1280 * and the fixed substr is ''$.
1281 * Since minlen is already taken into account, rx_origin+1 is
1282 * before strend; accidentally, minlen >= 1 guaranties no false
1283 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1284 * 0) below assumes that regstclass does not come from lookahead...
1285 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1286 * This leaves EXACTF-ish only, which are dealt with in
1287 * find_byclass().
1288 */
1289
7d2d37f5 1290 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
fa3bb21d 1291 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
000dfd2d
DM
1292 else if (prog->float_substr || prog->float_utf8) {
1293 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1294 endpos= HOP3c(rx_max_float, cl_l, strend);
1295 }
1de06328
YO
1296 else
1297 endpos= strend;
1298
6ad9a8ab 1299 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
147e3846
KW
1300 " looking for class: start_shift: %" IVdf " check_at: %" IVdf
1301 " rx_origin: %" IVdf " endpos: %" IVdf "\n",
1dc475d0 1302 (IV)start_shift, (IV)(check_at - strbeg),
c43b5520 1303 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
d8080198 1304
c43b5520 1305 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
f9176b44 1306 reginfo);
be778b1a 1307 if (!s) {
6eb5f6b9 1308 if (endpos == strend) {
6ad9a8ab 1309 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1dc475d0 1310 " Could not match STCLASS...\n") );
6eb5f6b9
JH
1311 goto fail;
1312 }
6ad9a8ab 1313 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1dc475d0 1314 " This position contradicts STCLASS...\n") );
e0eb31e7
DM
1315 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1316 && !(prog->intflags & PREGf_IMPLICIT))
653099ff 1317 goto fail;
9fed8d02 1318
6eb5f6b9 1319 /* Contradict one of substrings */
97136c8a
DM
1320 if (prog->anchored_substr || prog->anchored_utf8) {
1321 if (prog->substrs->check_ix == 1) { /* check is float */
1322 /* Have both, check_string is floating */
1323 assert(rx_origin + start_shift <= check_at);
1324 if (rx_origin + start_shift != check_at) {
1325 /* not at latest position float substr could match:
c75a3985
DM
1326 * Recheck anchored substring, but not floating.
1327 * The condition above is in bytes rather than
1328 * chars for efficiency. It's conservative, in
1329 * that it errs on the side of doing 'goto
88203927
DM
1330 * do_other_substr'. In this case, at worst,
1331 * an extra anchored search may get done, but in
1332 * practice the extra fbm_instr() is likely to
1333 * get skipped anyway. */
6ad9a8ab 1334 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
147e3846 1335 " about to retry anchored at offset %ld (rx_origin now %" IVdf ")...\n",
675e93ee
DM
1336 (long)(other_last - strbeg),
1337 (IV)(rx_origin - strbeg)
1338 ));
97136c8a 1339 goto do_other_substr;
3369914b 1340 }
3369914b
DM
1341 }
1342 }
97136c8a 1343 else {
9fed8d02
DM
1344 /* float-only */
1345
7d2d37f5 1346 if (ml_anch) {
c75a3985
DM
1347 /* In the presence of ml_anch, we might be able to
1348 * find another \n without breaking the current float
1349 * constraint. */
1350
1351 /* strictly speaking this should be HOP3c(..., 1, ...),
1352 * but since we goto a block of code that's going to
1353 * search for the next \n if any, its safe here */
9fed8d02 1354 rx_origin++;
6ad9a8ab 1355 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
675e93ee 1356 " about to look for /%s^%s/m starting at rx_origin %ld...\n",
9fed8d02 1357 PL_colors[0], PL_colors[1],
675e93ee 1358 (long)(rx_origin - strbeg)) );
9fed8d02 1359 goto postprocess_substr_matches;
ab60c45a 1360 }
c75a3985
DM
1361
1362 /* strictly speaking this can never be true; but might
1363 * be if we ever allow intuit without substrings */
1364 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
9fed8d02 1365 goto fail;
c75a3985 1366
000dfd2d 1367 rx_origin = rx_max_float;
9fed8d02
DM
1368 }
1369
c75a3985
DM
1370 /* at this point, any matching substrings have been
1371 * contradicted. Start again... */
1372
9fed8d02 1373 rx_origin = HOP3c(rx_origin, 1, strend);
557f47af
DM
1374
1375 /* uses bytes rather than char calculations for efficiency.
1376 * It's conservative: it errs on the side of doing 'goto restart',
1377 * where there is code that does a proper char-based test */
9fed8d02 1378 if (rx_origin + start_shift + end_shift > strend) {
6ad9a8ab 1379 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
9fed8d02
DM
1380 " Could not match STCLASS...\n") );
1381 goto fail;
1382 }
6ad9a8ab 1383 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
147e3846 1384 " about to look for %s substr starting at offset %ld (rx_origin now %" IVdf ")...\n",
9fed8d02 1385 (prog->substrs->check_ix ? "floating" : "anchored"),
675e93ee
DM
1386 (long)(rx_origin + start_shift - strbeg),
1387 (IV)(rx_origin - strbeg)
1388 ));
9fed8d02 1389 goto restart;
6eb5f6b9 1390 }
9fed8d02 1391
c75a3985
DM
1392 /* Success !!! */
1393
5f9c6575 1394 if (rx_origin != s) {
6ad9a8ab 1395 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0 1396 " By STCLASS: moving %ld --> %ld\n",
675e93ee 1397 (long)(rx_origin - strbeg), (long)(s - strbeg))
b7953727
JH
1398 );
1399 }
1400 else {
6ad9a8ab 1401 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1dc475d0 1402 " Does not contradict STCLASS...\n");
b7953727
JH
1403 );
1404 }
6eb5f6b9 1405 }
ffad1e6a
DM
1406
1407 /* Decide whether using the substrings helped */
1408
1409 if (rx_origin != strpos) {
1410 /* Fixed substring is found far enough so that the match
1411 cannot start at strpos. */
1412
6ad9a8ab 1413 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " try at offset...\n"));
ffad1e6a
DM
1414 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1415 }
1416 else {
70563e16
DM
1417 /* The found rx_origin position does not prohibit matching at
1418 * strpos, so calling intuit didn't gain us anything. Decrement
1419 * the BmUSEFUL() count on the check substring, and if we reach
1420 * zero, free it. */
1421 if (!(prog->intflags & PREGf_NAUGHTY)
ffad1e6a
DM
1422 && (utf8_target ? (
1423 prog->check_utf8 /* Could be deleted already */
1424 && --BmUSEFUL(prog->check_utf8) < 0
1425 && (prog->check_utf8 == prog->float_utf8)
1426 ) : (
1427 prog->check_substr /* Could be deleted already */
1428 && --BmUSEFUL(prog->check_substr) < 0
1429 && (prog->check_substr == prog->float_substr)
1430 )))
1431 {
1432 /* If flags & SOMETHING - do not do it many times on the same match */
6ad9a8ab 1433 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " ... Disabling check substring...\n"));
ffad1e6a
DM
1434 /* XXX Does the destruction order has to change with utf8_target? */
1435 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1436 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1437 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1438 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1439 check = NULL; /* abort */
ffad1e6a
DM
1440 /* XXXX This is a remnant of the old implementation. It
1441 looks wasteful, since now INTUIT can use many
1442 other heuristics. */
1443 prog->extflags &= ~RXf_USE_INTUIT;
ffad1e6a
DM
1444 }
1445 }
1446
6ad9a8ab 1447 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
ffad1e6a 1448 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
675e93ee 1449 PL_colors[4], PL_colors[5], (long)(rx_origin - strbeg)) );
ffad1e6a 1450
c765d6e0 1451 return rx_origin;
2c2d71f5
JH
1452
1453 fail_finish: /* Substring not found */
33b8afdf 1454 if (prog->check_substr || prog->check_utf8) /* could be removed already */
f2ed9b32 1455 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
cad2e5aa 1456 fail:
6ad9a8ab 1457 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch rejected by optimizer%s\n",
e4584336 1458 PL_colors[4], PL_colors[5]));
bd61b366 1459 return NULL;
cad2e5aa 1460}
9661b544 1461
70563e16 1462
a0a388a1 1463#define DECL_TRIE_TYPE(scan) \
e7fd4aa1 1464 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
a4525e78
KW
1465 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold, \
1466 trie_utf8l, trie_flu8 } \
e7fd4aa1
KW
1467 trie_type = ((scan->flags == EXACT) \
1468 ? (utf8_target ? trie_utf8 : trie_plain) \
a4525e78
KW
1469 : (scan->flags == EXACTL) \
1470 ? (utf8_target ? trie_utf8l : trie_plain) \
1471 : (scan->flags == EXACTFA) \
1472 ? (utf8_target \
1473 ? trie_utf8_exactfa_fold \
1474 : trie_latin_utf8_exactfa_fold) \
1475 : (scan->flags == EXACTFLU8 \
1476 ? trie_flu8 \
1477 : (utf8_target \
1478 ? trie_utf8_fold \
1479 : trie_latin_utf8_fold)))
fab2782b 1480
fd3249ee 1481#define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
baa60164 1482STMT_START { \
fab2782b 1483 STRLEN skiplen; \
baa60164 1484 U8 flags = FOLD_FLAGS_FULL; \
fab2782b 1485 switch (trie_type) { \
a4525e78 1486 case trie_flu8: \
780fcc9f 1487 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
613abc6d
KW
1488 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1489 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \
1490 } \
a4525e78 1491 goto do_trie_utf8_fold; \
31f05a37 1492 case trie_utf8_exactfa_fold: \
baa60164 1493 flags |= FOLD_FLAGS_NOMIX_ASCII; \
8e57b935 1494 /* FALLTHROUGH */ \
fab2782b 1495 case trie_utf8_fold: \
a4525e78 1496 do_trie_utf8_fold: \
fab2782b 1497 if ( foldlen>0 ) { \
c80e42f3 1498 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO
1499 foldlen -= len; \
1500 uscan += len; \
1501 len=0; \
1502 } else { \
567b353c 1503 uvc = _toFOLD_utf8_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
fab2782b 1504 len = UTF8SKIP(uc); \
5f560d8a 1505 skiplen = UVCHR_SKIP( uvc ); \
fab2782b
YO
1506 foldlen -= skiplen; \
1507 uscan = foldbuf + skiplen; \
1508 } \
1509 break; \
baa60164
KW
1510 case trie_latin_utf8_exactfa_fold: \
1511 flags |= FOLD_FLAGS_NOMIX_ASCII; \
8e57b935 1512 /* FALLTHROUGH */ \
fab2782b
YO
1513 case trie_latin_utf8_fold: \
1514 if ( foldlen>0 ) { \
c80e42f3 1515 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO
1516 foldlen -= len; \
1517 uscan += len; \
1518 len=0; \
1519 } else { \
1520 len = 1; \
31f05a37 1521 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
5f560d8a 1522 skiplen = UVCHR_SKIP( uvc ); \
fab2782b
YO
1523 foldlen -= skiplen; \
1524 uscan = foldbuf + skiplen; \
1525 } \
1526 break; \
a4525e78 1527 case trie_utf8l: \
780fcc9f 1528 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
613abc6d
KW
1529 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1530 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \
1531 } \
780fcc9f 1532 /* FALLTHROUGH */ \
fab2782b 1533 case trie_utf8: \
c80e42f3 1534 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO
1535 break; \
1536 case trie_plain: \
1537 uvc = (UV)*uc; \
1538 len = 1; \
1539 } \
1540 if (uvc < 256) { \
1541 charid = trie->charmap[ uvc ]; \
1542 } \
1543 else { \
1544 charid = 0; \
1545 if (widecharmap) { \
1546 SV** const svpp = hv_fetch(widecharmap, \
1547 (char*)&uvc, sizeof(UV), 0); \
1548 if (svpp) \
1549 charid = (U16)SvIV(*svpp); \
1550 } \
1551 } \
4cadc6a9
YO
1552} STMT_END
1553
cb41e5d6 1554#define DUMP_EXEC_POS(li,s,doutf8,depth) \
ae7c5b9b 1555 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
cb41e5d6 1556 startpos, doutf8, depth)
ae7c5b9b 1557
c84a03c5 1558#define REXEC_FBC_EXACTISH_SCAN(COND) \
4cadc6a9
YO
1559STMT_START { \
1560 while (s <= e) { \
c84a03c5 1561 if ( (COND) \
fac1af77 1562 && (ln == 1 || folder(s, pat_string, ln)) \
02d5137b 1563 && (reginfo->intuit || regtry(reginfo, &s)) )\
4cadc6a9
YO
1564 goto got_it; \
1565 s++; \
1566 } \
1567} STMT_END
1568
c84a03c5 1569#define REXEC_FBC_UTF8_SCAN(CODE) \
4cadc6a9 1570STMT_START { \
9a902117 1571 while (s < strend) { \
c84a03c5 1572 CODE \
9a902117 1573 s += UTF8SKIP(s); \
4cadc6a9
YO
1574 } \
1575} STMT_END
1576
c84a03c5 1577#define REXEC_FBC_SCAN(CODE) \
4cadc6a9
YO
1578STMT_START { \
1579 while (s < strend) { \
c84a03c5 1580 CODE \
4cadc6a9
YO
1581 s++; \
1582 } \
1583} STMT_END
1584
05bd126c
KW
1585#define REXEC_FBC_UTF8_CLASS_SCAN(COND) \
1586REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \
1587 if (COND) { \
1588 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1589 goto got_it; \
1590 else \
1591 tmp = doevery; \
1592 } \
1593 else \
1594 tmp = 1; \
4cadc6a9
YO
1595)
1596
05bd126c
KW
1597#define REXEC_FBC_CLASS_SCAN(COND) \
1598REXEC_FBC_SCAN( /* Loops while (s < strend) */ \
1599 if (COND) { \
1600 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1601 goto got_it; \
1602 else \
1603 tmp = doevery; \
1604 } \
1605 else \
1606 tmp = 1; \
4cadc6a9
YO
1607)
1608
c84a03c5 1609#define REXEC_FBC_CSCAN(CONDUTF8,COND) \
baa60164 1610 if (utf8_target) { \
c84a03c5 1611 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \
e1d1eefb
YO
1612 } \
1613 else { \
c84a03c5 1614 REXEC_FBC_CLASS_SCAN(COND); \
d981ef24 1615 }
05bd126c 1616
05bd126c
KW
1617/* The three macros below are slightly different versions of the same logic.
1618 *
1619 * The first is for /a and /aa when the target string is UTF-8. This can only
1620 * match ascii, but it must advance based on UTF-8. The other two handle the
1621 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1622 * for the boundary (or non-boundary) between a word and non-word character.
1623 * The utf8 and non-utf8 cases have the same logic, but the details must be
1624 * different. Find the "wordness" of the character just prior to this one, and
1625 * compare it with the wordness of this one. If they differ, we have a
1626 * boundary. At the beginning of the string, pretend that the previous
1627 * character was a new-line.
1628 *
1629 * All these macros uncleanly have side-effects with each other and outside
1630 * variables. So far it's been too much trouble to clean-up
1631 *
1632 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1633 * a word character or not.
1634 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1635 * word/non-word
1636 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1637 *
1638 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1639 * are looking for a boundary or for a non-boundary. If we are looking for a
1640 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1641 * see if this tentative match actually works, and if so, to quit the loop
1642 * here. And vice-versa if we are looking for a non-boundary.
1643 *
1644 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1645 * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of
1646 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1647 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1648 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1649 * complement. But in that branch we complement tmp, meaning that at the
1650 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1651 * which means at the top of the loop in the next iteration, it is
1652 * TEST_NON_UTF8(s-1) */
b2f4e957 1653#define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
05bd126c
KW
1654 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1655 tmp = TEST_NON_UTF8(tmp); \
1656 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1657 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1658 tmp = !tmp; \
1659 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1660 } \
1661 else { \
1662 IF_FAIL; \
1663 } \
1664 ); \
1665
1666/* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1667 * TEST_UTF8 is a macro that for the same input code points returns identically
1668 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
236d82fd 1669#define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
05bd126c
KW
1670 if (s == reginfo->strbeg) { \
1671 tmp = '\n'; \
1672 } \
1673 else { /* Back-up to the start of the previous character */ \
1674 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1675 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
3db24e1e 1676 0, UTF8_ALLOW_DEFAULT); \
05bd126c
KW
1677 } \
1678 tmp = TEST_UV(tmp); \
1679 LOAD_UTF8_CHARCLASS_ALNUM(); \
1680 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
7a207065 1681 if (tmp == ! (TEST_UTF8((U8 *) s, (U8 *) reginfo->strend))) { \
05bd126c
KW
1682 tmp = !tmp; \
1683 IF_SUCCESS; \
1684 } \
1685 else { \
1686 IF_FAIL; \
1687 } \
1688 );
cfaf538b 1689
05bd126c
KW
1690/* Like the above two macros. UTF8_CODE is the complete code for handling
1691 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1692 * macros below */
baa60164 1693#define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
63ac0dad 1694 if (utf8_target) { \
05bd126c 1695 UTF8_CODE \
63ac0dad
KW
1696 } \
1697 else { /* Not utf8 */ \
9d9163fb 1698 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
63ac0dad 1699 tmp = TEST_NON_UTF8(tmp); \
05bd126c 1700 REXEC_FBC_SCAN( /* advances s while s < strend */ \
63ac0dad 1701 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
63ac0dad 1702 IF_SUCCESS; \
760cfa8e 1703 tmp = !tmp; \
63ac0dad
KW
1704 } \
1705 else { \
1706 IF_FAIL; \
1707 } \
1708 ); \
1709 } \
c8519dc7
KW
1710 /* Here, things have been set up by the previous code so that tmp is the \
1711 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1712 * utf8ness of the target). We also have to check if this matches against \
1713 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1714 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1715 * string */ \
1716 if (tmp == ! TEST_NON_UTF8('\n')) { \
1717 IF_SUCCESS; \
1718 } \
1719 else { \
1720 IF_FAIL; \
1721 }
63ac0dad 1722
ae7c5b9b
KW
1723/* This is the macro to use when we want to see if something that looks like it
1724 * could match, actually does, and if so exits the loop */
1725#define REXEC_FBC_TRYIT \
1726 if ((reginfo->intuit || regtry(reginfo, &s))) \
1727 goto got_it
1728
1729/* The only difference between the BOUND and NBOUND cases is that
1730 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1731 * NBOUND. This is accomplished by passing it as either the if or else clause,
1732 * with the other one being empty (PLACEHOLDER is defined as empty).
1733 *
1734 * The TEST_FOO parameters are for operating on different forms of input, but
1735 * all should be ones that return identically for the same underlying code
1736 * points */
1737#define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1738 FBC_BOUND_COMMON( \
1739 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1740 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1741
44129e46 1742#define FBC_BOUND_A(TEST_NON_UTF8) \
ae7c5b9b
KW
1743 FBC_BOUND_COMMON( \
1744 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1745 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1746
1747#define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1748 FBC_BOUND_COMMON( \
1749 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1750 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1751
44129e46 1752#define FBC_NBOUND_A(TEST_NON_UTF8) \
ae7c5b9b
KW
1753 FBC_BOUND_COMMON( \
1754 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1755 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1756
8bde5eaf
JH
1757#ifdef DEBUGGING
1758static IV
1759S_get_break_val_cp_checked(SV* const invlist, const UV cp_in) {
1760 IV cp_out = Perl__invlist_search(invlist, cp_in);
1761 assert(cp_out >= 0);
1762 return cp_out;
1763}
1764# define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
1765 invmap[S_get_break_val_cp_checked(invlist, cp)]
1766#else
1767# define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
1768 invmap[_invlist_search(invlist, cp)]
1769#endif
1770
64935bc6
KW
1771/* Takes a pointer to an inversion list, a pointer to its corresponding
1772 * inversion map, and a code point, and returns the code point's value
1773 * according to the two arrays. It assumes that all code points have a value.
1774 * This is used as the base macro for macros for particular properties */
1775#define _generic_GET_BREAK_VAL_CP(invlist, invmap, cp) \
8bde5eaf 1776 _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp)
64935bc6
KW
1777
1778/* Same as above, but takes begin, end ptrs to a UTF-8 encoded string instead
1779 * of a code point, returning the value for the first code point in the string.
1780 * And it takes the particular macro name that finds the desired value given a
1781 * code point. Merely convert the UTF-8 to code point and call the cp macro */
1782#define _generic_GET_BREAK_VAL_UTF8(cp_macro, pos, strend) \
1783 (__ASSERT_(pos < strend) \
1784 /* Note assumes is valid UTF-8 */ \
1785 (cp_macro(utf8_to_uvchr_buf((pos), (strend), NULL))))
1786
1787/* Returns the GCB value for the input code point */
1788#define getGCB_VAL_CP(cp) \
1789 _generic_GET_BREAK_VAL_CP( \
1790 PL_GCB_invlist, \
02f811dd 1791 _Perl_GCB_invmap, \
64935bc6
KW
1792 (cp))
1793
1794/* Returns the GCB value for the first code point in the UTF-8 encoded string
1795 * bounded by pos and strend */
1796#define getGCB_VAL_UTF8(pos, strend) \
1797 _generic_GET_BREAK_VAL_UTF8(getGCB_VAL_CP, pos, strend)
05bd126c 1798
6b659339
KW
1799/* Returns the LB value for the input code point */
1800#define getLB_VAL_CP(cp) \
1801 _generic_GET_BREAK_VAL_CP( \
1802 PL_LB_invlist, \
1803 _Perl_LB_invmap, \
1804 (cp))
1805
1806/* Returns the LB value for the first code point in the UTF-8 encoded string
1807 * bounded by pos and strend */
1808#define getLB_VAL_UTF8(pos, strend) \
1809 _generic_GET_BREAK_VAL_UTF8(getLB_VAL_CP, pos, strend)
1810
06ae2722
KW
1811
1812/* Returns the SB value for the input code point */
1813#define getSB_VAL_CP(cp) \
1814 _generic_GET_BREAK_VAL_CP( \
1815 PL_SB_invlist, \
bf4268fa 1816 _Perl_SB_invmap, \
06ae2722
KW
1817 (cp))
1818
1819/* Returns the SB value for the first code point in the UTF-8 encoded string
1820 * bounded by pos and strend */
1821#define getSB_VAL_UTF8(pos, strend) \
1822 _generic_GET_BREAK_VAL_UTF8(getSB_VAL_CP, pos, strend)
1823
ae3bb8ea
KW
1824/* Returns the WB value for the input code point */
1825#define getWB_VAL_CP(cp) \
1826 _generic_GET_BREAK_VAL_CP( \
1827 PL_WB_invlist, \
bf4268fa 1828 _Perl_WB_invmap, \
ae3bb8ea
KW
1829 (cp))
1830
1831/* Returns the WB value for the first code point in the UTF-8 encoded string
1832 * bounded by pos and strend */
1833#define getWB_VAL_UTF8(pos, strend) \
1834 _generic_GET_BREAK_VAL_UTF8(getWB_VAL_CP, pos, strend)
1835
786e8c11 1836/* We know what class REx starts with. Try to find this position... */
02d5137b 1837/* if reginfo->intuit, its a dryrun */
786e8c11
YO
1838/* annoyingly all the vars in this routine have different names from their counterparts
1839 in regmatch. /grrr */
3c3eec57 1840STATIC char *
07be1b83 1841S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
f9176b44 1842 const char *strend, regmatch_info *reginfo)
a687059c 1843{
73104a1b
KW
1844 dVAR;
1845 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1846 char *pat_string; /* The pattern's exactish string */
1847 char *pat_end; /* ptr to end char of pat_string */
1848 re_fold_t folder; /* Function for computing non-utf8 folds */
1849 const U8 *fold_array; /* array for folding ords < 256 */
1850 STRLEN ln;
1851 STRLEN lnc;
73104a1b
KW
1852 U8 c1;
1853 U8 c2;
1854 char *e;
1855 I32 tmp = 1; /* Scratch variable? */
ba44c216 1856 const bool utf8_target = reginfo->is_utf8_target;
73104a1b 1857 UV utf8_fold_flags = 0;
f9176b44 1858 const bool is_utf8_pat = reginfo->is_utf8_pat;
3018b823
KW
1859 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1860 with a result inverts that result, as 0^1 =
1861 1 and 1^1 = 0 */
1862 _char_class_number classnum;
1863
73104a1b 1864 RXi_GET_DECL(prog,progi);
2f7f8cb1 1865
73104a1b 1866 PERL_ARGS_ASSERT_FIND_BYCLASS;
2f7f8cb1 1867
73104a1b
KW
1868 /* We know what class it must start with. */
1869 switch (OP(c)) {
a4525e78 1870 case ANYOFL:
780fcc9f 1871 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
a0bd1a30 1872
d1c40ef5 1873 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(c)) && ! IN_UTF8_CTYPE_LOCALE) {
a0bd1a30
KW
1874 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
1875 }
1876
780fcc9f 1877 /* FALLTHROUGH */
ac44c12e 1878 case ANYOFD:
73104a1b
KW
1879 case ANYOF:
1880 if (utf8_target) {
1881 REXEC_FBC_UTF8_CLASS_SCAN(
3db24e1e 1882 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
73104a1b 1883 }
1451f692
DM
1884 else if (ANYOF_FLAGS(c)) {
1885 REXEC_FBC_CLASS_SCAN(reginclass(prog,c, (U8*)s, (U8*)s+1, 0));
1886 }
73104a1b 1887 else {
1451f692 1888 REXEC_FBC_CLASS_SCAN(ANYOF_BITMAP_TEST(c, *((U8*)s)));
73104a1b
KW
1889 }
1890 break;
73104a1b 1891
098b07d5
KW
1892 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1893 assert(! is_utf8_pat);
924ba076 1894 /* FALLTHROUGH */
73104a1b 1895 case EXACTFA:
984e6dd1 1896 if (is_utf8_pat || utf8_target) {
73104a1b
KW
1897 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1898 goto do_exactf_utf8;
1899 }
1900 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1901 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1902 goto do_exactf_non_utf8; /* isn't dealt with by these */
77a6d856 1903
2fdb7295
KW
1904 case EXACTF: /* This node only generated for non-utf8 patterns */
1905 assert(! is_utf8_pat);
73104a1b 1906 if (utf8_target) {
73104a1b
KW
1907 utf8_fold_flags = 0;
1908 goto do_exactf_utf8;
1909 }
1910 fold_array = PL_fold;
1911 folder = foldEQ;
1912 goto do_exactf_non_utf8;
1913
1914 case EXACTFL:
780fcc9f 1915 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
31f05a37 1916 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
cea315b6 1917 utf8_fold_flags = FOLDEQ_LOCALE;
73104a1b
KW
1918 goto do_exactf_utf8;
1919 }
1920 fold_array = PL_fold_locale;
1921 folder = foldEQ_locale;
1922 goto do_exactf_non_utf8;
3c760661 1923
73104a1b 1924 case EXACTFU_SS:
984e6dd1 1925 if (is_utf8_pat) {
73104a1b
KW
1926 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1927 }
1928 goto do_exactf_utf8;
16d951b7 1929
a4525e78
KW
1930 case EXACTFLU8:
1931 if (! utf8_target) { /* All code points in this node require
1932 UTF-8 to express. */
1933 break;
1934 }
613abc6d
KW
1935 utf8_fold_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
1936 | FOLDEQ_S2_FOLDS_SANE;
a4525e78
KW
1937 goto do_exactf_utf8;
1938
73104a1b 1939 case EXACTFU:
984e6dd1
DM
1940 if (is_utf8_pat || utf8_target) {
1941 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
73104a1b
KW
1942 goto do_exactf_utf8;
1943 }
fac1af77 1944
73104a1b
KW
1945 /* Any 'ss' in the pattern should have been replaced by regcomp,
1946 * so we don't have to worry here about this single special case
1947 * in the Latin1 range */
1948 fold_array = PL_fold_latin1;
1949 folder = foldEQ_latin1;
1950
924ba076 1951 /* FALLTHROUGH */
73104a1b 1952
c52b8b12 1953 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
73104a1b
KW
1954 are no glitches with fold-length differences
1955 between the target string and pattern */
1956
1957 /* The idea in the non-utf8 EXACTF* cases is to first find the
1958 * first character of the EXACTF* node and then, if necessary,
1959 * case-insensitively compare the full text of the node. c1 is the
1960 * first character. c2 is its fold. This logic will not work for
1961 * Unicode semantics and the german sharp ss, which hence should
1962 * not be compiled into a node that gets here. */
1963 pat_string = STRING(c);
1964 ln = STR_LEN(c); /* length to match in octets/bytes */
1965
1966 /* We know that we have to match at least 'ln' bytes (which is the
1967 * same as characters, since not utf8). If we have to match 3
1968 * characters, and there are only 2 availabe, we know without
1969 * trying that it will fail; so don't start a match past the
1970 * required minimum number from the far end */
ea3daa5d 1971 e = HOP3c(strend, -((SSize_t)ln), s);
73104a1b 1972
02d5137b 1973 if (reginfo->intuit && e < s) {
73104a1b
KW
1974 e = s; /* Due to minlen logic of intuit() */
1975 }
fac1af77 1976
73104a1b
KW
1977 c1 = *pat_string;
1978 c2 = fold_array[c1];
1979 if (c1 == c2) { /* If char and fold are the same */
1980 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1981 }
1982 else {
1983 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1984 }
1985 break;
fac1af77 1986
c52b8b12
KW
1987 do_exactf_utf8:
1988 {
73104a1b
KW
1989 unsigned expansion;
1990
1991 /* If one of the operands is in utf8, we can't use the simpler folding
1992 * above, due to the fact that many different characters can have the
1993 * same fold, or portion of a fold, or different- length fold */
1994 pat_string = STRING(c);
1995 ln = STR_LEN(c); /* length to match in octets/bytes */
1996 pat_end = pat_string + ln;
984e6dd1 1997 lnc = is_utf8_pat /* length to match in characters */
73104a1b
KW
1998 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1999 : ln;
2000
2001 /* We have 'lnc' characters to match in the pattern, but because of
2002 * multi-character folding, each character in the target can match
2003 * up to 3 characters (Unicode guarantees it will never exceed
2004 * this) if it is utf8-encoded; and up to 2 if not (based on the
2005 * fact that the Latin 1 folds are already determined, and the
2006 * only multi-char fold in that range is the sharp-s folding to
2007 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
2008 * string character. Adjust lnc accordingly, rounding up, so that
2009 * if we need to match at least 4+1/3 chars, that really is 5. */
2010 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
2011 lnc = (lnc + expansion - 1) / expansion;
2012
2013 /* As in the non-UTF8 case, if we have to match 3 characters, and
2014 * only 2 are left, it's guaranteed to fail, so don't start a
2015 * match that would require us to go beyond the end of the string
2016 */
ea3daa5d 2017 e = HOP3c(strend, -((SSize_t)lnc), s);
73104a1b 2018
02d5137b 2019 if (reginfo->intuit && e < s) {
73104a1b
KW
2020 e = s; /* Due to minlen logic of intuit() */
2021 }
0658cdde 2022
73104a1b
KW
2023 /* XXX Note that we could recalculate e to stop the loop earlier,
2024 * as the worst case expansion above will rarely be met, and as we
2025 * go along we would usually find that e moves further to the left.
2026 * This would happen only after we reached the point in the loop
2027 * where if there were no expansion we should fail. Unclear if
2028 * worth the expense */
2029
2030 while (s <= e) {
2031 char *my_strend= (char *)strend;
2032 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
984e6dd1 2033 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
02d5137b 2034 && (reginfo->intuit || regtry(reginfo, &s)) )
73104a1b
KW
2035 {
2036 goto got_it;
2037 }
2038 s += (utf8_target) ? UTF8SKIP(s) : 1;
2039 }
2040 break;
2041 }
236d82fd 2042
73104a1b 2043 case BOUNDL:
780fcc9f 2044 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
64935bc6 2045 if (FLAGS(c) != TRADITIONAL_BOUND) {
89ad707a
KW
2046 if (! IN_UTF8_CTYPE_LOCALE) {
2047 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
64935bc6 2048 B_ON_NON_UTF8_LOCALE_IS_WRONG);
89ad707a 2049 }
64935bc6
KW
2050 goto do_boundu;
2051 }
2052
7a207065 2053 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
73104a1b 2054 break;
64935bc6 2055
73104a1b 2056 case NBOUNDL:
780fcc9f 2057 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
64935bc6 2058 if (FLAGS(c) != TRADITIONAL_BOUND) {
89ad707a
KW
2059 if (! IN_UTF8_CTYPE_LOCALE) {
2060 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
64935bc6 2061 B_ON_NON_UTF8_LOCALE_IS_WRONG);
89ad707a 2062 }
64935bc6
KW
2063 goto do_nboundu;
2064 }
2065
7a207065 2066 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
73104a1b 2067 break;
64935bc6
KW
2068
2069 case BOUND: /* regcomp.c makes sure that this only has the traditional \b
2070 meaning */
2071 assert(FLAGS(c) == TRADITIONAL_BOUND);
2072
7a207065 2073 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
73104a1b 2074 break;
64935bc6
KW
2075
2076 case BOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2077 meaning */
2078 assert(FLAGS(c) == TRADITIONAL_BOUND);
2079
44129e46 2080 FBC_BOUND_A(isWORDCHAR_A);
73104a1b 2081 break;
64935bc6
KW
2082
2083 case NBOUND: /* regcomp.c makes sure that this only has the traditional \b
2084 meaning */
2085 assert(FLAGS(c) == TRADITIONAL_BOUND);
2086
7a207065 2087 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
73104a1b 2088 break;
64935bc6
KW
2089
2090 case NBOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2091 meaning */
2092 assert(FLAGS(c) == TRADITIONAL_BOUND);
2093
44129e46 2094 FBC_NBOUND_A(isWORDCHAR_A);
73104a1b 2095 break;
64935bc6 2096
73104a1b 2097 case NBOUNDU:
64935bc6 2098 if ((bound_type) FLAGS(c) == TRADITIONAL_BOUND) {
7a207065 2099 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
64935bc6
KW
2100 break;
2101 }
2102
2103 do_nboundu:
2104
2105 to_complement = 1;
2106 /* FALLTHROUGH */
2107
2108 case BOUNDU:
2109 do_boundu:
2110 switch((bound_type) FLAGS(c)) {
2111 case TRADITIONAL_BOUND:
7a207065 2112 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
64935bc6
KW
2113 break;
2114 case GCB_BOUND:
a7a8bd1e 2115 if (s == reginfo->strbeg) {
67481c39 2116 if (reginfo->intuit || regtry(reginfo, &s))
64935bc6
KW
2117 {
2118 goto got_it;
2119 }
a7a8bd1e
KW
2120
2121 /* Didn't match. Try at the next position (if there is one) */
64935bc6 2122 s += (utf8_target) ? UTF8SKIP(s) : 1;
a7a8bd1e
KW
2123 if (UNLIKELY(s >= reginfo->strend)) {
2124 break;
2125 }
64935bc6
KW
2126 }
2127
2128 if (utf8_target) {
85e5f08b 2129 GCB_enum before = getGCB_VAL_UTF8(
64935bc6
KW
2130 reghop3((U8*)s, -1,
2131 (U8*)(reginfo->strbeg)),
2132 (U8*) reginfo->strend);
2133 while (s < strend) {
85e5f08b 2134 GCB_enum after = getGCB_VAL_UTF8((U8*) s,
64935bc6 2135 (U8*) reginfo->strend);
b0e24409
KW
2136 if ( (to_complement ^ isGCB(before,
2137 after,
2138 (U8*) reginfo->strbeg,
2139 (U8*) s,
2140 utf8_target))
00e3344b
KW
2141 && (reginfo->intuit || regtry(reginfo, &s)))
2142 {
2143 goto got_it;
64935bc6 2144 }
43a7bd62 2145 before = after;
64935bc6
KW
2146 s += UTF8SKIP(s);
2147 }
2148 }
2149 else { /* Not utf8. Everything is a GCB except between CR and
2150 LF */
2151 while (s < strend) {
00e3344b
KW
2152 if ((to_complement ^ ( UCHARAT(s - 1) != '\r'
2153 || UCHARAT(s) != '\n'))
2154 && (reginfo->intuit || regtry(reginfo, &s)))
64935bc6 2155 {
00e3344b 2156 goto got_it;
64935bc6 2157 }
43a7bd62 2158 s++;
64935bc6
KW
2159 }
2160 }
2161
6de80efc
KW
2162 /* And, since this is a bound, it can match after the final
2163 * character in the string */
67481c39 2164 if ((reginfo->intuit || regtry(reginfo, &s))) {
64935bc6
KW
2165 goto got_it;
2166 }
2167 break;
ae3bb8ea 2168
6b659339
KW
2169 case LB_BOUND:
2170 if (s == reginfo->strbeg) {
2171 if (reginfo->intuit || regtry(reginfo, &s)) {
2172 goto got_it;
2173 }
2174 s += (utf8_target) ? UTF8SKIP(s) : 1;
2175 if (UNLIKELY(s >= reginfo->strend)) {
2176 break;
2177 }
2178 }
2179
2180 if (utf8_target) {
2181 LB_enum before = getLB_VAL_UTF8(reghop3((U8*)s,
2182 -1,
2183 (U8*)(reginfo->strbeg)),
2184 (U8*) reginfo->strend);
2185 while (s < strend) {
2186 LB_enum after = getLB_VAL_UTF8((U8*) s, (U8*) reginfo->strend);
2187 if (to_complement ^ isLB(before,
2188 after,
2189 (U8*) reginfo->strbeg,
2190 (U8*) s,
2191 (U8*) reginfo->strend,
2192 utf8_target)
2193 && (reginfo->intuit || regtry(reginfo, &s)))
2194 {
2195 goto got_it;
2196 }
2197 before = after;
2198 s += UTF8SKIP(s);
2199 }
2200 }
2201 else { /* Not utf8. */
2202 LB_enum before = getLB_VAL_CP((U8) *(s -1));
2203 while (s < strend) {
2204 LB_enum after = getLB_VAL_CP((U8) *s);
2205 if (to_complement ^ isLB(before,
2206 after,
2207 (U8*) reginfo->strbeg,
2208 (U8*) s,
2209 (U8*) reginfo->strend,
2210 utf8_target)
2211 && (reginfo->intuit || regtry(reginfo, &s)))
2212 {
2213 goto got_it;
2214 }
2215 before = after;
2216 s++;
2217 }
2218 }
2219
2220 if (reginfo->intuit || regtry(reginfo, &s)) {
2221 goto got_it;
2222 }
2223
2224 break;
2225
06ae2722 2226 case SB_BOUND:
a7a8bd1e 2227 if (s == reginfo->strbeg) {
67481c39 2228 if (reginfo->intuit || regtry(reginfo, &s)) {
06ae2722
KW
2229 goto got_it;
2230 }
06ae2722 2231 s += (utf8_target) ? UTF8SKIP(s) : 1;
a7a8bd1e
KW
2232 if (UNLIKELY(s >= reginfo->strend)) {
2233 break;
2234 }
06ae2722
KW
2235 }
2236
2237 if (utf8_target) {
85e5f08b 2238 SB_enum before = getSB_VAL_UTF8(reghop3((U8*)s,
06ae2722
KW
2239 -1,
2240 (U8*)(reginfo->strbeg)),
2241 (U8*) reginfo->strend);
2242 while (s < strend) {
85e5f08b 2243 SB_enum after = getSB_VAL_UTF8((U8*) s,
06ae2722 2244 (U8*) reginfo->strend);
00e3344b
KW
2245 if ((to_complement ^ isSB(before,
2246 after,
2247 (U8*) reginfo->strbeg,
2248 (U8*) s,
2249 (U8*) reginfo->strend,
2250 utf8_target))
2251 && (reginfo->intuit || regtry(reginfo, &s)))
06ae2722 2252 {
00e3344b 2253 goto got_it;
06ae2722 2254 }
43a7bd62 2255 before = after;
06ae2722
KW
2256 s += UTF8SKIP(s);
2257 }
2258 }
2259 else { /* Not utf8. */
85e5f08b 2260 SB_enum before = getSB_VAL_CP((U8) *(s -1));
06ae2722 2261 while (s < strend) {
85e5f08b 2262 SB_enum after = getSB_VAL_CP((U8) *s);
00e3344b
KW
2263 if ((to_complement ^ isSB(before,
2264 after,
2265 (U8*) reginfo->strbeg,
2266 (U8*) s,
2267 (U8*) reginfo->strend,
2268 utf8_target))
2269 && (reginfo->intuit || regtry(reginfo, &s)))
06ae2722 2270 {
00e3344b 2271 goto got_it;
06ae2722 2272 }
43a7bd62 2273 before = after;
06ae2722
KW
2274 s++;
2275 }
2276 }
2277
2278 /* Here are at the final position in the target string. The SB
2279 * value is always true here, so matches, depending on other
2280 * constraints */
67481c39 2281 if (reginfo->intuit || regtry(reginfo, &s)) {
06ae2722
KW
2282 goto got_it;
2283 }
2284
2285 break;
2286
ae3bb8ea
KW
2287 case WB_BOUND:
2288 if (s == reginfo->strbeg) {
67481c39 2289 if (reginfo->intuit || regtry(reginfo, &s)) {
ae3bb8ea
KW
2290 goto got_it;
2291 }
2292 s += (utf8_target) ? UTF8SKIP(s) : 1;
a7a8bd1e
KW
2293 if (UNLIKELY(s >= reginfo->strend)) {
2294 break;
2295 }
ae3bb8ea
KW
2296 }
2297
2298 if (utf8_target) {
2299 /* We are at a boundary between char_sub_0 and char_sub_1.
2300 * We also keep track of the value for char_sub_-1 as we
2301 * loop through the line. Context may be needed to make a
2302 * determination, and if so, this can save having to
2303 * recalculate it */
85e5f08b
KW
2304 WB_enum previous = WB_UNKNOWN;
2305 WB_enum before = getWB_VAL_UTF8(
ae3bb8ea
KW
2306 reghop3((U8*)s,
2307 -1,
2308 (U8*)(reginfo->strbeg)),
2309 (U8*) reginfo->strend);
2310 while (s < strend) {
85e5f08b 2311 WB_enum after = getWB_VAL_UTF8((U8*) s,
ae3bb8ea 2312 (U8*) reginfo->strend);
00e3344b
KW
2313 if ((to_complement ^ isWB(previous,
2314 before,
2315 after,
2316 (U8*) reginfo->strbeg,
2317 (U8*) s,
2318 (U8*) reginfo->strend,
2319 utf8_target))
2320 && (reginfo->intuit || regtry(reginfo, &s)))
ae3bb8ea 2321 {
00e3344b 2322 goto got_it;
ae3bb8ea 2323 }
43a7bd62
KW
2324 previous = before;
2325 before = after;
ae3bb8ea
KW
2326 s += UTF8SKIP(s);
2327 }
2328 }
2329 else { /* Not utf8. */
85e5f08b
KW
2330 WB_enum previous = WB_UNKNOWN;
2331 WB_enum before = getWB_VAL_CP((U8) *(s -1));
ae3bb8ea 2332 while (s < strend) {
85e5f08b 2333 WB_enum after = getWB_VAL_CP((U8) *s);
00e3344b
KW
2334 if ((to_complement ^ isWB(previous,
2335 before,
2336 after,
2337 (U8*) reginfo->strbeg,
2338 (U8*) s,
2339 (U8*) reginfo->strend,
2340 utf8_target))
2341 && (reginfo->intuit || regtry(reginfo, &s)))
ae3bb8ea 2342 {
00e3344b 2343 goto got_it;
ae3bb8ea 2344 }
43a7bd62
KW
2345 previous = before;
2346 before = after;
ae3bb8ea
KW
2347 s++;
2348 }
2349 }
2350
67481c39 2351 if (reginfo->intuit || regtry(reginfo, &s)) {
ae3bb8ea
KW
2352 goto got_it;
2353 }
64935bc6 2354 }
73104a1b 2355 break;
64935bc6 2356
73104a1b
KW
2357 case LNBREAK:
2358 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
2359 is_LNBREAK_latin1_safe(s, strend)
2360 );
2361 break;
3018b823
KW
2362
2363 /* The argument to all the POSIX node types is the class number to pass to
2364 * _generic_isCC() to build a mask for searching in PL_charclass[] */
2365
2366 case NPOSIXL:
2367 to_complement = 1;
2368 /* FALLTHROUGH */
2369
2370 case POSIXL:
780fcc9f 2371 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3018b823
KW
2372 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
2373 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
73104a1b 2374 break;
3018b823
KW
2375
2376 case NPOSIXD:
2377 to_complement = 1;
2378 /* FALLTHROUGH */
2379
2380 case POSIXD:
2381 if (utf8_target) {
2382 goto posix_utf8;
2383 }
2384 goto posixa;
2385
2386 case NPOSIXA:
2387 if (utf8_target) {
2388 /* The complement of something that matches only ASCII matches all
837226c8 2389 * non-ASCII, plus everything in ASCII that isn't in the class. */
7a207065 2390 REXEC_FBC_UTF8_CLASS_SCAN( ! isASCII_utf8_safe(s, strend)
3018b823
KW
2391 || ! _generic_isCC_A(*s, FLAGS(c)));
2392 break;
2393 }
2394
2395 to_complement = 1;
2396 /* FALLTHROUGH */
2397
73104a1b 2398 case POSIXA:
3018b823 2399 posixa:
73104a1b 2400 /* Don't need to worry about utf8, as it can match only a single
3018b823
KW
2401 * byte invariant character. */
2402 REXEC_FBC_CLASS_SCAN(
2403 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
73104a1b 2404 break;
3018b823
KW
2405
2406 case NPOSIXU:
2407 to_complement = 1;
2408 /* FALLTHROUGH */
2409
2410 case POSIXU:
2411 if (! utf8_target) {
2412 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
2413 FLAGS(c))));
2414 }
2415 else {
2416
c52b8b12 2417 posix_utf8:
3018b823
KW
2418 classnum = (_char_class_number) FLAGS(c);
2419 if (classnum < _FIRST_NON_SWASH_CC) {
2420 while (s < strend) {
2421
2422 /* We avoid loading in the swash as long as possible, but
2423 * should we have to, we jump to a separate loop. This
2424 * extra 'if' statement is what keeps this code from being
2425 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
2426 if (UTF8_IS_ABOVE_LATIN1(*s)) {
2427 goto found_above_latin1;
2428 }
2429 if ((UTF8_IS_INVARIANT(*s)
2430 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
2431 classnum)))
2432 || (UTF8_IS_DOWNGRADEABLE_START(*s)
2433 && to_complement ^ cBOOL(
a62b247b 2434 _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*s,
94bb8c36 2435 *(s + 1)),
3018b823
KW
2436 classnum))))
2437 {
02d5137b 2438 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
3018b823
KW
2439 goto got_it;
2440 else {
2441 tmp = doevery;
2442 }
2443 }
2444 else {
2445 tmp = 1;
2446 }
2447 s += UTF8SKIP(s);
2448 }
2449 }
2450 else switch (classnum) { /* These classes are implemented as
2451 macros */
779cf272 2452 case _CC_ENUM_SPACE:
3018b823 2453 REXEC_FBC_UTF8_CLASS_SCAN(
7a207065 2454 to_complement ^ cBOOL(isSPACE_utf8_safe(s, strend)));
3018b823
KW
2455 break;
2456
2457 case _CC_ENUM_BLANK:
2458 REXEC_FBC_UTF8_CLASS_SCAN(
7a207065 2459 to_complement ^ cBOOL(isBLANK_utf8_safe(s, strend)));
3018b823
KW
2460 break;
2461
2462 case _CC_ENUM_XDIGIT:
2463 REXEC_FBC_UTF8_CLASS_SCAN(
7a207065 2464 to_complement ^ cBOOL(isXDIGIT_utf8_safe(s, strend)));
3018b823
KW
2465 break;
2466
2467 case _CC_ENUM_VERTSPACE:
2468 REXEC_FBC_UTF8_CLASS_SCAN(
7a207065 2469 to_complement ^ cBOOL(isVERTWS_utf8_safe(s, strend)));
3018b823
KW
2470 break;
2471
2472 case _CC_ENUM_CNTRL:
2473 REXEC_FBC_UTF8_CLASS_SCAN(
7a207065 2474 to_complement ^ cBOOL(isCNTRL_utf8_safe(s, strend)));
3018b823
KW
2475 break;
2476
2477 default:
2478 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
e5964223 2479 NOT_REACHED; /* NOTREACHED */
3018b823
KW
2480 }
2481 }
2482 break;
2483
2484 found_above_latin1: /* Here we have to load a swash to get the result
2485 for the current code point */
2486 if (! PL_utf8_swash_ptrs[classnum]) {
2487 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2488 PL_utf8_swash_ptrs[classnum] =
2a16ac92
KW
2489 _core_swash_init("utf8",
2490 "",
2491 &PL_sv_undef, 1, 0,
2492 PL_XPosix_ptrs[classnum], &flags);
3018b823
KW
2493 }
2494
2495 /* This is a copy of the loop above for swash classes, though using the
2496 * FBC macro instead of being expanded out. Since we've loaded the
2497 * swash, we don't have to check for that each time through the loop */
2498 REXEC_FBC_UTF8_CLASS_SCAN(
7a207065 2499 to_complement ^ cBOOL(_generic_utf8_safe(
3018b823
KW
2500 classnum,
2501 s,
7a207065 2502 strend,
3018b823
KW
2503 swash_fetch(PL_utf8_swash_ptrs[classnum],
2504 (U8 *) s, TRUE))));
73104a1b
KW
2505 break;
2506
2507 case AHOCORASICKC:
2508 case AHOCORASICK:
2509 {
2510 DECL_TRIE_TYPE(c);
2511 /* what trie are we using right now */
2512 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2513 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2514 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2515
2516 const char *last_start = strend - trie->minlen;
6148ee25 2517#ifdef DEBUGGING
73104a1b 2518 const char *real_start = s;
6148ee25 2519#endif
73104a1b
KW
2520 STRLEN maxlen = trie->maxlen;
2521 SV *sv_points;
2522 U8 **points; /* map of where we were in the input string
2523 when reading a given char. For ASCII this
2524 is unnecessary overhead as the relationship
2525 is always 1:1, but for Unicode, especially
2526 case folded Unicode this is not true. */
2527 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2528 U8 *bitmap=NULL;
2529
2530
2531 GET_RE_DEBUG_FLAGS_DECL;
2532
2533 /* We can't just allocate points here. We need to wrap it in
2534 * an SV so it gets freed properly if there is a croak while
2535 * running the match */
2536 ENTER;
2537 SAVETMPS;
2538 sv_points=newSV(maxlen * sizeof(U8 *));
2539 SvCUR_set(sv_points,
2540 maxlen * sizeof(U8 *));
2541 SvPOK_on(sv_points);
2542 sv_2mortal(sv_points);
2543 points=(U8**)SvPV_nolen(sv_points );
2544 if ( trie_type != trie_utf8_fold
2545 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2546 {
2547 if (trie->bitmap)
2548 bitmap=(U8*)trie->bitmap;
2549 else
2550 bitmap=(U8*)ANYOF_BITMAP(c);
2551 }
2552 /* this is the Aho-Corasick algorithm modified a touch
2553 to include special handling for long "unknown char" sequences.
2554 The basic idea being that we use AC as long as we are dealing
2555 with a possible matching char, when we encounter an unknown char
2556 (and we have not encountered an accepting state) we scan forward
2557 until we find a legal starting char.
2558 AC matching is basically that of trie matching, except that when
2559 we encounter a failing transition, we fall back to the current
2560 states "fail state", and try the current char again, a process
2561 we repeat until we reach the root state, state 1, or a legal
2562 transition. If we fail on the root state then we can either
2563 terminate if we have reached an accepting state previously, or
2564 restart the entire process from the beginning if we have not.
2565
2566 */
2567 while (s <= last_start) {
2568 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2569 U8 *uc = (U8*)s;
2570 U16 charid = 0;
2571 U32 base = 1;
2572 U32 state = 1;
2573 UV uvc = 0;
2574 STRLEN len = 0;
2575 STRLEN foldlen = 0;
2576 U8 *uscan = (U8*)NULL;
2577 U8 *leftmost = NULL;
2578#ifdef DEBUGGING
2579 U32 accepted_word= 0;
786e8c11 2580#endif
73104a1b
KW
2581 U32 pointpos = 0;
2582
2583 while ( state && uc <= (U8*)strend ) {
2584 int failed=0;
2585 U32 word = aho->states[ state ].wordnum;
2586
2587 if( state==1 ) {
2588 if ( bitmap ) {
2589 DEBUG_TRIE_EXECUTE_r(
2590 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2591 dump_exec_pos( (char *)uc, c, strend, real_start,
cb41e5d6 2592 (char *)uc, utf8_target, 0 );
6ad9a8ab 2593 Perl_re_printf( aTHX_
73104a1b
KW
2594 " Scanning for legal start char...\n");
2595 }
2596 );
2597 if (utf8_target) {
2598 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2599 uc += UTF8SKIP(uc);
2600 }
2601 } else {
2602 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2603 uc++;
2604 }
786e8c11 2605 }
73104a1b 2606 s= (char *)uc;
07be1b83 2607 }
73104a1b
KW
2608 if (uc >(U8*)last_start) break;
2609 }
2610
2611 if ( word ) {
2612 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2613 if (!leftmost || lpos < leftmost) {
2614 DEBUG_r(accepted_word=word);
2615 leftmost= lpos;
7016d6eb 2616 }
73104a1b 2617 if (base==0) break;
7016d6eb 2618
73104a1b
KW
2619 }
2620 points[pointpos++ % maxlen]= uc;
2621 if (foldlen || uc < (U8*)strend) {
2622 REXEC_TRIE_READ_CHAR(trie_type, trie,
2623 widecharmap, uc,
2624 uscan, len, uvc, charid, foldlen,
2625 foldbuf, uniflags);
2626 DEBUG_TRIE_EXECUTE_r({
2627 dump_exec_pos( (char *)uc, c, strend,
cb41e5d6 2628 real_start, s, utf8_target, 0);
6ad9a8ab 2629 Perl_re_printf( aTHX_
147e3846 2630 " Charid:%3u CP:%4" UVxf " ",
73104a1b
KW
2631 charid, uvc);
2632 });
2633 }
2634 else {
2635 len = 0;
2636 charid = 0;
2637 }
07be1b83 2638
73104a1b
KW
2639
2640 do {
6148ee25 2641#ifdef DEBUGGING
73104a1b 2642 word = aho->states[ state ].wordnum;
6148ee25 2643#endif
73104a1b
KW
2644 base = aho->states[ state ].trans.base;
2645
2646 DEBUG_TRIE_EXECUTE_r({
2647 if (failed)
2648 dump_exec_pos( (char *)uc, c, strend, real_start,
cb41e5d6 2649 s, utf8_target, 0 );
6ad9a8ab 2650 Perl_re_printf( aTHX_
147e3846 2651 "%sState: %4" UVxf ", word=%" UVxf,
73104a1b
KW
2652 failed ? " Fail transition to " : "",
2653 (UV)state, (UV)word);
2654 });
2655 if ( base ) {
2656 U32 tmp;
2657 I32 offset;
2658 if (charid &&
2659 ( ((offset = base + charid
2660 - 1 - trie->uniquecharcount)) >= 0)
2661 && ((U32)offset < trie->lasttrans)
2662 && trie->trans[offset].check == state
2663 && (tmp=trie->trans[offset].next))
2664 {
2665 DEBUG_TRIE_EXECUTE_r(
6ad9a8ab 2666 Perl_re_printf( aTHX_ " - legal\n"));
73104a1b
KW
2667 state = tmp;
2668 break;
07be1b83
YO
2669 }
2670 else {
786e8c11 2671 DEBUG_TRIE_EXECUTE_r(
6ad9a8ab 2672 Perl_re_printf( aTHX_ " - fail\n"));
786e8c11 2673 failed = 1;
73104a1b 2674 state = aho->fail[state];
07be1b83 2675 }
07be1b83 2676 }
73104a1b
KW
2677 else {
2678 /* we must be accepting here */
2679 DEBUG_TRIE_EXECUTE_r(
6ad9a8ab 2680 Perl_re_printf( aTHX_ " - accepting\n"));
73104a1b
KW
2681 failed = 1;
2682 break;
786e8c11 2683 }
73104a1b
KW
2684 } while(state);
2685 uc += len;
2686 if (failed) {
2687 if (leftmost)
2688 break;
2689 if (!state) state = 1;
07be1b83 2690 }
73104a1b
KW
2691 }
2692 if ( aho->states[ state ].wordnum ) {
2693 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2694 if (!leftmost || lpos < leftmost) {
2695 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2696 leftmost = lpos;
07be1b83
YO
2697 }
2698 }
73104a1b
KW
2699 if (leftmost) {
2700 s = (char*)leftmost;
2701 DEBUG_TRIE_EXECUTE_r({
147e3846 2702 Perl_re_printf( aTHX_ "Matches word #%" UVxf " at position %" IVdf ". Trying full pattern...\n",
73104a1b
KW
2703 (UV)accepted_word, (IV)(s - real_start)
2704 );
2705 });
02d5137b 2706 if (reginfo->intuit || regtry(reginfo, &s)) {
73104a1b
KW
2707 FREETMPS;
2708 LEAVE;
2709 goto got_it;
2710 }
2711 s = HOPc(s,1);
2712 DEBUG_TRIE_EXECUTE_r({
6ad9a8ab 2713 Perl_re_printf( aTHX_ "Pattern failed. Looking for new start point...\n");
73104a1b
KW
2714 });
2715 } else {
2716 DEBUG_TRIE_EXECUTE_r(
6ad9a8ab 2717 Perl_re_printf( aTHX_ "No match.\n"));
73104a1b
KW
2718 break;
2719 }
2720 }
2721 FREETMPS;
2722 LEAVE;
2723 }
2724 break;
2725 default:
2726 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
73104a1b
KW
2727 }
2728 return 0;
2729 got_it:
2730 return s;
6eb5f6b9
JH
2731}
2732
60165aa4
DM
2733/* set RX_SAVED_COPY, RX_SUBBEG etc.
2734 * flags have same meanings as with regexec_flags() */
2735
749f4950
DM
2736static void
2737S_reg_set_capture_string(pTHX_ REGEXP * const rx,
60165aa4
DM
2738 char *strbeg,
2739 char *strend,
2740 SV *sv,
2741 U32 flags,
2742 bool utf8_target)
2743{
2744 struct regexp *const prog = ReANY(rx);
2745
60165aa4
DM
2746 if (flags & REXEC_COPY_STR) {
2747#ifdef PERL_ANY_COW
2748 if (SvCANCOW(sv)) {
eb8fc9fe 2749 DEBUG_C(Perl_re_printf( aTHX_
60165aa4 2750 "Copy on write: regexp capture, type %d\n",
eb8fc9fe 2751 (int) SvTYPE(sv)));
5411a0e5
DM
2752 /* Create a new COW SV to share the match string and store
2753 * in saved_copy, unless the current COW SV in saved_copy
2754 * is valid and suitable for our purpose */
2755 if (( prog->saved_copy
2756 && SvIsCOW(prog->saved_copy)
2757 && SvPOKp(prog->saved_copy)
2758 && SvIsCOW(sv)
2759 && SvPOKp(sv)
2760 && SvPVX(sv) == SvPVX(prog->saved_copy)))
a76b0e90 2761 {
5411a0e5
DM
2762 /* just reuse saved_copy SV */
2763 if (RXp_MATCH_COPIED(prog)) {
2764 Safefree(prog->subbeg);
2765 RXp_MATCH_COPIED_off(prog);
2766 }
2767 }
2768 else {
2769 /* create new COW SV to share string */
a76b0e90
DM
2770 RX_MATCH_COPY_FREE(rx);
2771 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
a76b0e90 2772 }
5411a0e5
DM
2773 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2774 assert (SvPOKp(prog->saved_copy));
60165aa4
DM
2775 prog->sublen = strend - strbeg;
2776 prog->suboffset = 0;
2777 prog->subcoffset = 0;
2778 } else
2779#endif
2780 {
99a90e59
FC
2781 SSize_t min = 0;
2782 SSize_t max = strend - strbeg;
ea3daa5d 2783 SSize_t sublen;
60165aa4
DM
2784
2785 if ( (flags & REXEC_COPY_SKIP_POST)
e322109a 2786 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM
2787 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2788 ) { /* don't copy $' part of string */
2789 U32 n = 0;
2790 max = -1;
2791 /* calculate the right-most part of the string covered
f67a5002 2792 * by a capture. Due to lookahead, this may be to
60165aa4
DM
2793 * the right of $&, so we have to scan all captures */
2794 while (n <= prog->lastparen) {
2795 if (prog->offs[n].end > max)
2796 max = prog->offs[n].end;
2797 n++;
2798 }
2799 if (max == -1)
2800 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2801 ? prog->offs[0].start
2802 : 0;
2803 assert(max >= 0 && max <= strend - strbeg);
2804 }
2805
2806 if ( (flags & REXEC_COPY_SKIP_PRE)
e322109a 2807 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM
2808 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2809 ) { /* don't copy $` part of string */
2810 U32 n = 0;
2811 min = max;
2812 /* calculate the left-most part of the string covered
f67a5002 2813 * by a capture. Due to lookbehind, this may be to
60165aa4
DM
2814 * the left of $&, so we have to scan all captures */
2815 while (min && n <= prog->lastparen) {
2816 if ( prog->offs[n].start != -1
2817 && prog->offs[n].start < min)
2818 {
2819 min = prog->offs[n].start;
2820 }
2821 n++;
2822 }
2823 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2824 && min > prog->offs[0].end
2825 )
2826 min = prog->offs[0].end;
2827
2828 }
2829
2830 assert(min >= 0 && min <= max && min <= strend - strbeg);
2831 sublen = max - min;
2832
2833 if (RX_MATCH_COPIED(rx)) {
2834 if (sublen > prog->sublen)
2835 prog->subbeg =
2836 (char*)saferealloc(prog->subbeg, sublen+1);
2837 }
2838 else
2839 prog->subbeg = (char*)safemalloc(sublen+1);
2840 Copy(strbeg + min, prog->subbeg, sublen, char);
2841 prog->subbeg[sublen] = '\0';
2842 prog->suboffset = min;
2843 prog->sublen = sublen;
2844 RX_MATCH_COPIED_on(rx);
2845 }
2846 prog->subcoffset = prog->suboffset;
2847 if (prog->suboffset && utf8_target) {
2848 /* Convert byte offset to chars.
2849 * XXX ideally should only compute this if @-/@+
2850 * has been seen, a la PL_sawampersand ??? */
2851
2852 /* If there's a direct correspondence between the
2853 * string which we're matching and the original SV,
2854 * then we can use the utf8 len cache associated with
2855 * the SV. In particular, it means that under //g,
2856 * sv_pos_b2u() will use the previously cached
2857 * position to speed up working out the new length of
2858 * subcoffset, rather than counting from the start of
2859 * the string each time. This stops
2860 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2861 * from going quadratic */
2862 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
ea3daa5d
FC
2863 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2864 SV_GMAGIC|SV_CONST_RETURN);
60165aa4
DM
2865 else
2866 prog->subcoffset = utf8_length((U8*)strbeg,
2867 (U8*)(strbeg+prog->suboffset));
2868 }
2869 }
2870 else {
2871 RX_MATCH_COPY_FREE(rx);
2872 prog->subbeg = strbeg;
2873 prog->suboffset = 0;
2874 prog->subcoffset = 0;
2875 prog->sublen = strend - strbeg;
2876 }
2877}
2878
2879
2880
fae667d5 2881
6eb5f6b9
JH
2882/*
2883 - regexec_flags - match a regexp against a string
2884 */
2885I32
5aaab254 2886Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
ea3daa5d 2887 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
8fd1a950
DM
2888/* stringarg: the point in the string at which to begin matching */
2889/* strend: pointer to null at end of string */
2890/* strbeg: real beginning of string */
2891/* minend: end of match must be >= minend bytes after stringarg. */
2892/* sv: SV being matched: only used for utf8 flag, pos() etc; string
2893 * itself is accessed via the pointers above */
2894/* data: May be used for some additional optimizations.
d058ec57 2895 Currently unused. */
a340edde 2896/* flags: For optimizations. See REXEC_* in regexp.h */
8fd1a950 2897
6eb5f6b9 2898{
8d919b0a 2899 struct regexp *const prog = ReANY(rx);
5aaab254 2900 char *s;
eb578fdb 2901 regnode *c;
03c83e26 2902 char *startpos;
ea3daa5d
FC
2903 SSize_t minlen; /* must match at least this many chars */
2904 SSize_t dontbother = 0; /* how many characters not to try at end */
f2ed9b32 2905 const bool utf8_target = cBOOL(DO_UTF8(sv));
2757e526 2906 I32 multiline;
f8fc2ecf 2907 RXi_GET_DECL(prog,progi);
02d5137b
DM
2908 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2909 regmatch_info *const reginfo = &reginfo_buf;
e9105d30 2910 regexp_paren_pair *swap = NULL;
006f26b2 2911 I32 oldsave;
a3621e74
YO
2912 GET_RE_DEBUG_FLAGS_DECL;
2913
7918f24d 2914 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
9d4ba2ae 2915 PERL_UNUSED_ARG(data);
6eb5f6b9
JH
2916
2917 /* Be paranoid... */
3dc78631 2918 if (prog == NULL) {
6eb5f6b9 2919 Perl_croak(aTHX_ "NULL regexp parameter");
6eb5f6b9
JH
2920 }
2921
6c3fea77 2922 DEBUG_EXECUTE_r(
03c83e26 2923 debug_start_match(rx, utf8_target, stringarg, strend,
6c3fea77
DM
2924 "Matching");
2925 );
8adc0f72 2926
b342a604
DM
2927 startpos = stringarg;
2928
4cf1a867
DM
2929 /* set these early as they may be used by the HOP macros below */
2930 reginfo->strbeg = strbeg;
2931 reginfo->strend = strend;
2932 reginfo->is_utf8_target = cBOOL(utf8_target);
2933
58430ea8 2934 if (prog->intflags & PREGf_GPOS_SEEN) {
d307c076
DM
2935 MAGIC *mg;
2936
fef7148b
DM
2937 /* set reginfo->ganch, the position where \G can match */
2938
2939 reginfo->ganch =
2940 (flags & REXEC_IGNOREPOS)
2941 ? stringarg /* use start pos rather than pos() */
3dc78631 2942 : ((mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
25fdce4a
FC
2943 /* Defined pos(): */
2944 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
fef7148b
DM
2945 : strbeg; /* pos() not defined; use start of string */
2946
6ad9a8ab 2947 DEBUG_GPOS_r(Perl_re_printf( aTHX_
147e3846 2948 "GPOS ganch set to strbeg[%" IVdf "]\n", (IV)(reginfo->ganch - strbeg)));
fef7148b 2949
03c83e26
DM
2950 /* in the presence of \G, we may need to start looking earlier in
2951 * the string than the suggested start point of stringarg:
0b2c2a84 2952 * if prog->gofs is set, then that's a known, fixed minimum
03c83e26
DM
2953 * offset, such as
2954 * /..\G/: gofs = 2
2955 * /ab|c\G/: gofs = 1
2956 * or if the minimum offset isn't known, then we have to go back
2957 * to the start of the string, e.g. /w+\G/
2958 */
2bfbe302 2959
8e1490ee 2960 if (prog->intflags & PREGf_ANCH_GPOS) {
4cf1a867
DM
2961 if (prog->gofs) {
2962 startpos = HOPBACKc(reginfo->ganch, prog->gofs);
2963 if (!startpos ||
2964 ((flags & REXEC_FAIL_ON_UNDERFLOW) && startpos < stringarg))
2965 {
6ad9a8ab 2966 DEBUG_r(Perl_re_printf( aTHX_
4cf1a867
DM
2967 "fail: ganch-gofs before earliest possible start\n"));
2968 return 0;
2969 }
2bfbe302 2970 }
4cf1a867
DM
2971 else
2972 startpos = reginfo->ganch;
2bfbe302
DM
2973 }
2974 else if (prog->gofs) {
4cf1a867
DM
2975 startpos = HOPBACKc(startpos, prog->gofs);
2976 if (!startpos)
b342a604 2977 startpos = strbeg;
03c83e26 2978 }
58430ea8 2979 else if (prog->intflags & PREGf_GPOS_FLOAT)
b342a604 2980 startpos = strbeg;
03c83e26
DM
2981 }
2982
2983 minlen = prog->minlen;
b342a604 2984 if ((startpos + minlen) > strend || startpos < strbeg) {
6ad9a8ab 2985 DEBUG_r(Perl_re_printf( aTHX_
03c83e26
DM
2986 "Regex match can't succeed, so not even tried\n"));
2987 return 0;
2988 }
2989
63a3746a
DM
2990 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2991 * which will call destuctors to reset PL_regmatch_state, free higher
2992 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2993 * regmatch_info_aux_eval */
2994
2995 oldsave = PL_savestack_ix;
2996
dfa77d06
DM
2997 s = startpos;
2998
e322109a 2999 if ((prog->extflags & RXf_USE_INTUIT)
7fadf4a7
DM
3000 && !(flags & REXEC_CHECKED))
3001 {
dfa77d06 3002 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
7fadf4a7 3003 flags, NULL);
dfa77d06 3004 if (!s)
7fadf4a7
DM
3005 return 0;
3006
e322109a 3007 if (prog->extflags & RXf_CHECK_ALL) {
7fadf4a7
DM
3008 /* we can match based purely on the result of INTUIT.
3009 * Set up captures etc just for $& and $-[0]
3010 * (an intuit-only match wont have $1,$2,..) */
3011 assert(!prog->nparens);
d5e7783a
DM
3012
3013 /* s/// doesn't like it if $& is earlier than where we asked it to
3014 * start searching (which can happen on something like /.\G/) */
3015 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3016 && (s < stringarg))
3017 {
3018 /* this should only be possible under \G */
58430ea8 3019 assert(prog->intflags & PREGf_GPOS_SEEN);
6ad9a8ab 3020 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
d5e7783a
DM
3021 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3022 goto phooey;
3023 }
3024
7fadf4a7
DM
3025 /* match via INTUIT shouldn't have any captures.
3026 * Let @-, @+, $^N know */
3027 prog->lastparen = prog->lastcloseparen = 0;
3028 RX_MATCH_UTF8_set(rx, utf8_target);
3ff69bd6
DM
3029 prog->offs[0].start = s - strbeg;
3030 prog->offs[0].end = utf8_target
3031 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
3032 : s - strbeg + prog->minlenret;
7fadf4a7 3033 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 3034 S_reg_set_capture_string(aTHX_ rx,
7fadf4a7
DM
3035 strbeg, strend,
3036 sv, flags, utf8_target);
3037
7fadf4a7
DM
3038 return 1;
3039 }
3040 }
3041
6c3fea77 3042 multiline = prog->extflags & RXf_PMf_MULTILINE;
1de06328 3043
dfa77d06 3044 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
6ad9a8ab 3045 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
a72c7584
JH
3046 "String too short [regexec_flags]...\n"));
3047 goto phooey;
1aa99e6b 3048 }
1de06328 3049
6eb5f6b9 3050 /* Check validity of program. */
f8fc2ecf 3051 if (UCHARAT(progi->program) != REG_MAGIC) {
6eb5f6b9
JH
3052 Perl_croak(aTHX_ "corrupted regexp program");
3053 }
3054
1738e041 3055 RX_MATCH_TAINTED_off(rx);
ab4e48c1 3056 RX_MATCH_UTF8_set(rx, utf8_target);
1738e041 3057
6c3fea77
DM
3058 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
3059 reginfo->intuit = 0;
02d5137b
DM
3060 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
3061 reginfo->warned = FALSE;
02d5137b 3062 reginfo->sv = sv;
1cb48e53 3063 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
6eb5f6b9 3064 /* see how far we have to get to not match where we matched before */
fe3974be 3065 reginfo->till = stringarg + minend;
6eb5f6b9 3066
60779a30 3067 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
82c23608
FC
3068 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
3069 S_cleanup_regmatch_info_aux has executed (registered by
3070 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
3071 magic belonging to this SV.
3072 Not newSVsv, either, as it does not COW.
3073 */
3074 reginfo->sv = newSV(0);
4cba5ac0 3075 SvSetSV_nosteal(reginfo->sv, sv);
82c23608
FC
3076 SAVEFREESV(reginfo->sv);
3077 }
3078
331b2dcc
DM
3079 /* reserve next 2 or 3 slots in PL_regmatch_state:
3080 * slot N+0: may currently be in use: skip it
3081 * slot N+1: use for regmatch_info_aux struct
3082 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
3083 * slot N+3: ready for use by regmatch()
3084 */
bf2039a9 3085
331b2dcc
DM
3086 {
3087 regmatch_state *old_regmatch_state;
3088 regmatch_slab *old_regmatch_slab;
3089 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
3090
3091 /* on first ever match, allocate first slab */
3092 if (!PL_regmatch_slab) {
3093 Newx(PL_regmatch_slab, 1, regmatch_slab);
3094 PL_regmatch_slab->prev = NULL;
3095 PL_regmatch_slab->next = NULL;
3096 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3097 }
bf2039a9 3098
331b2dcc
DM
3099 old_regmatch_state = PL_regmatch_state;
3100 old_regmatch_slab = PL_regmatch_slab;
bf2039a9 3101
331b2dcc
DM
3102 for (i=0; i <= max; i++) {
3103 if (i == 1)
3104 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
3105 else if (i ==2)
3106 reginfo->info_aux_eval =
3107 reginfo->info_aux->info_aux_eval =
3108 &(PL_regmatch_state->u.info_aux_eval);
bf2039a9 3109
331b2dcc
DM
3110 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
3111 PL_regmatch_state = S_push_slab(aTHX);
3112 }
bf2039a9 3113
331b2dcc
DM
3114 /* note initial PL_regmatch_state position; at end of match we'll
3115 * pop back to there and free any higher slabs */
bf2039a9 3116
331b2dcc
DM
3117 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
3118 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2ac8ff4b 3119 reginfo->info_aux->poscache = NULL;
bf2039a9 3120
331b2dcc 3121 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
bf2039a9 3122
331b2dcc
DM
3123 if ((prog->extflags & RXf_EVAL_SEEN))
3124 S_setup_eval_state(aTHX_ reginfo);
3125 else
3126 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
bf2039a9 3127 }
d3aa529c 3128
6eb5f6b9 3129 /* If there is a "must appear" string, look for it. */
6eb5f6b9 3130
288b8c02 3131 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
e9105d30
GG
3132 /* We have to be careful. If the previous successful match
3133 was from this regex we don't want a subsequent partially
3134 successful match to clobber the old results.
3135 So when we detect this possibility we add a swap buffer
d8da0584
KW
3136 to the re, and switch the buffer each match. If we fail,
3137 we switch it back; otherwise we leave it swapped.
e9105d30
GG
3138 */
3139 swap = prog->offs;
3140 /* do we need a save destructor here for eval dies? */
3141 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2b1a3689 3142 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
147e3846 3143 "rex=0x%" UVxf " saving offs: orig=0x%" UVxf " new=0x%" UVxf "\n",
2b1a3689
YO
3144 0,
3145 PTR2UV(prog),
495f47a5
DM
3146 PTR2UV(swap),
3147 PTR2UV(prog->offs)
3148 ));
c74340f9 3149 }
6eb5f6b9 3150
ba6840fb
YO
3151 if (prog->recurse_locinput)
3152 Zero(prog->recurse_locinput,prog->nparens + 1, char *);
3153
0fa70a06
DM
3154 /* Simplest case: anchored match need be tried only once, or with
3155 * MBOL, only at the beginning of each line.
3156 *
3157 * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets
3158 * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't
3159 * match at the start of the string then it won't match anywhere else
3160 * either; while with /.*.../, if it doesn't match at the beginning,
3161 * the earliest it could match is at the start of the next line */
3162
8e1490ee 3163 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
0fa70a06
DM
3164 char *end;
3165
3166 if (regtry(reginfo, &s))
6eb5f6b9 3167 goto got_it;
0fa70a06
DM
3168
3169 if (!(prog->intflags & PREGf_ANCH_MBOL))
3170 goto phooey;
3171
3172 /* didn't match at start, try at other newline positions */
3173
3174 if (minlen)
3175 dontbother = minlen - 1;
3176 end = HOP3c(strend, -dontbother, strbeg) - 1;
3177
3178 /* skip to next newline */
3179
3180 while (s <= end) { /* note it could be possible to match at the end of the string */
3181 /* NB: newlines are the same in unicode as they are in latin */
3182 if (*s++ != '\n')
3183 continue;
3184 if (prog->check_substr || prog->check_utf8) {
3185 /* note that with PREGf_IMPLICIT, intuit can only fail
3186 * or return the start position, so it's of limited utility.
3187 * Nevertheless, I made the decision that the potential for
3188 * quick fail was still worth it - DAPM */
3189 s = re_intuit_start(rx, sv, strbeg, s, strend, flags, NULL);
3190 if (!s)
3191 goto phooey;
3192 }
3193 if (regtry(reginfo, &s))
3194 goto got_it;
3195 }
3196 goto phooey;
3197 } /* end anchored search */
3198
3199 if (prog->intflags & PREGf_ANCH_GPOS)
f9f4320a 3200 {
a8430a8b
YO
3201 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
3202 assert(prog->intflags & PREGf_GPOS_SEEN);
2bfbe302
DM
3203 /* For anchored \G, the only position it can match from is
3204 * (ganch-gofs); we already set startpos to this above; if intuit
3205 * moved us on from there, we can't possibly succeed */
4cf1a867 3206 assert(startpos == HOPBACKc(reginfo->ganch, prog->gofs));
2bfbe302 3207 if (s == startpos && regtry(reginfo, &s))
6eb5f6b9
JH
3208 goto got_it;
3209 goto phooey;
3210 }
3211
3212 /* Messy cases: unanchored match. */
bbe252da 3213 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
6eb5f6b9 3214 /* we have /x+whatever/ */
984e6dd1 3215 /* it must be a one character string (XXXX Except is_utf8_pat?) */
33b8afdf 3216 char ch;
bf93d4cc
GS
3217#ifdef DEBUGGING
3218 int did_match = 0;
3219#endif
f2ed9b32 3220 if (utf8_target) {
7e0d5ad7
KW
3221 if (! prog->anchored_utf8) {
3222 to_utf8_substr(prog);
3223 }
3224 ch = SvPVX_const(prog->anchored_utf8)[0];
4cadc6a9 3225 REXEC_FBC_SCAN(
6eb5f6b9 3226 if (*s == ch) {
a3621e74 3227 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 3228 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH
3229 s += UTF8SKIP(s);
3230 while (s < strend && *s == ch)
3231 s += UTF8SKIP(s);
3232 }
4cadc6a9 3233 );
7e0d5ad7 3234
6eb5f6b9
JH
3235 }
3236 else {
7e0d5ad7
KW
3237 if (! prog->anchored_substr) {
3238 if (! to_byte_substr(prog)) {
6b54ddc5 3239 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
3240 }
3241 }
3242 ch = SvPVX_const(prog->anchored_substr)[0];
4cadc6a9 3243 REXEC_FBC_SCAN(
6eb5f6b9 3244 if (*s == ch) {
a3621e74 3245 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 3246 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH
3247 s++;
3248 while (s < strend && *s == ch)
3249 s++;
3250 }
4cadc6a9 3251 );
6eb5f6b9 3252 }
a3621e74 3253 DEBUG_EXECUTE_r(if (!did_match)
6ad9a8ab 3254 Perl_re_printf( aTHX_
b7953727
JH
3255 "Did not find anchored character...\n")
3256 );
6eb5f6b9 3257 }
a0714e2c
SS
3258 else if (prog->anchored_substr != NULL
3259 || prog->anchored_utf8 != NULL
3260 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
33b8afdf
JH
3261 && prog->float_max_offset < strend - s)) {
3262 SV *must;
ea3daa5d
FC
3263 SSize_t back_max;
3264 SSize_t back_min;
33b8afdf 3265 char *last;
6eb5f6b9 3266 char *last1; /* Last position checked before */
bf93d4cc
GS
3267#ifdef DEBUGGING
3268 int did_match = 0;
3269#endif
33b8afdf 3270 if (prog->anchored_substr || prog->anchored_utf8) {
7e0d5ad7
KW
3271 if (utf8_target) {
3272 if (! prog->anchored_utf8) {
3273 to_utf8_substr(prog);
3274 }
3275 must = prog->anchored_utf8;
3276 }
3277 else {
3278 if (! prog->anchored_substr) {
3279 if (! to_byte_substr(prog)) {
6b54ddc5 3280 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
3281 }
3282 }
3283 must = prog->anchored_substr;
3284 }
33b8afdf
JH
3285 back_max = back_min = prog->anchored_offset;
3286 } else {
7e0d5ad7
KW
3287 if (utf8_target) {
3288 if (! prog->float_utf8) {
3289 to_utf8_substr(prog);
3290 }
3291 must = prog->float_utf8;
3292 }
3293 else {
3294 if (! prog->float_substr) {
3295 if (! to_byte_substr(prog)) {
6b54ddc5 3296 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
3297 }
3298 }
3299 must = prog->float_substr;
3300 }
33b8afdf
JH
3301 back_max = prog->float_max_offset;
3302 back_min = prog->float_min_offset;
3303 }
1de06328 3304
1de06328
YO
3305 if (back_min<0) {
3306 last = strend;
3307 } else {
3308 last = HOP3c(strend, /* Cannot start after this */
ea3daa5d 3309 -(SSize_t)(CHR_SVLEN(must)
1de06328
YO
3310 - (SvTAIL(must) != 0) + back_min), strbeg);
3311 }
9d9163fb 3312 if (s > reginfo->strbeg)
6eb5f6b9
JH
3313 last1 = HOPc(s, -1);
3314 else
3315 last1 = s - 1; /* bogus */
3316
a0288114 3317 /* XXXX check_substr already used to find "s", can optimize if
6eb5f6b9 3318 check_substr==must. */
bf05793b 3319 dontbother = 0;
6eb5f6b9
JH
3320 strend = HOPc(strend, -dontbother);
3321 while ( (s <= last) &&
e50d57d4 3322 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
9041c2e3 3323 (unsigned char*)strend, must,
c33e64f0 3324 multiline ? FBMrf_MULTILINE : 0)) ) {
a3621e74 3325 DEBUG_EXECUTE_r( did_match = 1 );
6eb5f6b9
JH
3326 if (HOPc(s, -back_max) > last1) {
3327 last1 = HOPc(s, -back_min);
3328 s = HOPc(s, -back_max);
3329 }
3330 else {
9d9163fb
DM
3331 char * const t = (last1 >= reginfo->strbeg)
3332 ? HOPc(last1, 1) : last1 + 1;
6eb5f6b9
JH
3333
3334 last1 = HOPc(s, -back_min);
52657f30 3335 s = t;
6eb5f6b9 3336 }
f2ed9b32 3337 if (utf8_target) {
6eb5f6b9 3338 while (s <= last1) {
02d5137b 3339 if (regtry(reginfo, &s))
6eb5f6b9 3340 goto got_it;
7016d6eb
DM
3341 if (s >= last1) {
3342 s++; /* to break out of outer loop */
3343 break;
3344 }
3345 s += UTF8SKIP(s);
6eb5f6b9
JH
3346 }
3347 }
3348 else {
3349 while (s <= last1) {
02d5137b 3350 if (regtry(reginfo, &s))
6eb5f6b9
JH
3351 goto got_it;
3352 s++;
3353 }
3354 }
3355 }
ab3bbdeb 3356 DEBUG_EXECUTE_r(if (!did_match) {
f2ed9b32 3357 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
ab3bbdeb 3358 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
6ad9a8ab 3359 Perl_re_printf( aTHX_ "Did not find %s substr %s%s...\n",
33b8afdf 3360 ((must == prog->anchored_substr || must == prog->anchored_utf8)
bf93d4cc 3361 ? "anchored" : "floating"),
ab3bbdeb
YO
3362 quoted, RE_SV_TAIL(must));
3363 });
6eb5f6b9
JH
3364 goto phooey;
3365 }
f8fc2ecf 3366 else if ( (c = progi->regstclass) ) {
f14c76ed 3367 if (minlen) {
f8fc2ecf 3368 const OPCODE op = OP(progi->regstclass);
66e933ab 3369 /* don't bother with what can't match */
33c28ab2 3370 if (PL_regkind[op] != EXACT && PL_regkind[op] != TRIE)
f14c76ed
RGS
3371 strend = HOPc(strend, -(minlen - 1));
3372 }
a3621e74 3373 DEBUG_EXECUTE_r({
be8e71aa 3374 SV * const prop = sv_newmortal();
8b9781c9 3375 regprop(prog, prop, c, reginfo, NULL);
0df25f3d 3376 {
f2ed9b32 3377 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb 3378 s,strend-s,60);
6ad9a8ab 3379 Perl_re_printf( aTHX_
1c8f8eb1 3380 "Matching stclass %.*s against %s (%d bytes)\n",
e4f74956 3381 (int)SvCUR(prop), SvPVX_const(prop),
ab3bbdeb 3382 quoted, (int)(strend - s));
0df25f3d 3383 }
ffc61ed2 3384 });
f9176b44 3385 if (find_byclass(prog, c, s, strend, reginfo))
6eb5f6b9 3386 goto got_it;
6ad9a8ab 3387 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "Contradicts stclass... [regexec_flags]\n"));
d6a28714
JH
3388 }
3389 else {
3390 dontbother = 0;
a0714e2c 3391 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
33b8afdf 3392 /* Trim the end. */
6af40bd7 3393 char *last= NULL;
33b8afdf 3394 SV* float_real;
c33e64f0
FC
3395 STRLEN len;
3396 const char *little;
33b8afdf 3397
7e0d5ad7
KW
3398 if (utf8_target) {
3399 if (! prog->float_utf8) {
3400 to_utf8_substr(prog);
3401 }
3402 float_real = prog->float_utf8;
3403 }
3404 else {
3405 if (! prog->float_substr) {
3406 if (! to_byte_substr(prog)) {
6b54ddc5 3407 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
3408 }
3409 }
3410 float_real = prog->float_substr;
3411 }
d6a28714 3412
c33e64f0
FC
3413 little = SvPV_const(float_real, len);
3414 if (SvTAIL(float_real)) {
7f18ad16
KW
3415 /* This means that float_real contains an artificial \n on
3416 * the end due to the presence of something like this:
3417 * /foo$/ where we can match both "foo" and "foo\n" at the
3418 * end of the string. So we have to compare the end of the
3419 * string first against the float_real without the \n and
3420 * then against the full float_real with the string. We
3421 * have to watch out for cases where the string might be
3422 * smaller than the float_real or the float_real without
3423 * the \n. */
1a13b075
YO
3424 char *checkpos= strend - len;
3425 DEBUG_OPTIMISE_r(
6ad9a8ab 3426 Perl_re_printf( aTHX_
1a13b075
YO
3427 "%sChecking for float_real.%s\n",
3428 PL_colors[4], PL_colors[5]));
3429 if (checkpos + 1 < strbeg) {
7f18ad16
KW
3430 /* can't match, even if we remove the trailing \n
3431 * string is too short to match */
1a13b075 3432 DEBUG_EXECUTE_r(
6ad9a8ab 3433 Perl_re_printf( aTHX_
1a13b075
YO
3434 "%sString shorter than required trailing substring, cannot match.%s\n",
3435 PL_colors[4], PL_colors[5]));
3436 goto phooey;
3437 } else if (memEQ(checkpos + 1, little, len - 1)) {
7f18ad16
KW
3438 /* can match, the end of the string matches without the
3439 * "\n" */
1a13b075
YO
3440 last = checkpos + 1;
3441 } else if (checkpos < strbeg) {
7f18ad16
KW
3442 /* cant match, string is too short when the "\n" is
3443 * included */
1a13b075 3444 DEBUG_EXECUTE_r(
6ad9a8ab 3445 Perl_re_printf( aTHX_
1a13b075
YO
3446 "%sString does not contain required trailing substring, cannot match.%s\n",
3447 PL_colors[4], PL_colors[5]));
3448 goto phooey;
3449 } else if (!multiline) {
7f18ad16
KW
3450 /* non multiline match, so compare with the "\n" at the
3451 * end of the string */
1a13b075
YO
3452 if (memEQ(checkpos, little, len)) {
3453 last= checkpos;
3454 } else {
3455 DEBUG_EXECUTE_r(
6ad9a8ab 3456 Perl_re_printf( aTHX_
1a13b075
YO
3457 "%sString does not contain required trailing substring, cannot match.%s\n",
3458 PL_colors[4], PL_colors[5]));
3459 goto phooey;
3460 }
3461 } else {
7f18ad16
KW
3462 /* multiline match, so we have to search for a place