This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
https://perl5.git.perl.org / perl5.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
regen/mk_invlists.pl: Add GCB tables
[perl5.git] / regexec.c
CommitLineData
a0d0e21e
LW		 1/* regexec.c
2 */
3
4/*
4ac71550
TC		 5 * One Ring to rule them all, One Ring to find them
6 &
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
a0d0e21e
LW		 10 */
11
61296642
DM		 12/* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
166f8a29 14 * a regular expression.
e4a054ea
DM		 15 *
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
166f8a29
DM		 20 */
21
a687059c
LW		 22/* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
24 */
25
26/* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
29 */
30
e50aee73
AD		 31/* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
34*/
35
b9d5759e 36#ifdef PERL_EXT_RE_BUILD
54df2634 37#include "re_top.h"
9041c2e3 38#endif
56953603 39
a687059c 40/*
e50aee73 41 * pregcomp and pregexec -- regsub and regerror are not used in perl
a687059c
LW		 42 *
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
45 *
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
49 *
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
52 * from defects in it.
53 *
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
56 *
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
59 *
60 **** Alterations to Henry's code are...
61 ****
4bb101f2 62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
1129b882
NC		 63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
a687059c 65 ****
9ef589d8
LW		 66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
a687059c
LW		 68 *
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
72 */
73#include "EXTERN.h"
864dbfa3 74#define PERL_IN_REGEXEC_C
a687059c 75#include "perl.h"
0f5d15d6 76
54df2634
NC		 77#ifdef PERL_IN_XSUB_RE
78# include "re_comp.h"
79#else
80# include "regcomp.h"
81#endif
a687059c 82
81e983c1 83#include "inline_invlist.c"
1b0f46bf 84#include "unicode_constants.h"
81e983c1 85
e1cf74e3
CB		 86#ifdef DEBUGGING
87/* At least one required character in the target string is expressible only in
88 * UTF-8. */
89static const char* const non_utf8_target_but_utf8_required
90 = "Can't match, because target string needs to be in UTF-8\n";
91#endif
92
93#define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
94 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\
95 goto target; \
96} STMT_END
97
c74f6de9
KW		 98#define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
99
a687059c
LW		 100#ifndef STATIC
101#define STATIC static
102#endif
103
2d66f61e 104/* Valid only for non-utf8 strings: avoids the reginclass
7e2509c1
KW		 105 * call if there are no complications: i.e., if everything matchable is
106 * straight forward in the bitmap */
3db24e1e 107#define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,c+1,0) \
af364d03 108 : ANYOF_BITMAP_TEST(p,*(c)))
7d3e948e 109
c277df42
IZ		 110/*
111 * Forwards.
112 */
113
f2ed9b32 114#define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
ba44c216 115#define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b)
a0ed51b3 116
3dab1dad 117#define HOPc(pos,off) \
ba44c216 118 (char *)(reginfo->is_utf8_target \
220db18a 119 ? reghop3((U8*)pos, off, \
9d9163fb 120 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
3dab1dad 121 : (U8*)(pos + off))
557f47af 122
3dab1dad 123#define HOPBACKc(pos, off) \
ba44c216 124 (char*)(reginfo->is_utf8_target \
9d9163fb
DM		 125 ? reghopmaybe3((U8*)pos, -off, (U8*)(reginfo->strbeg)) \
126 : (pos - off >= reginfo->strbeg) \
8e11feef 127 ? (U8*)pos - off \
3dab1dad 128 : NULL)
efb30f32 129
ba44c216 130#define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
1aa99e6b 131#define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
1aa99e6b 132
557f47af
DM		 133/* lim must be +ve. Returns NULL on overshoot */
134#define HOPMAYBE3(pos,off,lim) \
135 (reginfo->is_utf8_target \
136 ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \
137 : ((U8*)pos + off <= lim) \
138 ? (U8*)pos + off \
139 : NULL)
140
8e9f2289
DM		 141/* like HOP3, but limits the result to <= lim even for the non-utf8 case.
142 * off must be >=0; args should be vars rather than expressions */
143#define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \
144 ? reghop3((U8*)(pos), off, (U8*)(lim)) \
145 : (U8*)((pos + off) > lim ? lim : (pos + off)))
146
2974eaec
DM		 147#define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \
148 ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \
149 : (U8*)(pos + off))
150#define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim))
7016d6eb
DM		 151
152#define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
153#define NEXTCHR_IS_EOS (nextchr < 0)
154
155#define SET_nextchr \
220db18a 156 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
7016d6eb
DM		 157
158#define SET_locinput(p) \
159 locinput = (p); \
160 SET_nextchr
161
162
2a16ac92 163#define LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) STMT_START { \
c7304fe2
KW		 164 if (!swash_ptr) { \
165 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
c7304fe2 166 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
2a16ac92 167 1, 0, invlist, &flags); \
c7304fe2
KW		 168 assert(swash_ptr); \
169 } \
170 } STMT_END
171
172/* If in debug mode, we test that a known character properly matches */
173#ifdef DEBUGGING
174# define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
175 property_name, \
2a16ac92 176 invlist, \
c7304fe2 177 utf8_char_in_property) \
2a16ac92 178 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist); \
c7304fe2
KW		 179 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
180#else
181# define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
182 property_name, \
2a16ac92 183 invlist, \
c7304fe2 184 utf8_char_in_property) \
2a16ac92 185 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist)
c7304fe2 186#endif
d1eb3177 187
c7304fe2
KW		 188#define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
189 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
2a16ac92
KW		 190 "", \
191 PL_XPosix_ptrs[_CC_WORDCHAR], \
df38da56 192 LATIN_CAPITAL_LETTER_SHARP_S_UTF8);
c7304fe2
KW		 193
194#define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
195 STMT_START { \
196 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \
197 "_X_regular_begin", \
2a16ac92 198 NULL, \
df38da56 199 LATIN_CAPITAL_LETTER_SHARP_S_UTF8); \
c7304fe2
KW		 200 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \
201 "_X_extend", \
2a16ac92 202 NULL, \
c7304fe2
KW		 203 COMBINING_GRAVE_ACCENT_UTF8); \
204 } STMT_END
d1eb3177 205
c7304fe2 206#define PLACEHOLDER /* Something for the preprocessor to grab onto */
3dab1dad
YO		 207/* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
208
5f80c4cf 209/* for use after a quantifier and before an EXACT-like node -- japhy */
c35dcbe2
YO		 210/* it would be nice to rework regcomp.sym to generate this stuff. sigh
211 *
212 * NOTE that *nothing* that affects backtracking should be in here, specifically
213 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
214 * node that is in between two EXACT like nodes when ascertaining what the required
215 * "follow" character is. This should probably be moved to regex compile time
216 * although it may be done at run time beause of the REF possibility - more
217 * investigation required. -- demerphq
218*/
baa60164
KW		 219#define JUMPABLE(rn) ( \
220 OP(rn) == OPEN || \
3e901dc0 221 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
baa60164
KW		 222 OP(rn) == EVAL || \
223 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
224 OP(rn) == PLUS || OP(rn) == MINMOD || \
225 OP(rn) == KEEPS || \
226 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
e2d8ce26 227)
ee9b8eae 228#define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
e2d8ce26 229
ee9b8eae
YO		 230#define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
231
232#if 0
233/* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
a4525e78 234 we don't need this definition. XXX These are now out-of-sync*/
ee9b8eae 235#define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
098b07d5 236#define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFA || OP(rn)==EXACTFA_NO_TRIE || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
ee9b8eae
YO		 237#define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
238
239#else
240/* ... so we use this as its faster. */
a4525e78
KW		 241#define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==EXACTL )
242#define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFLU8 || OP(rn)==EXACTFU_SS || OP(rn) == EXACTFA || OP(rn) == EXACTFA_NO_TRIE)
ee9b8eae
YO		 243#define IS_TEXTF(rn) ( OP(rn)==EXACTF )
244#define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
245
246#endif
e2d8ce26 247
a84d97b6
HS		 248/*
249 Search for mandatory following text node; for lookahead, the text must
250 follow but for lookbehind (rn->flags != 0) we skip to the next step.
251*/
baa60164 252#define FIND_NEXT_IMPT(rn) STMT_START { \
3dab1dad
YO		 253 while (JUMPABLE(rn)) { \
254 const OPCODE type = OP(rn); \
255 if (type == SUSPEND || PL_regkind[type] == CURLY) \
e2d8ce26 256 rn = NEXTOPER(NEXTOPER(rn)); \
3dab1dad 257 else if (type == PLUS) \
cca55fe3 258 rn = NEXTOPER(rn); \
3dab1dad 259 else if (type == IFMATCH) \
a84d97b6 260 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
e2d8ce26 261 else rn += NEXT_OFF(rn); \
3dab1dad 262 } \
5f80c4cf 263} STMT_END
74750237 264
22913b96
KW		 265/* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode.
266 * These are for the pre-composed Hangul syllables, which are all in a
267 * contiguous block and arranged there in such a way so as to facilitate
268 * alorithmic determination of their characteristics. As such, they don't need
269 * a swash, but can be determined by simple arithmetic. Almost all are
270 * GCB=LVT, but every 28th one is a GCB=LV */
271#define SBASE 0xAC00 /* Start of block */
272#define SCount 11172 /* Length of block */
273#define TCount 28
c476f425 274
006f26b2
DM		 275#define SLAB_FIRST(s) (&(s)->states[0])
276#define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
277
a75351a1 278static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
bf2039a9 279static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
bf2039a9 280static regmatch_state * S_push_slab(pTHX);
51371543 281
87c0511b 282#define REGCP_PAREN_ELEMS 3
f067efbf 283#define REGCP_OTHER_ELEMS 3
e0fa7e2b 284#define REGCP_FRAME_ELEMS 1
620d5b66
NC		 285/* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
286 * are needed for the regexp context stack bookkeeping. */
287
76e3520e 288STATIC CHECKPOINT
92da3157 289S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
a0d0e21e 290{
a3b680e6 291 const int retval = PL_savestack_ix;
92da3157
DM		 292 const int paren_elems_to_push =
293 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
e0fa7e2b
NC		 294 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
295 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
87c0511b 296 I32 p;
40a82448 297 GET_RE_DEBUG_FLAGS_DECL;
a0d0e21e 298
b93070ed
DM		 299 PERL_ARGS_ASSERT_REGCPPUSH;
300
e49a9654 301 if (paren_elems_to_push < 0)
e8a85d26
JH		 302 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
303 (int)paren_elems_to_push, (int)maxopenparen,
304 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
e49a9654 305
e0fa7e2b
NC		 306 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
307 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
5df417d0 308 " out of range (%lu-%ld)",
92da3157
DM		 309 total_elems,
310 (unsigned long)maxopenparen,
311 (long)parenfloor);
e0fa7e2b 312
620d5b66 313 SSGROW(total_elems + REGCP_FRAME_ELEMS);
7f69552c 314
495f47a5 315 DEBUG_BUFFERS_r(
92da3157 316 if ((int)maxopenparen > (int)parenfloor)
495f47a5
DM		 317 PerlIO_printf(Perl_debug_log,
318 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
319 PTR2UV(rex),
320 PTR2UV(rex->offs)
321 );
322 );
92da3157 323 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
b1ce53c5 324/* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
99a90e59
FC		 325 SSPUSHIV(rex->offs[p].end);
326 SSPUSHIV(rex->offs[p].start);
1ca2007e 327 SSPUSHINT(rex->offs[p].start_tmp);
e7707071 328 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
495f47a5
DM		 329 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
330 (UV)p,
331 (IV)rex->offs[p].start,
332 (IV)rex->offs[p].start_tmp,
333 (IV)rex->offs[p].end
40a82448 334 ));
a0d0e21e 335 }
b1ce53c5 336/* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
92da3157 337 SSPUSHINT(maxopenparen);
b93070ed
DM		 338 SSPUSHINT(rex->lastparen);
339 SSPUSHINT(rex->lastcloseparen);
e0fa7e2b 340 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
41123dfd 341
a0d0e21e
LW		 342 return retval;
343}
344
c277df42 345/* These are needed since we do not localize EVAL nodes: */
ab3bbdeb
YO		 346#define REGCP_SET(cp) \
347 DEBUG_STATE_r( \
ab3bbdeb 348 PerlIO_printf(Perl_debug_log, \
e4f74956 349 " Setting an EVAL scope, savestack=%"IVdf"\n", \
ab3bbdeb
YO		 350 (IV)PL_savestack_ix)); \
351 cp = PL_savestack_ix
c3464db5 352
ab3bbdeb 353#define REGCP_UNWIND(cp) \
e4f74956 354 DEBUG_STATE_r( \
ab3bbdeb 355 if (cp != PL_savestack_ix) \
e4f74956
YO		 356 PerlIO_printf(Perl_debug_log, \
357 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
ab3bbdeb
YO		 358 (IV)(cp), (IV)PL_savestack_ix)); \
359 regcpblow(cp)
c277df42 360
a8d1f4b4
DM		 361#define UNWIND_PAREN(lp, lcp) \
362 for (n = rex->lastparen; n > lp; n--) \
363 rex->offs[n].end = -1; \
364 rex->lastparen = n; \
365 rex->lastcloseparen = lcp;
366
367
f067efbf 368STATIC void
92da3157 369S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
a0d0e21e 370{
e0fa7e2b 371 UV i;
87c0511b 372 U32 paren;
a3621e74
YO		 373 GET_RE_DEBUG_FLAGS_DECL;
374
7918f24d
NC		 375 PERL_ARGS_ASSERT_REGCPPOP;
376
b1ce53c5 377 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
c6bf6a65 378 i = SSPOPUV;
e0fa7e2b
NC		 379 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
380 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
b93070ed
DM		 381 rex->lastcloseparen = SSPOPINT;
382 rex->lastparen = SSPOPINT;
92da3157 383 *maxopenparen_p = SSPOPINT;
b1ce53c5 384
620d5b66 385 i -= REGCP_OTHER_ELEMS;
b1ce53c5 386 /* Now restore the parentheses context. */
495f47a5
DM		 387 DEBUG_BUFFERS_r(
388 if (i || rex->lastparen + 1 <= rex->nparens)
389 PerlIO_printf(Perl_debug_log,
390 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
391 PTR2UV(rex),
392 PTR2UV(rex->offs)
393 );
394 );
92da3157 395 paren = *maxopenparen_p;
620d5b66 396 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
ea3daa5d 397 SSize_t tmps;
1ca2007e 398 rex->offs[paren].start_tmp = SSPOPINT;
99a90e59
FC		 399 rex->offs[paren].start = SSPOPIV;
400 tmps = SSPOPIV;
b93070ed
DM		 401 if (paren <= rex->lastparen)
402 rex->offs[paren].end = tmps;
495f47a5
DM		 403 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
404 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
405 (UV)paren,
406 (IV)rex->offs[paren].start,
407 (IV)rex->offs[paren].start_tmp,
408 (IV)rex->offs[paren].end,
409 (paren > rex->lastparen ? "(skipped)" : ""));
c277df42 410 );
87c0511b 411 paren--;
a0d0e21e 412 }
daf18116 413#if 1
dafc8851
JH		 414 /* It would seem that the similar code in regtry()
415 * already takes care of this, and in fact it is in
416 * a better location to since this code can #if 0-ed out
417 * but the code in regtry() is needed or otherwise tests
418 * requiring null fields (pat.t#187 and split.t#{13,14}
daf18116
JH		 419 * (as of patchlevel 7877) will fail. Then again,
420 * this code seems to be necessary or otherwise
225593e1
DM		 421 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
422 * --jhi updated by dapm */
b93070ed 423 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
92da3157 424 if (i > *maxopenparen_p)
b93070ed
DM		 425 rex->offs[i].start = -1;
426 rex->offs[i].end = -1;
495f47a5
DM		 427 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
428 " \\%"UVuf": %s ..-1 undeffing\n",
429 (UV)i,
92da3157 430 (i > *maxopenparen_p) ? "-1" : " "
495f47a5 431 ));
a0d0e21e 432 }
dafc8851 433#endif
a0d0e21e
LW		 434}
435
74088413
DM		 436/* restore the parens and associated vars at savestack position ix,
437 * but without popping the stack */
438
439STATIC void
92da3157 440S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
74088413
DM		 441{
442 I32 tmpix = PL_savestack_ix;
443 PL_savestack_ix = ix;
92da3157 444 regcppop(rex, maxopenparen_p);
74088413
DM		 445 PL_savestack_ix = tmpix;
446}
447
02db2b7b 448#define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
a0d0e21e 449
31c7f561
KW		 450STATIC bool
451S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
452{
453 /* Returns a boolean as to whether or not 'character' is a member of the
454 * Posix character class given by 'classnum' that should be equivalent to a
455 * value in the typedef '_char_class_number'.
456 *
457 * Ideally this could be replaced by a just an array of function pointers
458 * to the C library functions that implement the macros this calls.
459 * However, to compile, the precise function signatures are required, and
460 * these may vary from platform to to platform. To avoid having to figure
461 * out what those all are on each platform, I (khw) am using this method,
7aee35ff
KW		 462 * which adds an extra layer of function call overhead (unless the C
463 * optimizer strips it away). But we don't particularly care about
464 * performance with locales anyway. */
31c7f561
KW		 465
466 switch ((_char_class_number) classnum) {
15861f94 467 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
31c7f561 468 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
e8d596e0
KW		 469 case _CC_ENUM_ASCII: return isASCII_LC(character);
470 case _CC_ENUM_BLANK: return isBLANK_LC(character);
b0d691b2
KW		 471 case _CC_ENUM_CASED: return isLOWER_LC(character)
472 || isUPPER_LC(character);
e8d596e0 473 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
31c7f561
KW		 474 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
475 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
476 case _CC_ENUM_LOWER: return isLOWER_LC(character);
477 case _CC_ENUM_PRINT: return isPRINT_LC(character);
e8d596e0 478 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
31c7f561 479 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
e8d596e0 480 case _CC_ENUM_SPACE: return isSPACE_LC(character);
31c7f561
KW		 481 case _CC_ENUM_UPPER: return isUPPER_LC(character);
482 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
31c7f561 483 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
31c7f561
KW		 484 default: /* VERTSPACE should never occur in locales */
485 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
486 }
487
e5964223 488 NOT_REACHED; /* NOTREACHED */
31c7f561
KW		 489 return FALSE;
490}
491
3018b823
KW		 492STATIC bool
493S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
494{
495 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
496 * 'character' is a member of the Posix character class given by 'classnum'
497 * that should be equivalent to a value in the typedef
498 * '_char_class_number'.
499 *
500 * This just calls isFOO_lc on the code point for the character if it is in
2f306ab9 501 * the range 0-255. Outside that range, all characters use Unicode
3018b823
KW		 502 * rules, ignoring any locale. So use the Unicode function if this class
503 * requires a swash, and use the Unicode macro otherwise. */
504
505 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
506
507 if (UTF8_IS_INVARIANT(*character)) {
508 return isFOO_lc(classnum, *character);
509 }
510 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
511 return isFOO_lc(classnum,
94bb8c36 512 TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1)));
3018b823
KW		 513 }
514
613abc6d
KW		 515 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(character, character + UTF8SKIP(character));
516
3018b823
KW		 517 if (classnum < _FIRST_NON_SWASH_CC) {
518
519 /* Initialize the swash unless done already */
520 if (! PL_utf8_swash_ptrs[classnum]) {
521 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2a16ac92
KW		 522 PL_utf8_swash_ptrs[classnum] =
523 _core_swash_init("utf8",
524 "",
525 &PL_sv_undef, 1, 0,
526 PL_XPosix_ptrs[classnum], &flags);
3018b823
KW		 527 }
528
92a2046b
KW		 529 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
530 character,
531 TRUE /* is UTF */ ));
3018b823
KW		 532 }
533
534 switch ((_char_class_number) classnum) {
535 case _CC_ENUM_SPACE:
536 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
537
538 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
539 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
540 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
e1ee3960 541 default: break;
3018b823
KW		 542 }
543
e1ee3960 544 return FALSE; /* Things like CNTRL are always below 256 */
3018b823
KW		 545}
546
a687059c 547/*
e50aee73 548 * pregexec and friends
a687059c
LW		 549 */
550
76234dfb 551#ifndef PERL_IN_XSUB_RE
a687059c 552/*
c277df42 553 - pregexec - match a regexp against a string
a687059c 554 */
c277df42 555I32
5aaab254 556Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
ea3daa5d 557 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
8fd1a950
DM		 558/* stringarg: the point in the string at which to begin matching */
559/* strend: pointer to null at end of string */
560/* strbeg: real beginning of string */
561/* minend: end of match must be >= minend bytes after stringarg. */
562/* screamer: SV being matched: only used for utf8 flag, pos() etc; string
563 * itself is accessed via the pointers above */
564/* nosave: For optimizations. */
c277df42 565{
7918f24d
NC		 566 PERL_ARGS_ASSERT_PREGEXEC;
567
c277df42 568 return
9041c2e3 569 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
c277df42
IZ		 570 nosave ? 0 : REXEC_COPY_STR);
571}
76234dfb 572#endif
22e551b9 573
cad2e5aa 574
6eb5f6b9 575
1a4edc3c
DM		 576/* re_intuit_start():
577 *
578 * Based on some optimiser hints, try to find the earliest position in the
579 * string where the regex could match.
580 *
581 * rx: the regex to match against
582 * sv: the SV being matched: only used for utf8 flag; the string
583 * itself is accessed via the pointers below. Note that on
584 * something like an overloaded SV, SvPOK(sv) may be false
585 * and the string pointers may point to something unrelated to
586 * the SV itself.
587 * strbeg: real beginning of string
588 * strpos: the point in the string at which to begin matching
589 * strend: pointer to the byte following the last char of the string
590 * flags currently unused; set to 0
591 * data: currently unused; set to NULL
592 *
593 * The basic idea of re_intuit_start() is to use some known information
594 * about the pattern, namely:
595 *
596 * a) the longest known anchored substring (i.e. one that's at a
597 * constant offset from the beginning of the pattern; but not
598 * necessarily at a fixed offset from the beginning of the
599 * string);
600 * b) the longest floating substring (i.e. one that's not at a constant
601 * offset from the beginning of the pattern);
602 * c) Whether the pattern is anchored to the string; either
603 * an absolute anchor: /^../, or anchored to \n: /^.../m,
604 * or anchored to pos(): /\G/;
605 * d) A start class: a real or synthetic character class which
606 * represents which characters are legal at the start of the pattern;
607 *
608 * to either quickly reject the match, or to find the earliest position
609 * within the string at which the pattern might match, thus avoiding
610 * running the full NFA engine at those earlier locations, only to
611 * eventually fail and retry further along.
612 *
613 * Returns NULL if the pattern can't match, or returns the address within
614 * the string which is the earliest place the match could occur.
615 *
616 * The longest of the anchored and floating substrings is called 'check'
617 * and is checked first. The other is called 'other' and is checked
618 * second. The 'other' substring may not be present. For example,
619 *
620 * /(abc|xyz)ABC\d{0,3}DEFG/
621 *
622 * will have
623 *
624 * check substr (float) = "DEFG", offset 6..9 chars
625 * other substr (anchored) = "ABC", offset 3..3 chars
626 * stclass = [ax]
627 *
628 * Be aware that during the course of this function, sometimes 'anchored'
629 * refers to a substring being anchored relative to the start of the
630 * pattern, and sometimes to the pattern itself being anchored relative to
631 * the string. For example:
632 *
633 * /\dabc/: "abc" is anchored to the pattern;
634 * /^\dabc/: "abc" is anchored to the pattern and the string;
635 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
636 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
637 * but the pattern is anchored to the string.
52a21eb3
DM		 638 */
639
cad2e5aa 640char *
52a21eb3
DM		 641Perl_re_intuit_start(pTHX_
642 REGEXP * const rx,
643 SV *sv,
644 const char * const strbeg,
645 char *strpos,
646 char *strend,
647 const U32 flags,
648 re_scream_pos_data *data)
cad2e5aa 649{
8d919b0a 650 struct regexp *const prog = ReANY(rx);
6b071d16 651 SSize_t start_shift = prog->check_offset_min;
cad2e5aa 652 /* Should be nonnegative! */
ea3daa5d 653 SSize_t end_shift = 0;
0fc004dd
DM		 654 /* current lowest pos in string where the regex can start matching */
655 char *rx_origin = strpos;
eb578fdb 656 SV *check;
f2ed9b32 657 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
6480a6c4 658 U8 other_ix = 1 - prog->substrs->check_ix;
6ad5ffb3 659 bool ml_anch = 0;
8f4bf5fc 660 char *other_last = strpos;/* latest pos 'other' substr already checked to */
bd61b366 661 char *check_at = NULL; /* check substr found at this pos */
bbe252da 662 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
f8fc2ecf 663 RXi_GET_DECL(prog,progi);
02d5137b
DM		 664 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
665 regmatch_info *const reginfo = &reginfo_buf;
a3621e74
YO		 666 GET_RE_DEBUG_FLAGS_DECL;
667
7918f24d 668 PERL_ARGS_ASSERT_RE_INTUIT_START;
c33e64f0
FC		 669 PERL_UNUSED_ARG(flags);
670 PERL_UNUSED_ARG(data);
7918f24d 671
1dc475d0
DM		 672 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
673 "Intuit: trying to determine minimum start position...\n"));
674
fb9bbddb
DM		 675 /* for now, assume that all substr offsets are positive. If at some point
676 * in the future someone wants to do clever things with look-behind and
677 * -ve offsets, they'll need to fix up any code in this function
678 * which uses these offsets. See the thread beginning
679 * <20140113145929.GF27210@iabyn.com>
680 */
681 assert(prog->substrs->data[0].min_offset >= 0);
682 assert(prog->substrs->data[0].max_offset >= 0);
683 assert(prog->substrs->data[1].min_offset >= 0);
684 assert(prog->substrs->data[1].max_offset >= 0);
685 assert(prog->substrs->data[2].min_offset >= 0);
686 assert(prog->substrs->data[2].max_offset >= 0);
687
f7022b5a 688 /* for now, assume that if both present, that the floating substring
83f2232d 689 * doesn't start before the anchored substring.
f7022b5a
DM		 690 * If you break this assumption (e.g. doing better optimisations
691 * with lookahead/behind), then you'll need to audit the code in this
692 * function carefully first
693 */
694 assert(
695 ! ( (prog->anchored_utf8 || prog->anchored_substr)
696 && (prog->float_utf8 || prog->float_substr))
697 || (prog->float_min_offset >= prog->anchored_offset));
698
1a4edc3c
DM		 699 /* byte rather than char calculation for efficiency. It fails
700 * to quickly reject some cases that can't match, but will reject
701 * them later after doing full char arithmetic */
c344f387 702 if (prog->minlen > strend - strpos) {
a3621e74 703 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 704 " String too short...\n"));
cad2e5aa 705 goto fail;
2c2d71f5 706 }
d8da0584 707
ab4e48c1 708 RX_MATCH_UTF8_set(rx,utf8_target);
6c3fea77 709 reginfo->is_utf8_target = cBOOL(utf8_target);
bf2039a9 710 reginfo->info_aux = NULL;
9d9163fb 711 reginfo->strbeg = strbeg;
220db18a 712 reginfo->strend = strend;
aed7b151 713 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
02d5137b 714 reginfo->intuit = 1;
1cb48e53
DM		 715 /* not actually used within intuit, but zero for safety anyway */
716 reginfo->poscache_maxiter = 0;
02d5137b 717
f2ed9b32 718 if (utf8_target) {
33b8afdf
JH		 719 if (!prog->check_utf8 && prog->check_substr)
720 to_utf8_substr(prog);
721 check = prog->check_utf8;
722 } else {
7e0d5ad7
KW		 723 if (!prog->check_substr && prog->check_utf8) {
724 if (! to_byte_substr(prog)) {
6b54ddc5 725 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
7e0d5ad7
KW		 726 }
727 }
33b8afdf
JH		 728 check = prog->check_substr;
729 }
274cd312 730
1dc475d0
DM		 731 /* dump the various substring data */
732 DEBUG_OPTIMISE_MORE_r({
733 int i;
734 for (i=0; i<=2; i++) {
735 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
736 : prog->substrs->data[i].substr);
737 if (!sv)
738 continue;
739
740 PerlIO_printf(Perl_debug_log,
741 " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf
742 " useful=%"IVdf" utf8=%d [%s]\n",
743 i,
744 (IV)prog->substrs->data[i].min_offset,
745 (IV)prog->substrs->data[i].max_offset,
746 (IV)prog->substrs->data[i].end_shift,
747 BmUSEFUL(sv),
748 utf8_target ? 1 : 0,
749 SvPEEK(sv));
750 }
751 });
752
8e1490ee 753 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
9fc7410e
DM		 754
755 /* ml_anch: check after \n?
756 *
0fa70a06 757 * A note about PREGf_IMPLICIT: on an un-anchored pattern beginning
9fc7410e
DM		 758 * with /.*.../, these flags will have been added by the
759 * compiler:
760 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
761 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
762 */
7d2d37f5
DM		 763 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
764 && !(prog->intflags & PREGf_IMPLICIT);
cad2e5aa 765
343c8a29 766 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
c889ccc8
DM		 767 /* we are only allowed to match at BOS or \G */
768
57fcbfa7 769 /* trivially reject if there's a BOS anchor and we're not at BOS.
7bb3b9eb
DM		 770 *
771 * Note that we don't try to do a similar quick reject for
772 * \G, since generally the caller will have calculated strpos
773 * based on pos() and gofs, so the string is already correctly
774 * anchored by definition; and handling the exceptions would
775 * be too fiddly (e.g. REXEC_IGNOREPOS).
57fcbfa7 776 */
7bb3b9eb 777 if ( strpos != strbeg
d3d47aac 778 && (prog->intflags & PREGf_ANCH_SBOL))
c889ccc8
DM		 779 {
780 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 781 " Not at start...\n"));
c889ccc8
DM		 782 goto fail;
783 }
784
a5d12a4b
DM		 785 /* in the presence of an anchor, the anchored (relative to the
786 * start of the regex) substr must also be anchored relative
66b7ec5c
DM		 787 * to strpos. So quickly reject if substr isn't found there.
788 * This works for \G too, because the caller will already have
789 * subtracted gofs from pos, and gofs is the offset from the
790 * \G to the start of the regex. For example, in /.abc\Gdef/,
791 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
792 * caller will have set strpos=pos()-4; we look for the substr
793 * at position pos()-4+1, which lines up with the "a" */
a5d12a4b 794
c889ccc8 795 if (prog->check_offset_min == prog->check_offset_max
d0d44648 796 && !(prog->intflags & PREGf_CANY_SEEN))
c889ccc8
DM		 797 {
798 /* Substring at constant offset from beg-of-str... */
d307bf57 799 SSize_t slen = SvCUR(check);
343c8a29 800 char *s = HOP3c(strpos, prog->check_offset_min, strend);
1de06328 801
1dc475d0
DM		 802 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
803 " Looking for check substr at fixed offset %"IVdf"...\n",
804 (IV)prog->check_offset_min));
805
7742aa66
DM		 806 if (SvTAIL(check)) {
807 /* In this case, the regex is anchored at the end too.
808 * Unless it's a multiline match, the lengths must match
809 * exactly, give or take a \n. NB: slen >= 1 since
810 * the last char of check is \n */
811 if (!multiline
812 && ( strend - s > slen
813 || strend - s < slen - 1
814 || (strend - s == slen && strend[-1] != '\n')))
c889ccc8
DM		 815 {
816 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 817 " String too long...\n"));
c889ccc8
DM		 818 goto fail_finish;
819 }
820 /* Now should match s[0..slen-2] */
821 slen--;
c889ccc8 822 }
d307bf57
DM		 823 if (slen && (*SvPVX_const(check) != *s
824 || (slen > 1 && memNE(SvPVX_const(check), s, slen))))
825 {
826 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 827 " String not equal...\n"));
d307bf57
DM		 828 goto fail_finish;
829 }
c889ccc8
DM		 830
831 check_at = s;
832 goto success_at_start;
cad2e5aa 833 }
cad2e5aa 834 }
cad2e5aa 835 }
0fc004dd 836
c0e0ec46 837 end_shift = prog->check_end_shift;
cad2e5aa 838
19188028 839#ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
0033605d 840 if (end_shift < 0)
1de06328 841 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
220fc49f 842 (IV)end_shift, RX_PRECOMP(prog));
2c2d71f5
JH		 843#endif
844
2c2d71f5 845 restart:
1de06328 846
66b7ec5c
DM		 847 /* This is the (re)entry point of the main loop in this function.
848 * The goal of this loop is to:
849 * 1) find the "check" substring in the region rx_origin..strend
850 * (adjusted by start_shift / end_shift). If not found, reject
851 * immediately.
852 * 2) If it exists, look for the "other" substr too if defined; for
853 * example, if the check substr maps to the anchored substr, then
854 * check the floating substr, and vice-versa. If not found, go
855 * back to (1) with rx_origin suitably incremented.
856 * 3) If we find an rx_origin position that doesn't contradict
857 * either of the substrings, then check the possible additional
858 * constraints on rx_origin of /^.../m or a known start class.
859 * If these fail, then depending on which constraints fail, jump
860 * back to here, or to various other re-entry points further along
861 * that skip some of the first steps.
862 * 4) If we pass all those tests, update the BmUSEFUL() count on the
863 * substring. If the start position was determined to be at the
864 * beginning of the string - so, not rejected, but not optimised,
865 * since we have to run regmatch from position 0 - decrement the
866 * BmUSEFUL() count. Otherwise increment it.
867 */
868
1a4edc3c
DM		 869
870 /* first, look for the 'check' substring */
871
1de06328 872 {
c33e64f0
FC		 873 U8* start_point;
874 U8* end_point;
c889ccc8 875
c889ccc8 876 DEBUG_OPTIMISE_MORE_r({
1dc475d0 877 PerlIO_printf(Perl_debug_log,
ae5d4331 878 " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf
1dc475d0 879 " Start shift: %"IVdf" End shift %"IVdf
4d281893 880 " Real end Shift: %"IVdf"\n",
197641b6 881 (IV)(rx_origin - strpos),
c889ccc8 882 (IV)prog->check_offset_min,
12fbc530
DM		 883 (IV)start_shift,
884 (IV)end_shift,
c889ccc8
DM		 885 (IV)prog->check_end_shift);
886 });
1de06328 887
0d331aaf 888 if (prog->intflags & PREGf_CANY_SEEN) {
0fc004dd 889 start_point= (U8*)(rx_origin + start_shift);
12fbc530 890 end_point= (U8*)(strend - end_shift);
557f47af
DM		 891 if (start_point > end_point)
892 goto fail_finish;
1de06328 893 } else {
557f47af
DM		 894 end_point = HOP3(strend, -end_shift, strbeg);
895 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
896 if (!start_point)
897 goto fail_finish;
1de06328 898 }
c889ccc8 899
557f47af 900
e0362b86 901 /* If the regex is absolutely anchored to either the start of the
d3d47aac 902 * string (SBOL) or to pos() (ANCH_GPOS), then
e0362b86
DM		 903 * check_offset_max represents an upper bound on the string where
904 * the substr could start. For the ANCH_GPOS case, we assume that
905 * the caller of intuit will have already set strpos to
906 * pos()-gofs, so in this case strpos + offset_max will still be
907 * an upper bound on the substr.
908 */
c19c836a
DM		 909 if (!ml_anch
910 && prog->intflags & PREGf_ANCH
e0362b86 911 && prog->check_offset_max != SSize_t_MAX)
c19c836a 912 {
1a08ba3a 913 SSize_t len = SvCUR(check) - !!SvTAIL(check);
e0362b86
DM		 914 const char * const anchor =
915 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
916
917 /* do a bytes rather than chars comparison. It's conservative;
918 * so it skips doing the HOP if the result can't possibly end
919 * up earlier than the old value of end_point.
920 */
921 if ((char*)end_point - anchor > prog->check_offset_max) {
922 end_point = HOP3lim((U8*)anchor,
923 prog->check_offset_max,
924 end_point -len)
925 + len;
926 }
d6ef1678
DM		 927 }
928
6bda09f9 929 DEBUG_OPTIMISE_MORE_r({
1dc475d0 930 PerlIO_printf(Perl_debug_log, " fbm_instr len=%d str=<%.*s>\n",
1de06328 931 (int)(end_point - start_point),
fc8cd66c 932 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
1de06328
YO		 933 start_point);
934 });
935
ae5d4331 936 check_at = fbm_instr( start_point, end_point,
7fba1cd6 937 check, multiline ? FBMrf_MULTILINE : 0);
c889ccc8 938
8fd34720
DM		 939 /* Update the count-of-usability, remove useless subpatterns,
940 unshift s. */
941
942 DEBUG_EXECUTE_r({
943 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
944 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
945 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s",
946 (check_at ? "Found" : "Did not find"),
947 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
948 ? "anchored" : "floating"),
949 quoted,
950 RE_SV_TAIL(check),
951 (check_at ? " at offset " : "...\n") );
952 });
2c2d71f5 953
8fd34720
DM		 954 if (!check_at)
955 goto fail_finish;
956 /* Finish the diagnostic message */
957 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(check_at - strpos)) );
2c2d71f5 958
8fd34720
DM		 959 /* set rx_origin to the minimum position where the regex could start
960 * matching, given the constraint of the just-matched check substring.
961 * But don't set it lower than previously.
962 */
fdc003fd 963
8fd34720
DM		 964 if (check_at - rx_origin > prog->check_offset_max)
965 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
966 }
fdc003fd
DM		 967
968
1a4edc3c 969 /* now look for the 'other' substring if defined */
2c2d71f5 970
6480a6c4
DM		 971 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
972 : prog->substrs->data[other_ix].substr)
1de06328 973 {
30944b6d 974 /* Take into account the "other" substring. */
6c3343a6
DM		 975 char *last, *last1;
976 char *s;
977 SV* must;
978 struct reg_substr_datum *other;
979
980 do_other_substr:
981 other = &prog->substrs->data[other_ix];
982
983 /* if "other" is anchored:
984 * we've previously found a floating substr starting at check_at.
985 * This means that the regex origin must lie somewhere
986 * between min (rx_origin): HOP3(check_at, -check_offset_max)
987 * and max: HOP3(check_at, -check_offset_min)
988 * (except that min will be >= strpos)
989 * So the fixed substr must lie somewhere between
990 * HOP3(min, anchored_offset)
991 * HOP3(max, anchored_offset) + SvCUR(substr)
992 */
993
994 /* if "other" is floating
995 * Calculate last1, the absolute latest point where the
996 * floating substr could start in the string, ignoring any
997 * constraints from the earlier fixed match. It is calculated
998 * as follows:
999 *
1000 * strend - prog->minlen (in chars) is the absolute latest
1001 * position within the string where the origin of the regex
1002 * could appear. The latest start point for the floating
1003 * substr is float_min_offset(*) on from the start of the
1004 * regex. last1 simply combines thee two offsets.
1005 *
1006 * (*) You might think the latest start point should be
1007 * float_max_offset from the regex origin, and technically
1008 * you'd be correct. However, consider
1009 * /a\d{2,4}bcd\w/
1010 * Here, float min, max are 3,5 and minlen is 7.
1011 * This can match either
1012 * /a\d\dbcd\w/
1013 * /a\d\d\dbcd\w/
1014 * /a\d\d\d\dbcd\w/
1015 * In the first case, the regex matches minlen chars; in the
1016 * second, minlen+1, in the third, minlen+2.
1017 * In the first case, the floating offset is 3 (which equals
1018 * float_min), in the second, 4, and in the third, 5 (which
1019 * equals float_max). In all cases, the floating string bcd
1020 * can never start more than 4 chars from the end of the
1021 * string, which equals minlen - float_min. As the substring
1022 * starts to match more than float_min from the start of the
1023 * regex, it makes the regex match more than minlen chars,
1024 * and the two cancel each other out. So we can always use
1025 * float_min - minlen, rather than float_max - minlen for the
1026 * latest position in the string.
1027 *
1028 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1029 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1030 */
1031
e7a14a9c 1032 assert(prog->minlen >= other->min_offset);
6c3343a6
DM		 1033 last1 = HOP3c(strend,
1034 other->min_offset - prog->minlen, strbeg);
1035
4d006249 1036 if (other_ix) {/* i.e. if (other-is-float) */
6c3343a6
DM		 1037 /* last is the latest point where the floating substr could
1038 * start, *given* any constraints from the earlier fixed
1039 * match. This constraint is that the floating string starts
1040 * <= float_max_offset chars from the regex origin (rx_origin).
1041 * If this value is less than last1, use it instead.
eb3831ce 1042 */
6c3343a6
DM		 1043 assert(rx_origin <= last1);
1044 last =
1045 /* this condition handles the offset==infinity case, and
1046 * is a short-cut otherwise. Although it's comparing a
1047 * byte offset to a char length, it does so in a safe way,
1048 * since 1 char always occupies 1 or more bytes,
1049 * so if a string range is (last1 - rx_origin) bytes,
1050 * it will be less than or equal to (last1 - rx_origin)
1051 * chars; meaning it errs towards doing the accurate HOP3
1052 * rather than just using last1 as a short-cut */
1053 (last1 - rx_origin) < other->max_offset
1054 ? last1
1055 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1056 }
1057 else {
1058 assert(strpos + start_shift <= check_at);
1059 last = HOP4c(check_at, other->min_offset - start_shift,
1060 strbeg, strend);
1061 }
ead917d0 1062
6c3343a6
DM		 1063 s = HOP3c(rx_origin, other->min_offset, strend);
1064 if (s < other_last) /* These positions already checked */
1065 s = other_last;
1066
1067 must = utf8_target ? other->utf8_substr : other->substr;
1068 assert(SvPOK(must));
1069 s = fbm_instr(
1070 (unsigned char*)s,
1071 (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0),
1072 must,
1073 multiline ? FBMrf_MULTILINE : 0
1074 );
1075 DEBUG_EXECUTE_r({
1076 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1077 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1078 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s",
1079 s ? "Found" : "Contradicts",
1080 other_ix ? "floating" : "anchored",
1081 quoted, RE_SV_TAIL(must));
1082 });
ead917d0 1083
ead917d0 1084
6c3343a6
DM		 1085 if (!s) {
1086 /* last1 is latest possible substr location. If we didn't
1087 * find it before there, we never will */
1088 if (last >= last1) {
1089 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1090 ", giving up...\n"));
1091 goto fail_finish;
ead917d0
DM		 1092 }
1093
6c3343a6
DM		 1094 /* try to find the check substr again at a later
1095 * position. Maybe next time we'll find the "other" substr
1096 * in range too */
1097 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1098 ", trying %s at offset %ld...\n",
1099 (other_ix ? "floating" : "anchored"),
197641b6 1100 (long)(HOP3c(check_at, 1, strend) - strpos)));
6c3343a6
DM		 1101
1102 other_last = HOP3c(last, 1, strend) /* highest failure */;
1103 rx_origin =
4d006249 1104 other_ix /* i.e. if other-is-float */
6c3343a6
DM		 1105 ? HOP3c(rx_origin, 1, strend)
1106 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1107 goto restart;
1108 }
1109 else {
1110 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
197641b6 1111 (long)(s - strpos)));
ead917d0 1112
4d006249 1113 if (other_ix) { /* if (other-is-float) */
6c3343a6
DM		 1114 /* other_last is set to s, not s+1, since its possible for
1115 * a floating substr to fail first time, then succeed
1116 * second time at the same floating position; e.g.:
1117 * "-AB--AABZ" =~ /\wAB\d*Z/
1118 * The first time round, anchored and float match at
1119 * "-(AB)--AAB(Z)" then fail on the initial \w character
1120 * class. Second time round, they match at "-AB--A(AB)(Z)".
1121 */
1122 other_last = s;
ead917d0
DM		 1123 }
1124 else {
6c3343a6
DM		 1125 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1126 other_last = HOP3c(s, 1, strend);
ead917d0 1127 }
6c3343a6 1128 }
cad2e5aa 1129 }
acba93e8
DM		 1130 else {
1131 DEBUG_OPTIMISE_MORE_r(
1132 PerlIO_printf(Perl_debug_log,
1133 " Check-only match: offset min:%"IVdf" max:%"IVdf
1c1c599d 1134 " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf
acba93e8
DM		 1135 " strend-strpos:%"IVdf"\n",
1136 (IV)prog->check_offset_min,
1137 (IV)prog->check_offset_max,
1c1c599d 1138 (IV)(check_at-strpos),
acba93e8 1139 (IV)(rx_origin-strpos),
1c1c599d 1140 (IV)(rx_origin-check_at),
acba93e8
DM		 1141 (IV)(strend-strpos)
1142 )
1143 );
1144 }
2c2d71f5 1145
acba93e8 1146 postprocess_substr_matches:
0991020e 1147
1a4edc3c 1148 /* handle the extra constraint of /^.../m if present */
e3c6feb0 1149
7d2d37f5 1150 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
4620cb61
DM		 1151 char *s;
1152
a62659bd
DM		 1153 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1154 " looking for /^/m anchor"));
d0880ea7
DM		 1155
1156 /* we have failed the constraint of a \n before rx_origin.
2e759faa
DM		 1157 * Find the next \n, if any, even if it's beyond the current
1158 * anchored and/or floating substrings. Whether we should be
1159 * scanning ahead for the next \n or the next substr is debatable.
1160 * On the one hand you'd expect rare substrings to appear less
1161 * often than \n's. On the other hand, searching for \n means
1162 * we're effectively flipping been check_substr and "\n" on each
1163 * iteration as the current "rarest" string candidate, which
1164 * means for example that we'll quickly reject the whole string if
1165 * hasn't got a \n, rather than trying every substr position
1166 * first
1167 */
d0880ea7 1168
4620cb61
DM		 1169 s = HOP3c(strend, - prog->minlen, strpos);
1170 if (s <= rx_origin ||
1171 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1172 {
d0880ea7
DM		 1173 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1174 " Did not find /%s^%s/m...\n",
1175 PL_colors[0], PL_colors[1]));
a62659bd
DM		 1176 goto fail_finish;
1177 }
d0880ea7 1178
4ada1233
DM		 1179 /* earliest possible origin is 1 char after the \n.
1180 * (since *rx_origin == '\n', it's safe to ++ here rather than
1181 * HOP(rx_origin, 1)) */
1182 rx_origin++;
d0880ea7 1183
f4f115de 1184 if (prog->substrs->check_ix == 0 /* check is anchored */
4ada1233 1185 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
f4f115de 1186 {
d0880ea7
DM		 1187 /* Position contradicts check-string; either because
1188 * check was anchored (and thus has no wiggle room),
4ada1233 1189 * or check was float and rx_origin is above the float range */
d0880ea7
DM		 1190 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1191 " Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
4ada1233 1192 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
d0880ea7
DM		 1193 goto restart;
1194 }
1195
1196 /* if we get here, the check substr must have been float,
2e759faa 1197 * is in range, and we may or may not have had an anchored
d0880ea7
DM		 1198 * "other" substr which still contradicts */
1199 assert(prog->substrs->check_ix); /* check is float */
1200
1201 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1202 /* whoops, the anchored "other" substr exists, so we still
1203 * contradict. On the other hand, the float "check" substr
1204 * didn't contradict, so just retry the anchored "other"
1205 * substr */
1206 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1207 " Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1208 PL_colors[0], PL_colors[1],
1209 (long)(rx_origin - strpos),
1210 (long)(rx_origin - strpos + prog->anchored_offset)));
1211 goto do_other_substr;
1212 }
1213
1214 /* success: we don't contradict the found floating substring
1215 * (and there's no anchored substr). */
d0880ea7
DM		 1216 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1217 " Found /%s^%s/m at offset %ld...\n",
1218 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
e3c6feb0
DM		 1219 }
1220 else {
2e759faa 1221 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
fe4f3442 1222 " (multiline anchor test skipped)\n"));
e3c6feb0
DM		 1223 }
1224
ffad1e6a 1225 success_at_start:
e3c6feb0 1226
cad2e5aa 1227
dd170ff5
DM		 1228 /* if we have a starting character class, then test that extra constraint.
1229 * (trie stclasses are too expensive to use here, we are better off to
1230 * leave it to regmatch itself) */
1231
f8fc2ecf 1232 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
f8fc2ecf 1233 const U8* const str = (U8*)STRING(progi->regstclass);
0991020e 1234
2c75e362 1235 /* XXX this value could be pre-computed */
f8fc2ecf 1236 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
2c75e362
DM		 1237 ? (reginfo->is_utf8_pat
1238 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1239 : STR_LEN(progi->regstclass))
66e933ab 1240 : 1);
1de06328 1241 char * endpos;
fa3bb21d 1242 char *s;
000dfd2d
DM		 1243 /* latest pos that a matching float substr constrains rx start to */
1244 char *rx_max_float = NULL;
1245
c75a3985
DM		 1246 /* if the current rx_origin is anchored, either by satisfying an
1247 * anchored substring constraint, or a /^.../m constraint, then we
1248 * can reject the current origin if the start class isn't found
1249 * at the current position. If we have a float-only match, then
1250 * rx_origin is constrained to a range; so look for the start class
1251 * in that range. if neither, then look for the start class in the
1252 * whole rest of the string */
1253
dd170ff5
DM		 1254 /* XXX DAPM it's not clear what the minlen test is for, and why
1255 * it's not used in the floating case. Nothing in the test suite
1256 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1257 * Here are some old comments, which may or may not be correct:
1258 *
1259 * minlen == 0 is possible if regstclass is \b or \B,
1260 * and the fixed substr is ''$.
1261 * Since minlen is already taken into account, rx_origin+1 is
1262 * before strend; accidentally, minlen >= 1 guaranties no false
1263 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1264 * 0) below assumes that regstclass does not come from lookahead...
1265 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1266 * This leaves EXACTF-ish only, which are dealt with in
1267 * find_byclass().
1268 */
1269
7d2d37f5 1270 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
fa3bb21d 1271 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
000dfd2d
DM		 1272 else if (prog->float_substr || prog->float_utf8) {
1273 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1274 endpos= HOP3c(rx_max_float, cl_l, strend);
1275 }
1de06328
YO		 1276 else
1277 endpos= strend;
1278
1dc475d0
DM		 1279 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1280 " looking for class: start_shift: %"IVdf" check_at: %"IVdf
c43b5520 1281 " rx_origin: %"IVdf" endpos: %"IVdf"\n",
1dc475d0 1282 (IV)start_shift, (IV)(check_at - strbeg),
c43b5520 1283 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
d8080198 1284
c43b5520 1285 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
f9176b44 1286 reginfo);
be778b1a 1287 if (!s) {
6eb5f6b9 1288 if (endpos == strend) {
a3621e74 1289 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1dc475d0 1290 " Could not match STCLASS...\n") );
6eb5f6b9
JH		 1291 goto fail;
1292 }
a3621e74 1293 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1dc475d0 1294 " This position contradicts STCLASS...\n") );
e0eb31e7
DM		 1295 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1296 && !(prog->intflags & PREGf_IMPLICIT))
653099ff 1297 goto fail;
9fed8d02 1298
6eb5f6b9 1299 /* Contradict one of substrings */
97136c8a
DM		 1300 if (prog->anchored_substr || prog->anchored_utf8) {
1301 if (prog->substrs->check_ix == 1) { /* check is float */
1302 /* Have both, check_string is floating */
1303 assert(rx_origin + start_shift <= check_at);
1304 if (rx_origin + start_shift != check_at) {
1305 /* not at latest position float substr could match:
c75a3985
DM		 1306 * Recheck anchored substring, but not floating.
1307 * The condition above is in bytes rather than
1308 * chars for efficiency. It's conservative, in
1309 * that it errs on the side of doing 'goto
1310 * do_other_substr', where a more accurate
1311 * char-based calculation will be done */
97136c8a
DM		 1312 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1313 " Looking for anchored substr starting at offset %ld...\n",
1314 (long)(other_last - strpos)) );
1315 goto do_other_substr;
3369914b 1316 }
3369914b
DM		 1317 }
1318 }
97136c8a 1319 else {
9fed8d02
DM		 1320 /* float-only */
1321
7d2d37f5 1322 if (ml_anch) {
c75a3985
DM		 1323 /* In the presence of ml_anch, we might be able to
1324 * find another \n without breaking the current float
1325 * constraint. */
1326
1327 /* strictly speaking this should be HOP3c(..., 1, ...),
1328 * but since we goto a block of code that's going to
1329 * search for the next \n if any, its safe here */
9fed8d02 1330 rx_origin++;
9fed8d02
DM		 1331 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1332 " Looking for /%s^%s/m starting at offset %ld...\n",
1333 PL_colors[0], PL_colors[1],
1334 (long)(rx_origin - strpos)) );
9fed8d02 1335 goto postprocess_substr_matches;
ab60c45a 1336 }
c75a3985
DM		 1337
1338 /* strictly speaking this can never be true; but might
1339 * be if we ever allow intuit without substrings */
1340 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
9fed8d02 1341 goto fail;
c75a3985 1342
000dfd2d 1343 rx_origin = rx_max_float;
9fed8d02
DM		 1344 }
1345
c75a3985
DM		 1346 /* at this point, any matching substrings have been
1347 * contradicted. Start again... */
1348
9fed8d02 1349 rx_origin = HOP3c(rx_origin, 1, strend);
557f47af
DM		 1350
1351 /* uses bytes rather than char calculations for efficiency.
1352 * It's conservative: it errs on the side of doing 'goto restart',
1353 * where there is code that does a proper char-based test */
9fed8d02 1354 if (rx_origin + start_shift + end_shift > strend) {
40268e5b 1355 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
9fed8d02
DM		 1356 " Could not match STCLASS...\n") );
1357 goto fail;
1358 }
1359 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1360 " Looking for %s substr starting at offset %ld...\n",
1361 (prog->substrs->check_ix ? "floating" : "anchored"),
1362 (long)(rx_origin + start_shift - strpos)) );
1363 goto restart;
6eb5f6b9 1364 }
9fed8d02 1365
c75a3985
DM		 1366 /* Success !!! */
1367
5f9c6575 1368 if (rx_origin != s) {
a3621e74 1369 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 1370 " By STCLASS: moving %ld --> %ld\n",
5f9c6575 1371 (long)(rx_origin - strpos), (long)(s - strpos))
b7953727
JH		 1372 );
1373 }
1374 else {
a3621e74 1375 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 1376 " Does not contradict STCLASS...\n");
b7953727
JH		 1377 );
1378 }
6eb5f6b9 1379 }
ffad1e6a
DM		 1380
1381 /* Decide whether using the substrings helped */
1382
1383 if (rx_origin != strpos) {
1384 /* Fixed substring is found far enough so that the match
1385 cannot start at strpos. */
1386
1387 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n"));
1388 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1389 }
1390 else {
70563e16
DM		 1391 /* The found rx_origin position does not prohibit matching at
1392 * strpos, so calling intuit didn't gain us anything. Decrement
1393 * the BmUSEFUL() count on the check substring, and if we reach
1394 * zero, free it. */
1395 if (!(prog->intflags & PREGf_NAUGHTY)
ffad1e6a
DM		 1396 && (utf8_target ? (
1397 prog->check_utf8 /* Could be deleted already */
1398 && --BmUSEFUL(prog->check_utf8) < 0
1399 && (prog->check_utf8 == prog->float_utf8)
1400 ) : (
1401 prog->check_substr /* Could be deleted already */
1402 && --BmUSEFUL(prog->check_substr) < 0
1403 && (prog->check_substr == prog->float_substr)
1404 )))
1405 {
1406 /* If flags & SOMETHING - do not do it many times on the same match */
1407 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n"));
1408 /* XXX Does the destruction order has to change with utf8_target? */
1409 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1410 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1411 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1412 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1413 check = NULL; /* abort */
ffad1e6a
DM		 1414 /* XXXX This is a remnant of the old implementation. It
1415 looks wasteful, since now INTUIT can use many
1416 other heuristics. */
1417 prog->extflags &= ~RXf_USE_INTUIT;
ffad1e6a
DM		 1418 }
1419 }
1420
1421 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1422 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1423 PL_colors[4], PL_colors[5], (long)(rx_origin - strpos)) );
1424
c765d6e0 1425 return rx_origin;
2c2d71f5
JH		 1426
1427 fail_finish: /* Substring not found */
33b8afdf 1428 if (prog->check_substr || prog->check_utf8) /* could be removed already */
f2ed9b32 1429 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
cad2e5aa 1430 fail:
a3621e74 1431 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
e4584336 1432 PL_colors[4], PL_colors[5]));
bd61b366 1433 return NULL;
cad2e5aa 1434}
9661b544 1435
70563e16 1436
a0a388a1 1437#define DECL_TRIE_TYPE(scan) \
e7fd4aa1 1438 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
a4525e78
KW		 1439 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold, \
1440 trie_utf8l, trie_flu8 } \
e7fd4aa1
KW		 1441 trie_type = ((scan->flags == EXACT) \
1442 ? (utf8_target ? trie_utf8 : trie_plain) \
a4525e78
KW		 1443 : (scan->flags == EXACTL) \
1444 ? (utf8_target ? trie_utf8l : trie_plain) \
1445 : (scan->flags == EXACTFA) \
1446 ? (utf8_target \
1447 ? trie_utf8_exactfa_fold \
1448 : trie_latin_utf8_exactfa_fold) \
1449 : (scan->flags == EXACTFLU8 \
1450 ? trie_flu8 \
1451 : (utf8_target \
1452 ? trie_utf8_fold \
1453 : trie_latin_utf8_fold)))
fab2782b 1454
fd3249ee 1455#define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
baa60164 1456STMT_START { \
fab2782b 1457 STRLEN skiplen; \
baa60164 1458 U8 flags = FOLD_FLAGS_FULL; \
fab2782b 1459 switch (trie_type) { \
a4525e78 1460 case trie_flu8: \
780fcc9f 1461 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
613abc6d
KW		 1462 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1463 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \
1464 } \
a4525e78 1465 goto do_trie_utf8_fold; \
31f05a37 1466 case trie_utf8_exactfa_fold: \
baa60164 1467 flags |= FOLD_FLAGS_NOMIX_ASCII; \
8e57b935 1468 /* FALLTHROUGH */ \
fab2782b 1469 case trie_utf8_fold: \
a4525e78 1470 do_trie_utf8_fold: \
fab2782b 1471 if ( foldlen>0 ) { \
c80e42f3 1472 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO		 1473 foldlen -= len; \
1474 uscan += len; \
1475 len=0; \
1476 } else { \
445bf929 1477 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
fab2782b
YO		 1478 len = UTF8SKIP(uc); \
1479 skiplen = UNISKIP( uvc ); \
1480 foldlen -= skiplen; \
1481 uscan = foldbuf + skiplen; \
1482 } \
1483 break; \
baa60164
KW		 1484 case trie_latin_utf8_exactfa_fold: \
1485 flags |= FOLD_FLAGS_NOMIX_ASCII; \
8e57b935 1486 /* FALLTHROUGH */ \
fab2782b
YO		 1487 case trie_latin_utf8_fold: \
1488 if ( foldlen>0 ) { \
c80e42f3 1489 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO		 1490 foldlen -= len; \
1491 uscan += len; \
1492 len=0; \
1493 } else { \
1494 len = 1; \
31f05a37 1495 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
fab2782b
YO		 1496 skiplen = UNISKIP( uvc ); \
1497 foldlen -= skiplen; \
1498 uscan = foldbuf + skiplen; \
1499 } \
1500 break; \
a4525e78 1501 case trie_utf8l: \
780fcc9f 1502 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
613abc6d
KW		 1503 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1504 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \
1505 } \
780fcc9f 1506 /* FALLTHROUGH */ \
fab2782b 1507 case trie_utf8: \
c80e42f3 1508 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO		 1509 break; \
1510 case trie_plain: \
1511 uvc = (UV)*uc; \
1512 len = 1; \
1513 } \
1514 if (uvc < 256) { \
1515 charid = trie->charmap[ uvc ]; \
1516 } \
1517 else { \
1518 charid = 0; \
1519 if (widecharmap) { \
1520 SV** const svpp = hv_fetch(widecharmap, \
1521 (char*)&uvc, sizeof(UV), 0); \
1522 if (svpp) \
1523 charid = (U16)SvIV(*svpp); \
1524 } \
1525 } \
4cadc6a9
YO		 1526} STMT_END
1527
ae7c5b9b
KW		 1528#define DUMP_EXEC_POS(li,s,doutf8) \
1529 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1530 startpos, doutf8)
1531
c84a03c5 1532#define REXEC_FBC_EXACTISH_SCAN(COND) \
4cadc6a9
YO		 1533STMT_START { \
1534 while (s <= e) { \
c84a03c5 1535 if ( (COND) \
fac1af77 1536 && (ln == 1 || folder(s, pat_string, ln)) \
02d5137b 1537 && (reginfo->intuit || regtry(reginfo, &s)) )\
4cadc6a9
YO		 1538 goto got_it; \
1539 s++; \
1540 } \
1541} STMT_END
1542
c84a03c5 1543#define REXEC_FBC_UTF8_SCAN(CODE) \
4cadc6a9 1544STMT_START { \
9a902117 1545 while (s < strend) { \
c84a03c5 1546 CODE \
9a902117 1547 s += UTF8SKIP(s); \
4cadc6a9
YO		 1548 } \
1549} STMT_END
1550
c84a03c5 1551#define REXEC_FBC_SCAN(CODE) \
4cadc6a9
YO		 1552STMT_START { \
1553 while (s < strend) { \
c84a03c5 1554 CODE \
4cadc6a9
YO		 1555 s++; \
1556 } \
1557} STMT_END
1558
05bd126c
KW		 1559#define REXEC_FBC_UTF8_CLASS_SCAN(COND) \
1560REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \
1561 if (COND) { \
1562 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1563 goto got_it; \
1564 else \
1565 tmp = doevery; \
1566 } \
1567 else \
1568 tmp = 1; \
4cadc6a9
YO		 1569)
1570
05bd126c
KW		 1571#define REXEC_FBC_CLASS_SCAN(COND) \
1572REXEC_FBC_SCAN( /* Loops while (s < strend) */ \
1573 if (COND) { \
1574 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1575 goto got_it; \
1576 else \
1577 tmp = doevery; \
1578 } \
1579 else \
1580 tmp = 1; \
4cadc6a9
YO		 1581)
1582
c84a03c5 1583#define REXEC_FBC_CSCAN(CONDUTF8,COND) \
baa60164 1584 if (utf8_target) { \
c84a03c5 1585 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \
e1d1eefb
YO		 1586 } \
1587 else { \
c84a03c5 1588 REXEC_FBC_CLASS_SCAN(COND); \
d981ef24 1589 }
05bd126c 1590
05bd126c
KW		 1591/* The three macros below are slightly different versions of the same logic.
1592 *
1593 * The first is for /a and /aa when the target string is UTF-8. This can only
1594 * match ascii, but it must advance based on UTF-8. The other two handle the
1595 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1596 * for the boundary (or non-boundary) between a word and non-word character.
1597 * The utf8 and non-utf8 cases have the same logic, but the details must be
1598 * different. Find the "wordness" of the character just prior to this one, and
1599 * compare it with the wordness of this one. If they differ, we have a
1600 * boundary. At the beginning of the string, pretend that the previous
1601 * character was a new-line.
1602 *
1603 * All these macros uncleanly have side-effects with each other and outside
1604 * variables. So far it's been too much trouble to clean-up
1605 *
1606 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1607 * a word character or not.
1608 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1609 * word/non-word
1610 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1611 *
1612 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1613 * are looking for a boundary or for a non-boundary. If we are looking for a
1614 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1615 * see if this tentative match actually works, and if so, to quit the loop
1616 * here. And vice-versa if we are looking for a non-boundary.
1617 *
1618 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1619 * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of
1620 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1621 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1622 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1623 * complement. But in that branch we complement tmp, meaning that at the
1624 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1625 * which means at the top of the loop in the next iteration, it is
1626 * TEST_NON_UTF8(s-1) */
b2f4e957 1627#define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
05bd126c
KW		 1628 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1629 tmp = TEST_NON_UTF8(tmp); \
1630 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1631 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1632 tmp = !tmp; \
1633 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1634 } \
1635 else { \
1636 IF_FAIL; \
1637 } \
1638 ); \
1639
1640/* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1641 * TEST_UTF8 is a macro that for the same input code points returns identically
1642 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
236d82fd 1643#define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
05bd126c
KW		 1644 if (s == reginfo->strbeg) { \
1645 tmp = '\n'; \
1646 } \
1647 else { /* Back-up to the start of the previous character */ \
1648 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1649 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
3db24e1e 1650 0, UTF8_ALLOW_DEFAULT); \
05bd126c
KW		 1651 } \
1652 tmp = TEST_UV(tmp); \
1653 LOAD_UTF8_CHARCLASS_ALNUM(); \
1654 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1655 if (tmp == ! (TEST_UTF8((U8 *) s))) { \
1656 tmp = !tmp; \
1657 IF_SUCCESS; \
1658 } \
1659 else { \
1660 IF_FAIL; \
1661 } \
1662 );
cfaf538b 1663
05bd126c
KW		 1664/* Like the above two macros. UTF8_CODE is the complete code for handling
1665 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1666 * macros below */
baa60164 1667#define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
63ac0dad 1668 if (utf8_target) { \
05bd126c 1669 UTF8_CODE \
63ac0dad
KW		 1670 } \
1671 else { /* Not utf8 */ \
9d9163fb 1672 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
63ac0dad 1673 tmp = TEST_NON_UTF8(tmp); \
05bd126c 1674 REXEC_FBC_SCAN( /* advances s while s < strend */ \
63ac0dad 1675 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
63ac0dad 1676 IF_SUCCESS; \
760cfa8e 1677 tmp = !tmp; \
63ac0dad
KW		 1678 } \
1679 else { \
1680 IF_FAIL; \
1681 } \
1682 ); \
1683 } \
c8519dc7
KW		 1684 /* Here, things have been set up by the previous code so that tmp is the \
1685 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1686 * utf8ness of the target). We also have to check if this matches against \
1687 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1688 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1689 * string */ \
1690 if (tmp == ! TEST_NON_UTF8('\n')) { \
1691 IF_SUCCESS; \
1692 } \
1693 else { \
1694 IF_FAIL; \
1695 }
63ac0dad 1696
ae7c5b9b
KW		 1697/* This is the macro to use when we want to see if something that looks like it
1698 * could match, actually does, and if so exits the loop */
1699#define REXEC_FBC_TRYIT \
1700 if ((reginfo->intuit || regtry(reginfo, &s))) \
1701 goto got_it
1702
1703/* The only difference between the BOUND and NBOUND cases is that
1704 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1705 * NBOUND. This is accomplished by passing it as either the if or else clause,
1706 * with the other one being empty (PLACEHOLDER is defined as empty).
1707 *
1708 * The TEST_FOO parameters are for operating on different forms of input, but
1709 * all should be ones that return identically for the same underlying code
1710 * points */
1711#define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1712 FBC_BOUND_COMMON( \
1713 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1714 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1715
44129e46 1716#define FBC_BOUND_A(TEST_NON_UTF8) \
ae7c5b9b
KW		 1717 FBC_BOUND_COMMON( \
1718 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1719 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1720
1721#define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1722 FBC_BOUND_COMMON( \
1723 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1724 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1725
44129e46 1726#define FBC_NBOUND_A(TEST_NON_UTF8) \
ae7c5b9b
KW		 1727 FBC_BOUND_COMMON( \
1728 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1729 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1730
05bd126c 1731
786e8c11 1732/* We know what class REx starts with. Try to find this position... */
02d5137b 1733/* if reginfo->intuit, its a dryrun */
786e8c11
YO		 1734/* annoyingly all the vars in this routine have different names from their counterparts
1735 in regmatch. /grrr */
3c3eec57 1736STATIC char *
07be1b83 1737S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
f9176b44 1738 const char *strend, regmatch_info *reginfo)
a687059c 1739{
73104a1b
KW		 1740 dVAR;
1741 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1742 char *pat_string; /* The pattern's exactish string */
1743 char *pat_end; /* ptr to end char of pat_string */
1744 re_fold_t folder; /* Function for computing non-utf8 folds */
1745 const U8 *fold_array; /* array for folding ords < 256 */
1746 STRLEN ln;
1747 STRLEN lnc;
73104a1b
KW		 1748 U8 c1;
1749 U8 c2;
1750 char *e;
1751 I32 tmp = 1; /* Scratch variable? */
ba44c216 1752 const bool utf8_target = reginfo->is_utf8_target;
73104a1b 1753 UV utf8_fold_flags = 0;
f9176b44 1754 const bool is_utf8_pat = reginfo->is_utf8_pat;
3018b823
KW		 1755 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1756 with a result inverts that result, as 0^1 =
1757 1 and 1^1 = 0 */
1758 _char_class_number classnum;
1759
73104a1b 1760 RXi_GET_DECL(prog,progi);
2f7f8cb1 1761
73104a1b 1762 PERL_ARGS_ASSERT_FIND_BYCLASS;
2f7f8cb1 1763
73104a1b
KW		 1764 /* We know what class it must start with. */
1765 switch (OP(c)) {
a4525e78 1766 case ANYOFL:
780fcc9f
KW		 1767 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1768 /* FALLTHROUGH */
73104a1b
KW		 1769 case ANYOF:
1770 if (utf8_target) {
1771 REXEC_FBC_UTF8_CLASS_SCAN(
3db24e1e 1772 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
73104a1b
KW		 1773 }
1774 else {
1775 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1776 }
1777 break;
1778 case CANY:
1779 REXEC_FBC_SCAN(
02d5137b 1780 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
73104a1b
KW		 1781 goto got_it;
1782 else
1783 tmp = doevery;
1784 );
1785 break;
1786
098b07d5
KW		 1787 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1788 assert(! is_utf8_pat);
924ba076 1789 /* FALLTHROUGH */
73104a1b 1790 case EXACTFA:
984e6dd1 1791 if (is_utf8_pat || utf8_target) {
73104a1b
KW		 1792 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1793 goto do_exactf_utf8;
1794 }
1795 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1796 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1797 goto do_exactf_non_utf8; /* isn't dealt with by these */
77a6d856 1798
2fdb7295
KW		 1799 case EXACTF: /* This node only generated for non-utf8 patterns */
1800 assert(! is_utf8_pat);
73104a1b 1801 if (utf8_target) {
73104a1b
KW		 1802 utf8_fold_flags = 0;
1803 goto do_exactf_utf8;
1804 }
1805 fold_array = PL_fold;
1806 folder = foldEQ;
1807 goto do_exactf_non_utf8;
1808
1809 case EXACTFL:
780fcc9f 1810 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
31f05a37 1811 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
cea315b6 1812 utf8_fold_flags = FOLDEQ_LOCALE;
73104a1b
KW		 1813 goto do_exactf_utf8;
1814 }
1815 fold_array = PL_fold_locale;
1816 folder = foldEQ_locale;
1817 goto do_exactf_non_utf8;
3c760661 1818
73104a1b 1819 case EXACTFU_SS:
984e6dd1 1820 if (is_utf8_pat) {
73104a1b
KW		 1821 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1822 }
1823 goto do_exactf_utf8;
16d951b7 1824
a4525e78
KW		 1825 case EXACTFLU8:
1826 if (! utf8_target) { /* All code points in this node require
1827 UTF-8 to express. */
1828 break;
1829 }
613abc6d
KW		 1830 utf8_fold_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
1831 | FOLDEQ_S2_FOLDS_SANE;
a4525e78
KW		 1832 goto do_exactf_utf8;
1833
73104a1b 1834 case EXACTFU:
984e6dd1
DM		 1835 if (is_utf8_pat || utf8_target) {
1836 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
73104a1b
KW		 1837 goto do_exactf_utf8;
1838 }
fac1af77 1839
73104a1b
KW		 1840 /* Any 'ss' in the pattern should have been replaced by regcomp,
1841 * so we don't have to worry here about this single special case
1842 * in the Latin1 range */
1843 fold_array = PL_fold_latin1;
1844 folder = foldEQ_latin1;
1845
924ba076 1846 /* FALLTHROUGH */
73104a1b 1847
c52b8b12 1848 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
73104a1b
KW		 1849 are no glitches with fold-length differences
1850 between the target string and pattern */
1851
1852 /* The idea in the non-utf8 EXACTF* cases is to first find the
1853 * first character of the EXACTF* node and then, if necessary,
1854 * case-insensitively compare the full text of the node. c1 is the
1855 * first character. c2 is its fold. This logic will not work for
1856 * Unicode semantics and the german sharp ss, which hence should
1857 * not be compiled into a node that gets here. */
1858 pat_string = STRING(c);
1859 ln = STR_LEN(c); /* length to match in octets/bytes */
1860
1861 /* We know that we have to match at least 'ln' bytes (which is the
1862 * same as characters, since not utf8). If we have to match 3
1863 * characters, and there are only 2 availabe, we know without
1864 * trying that it will fail; so don't start a match past the
1865 * required minimum number from the far end */
ea3daa5d 1866 e = HOP3c(strend, -((SSize_t)ln), s);
73104a1b 1867
02d5137b 1868 if (reginfo->intuit && e < s) {
73104a1b
KW		 1869 e = s; /* Due to minlen logic of intuit() */
1870 }
fac1af77 1871
73104a1b
KW		 1872 c1 = *pat_string;
1873 c2 = fold_array[c1];
1874 if (c1 == c2) { /* If char and fold are the same */
1875 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1876 }
1877 else {
1878 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1879 }
1880 break;
fac1af77 1881
c52b8b12
KW		 1882 do_exactf_utf8:
1883 {
73104a1b
KW		 1884 unsigned expansion;
1885
1886 /* If one of the operands is in utf8, we can't use the simpler folding
1887 * above, due to the fact that many different characters can have the
1888 * same fold, or portion of a fold, or different- length fold */
1889 pat_string = STRING(c);
1890 ln = STR_LEN(c); /* length to match in octets/bytes */
1891 pat_end = pat_string + ln;
984e6dd1 1892 lnc = is_utf8_pat /* length to match in characters */
73104a1b
KW		 1893 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1894 : ln;
1895
1896 /* We have 'lnc' characters to match in the pattern, but because of
1897 * multi-character folding, each character in the target can match
1898 * up to 3 characters (Unicode guarantees it will never exceed
1899 * this) if it is utf8-encoded; and up to 2 if not (based on the
1900 * fact that the Latin 1 folds are already determined, and the
1901 * only multi-char fold in that range is the sharp-s folding to
1902 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1903 * string character. Adjust lnc accordingly, rounding up, so that
1904 * if we need to match at least 4+1/3 chars, that really is 5. */
1905 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1906 lnc = (lnc + expansion - 1) / expansion;
1907
1908 /* As in the non-UTF8 case, if we have to match 3 characters, and
1909 * only 2 are left, it's guaranteed to fail, so don't start a
1910 * match that would require us to go beyond the end of the string
1911 */
ea3daa5d 1912 e = HOP3c(strend, -((SSize_t)lnc), s);
73104a1b 1913
02d5137b 1914 if (reginfo->intuit && e < s) {
73104a1b
KW		 1915 e = s; /* Due to minlen logic of intuit() */
1916 }
0658cdde 1917
73104a1b
KW		 1918 /* XXX Note that we could recalculate e to stop the loop earlier,
1919 * as the worst case expansion above will rarely be met, and as we
1920 * go along we would usually find that e moves further to the left.
1921 * This would happen only after we reached the point in the loop
1922 * where if there were no expansion we should fail. Unclear if
1923 * worth the expense */
1924
1925 while (s <= e) {
1926 char *my_strend= (char *)strend;
1927 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
984e6dd1 1928 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
02d5137b 1929 && (reginfo->intuit || regtry(reginfo, &s)) )
73104a1b
KW		 1930 {
1931 goto got_it;
1932 }
1933 s += (utf8_target) ? UTF8SKIP(s) : 1;
1934 }
1935 break;
1936 }
236d82fd 1937
73104a1b 1938 case BOUNDL:
780fcc9f 1939 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
236d82fd 1940 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
73104a1b
KW		 1941 break;
1942 case NBOUNDL:
780fcc9f 1943 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
236d82fd 1944 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
73104a1b
KW		 1945 break;
1946 case BOUND:
236d82fd 1947 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
73104a1b
KW		 1948 break;
1949 case BOUNDA:
44129e46 1950 FBC_BOUND_A(isWORDCHAR_A);
73104a1b
KW		 1951 break;
1952 case NBOUND:
236d82fd 1953 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
73104a1b
KW		 1954 break;
1955 case NBOUNDA:
44129e46 1956 FBC_NBOUND_A(isWORDCHAR_A);
73104a1b
KW		 1957 break;
1958 case BOUNDU:
236d82fd 1959 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
73104a1b
KW		 1960 break;
1961 case NBOUNDU:
236d82fd 1962 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
73104a1b 1963 break;
73104a1b
KW		 1964 case LNBREAK:
1965 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1966 is_LNBREAK_latin1_safe(s, strend)
1967 );
1968 break;
3018b823
KW		 1969
1970 /* The argument to all the POSIX node types is the class number to pass to
1971 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1972
1973 case NPOSIXL:
1974 to_complement = 1;
1975 /* FALLTHROUGH */
1976
1977 case POSIXL:
780fcc9f 1978 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3018b823
KW		 1979 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1980 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
73104a1b 1981 break;
3018b823
KW		 1982
1983 case NPOSIXD:
1984 to_complement = 1;
1985 /* FALLTHROUGH */
1986
1987 case POSIXD:
1988 if (utf8_target) {
1989 goto posix_utf8;
1990 }
1991 goto posixa;
1992
1993 case NPOSIXA:
1994 if (utf8_target) {
1995 /* The complement of something that matches only ASCII matches all
837226c8
KW		 1996 * non-ASCII, plus everything in ASCII that isn't in the class. */
1997 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
3018b823
KW		 1998 || ! _generic_isCC_A(*s, FLAGS(c)));
1999 break;
2000 }
2001
2002 to_complement = 1;
2003 /* FALLTHROUGH */
2004
73104a1b 2005 case POSIXA:
3018b823 2006 posixa:
73104a1b 2007 /* Don't need to worry about utf8, as it can match only a single
3018b823
KW		 2008 * byte invariant character. */
2009 REXEC_FBC_CLASS_SCAN(
2010 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
73104a1b 2011 break;
3018b823
KW		 2012
2013 case NPOSIXU:
2014 to_complement = 1;
2015 /* FALLTHROUGH */
2016
2017 case POSIXU:
2018 if (! utf8_target) {
2019 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
2020 FLAGS(c))));
2021 }
2022 else {
2023
c52b8b12 2024 posix_utf8:
3018b823
KW		 2025 classnum = (_char_class_number) FLAGS(c);
2026 if (classnum < _FIRST_NON_SWASH_CC) {
2027 while (s < strend) {
2028
2029 /* We avoid loading in the swash as long as possible, but
2030 * should we have to, we jump to a separate loop. This
2031 * extra 'if' statement is what keeps this code from being
2032 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
2033 if (UTF8_IS_ABOVE_LATIN1(*s)) {
2034 goto found_above_latin1;
2035 }
2036 if ((UTF8_IS_INVARIANT(*s)
2037 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
2038 classnum)))
2039 || (UTF8_IS_DOWNGRADEABLE_START(*s)
2040 && to_complement ^ cBOOL(
94bb8c36
KW		 2041 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
2042 *(s + 1)),
3018b823
KW		 2043 classnum))))
2044 {
02d5137b 2045 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
3018b823
KW		 2046 goto got_it;
2047 else {
2048 tmp = doevery;
2049 }
2050 }
2051 else {
2052 tmp = 1;
2053 }
2054 s += UTF8SKIP(s);
2055 }
2056 }
2057 else switch (classnum) { /* These classes are implemented as
2058 macros */
2059 case _CC_ENUM_SPACE: /* XXX would require separate code if we
2060 revert the change of \v matching this */
924ba076 2061 /* FALLTHROUGH */
3018b823
KW		 2062
2063 case _CC_ENUM_PSXSPC:
2064 REXEC_FBC_UTF8_CLASS_SCAN(
2065 to_complement ^ cBOOL(isSPACE_utf8(s)));
2066 break;
2067
2068 case _CC_ENUM_BLANK:
2069 REXEC_FBC_UTF8_CLASS_SCAN(
2070 to_complement ^ cBOOL(isBLANK_utf8(s)));
2071 break;
2072
2073 case _CC_ENUM_XDIGIT:
2074 REXEC_FBC_UTF8_CLASS_SCAN(
2075 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
2076 break;
2077
2078 case _CC_ENUM_VERTSPACE:
2079 REXEC_FBC_UTF8_CLASS_SCAN(
2080 to_complement ^ cBOOL(isVERTWS_utf8(s)));
2081 break;
2082
2083 case _CC_ENUM_CNTRL:
2084 REXEC_FBC_UTF8_CLASS_SCAN(
2085 to_complement ^ cBOOL(isCNTRL_utf8(s)));
2086 break;
2087
2088 default:
2089 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
e5964223 2090 NOT_REACHED; /* NOTREACHED */
3018b823
KW		 2091 }
2092 }
2093 break;
2094
2095 found_above_latin1: /* Here we have to load a swash to get the result
2096 for the current code point */
2097 if (! PL_utf8_swash_ptrs[classnum]) {
2098 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2099 PL_utf8_swash_ptrs[classnum] =
2a16ac92
KW		 2100 _core_swash_init("utf8",
2101 "",
2102 &PL_sv_undef, 1, 0,
2103 PL_XPosix_ptrs[classnum], &flags);
3018b823
KW		 2104 }
2105
2106 /* This is a copy of the loop above for swash classes, though using the
2107 * FBC macro instead of being expanded out. Since we've loaded the
2108 * swash, we don't have to check for that each time through the loop */
2109 REXEC_FBC_UTF8_CLASS_SCAN(
2110 to_complement ^ cBOOL(_generic_utf8(
2111 classnum,
2112 s,
2113 swash_fetch(PL_utf8_swash_ptrs[classnum],
2114 (U8 *) s, TRUE))));
73104a1b
KW		 2115 break;
2116
2117 case AHOCORASICKC:
2118 case AHOCORASICK:
2119 {
2120 DECL_TRIE_TYPE(c);
2121 /* what trie are we using right now */
2122 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2123 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2124 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2125
2126 const char *last_start = strend - trie->minlen;
6148ee25 2127#ifdef DEBUGGING
73104a1b 2128 const char *real_start = s;
6148ee25 2129#endif
73104a1b
KW		 2130 STRLEN maxlen = trie->maxlen;
2131 SV *sv_points;
2132 U8 **points; /* map of where we were in the input string
2133 when reading a given char. For ASCII this
2134 is unnecessary overhead as the relationship
2135 is always 1:1, but for Unicode, especially
2136 case folded Unicode this is not true. */
2137 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2138 U8 *bitmap=NULL;
2139
2140
2141 GET_RE_DEBUG_FLAGS_DECL;
2142
2143 /* We can't just allocate points here. We need to wrap it in
2144 * an SV so it gets freed properly if there is a croak while
2145 * running the match */
2146 ENTER;
2147 SAVETMPS;
2148 sv_points=newSV(maxlen * sizeof(U8 *));
2149 SvCUR_set(sv_points,
2150 maxlen * sizeof(U8 *));
2151 SvPOK_on(sv_points);
2152 sv_2mortal(sv_points);
2153 points=(U8**)SvPV_nolen(sv_points );
2154 if ( trie_type != trie_utf8_fold
2155 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2156 {
2157 if (trie->bitmap)
2158 bitmap=(U8*)trie->bitmap;
2159 else
2160 bitmap=(U8*)ANYOF_BITMAP(c);
2161 }
2162 /* this is the Aho-Corasick algorithm modified a touch
2163 to include special handling for long "unknown char" sequences.
2164 The basic idea being that we use AC as long as we are dealing
2165 with a possible matching char, when we encounter an unknown char
2166 (and we have not encountered an accepting state) we scan forward
2167 until we find a legal starting char.
2168 AC matching is basically that of trie matching, except that when
2169 we encounter a failing transition, we fall back to the current
2170 states "fail state", and try the current char again, a process
2171 we repeat until we reach the root state, state 1, or a legal
2172 transition. If we fail on the root state then we can either
2173 terminate if we have reached an accepting state previously, or
2174 restart the entire process from the beginning if we have not.
2175
2176 */
2177 while (s <= last_start) {
2178 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2179 U8 *uc = (U8*)s;
2180 U16 charid = 0;
2181 U32 base = 1;
2182 U32 state = 1;
2183 UV uvc = 0;
2184 STRLEN len = 0;
2185 STRLEN foldlen = 0;
2186 U8 *uscan = (U8*)NULL;
2187 U8 *leftmost = NULL;
2188#ifdef DEBUGGING
2189 U32 accepted_word= 0;
786e8c11 2190#endif
73104a1b
KW		 2191 U32 pointpos = 0;
2192
2193 while ( state && uc <= (U8*)strend ) {
2194 int failed=0;
2195 U32 word = aho->states[ state ].wordnum;
2196
2197 if( state==1 ) {
2198 if ( bitmap ) {
2199 DEBUG_TRIE_EXECUTE_r(
2200 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2201 dump_exec_pos( (char *)uc, c, strend, real_start,
2202 (char *)uc, utf8_target );
2203 PerlIO_printf( Perl_debug_log,
2204 " Scanning for legal start char...\n");
2205 }
2206 );
2207 if (utf8_target) {
2208 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2209 uc += UTF8SKIP(uc);
2210 }
2211 } else {
2212 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2213 uc++;
2214 }
786e8c11 2215 }
73104a1b 2216 s= (char *)uc;
07be1b83 2217 }
73104a1b
KW		 2218 if (uc >(U8*)last_start) break;
2219 }
2220
2221 if ( word ) {
2222 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2223 if (!leftmost || lpos < leftmost) {
2224 DEBUG_r(accepted_word=word);
2225 leftmost= lpos;
7016d6eb 2226 }
73104a1b 2227 if (base==0) break;
7016d6eb 2228
73104a1b
KW		 2229 }
2230 points[pointpos++ % maxlen]= uc;
2231 if (foldlen || uc < (U8*)strend) {
2232 REXEC_TRIE_READ_CHAR(trie_type, trie,
2233 widecharmap, uc,
2234 uscan, len, uvc, charid, foldlen,
2235 foldbuf, uniflags);
2236 DEBUG_TRIE_EXECUTE_r({
2237 dump_exec_pos( (char *)uc, c, strend,
2238 real_start, s, utf8_target);
2239 PerlIO_printf(Perl_debug_log,
2240 " Charid:%3u CP:%4"UVxf" ",
2241 charid, uvc);
2242 });
2243 }
2244 else {
2245 len = 0;
2246 charid = 0;
2247 }
07be1b83 2248
73104a1b
KW		 2249
2250 do {
6148ee25 2251#ifdef DEBUGGING
73104a1b 2252 word = aho->states[ state ].wordnum;
6148ee25 2253#endif
73104a1b
KW		 2254 base = aho->states[ state ].trans.base;
2255
2256 DEBUG_TRIE_EXECUTE_r({
2257 if (failed)
2258 dump_exec_pos( (char *)uc, c, strend, real_start,
2259 s, utf8_target );
2260 PerlIO_printf( Perl_debug_log,
2261 "%sState: %4"UVxf", word=%"UVxf,
2262 failed ? " Fail transition to " : "",
2263 (UV)state, (UV)word);
2264 });
2265 if ( base ) {
2266 U32 tmp;
2267 I32 offset;
2268 if (charid &&
2269 ( ((offset = base + charid
2270 - 1 - trie->uniquecharcount)) >= 0)
2271 && ((U32)offset < trie->lasttrans)
2272 && trie->trans[offset].check == state
2273 && (tmp=trie->trans[offset].next))
2274 {
2275 DEBUG_TRIE_EXECUTE_r(
2276 PerlIO_printf( Perl_debug_log," - legal\n"));
2277 state = tmp;
2278 break;
07be1b83
YO		 2279 }
2280 else {
786e8c11 2281 DEBUG_TRIE_EXECUTE_r(
73104a1b 2282 PerlIO_printf( Perl_debug_log," - fail\n"));
786e8c11 2283 failed = 1;
73104a1b 2284 state = aho->fail[state];
07be1b83 2285 }
07be1b83 2286 }
73104a1b
KW		 2287 else {
2288 /* we must be accepting here */
2289 DEBUG_TRIE_EXECUTE_r(
2290 PerlIO_printf( Perl_debug_log," - accepting\n"));
2291 failed = 1;
2292 break;
786e8c11 2293 }
73104a1b
KW		 2294 } while(state);
2295 uc += len;
2296 if (failed) {
2297 if (leftmost)
2298 break;
2299 if (!state) state = 1;
07be1b83 2300 }
73104a1b
KW		 2301 }
2302 if ( aho->states[ state ].wordnum ) {
2303 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2304 if (!leftmost || lpos < leftmost) {
2305 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2306 leftmost = lpos;
07be1b83
YO		 2307 }
2308 }
73104a1b
KW		 2309 if (leftmost) {
2310 s = (char*)leftmost;
2311 DEBUG_TRIE_EXECUTE_r({
2312 PerlIO_printf(
2313 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2314 (UV)accepted_word, (IV)(s - real_start)
2315 );
2316 });
02d5137b 2317 if (reginfo->intuit || regtry(reginfo, &s)) {
73104a1b
KW		 2318 FREETMPS;
2319 LEAVE;
2320 goto got_it;
2321 }
2322 s = HOPc(s,1);
2323 DEBUG_TRIE_EXECUTE_r({
2324 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2325 });
2326 } else {
2327 DEBUG_TRIE_EXECUTE_r(
2328 PerlIO_printf( Perl_debug_log,"No match.\n"));
2329 break;
2330 }
2331 }
2332 FREETMPS;
2333 LEAVE;
2334 }
2335 break;
2336 default:
2337 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
73104a1b
KW		 2338 }
2339 return 0;
2340 got_it:
2341 return s;
6eb5f6b9
JH		 2342}
2343
60165aa4
DM		 2344/* set RX_SAVED_COPY, RX_SUBBEG etc.
2345 * flags have same meanings as with regexec_flags() */
2346
749f4950
DM		 2347static void
2348S_reg_set_capture_string(pTHX_ REGEXP * const rx,
60165aa4
DM		 2349 char *strbeg,
2350 char *strend,
2351 SV *sv,
2352 U32 flags,
2353 bool utf8_target)
2354{
2355 struct regexp *const prog = ReANY(rx);
2356
60165aa4
DM		 2357 if (flags & REXEC_COPY_STR) {
2358#ifdef PERL_ANY_COW
2359 if (SvCANCOW(sv)) {
2360 if (DEBUG_C_TEST) {
2361 PerlIO_printf(Perl_debug_log,
2362 "Copy on write: regexp capture, type %d\n",
2363 (int) SvTYPE(sv));
2364 }
5411a0e5
DM		 2365 /* Create a new COW SV to share the match string and store
2366 * in saved_copy, unless the current COW SV in saved_copy
2367 * is valid and suitable for our purpose */
2368 if (( prog->saved_copy
2369 && SvIsCOW(prog->saved_copy)
2370 && SvPOKp(prog->saved_copy)
2371 && SvIsCOW(sv)
2372 && SvPOKp(sv)
2373 && SvPVX(sv) == SvPVX(prog->saved_copy)))
a76b0e90 2374 {
5411a0e5
DM		 2375 /* just reuse saved_copy SV */
2376 if (RXp_MATCH_COPIED(prog)) {
2377 Safefree(prog->subbeg);
2378 RXp_MATCH_COPIED_off(prog);
2379 }
2380 }
2381 else {
2382 /* create new COW SV to share string */
a76b0e90
DM		 2383 RX_MATCH_COPY_FREE(rx);
2384 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
a76b0e90 2385 }
5411a0e5
DM		 2386 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2387 assert (SvPOKp(prog->saved_copy));
60165aa4
DM		 2388 prog->sublen = strend - strbeg;
2389 prog->suboffset = 0;
2390 prog->subcoffset = 0;
2391 } else
2392#endif
2393 {
99a90e59
FC		 2394 SSize_t min = 0;
2395 SSize_t max = strend - strbeg;
ea3daa5d 2396 SSize_t sublen;
60165aa4
DM		 2397
2398 if ( (flags & REXEC_COPY_SKIP_POST)
e322109a 2399 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM		 2400 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2401 ) { /* don't copy $' part of string */
2402 U32 n = 0;
2403 max = -1;
2404 /* calculate the right-most part of the string covered
2405 * by a capture. Due to look-ahead, this may be to
2406 * the right of $&, so we have to scan all captures */
2407 while (n <= prog->lastparen) {
2408 if (prog->offs[n].end > max)
2409 max = prog->offs[n].end;
2410 n++;
2411 }
2412 if (max == -1)
2413 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2414 ? prog->offs[0].start
2415 : 0;
2416 assert(max >= 0 && max <= strend - strbeg);
2417 }
2418
2419 if ( (flags & REXEC_COPY_SKIP_PRE)
e322109a 2420 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM		 2421 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2422 ) { /* don't copy $` part of string */
2423 U32 n = 0;
2424 min = max;
2425 /* calculate the left-most part of the string covered
2426 * by a capture. Due to look-behind, this may be to
2427 * the left of $&, so we have to scan all captures */
2428 while (min && n <= prog->lastparen) {
2429 if ( prog->offs[n].start != -1
2430 && prog->offs[n].start < min)
2431 {
2432 min = prog->offs[n].start;
2433 }
2434 n++;
2435 }
2436 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2437 && min > prog->offs[0].end
2438 )
2439 min = prog->offs[0].end;
2440
2441 }
2442
2443 assert(min >= 0 && min <= max && min <= strend - strbeg);
2444 sublen = max - min;
2445
2446 if (RX_MATCH_COPIED(rx)) {
2447 if (sublen > prog->sublen)
2448 prog->subbeg =
2449 (char*)saferealloc(prog->subbeg, sublen+1);
2450 }
2451 else
2452 prog->subbeg = (char*)safemalloc(sublen+1);
2453 Copy(strbeg + min, prog->subbeg, sublen, char);
2454 prog->subbeg[sublen] = '\0';
2455 prog->suboffset = min;
2456 prog->sublen = sublen;
2457 RX_MATCH_COPIED_on(rx);
2458 }
2459 prog->subcoffset = prog->suboffset;
2460 if (prog->suboffset && utf8_target) {
2461 /* Convert byte offset to chars.
2462 * XXX ideally should only compute this if @-/@+
2463 * has been seen, a la PL_sawampersand ??? */
2464
2465 /* If there's a direct correspondence between the
2466 * string which we're matching and the original SV,
2467 * then we can use the utf8 len cache associated with
2468 * the SV. In particular, it means that under //g,
2469 * sv_pos_b2u() will use the previously cached
2470 * position to speed up working out the new length of
2471 * subcoffset, rather than counting from the start of
2472 * the string each time. This stops
2473 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2474 * from going quadratic */
2475 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
ea3daa5d
FC		 2476 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2477 SV_GMAGIC|SV_CONST_RETURN);
60165aa4
DM		 2478 else
2479 prog->subcoffset = utf8_length((U8*)strbeg,
2480 (U8*)(strbeg+prog->suboffset));
2481 }
2482 }
2483 else {
2484 RX_MATCH_COPY_FREE(rx);
2485 prog->subbeg = strbeg;
2486 prog->suboffset = 0;
2487 prog->subcoffset = 0;
2488 prog->sublen = strend - strbeg;
2489 }
2490}
2491
2492
2493
fae667d5 2494
6eb5f6b9
JH		 2495/*
2496 - regexec_flags - match a regexp against a string
2497 */
2498I32
5aaab254 2499Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
ea3daa5d 2500 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
8fd1a950
DM		 2501/* stringarg: the point in the string at which to begin matching */
2502/* strend: pointer to null at end of string */
2503/* strbeg: real beginning of string */
2504/* minend: end of match must be >= minend bytes after stringarg. */
2505/* sv: SV being matched: only used for utf8 flag, pos() etc; string
2506 * itself is accessed via the pointers above */
2507/* data: May be used for some additional optimizations.
d058ec57 2508 Currently unused. */
a340edde 2509/* flags: For optimizations. See REXEC_* in regexp.h */
8fd1a950 2510
6eb5f6b9 2511{
8d919b0a 2512 struct regexp *const prog = ReANY(rx);
5aaab254 2513 char *s;
eb578fdb 2514 regnode *c;
03c83e26 2515 char *startpos;
ea3daa5d
FC		 2516 SSize_t minlen; /* must match at least this many chars */
2517 SSize_t dontbother = 0; /* how many characters not to try at end */
f2ed9b32 2518 const bool utf8_target = cBOOL(DO_UTF8(sv));
2757e526 2519 I32 multiline;
f8fc2ecf 2520 RXi_GET_DECL(prog,progi);
02d5137b
DM		 2521 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2522 regmatch_info *const reginfo = &reginfo_buf;
e9105d30 2523 regexp_paren_pair *swap = NULL;
006f26b2 2524 I32 oldsave;
a3621e74
YO		 2525 GET_RE_DEBUG_FLAGS_DECL;
2526
7918f24d 2527 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
9d4ba2ae 2528 PERL_UNUSED_ARG(data);
6eb5f6b9
JH		 2529
2530 /* Be paranoid... */
03c83e26 2531 if (prog == NULL || stringarg == NULL) {
6eb5f6b9 2532 Perl_croak(aTHX_ "NULL regexp parameter");
6eb5f6b9
JH		 2533 }
2534
6c3fea77 2535 DEBUG_EXECUTE_r(
03c83e26 2536 debug_start_match(rx, utf8_target, stringarg, strend,
6c3fea77
DM		 2537 "Matching");
2538 );
8adc0f72 2539
b342a604
DM		 2540 startpos = stringarg;
2541
58430ea8 2542 if (prog->intflags & PREGf_GPOS_SEEN) {
d307c076
DM		 2543 MAGIC *mg;
2544
fef7148b
DM		 2545 /* set reginfo->ganch, the position where \G can match */
2546
2547 reginfo->ganch =
2548 (flags & REXEC_IGNOREPOS)
2549 ? stringarg /* use start pos rather than pos() */
2550 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
25fdce4a
FC		 2551 /* Defined pos(): */
2552 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
fef7148b
DM		 2553 : strbeg; /* pos() not defined; use start of string */
2554
2555 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
7b0eb0b8 2556 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg)));
fef7148b 2557
03c83e26
DM		 2558 /* in the presence of \G, we may need to start looking earlier in
2559 * the string than the suggested start point of stringarg:
0b2c2a84 2560 * if prog->gofs is set, then that's a known, fixed minimum
03c83e26
DM		 2561 * offset, such as
2562 * /..\G/: gofs = 2
2563 * /ab|c\G/: gofs = 1
2564 * or if the minimum offset isn't known, then we have to go back
2565 * to the start of the string, e.g. /w+\G/
2566 */
2bfbe302 2567
8e1490ee 2568 if (prog->intflags & PREGf_ANCH_GPOS) {
2bfbe302
DM		 2569 startpos = reginfo->ganch - prog->gofs;
2570 if (startpos <
2571 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2572 {
2573 DEBUG_r(PerlIO_printf(Perl_debug_log,
2574 "fail: ganch-gofs before earliest possible start\n"));
2575 return 0;
2576 }
2577 }
2578 else if (prog->gofs) {
b342a604
DM		 2579 if (startpos - prog->gofs < strbeg)
2580 startpos = strbeg;
2581 else
2582 startpos -= prog->gofs;
03c83e26 2583 }
58430ea8 2584 else if (prog->intflags & PREGf_GPOS_FLOAT)
b342a604 2585 startpos = strbeg;
03c83e26
DM		 2586 }
2587
2588 minlen = prog->minlen;
b342a604 2589 if ((startpos + minlen) > strend || startpos < strbeg) {
03c83e26
DM		 2590 DEBUG_r(PerlIO_printf(Perl_debug_log,
2591 "Regex match can't succeed, so not even tried\n"));
2592 return 0;
2593 }
2594
63a3746a
DM		 2595 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2596 * which will call destuctors to reset PL_regmatch_state, free higher
2597 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2598 * regmatch_info_aux_eval */
2599
2600 oldsave = PL_savestack_ix;
2601
dfa77d06
DM		 2602 s = startpos;
2603
e322109a 2604 if ((prog->extflags & RXf_USE_INTUIT)
7fadf4a7
DM		 2605 && !(flags & REXEC_CHECKED))
2606 {
dfa77d06 2607 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
7fadf4a7 2608 flags, NULL);
dfa77d06 2609 if (!s)
7fadf4a7
DM		 2610 return 0;
2611
e322109a 2612 if (prog->extflags & RXf_CHECK_ALL) {
7fadf4a7
DM		 2613 /* we can match based purely on the result of INTUIT.
2614 * Set up captures etc just for $& and $-[0]
2615 * (an intuit-only match wont have $1,$2,..) */
2616 assert(!prog->nparens);
d5e7783a
DM		 2617
2618 /* s/// doesn't like it if $& is earlier than where we asked it to
2619 * start searching (which can happen on something like /.\G/) */
2620 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2621 && (s < stringarg))
2622 {
2623 /* this should only be possible under \G */
58430ea8 2624 assert(prog->intflags & PREGf_GPOS_SEEN);
d5e7783a
DM		 2625 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2626 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2627 goto phooey;
2628 }
2629
7fadf4a7
DM		 2630 /* match via INTUIT shouldn't have any captures.
2631 * Let @-, @+, $^N know */
2632 prog->lastparen = prog->lastcloseparen = 0;
2633 RX_MATCH_UTF8_set(rx, utf8_target);
3ff69bd6
DM		 2634 prog->offs[0].start = s - strbeg;
2635 prog->offs[0].end = utf8_target
2636 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2637 : s - strbeg + prog->minlenret;
7fadf4a7 2638 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 2639 S_reg_set_capture_string(aTHX_ rx,
7fadf4a7
DM		 2640 strbeg, strend,
2641 sv, flags, utf8_target);
2642
7fadf4a7
DM		 2643 return 1;
2644 }
2645 }
2646
6c3fea77 2647 multiline = prog->extflags & RXf_PMf_MULTILINE;
1de06328 2648
dfa77d06 2649 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
a3621e74 2650 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
a72c7584
JH		 2651 "String too short [regexec_flags]...\n"));
2652 goto phooey;
1aa99e6b 2653 }
1de06328 2654
6eb5f6b9 2655 /* Check validity of program. */
f8fc2ecf 2656 if (UCHARAT(progi->program) != REG_MAGIC) {
6eb5f6b9
JH		 2657 Perl_croak(aTHX_ "corrupted regexp program");
2658 }
2659
1738e041 2660 RX_MATCH_TAINTED_off(rx);
ab4e48c1 2661 RX_MATCH_UTF8_set(rx, utf8_target);
1738e041 2662
6c3fea77
DM		 2663 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2664 reginfo->intuit = 0;
2665 reginfo->is_utf8_target = cBOOL(utf8_target);
02d5137b
DM		 2666 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2667 reginfo->warned = FALSE;
9d9163fb 2668 reginfo->strbeg = strbeg;
02d5137b 2669 reginfo->sv = sv;
1cb48e53 2670 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
220db18a 2671 reginfo->strend = strend;
6eb5f6b9 2672 /* see how far we have to get to not match where we matched before */
fe3974be 2673 reginfo->till = stringarg + minend;
6eb5f6b9 2674
60779a30 2675 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
82c23608
FC		 2676 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2677 S_cleanup_regmatch_info_aux has executed (registered by
2678 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2679 magic belonging to this SV.
2680 Not newSVsv, either, as it does not COW.
2681 */
2682 reginfo->sv = newSV(0);
4cba5ac0 2683 SvSetSV_nosteal(reginfo->sv, sv);
82c23608
FC		 2684 SAVEFREESV(reginfo->sv);
2685 }
2686
331b2dcc
DM		 2687 /* reserve next 2 or 3 slots in PL_regmatch_state:
2688 * slot N+0: may currently be in use: skip it
2689 * slot N+1: use for regmatch_info_aux struct
2690 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2691 * slot N+3: ready for use by regmatch()
2692 */
bf2039a9 2693
331b2dcc
DM		 2694 {
2695 regmatch_state *old_regmatch_state;
2696 regmatch_slab *old_regmatch_slab;
2697 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2698
2699 /* on first ever match, allocate first slab */
2700 if (!PL_regmatch_slab) {
2701 Newx(PL_regmatch_slab, 1, regmatch_slab);
2702 PL_regmatch_slab->prev = NULL;
2703 PL_regmatch_slab->next = NULL;
2704 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2705 }
bf2039a9 2706
331b2dcc
DM		 2707 old_regmatch_state = PL_regmatch_state;
2708 old_regmatch_slab = PL_regmatch_slab;
bf2039a9 2709
331b2dcc
DM		 2710 for (i=0; i <= max; i++) {
2711 if (i == 1)
2712 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2713 else if (i ==2)
2714 reginfo->info_aux_eval =
2715 reginfo->info_aux->info_aux_eval =
2716 &(PL_regmatch_state->u.info_aux_eval);
bf2039a9 2717
331b2dcc
DM		 2718 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2719 PL_regmatch_state = S_push_slab(aTHX);
2720 }
bf2039a9 2721
331b2dcc
DM		 2722 /* note initial PL_regmatch_state position; at end of match we'll
2723 * pop back to there and free any higher slabs */
bf2039a9 2724
331b2dcc
DM		 2725 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2726 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2ac8ff4b 2727 reginfo->info_aux->poscache = NULL;
bf2039a9 2728
331b2dcc 2729 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
bf2039a9 2730
331b2dcc
DM		 2731 if ((prog->extflags & RXf_EVAL_SEEN))
2732 S_setup_eval_state(aTHX_ reginfo);
2733 else
2734 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
bf2039a9 2735 }
d3aa529c 2736
6eb5f6b9 2737 /* If there is a "must appear" string, look for it. */
6eb5f6b9 2738
288b8c02 2739 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
e9105d30
GG		 2740 /* We have to be careful. If the previous successful match
2741 was from this regex we don't want a subsequent partially
2742 successful match to clobber the old results.
2743 So when we detect this possibility we add a swap buffer
d8da0584
KW		 2744 to the re, and switch the buffer each match. If we fail,
2745 we switch it back; otherwise we leave it swapped.
e9105d30
GG		 2746 */
2747 swap = prog->offs;
2748 /* do we need a save destructor here for eval dies? */
2749 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
495f47a5
DM		 2750 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2751 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2752 PTR2UV(prog),
2753 PTR2UV(swap),
2754 PTR2UV(prog->offs)
2755 ));
c74340f9 2756 }
6eb5f6b9 2757
0fa70a06
DM		 2758 /* Simplest case: anchored match need be tried only once, or with
2759 * MBOL, only at the beginning of each line.
2760 *
2761 * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets
2762 * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't
2763 * match at the start of the string then it won't match anywhere else
2764 * either; while with /.*.../, if it doesn't match at the beginning,
2765 * the earliest it could match is at the start of the next line */
2766
8e1490ee 2767 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
0fa70a06
DM		 2768 char *end;
2769
2770 if (regtry(reginfo, &s))
6eb5f6b9 2771 goto got_it;
0fa70a06
DM		 2772
2773 if (!(prog->intflags & PREGf_ANCH_MBOL))
2774 goto phooey;
2775
2776 /* didn't match at start, try at other newline positions */
2777
2778 if (minlen)
2779 dontbother = minlen - 1;
2780 end = HOP3c(strend, -dontbother, strbeg) - 1;
2781
2782 /* skip to next newline */
2783
2784 while (s <= end) { /* note it could be possible to match at the end of the string */
2785 /* NB: newlines are the same in unicode as they are in latin */
2786 if (*s++ != '\n')
2787 continue;
2788 if (prog->check_substr || prog->check_utf8) {
2789 /* note that with PREGf_IMPLICIT, intuit can only fail
2790 * or return the start position, so it's of limited utility.
2791 * Nevertheless, I made the decision that the potential for
2792 * quick fail was still worth it - DAPM */
2793 s = re_intuit_start(rx, sv, strbeg, s, strend, flags, NULL);
2794 if (!s)
2795 goto phooey;
2796 }
2797 if (regtry(reginfo, &s))
2798 goto got_it;
2799 }
2800 goto phooey;
2801 } /* end anchored search */
2802
2803 if (prog->intflags & PREGf_ANCH_GPOS)
f9f4320a 2804 {
a8430a8b
YO		 2805 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
2806 assert(prog->intflags & PREGf_GPOS_SEEN);
2bfbe302
DM		 2807 /* For anchored \G, the only position it can match from is
2808 * (ganch-gofs); we already set startpos to this above; if intuit
2809 * moved us on from there, we can't possibly succeed */
2810 assert(startpos == reginfo->ganch - prog->gofs);
2811 if (s == startpos && regtry(reginfo, &s))
6eb5f6b9
JH		 2812 goto got_it;
2813 goto phooey;
2814 }
2815
2816 /* Messy cases: unanchored match. */
bbe252da 2817 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
6eb5f6b9 2818 /* we have /x+whatever/ */
984e6dd1 2819 /* it must be a one character string (XXXX Except is_utf8_pat?) */
33b8afdf 2820 char ch;
bf93d4cc
GS		 2821#ifdef DEBUGGING
2822 int did_match = 0;
2823#endif
f2ed9b32 2824 if (utf8_target) {
7e0d5ad7
KW		 2825 if (! prog->anchored_utf8) {
2826 to_utf8_substr(prog);
2827 }
2828 ch = SvPVX_const(prog->anchored_utf8)[0];
4cadc6a9 2829 REXEC_FBC_SCAN(
6eb5f6b9 2830 if (*s == ch) {
a3621e74 2831 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 2832 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH		 2833 s += UTF8SKIP(s);
2834 while (s < strend && *s == ch)
2835 s += UTF8SKIP(s);
2836 }
4cadc6a9 2837 );
7e0d5ad7 2838
6eb5f6b9
JH		 2839 }
2840 else {
7e0d5ad7
KW		 2841 if (! prog->anchored_substr) {
2842 if (! to_byte_substr(prog)) {
6b54ddc5 2843 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW		 2844 }
2845 }
2846 ch = SvPVX_const(prog->anchored_substr)[0];
4cadc6a9 2847 REXEC_FBC_SCAN(
6eb5f6b9 2848 if (*s == ch) {
a3621e74 2849 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 2850 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH		 2851 s++;
2852 while (s < strend && *s == ch)
2853 s++;
2854 }
4cadc6a9 2855 );
6eb5f6b9 2856 }
a3621e74 2857 DEBUG_EXECUTE_r(if (!did_match)
bf93d4cc 2858 PerlIO_printf(Perl_debug_log,
b7953727
JH		 2859 "Did not find anchored character...\n")
2860 );
6eb5f6b9 2861 }
a0714e2c
SS		 2862 else if (prog->anchored_substr != NULL
2863 || prog->anchored_utf8 != NULL
2864 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
33b8afdf
JH		 2865 && prog->float_max_offset < strend - s)) {
2866 SV *must;
ea3daa5d
FC		 2867 SSize_t back_max;
2868 SSize_t back_min;
33b8afdf 2869 char *last;
6eb5f6b9 2870 char *last1; /* Last position checked before */
bf93d4cc
GS		 2871#ifdef DEBUGGING
2872 int did_match = 0;
2873#endif
33b8afdf 2874 if (prog->anchored_substr || prog->anchored_utf8) {
7e0d5ad7
KW		 2875 if (utf8_target) {
2876 if (! prog->anchored_utf8) {
2877 to_utf8_substr(prog);
2878 }
2879 must = prog->anchored_utf8;
2880 }
2881 else {
2882 if (! prog->anchored_substr) {
2883 if (! to_byte_substr(prog)) {
6b54ddc5 2884 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW		 2885 }
2886 }
2887 must = prog->anchored_substr;
2888 }
33b8afdf
JH		 2889 back_max = back_min = prog->anchored_offset;
2890 } else {
7e0d5ad7
KW		 2891 if (utf8_target) {
2892 if (! prog->float_utf8) {
2893 to_utf8_substr(prog);
2894 }
2895 must = prog->float_utf8;
2896 }
2897 else {
2898 if (! prog->float_substr) {
2899 if (! to_byte_substr(prog)) {
6b54ddc5 2900 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW		 2901 }
2902 }
2903 must = prog->float_substr;
2904 }
33b8afdf
JH		 2905 back_max = prog->float_max_offset;
2906 back_min = prog->float_min_offset;
2907 }
1de06328 2908
1de06328
YO		 2909 if (back_min<0) {
2910 last = strend;
2911 } else {
2912 last = HOP3c(strend, /* Cannot start after this */
ea3daa5d 2913 -(SSize_t)(CHR_SVLEN(must)
1de06328
YO		 2914 - (SvTAIL(must) != 0) + back_min), strbeg);
2915 }
9d9163fb 2916 if (s > reginfo->strbeg)
6eb5f6b9
JH		 2917 last1 = HOPc(s, -1);
2918 else
2919 last1 = s - 1; /* bogus */
2920
a0288114 2921 /* XXXX check_substr already used to find "s", can optimize if
6eb5f6b9 2922 check_substr==must. */
bf05793b 2923 dontbother = 0;
6eb5f6b9
JH		 2924 strend = HOPc(strend, -dontbother);
2925 while ( (s <= last) &&
e50d57d4 2926 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
9041c2e3 2927 (unsigned char*)strend, must,
c33e64f0 2928 multiline ? FBMrf_MULTILINE : 0)) ) {
a3621e74 2929 DEBUG_EXECUTE_r( did_match = 1 );
6eb5f6b9
JH		 2930 if (HOPc(s, -back_max) > last1) {
2931 last1 = HOPc(s, -back_min);
2932 s = HOPc(s, -back_max);
2933 }
2934 else {
9d9163fb
DM		 2935 char * const t = (last1 >= reginfo->strbeg)
2936 ? HOPc(last1, 1) : last1 + 1;
6eb5f6b9
JH		 2937
2938 last1 = HOPc(s, -back_min);
52657f30 2939 s = t;
6eb5f6b9 2940 }
f2ed9b32 2941 if (utf8_target) {
6eb5f6b9 2942 while (s <= last1) {
02d5137b 2943 if (regtry(reginfo, &s))
6eb5f6b9 2944 goto got_it;
7016d6eb
DM		 2945 if (s >= last1) {
2946 s++; /* to break out of outer loop */
2947 break;
2948 }
2949 s += UTF8SKIP(s);
6eb5f6b9
JH		 2950 }
2951 }
2952 else {
2953 while (s <= last1) {
02d5137b 2954 if (regtry(reginfo, &s))
6eb5f6b9
JH		 2955 goto got_it;
2956 s++;
2957 }
2958 }
2959 }
ab3bbdeb 2960 DEBUG_EXECUTE_r(if (!did_match) {
f2ed9b32 2961 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
ab3bbdeb
YO		 2962 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2963 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
33b8afdf 2964 ((must == prog->anchored_substr || must == prog->anchored_utf8)
bf93d4cc 2965 ? "anchored" : "floating"),
ab3bbdeb
YO		 2966 quoted, RE_SV_TAIL(must));
2967 });
6eb5f6b9
JH		 2968 goto phooey;
2969 }
f8fc2ecf 2970 else if ( (c = progi->regstclass) ) {
f14c76ed 2971 if (minlen) {
f8fc2ecf 2972 const OPCODE op = OP(progi->regstclass);
66e933ab 2973 /* don't bother with what can't match */
786e8c11 2974 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
f14c76ed
RGS		 2975 strend = HOPc(strend, -(minlen - 1));
2976 }
a3621e74 2977 DEBUG_EXECUTE_r({
be8e71aa 2978 SV * const prop = sv_newmortal();
8b9781c9 2979 regprop(prog, prop, c, reginfo, NULL);
0df25f3d 2980 {
f2ed9b32 2981 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb 2982 s,strend-s,60);
0df25f3d 2983 PerlIO_printf(Perl_debug_log,
1c8f8eb1 2984 "Matching stclass %.*s against %s (%d bytes)\n",
e4f74956 2985 (int)SvCUR(prop), SvPVX_const(prop),
ab3bbdeb 2986 quoted, (int)(strend - s));
0df25f3d 2987 }
ffc61ed2 2988 });
f9176b44 2989 if (find_byclass(prog, c, s, strend, reginfo))
6eb5f6b9 2990 goto got_it;
07be1b83 2991 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
d6a28714
JH		 2992 }
2993 else {
2994 dontbother = 0;
a0714e2c 2995 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
33b8afdf 2996 /* Trim the end. */
6af40bd7 2997 char *last= NULL;
33b8afdf 2998 SV* float_real;
c33e64f0
FC		 2999 STRLEN len;
3000 const char *little;
33b8afdf 3001
7e0d5ad7
KW		 3002 if (utf8_target) {
3003 if (! prog->float_utf8) {
3004 to_utf8_substr(prog);
3005 }
3006 float_real = prog->float_utf8;
3007 }
3008 else {
3009 if (! prog->float_substr) {
3010 if (! to_byte_substr(prog)) {
6b54ddc5 3011 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW		 3012 }
3013 }
3014 float_real = prog->float_substr;
3015 }
d6a28714 3016
c33e64f0
FC		 3017 little = SvPV_const(float_real, len);
3018 if (SvTAIL(float_real)) {
7f18ad16
KW		 3019 /* This means that float_real contains an artificial \n on
3020 * the end due to the presence of something like this:
3021 * /foo$/ where we can match both "foo" and "foo\n" at the
3022 * end of the string. So we have to compare the end of the
3023 * string first against the float_real without the \n and
3024 * then against the full float_real with the string. We
3025 * have to watch out for cases where the string might be
3026 * smaller than the float_real or the float_real without
3027 * the \n. */
1a13b075
YO		 3028 char *checkpos= strend - len;
3029 DEBUG_OPTIMISE_r(
3030 PerlIO_printf(Perl_debug_log,
3031 "%sChecking for float_real.%s\n",
3032 PL_colors[4], PL_colors[5]));
3033 if (checkpos + 1 < strbeg) {
7f18ad16
KW		 3034 /* can't match, even if we remove the trailing \n
3035 * string is too short to match */
1a13b075
YO		 3036 DEBUG_EXECUTE_r(
3037 PerlIO_printf(Perl_debug_log,
3038 "%sString shorter than required trailing substring, cannot match.%s\n",
3039 PL_colors[4], PL_colors[5]));
3040 goto phooey;
3041 } else if (memEQ(checkpos + 1, little, len - 1)) {
7f18ad16
KW		 3042 /* can match, the end of the string matches without the
3043 * "\n" */
1a13b075
YO		 3044 last = checkpos + 1;
3045 } else if (checkpos < strbeg) {
7f18ad16
KW		 3046 /* cant match, string is too short when the "\n" is
3047 * included */
1a13b075
YO		 3048 DEBUG_EXECUTE_r(
3049 PerlIO_printf(Perl_debug_log,
3050 "%sString does not contain required trailing substring, cannot match.%s\n",
3051 PL_colors[4], PL_colors[5]));
3052 goto phooey;
3053 } else if (!multiline) {
7f18ad16
KW		 3054 /* non multiline match, so compare with the "\n" at the
3055 * end of the string */
1a13b075
YO		 3056 if (memEQ(checkpos, little, len)) {
3057 last= checkpos;
3058 } else {
3059 DEBUG_EXECUTE_r(
3060 PerlIO_printf(Perl_debug_log,
3061 "%sString does not contain required trailing substring, cannot match.%s\n",
3062 PL_colors[4], PL_colors[5]));
3063 goto phooey;
3064 }
3065 } else {
7f18ad16
KW		 3066 /* multiline match, so we have to search for a place
3067 * where the full string is located */
d6a28714 3068 goto find_last;
1a13b075 3069 }
c33e64f0 3070 } else {
d6a28714 3071 find_last:
9041c2e3 3072 if (len)
d6a28714 3073 last = rninstr(s, strend, little, little + len);
b8c5462f 3074 else
a0288114 3075 last = strend; /* matching "$" */
b8c5462f 3076 }
6af40bd7 3077 if (!last) {
7f18ad16
KW		 3078 /* at one point this block contained a comment which was
3079 * probably incorrect, which said that this was a "should not
3080 * happen" case. Even if it was true when it was written I am
3081 * pretty sure it is not anymore, so I have removed the comment
3082 * and replaced it with this one. Yves */
6bda09f9
YO		 3083 DEBUG_EXECUTE_r(
3084 PerlIO_printf(Perl_debug_log,
b729e729
YO		 3085 "%sString does not contain required substring, cannot match.%s\n",
3086 PL_colors[4], PL_colors[5]
6af40bd7
YO		 3087 ));
3088 goto phooey;
bf93d4cc 3089 }
d6a28714
JH		 3090 dontbother = strend - last + prog->float_min_offset;
3091 }
3092 if (minlen && (dontbother < minlen))
3093 dontbother = minlen - 1;
3094 strend -= dontbother; /* this one's always in bytes! */
3095 /* We don't know much -- general case. */
f2ed9b32 3096 if (utf8_target) {
d6a28714 3097 for (;;) {
02d5137b 3098 if (regtry(reginfo, &s))
d6a28714
JH		 3099 goto got_it;
3100 if (s >= strend)
3101 break;
b8c5462f 3102 s += UTF8SKIP(s);
d6a28714
JH		 3103 };
3104 }
3105 else {
3106 do {
02d5137b 3107 if (regtry(reginfo, &s))
d6a28714
JH		 3108 goto got_it;
3109 } while (s++ < strend);
3110 }
3111 }
3112
3113 /* Failure. */
3114 goto phooey;
3115
7b52d656 3116 got_it:
d5e7783a
DM		 3117 /* s/// doesn't like it if $& is earlier than where we asked it to
3118 * start searching (which can happen on something like /.\G/) */
3119 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3120 && (prog->offs[0].start < stringarg - strbeg))
3121 {
3122 /* this should only be possible under \G */
58430ea8 3123 assert(prog->intflags & PREGf_GPOS_SEEN);
d5e7783a
DM		 3124 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3125 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3126 goto phooey;
3127 }
3128
495f47a5
DM		 3129 DEBUG_BUFFERS_r(
3130 if (swap)
3131 PerlIO_printf(Perl_debug_log,
3132 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
3133 PTR2UV(prog),
3134 PTR2UV(swap)
3135 );
3136 );
e9105d30 3137 Safefree(swap);
d6a28714 3138
bf2039a9
DM		 3139 /* clean up; this will trigger destructors that will free all slabs
3140 * above the current one, and cleanup the regmatch_info_aux
3141 * and regmatch_info_aux_eval sructs */
8adc0f72 3142
006f26b2
DM		 3143 LEAVE_SCOPE(oldsave);
3144
5daac39c
NC		 3145 if (RXp_PAREN_NAMES(prog))
3146 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
d6a28714
JH		 3147
3148 /* make sure $`, $&, $', and $digit will work later */
60165aa4 3149 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 3150 S_reg_set_capture_string(aTHX_ rx,
60165aa4
DM		 3151 strbeg, reginfo->strend,
3152 sv, flags, utf8_target);
9041c2e3 3153
d6a28714
JH		 3154 return 1;
3155
7b52d656 3156 phooey:
a3621e74 3157 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
e4584336 3158 PL_colors[4], PL_colors[5]));
8adc0f72 3159
bf2039a9
DM		 3160 /* clean up; this will trigger destructors that will free all slabs
3161 * above the current one, and cleanup the regmatch_info_aux
3162 * and regmatch_info_aux_eval sructs */
8adc0f72 3163
006f26b2
DM		 3164 LEAVE_SCOPE(oldsave);
3165
e9105d30 3166 if (swap) {
c74340f9 3167 /* we failed :-( roll it back */
495f47a5
DM		 3168 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
3169 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
3170 PTR2UV(prog),
3171 PTR2UV(prog->offs),
3172 PTR2UV(swap)
3173 ));
e9105d30
GG		 3174 Safefree(prog->offs);
3175 prog->offs = swap;
3176 }
d6a28714
JH		 3177 return 0;
3178}
3179
6bda09f9 3180
b3d298be 3181/* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
ec43f78b 3182 * Do inc before dec, in case old and new rex are the same */
baa60164 3183#define SET_reg_curpm(Re2) \
bf2039a9 3184 if (reginfo->info_aux_eval) { \
ec43f78b
DM		 3185 (void)ReREFCNT_inc(Re2); \
3186 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3187 PM_SETRE((PL_reg_curpm), (Re2)); \
3188 }
3189
3190
d6a28714
JH		 3191/*
3192 - regtry - try match at specific point
3193 */
3194STATIC I32 /* 0 failure, 1 success */
f73aaa43 3195S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
d6a28714 3196{
d6a28714 3197 CHECKPOINT lastcp;
288b8c02 3198 REGEXP *const rx = reginfo->prog;
8d919b0a 3199 regexp *const prog = ReANY(rx);
99a90e59 3200 SSize_t result;
f8fc2ecf 3201 RXi_GET_DECL(prog,progi);
a3621e74 3202 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC		 3203
3204 PERL_ARGS_ASSERT_REGTRY;
3205
24b23f37 3206 reginfo->cutpoint=NULL;
d6a28714 3207
9d9163fb 3208 prog->offs[0].start = *startposp - reginfo->strbeg;
d6a28714 3209 prog->lastparen = 0;
03994de8 3210 prog->lastcloseparen = 0;
d6a28714
JH		 3211
3212 /* XXXX What this code is doing here?!!! There should be no need
b93070ed 3213 to do this again and again, prog->lastparen should take care of
3dd2943c 3214 this! --ilya*/
dafc8851
JH		 3215
3216 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3217 * Actually, the code in regcppop() (which Ilya may be meaning by
b93070ed 3218 * prog->lastparen), is not needed at all by the test suite
225593e1
DM		 3219 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3220 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3221 * Meanwhile, this code *is* needed for the
daf18116
JH		 3222 * above-mentioned test suite tests to succeed. The common theme
3223 * on those tests seems to be returning null fields from matches.
225593e1 3224 * --jhi updated by dapm */
dafc8851 3225#if 1
d6a28714 3226 if (prog->nparens) {
b93070ed 3227 regexp_paren_pair *pp = prog->offs;
eb578fdb 3228 I32 i;
b93070ed 3229 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
f0ab9afb
NC		 3230 ++pp;
3231 pp->start = -1;
3232 pp->end = -1;
d6a28714
JH		 3233 }
3234 }
dafc8851 3235#endif
02db2b7b 3236 REGCP_SET(lastcp);
f73aaa43
DM		 3237 result = regmatch(reginfo, *startposp, progi->program + 1);
3238 if (result != -1) {
3239 prog->offs[0].end = result;
d6a28714
JH		 3240 return 1;
3241 }
24b23f37 3242 if (reginfo->cutpoint)
f73aaa43 3243 *startposp= reginfo->cutpoint;
02db2b7b 3244 REGCP_UNWIND(lastcp);
d6a28714
JH		 3245 return 0;
3246}
3247
02db2b7b 3248
8ba1375e
MJD		 3249#define sayYES goto yes
3250#define sayNO goto no
262b90c4 3251#define sayNO_SILENT goto no_silent
8ba1375e 3252
f9f4320a
YO		 3253/* we dont use STMT_START/END here because it leads to
3254 "unreachable code" warnings, which are bogus, but distracting. */
3255#define CACHEsayNO \
c476f425 3256 if (ST.cache_mask) \
2ac8ff4b 3257 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
f9f4320a 3258 sayNO
3298f257 3259
a3621e74 3260/* this is used to determine how far from the left messages like
265c4333
YO		 3261 'failed...' are printed. It should be set such that messages
3262 are inline with the regop output that created them.
a3621e74 3263*/
265c4333 3264#define REPORT_CODE_OFF 32
a3621e74
YO		 3265
3266
40a82448
DM		 3267#define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3268#define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
79a2a0e8
KW		 3269#define CHRTEST_NOT_A_CP_1 -999
3270#define CHRTEST_NOT_A_CP_2 -998
9e137952 3271
5d9a96ca
DM		 3272/* grab a new slab and return the first slot in it */
3273
3274STATIC regmatch_state *
3275S_push_slab(pTHX)
3276{
a35a87e7 3277#if PERL_VERSION < 9 && !defined(PERL_CORE)
54df2634
NC		 3278 dMY_CXT;
3279#endif
5d9a96ca
DM		 3280 regmatch_slab *s = PL_regmatch_slab->next;
3281 if (!s) {
3282 Newx(s, 1, regmatch_slab);
3283 s->prev = PL_regmatch_slab;
3284 s->next = NULL;
3285 PL_regmatch_slab->next = s;
3286 }
3287 PL_regmatch_slab = s;
86545054 3288 return SLAB_FIRST(s);
5d9a96ca 3289}
5b47454d 3290
95b24440 3291
40a82448
DM		 3292/* push a new state then goto it */
3293
4d5016e5
DM		 3294#define PUSH_STATE_GOTO(state, node, input) \
3295 pushinput = input; \
40a82448
DM		 3296 scan = node; \
3297 st->resume_state = state; \
3298 goto push_state;
3299
3300/* push a new state with success backtracking, then goto it */
3301
4d5016e5
DM		 3302#define PUSH_YES_STATE_GOTO(state, node, input) \
3303 pushinput = input; \
40a82448
DM		 3304 scan = node; \
3305 st->resume_state = state; \
3306 goto push_yes_state;
3307
aa283a38 3308
aa283a38 3309
4d5016e5 3310
d6a28714 3311/*
95b24440 3312
bf1f174e
DM		 3313regmatch() - main matching routine
3314
3315This is basically one big switch statement in a loop. We execute an op,
3316set 'next' to point the next op, and continue. If we come to a point which
3317we may need to backtrack to on failure such as (A|B|C), we push a
3318backtrack state onto the backtrack stack. On failure, we pop the top
3319state, and re-enter the loop at the state indicated. If there are no more
3320states to pop, we return failure.
3321
3322Sometimes we also need to backtrack on success; for example /A+/, where
3323after successfully matching one A, we need to go back and try to
3324match another one; similarly for lookahead assertions: if the assertion
3325completes successfully, we backtrack to the state just before the assertion
3326and then carry on. In these cases, the pushed state is marked as
3327'backtrack on success too'. This marking is in fact done by a chain of
3328pointers, each pointing to the previous 'yes' state. On success, we pop to
3329the nearest yes state, discarding any intermediate failure-only states.
3330Sometimes a yes state is pushed just to force some cleanup code to be
3331called at the end of a successful match or submatch; e.g. (??{$re}) uses
3332it to free the inner regex.
3333
3334Note that failure backtracking rewinds the cursor position, while
3335success backtracking leaves it alone.
3336
3337A pattern is complete when the END op is executed, while a subpattern
3338such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3339ops trigger the "pop to last yes state if any, otherwise return true"
3340behaviour.
3341
3342A common convention in this function is to use A and B to refer to the two
3343subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3344the subpattern to be matched possibly multiple times, while B is the entire
3345rest of the pattern. Variable and state names reflect this convention.
3346
3347The states in the main switch are the union of ops and failure/success of
3348substates associated with with that op. For example, IFMATCH is the op
3349that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3350'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3351successfully matched A and IFMATCH_A_fail is a state saying that we have
3352just failed to match A. Resume states always come in pairs. The backtrack
3353state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3354at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3355on success or failure.
3356
3357The struct that holds a backtracking state is actually a big union, with
3358one variant for each major type of op. The variable st points to the
3359top-most backtrack struct. To make the code clearer, within each
3360block of code we #define ST to alias the relevant union.
3361
3362Here's a concrete example of a (vastly oversimplified) IFMATCH
3363implementation:
3364
3365 switch (state) {
3366 ....
3367
3368#define ST st->u.ifmatch
3369
3370 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3371 ST.foo = ...; // some state we wish to save
95b24440 3372 ...
bf1f174e
DM		 3373 // push a yes backtrack state with a resume value of
3374 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3375 // first node of A:
4d5016e5 3376 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
bf1f174e
DM		 3377 // NOTREACHED
3378
3379 case IFMATCH_A: // we have successfully executed A; now continue with B
3380 next = B;
3381 bar = ST.foo; // do something with the preserved value
3382 break;
3383
3384 case IFMATCH_A_fail: // A failed, so the assertion failed
3385 ...; // do some housekeeping, then ...
3386 sayNO; // propagate the failure
3387
3388#undef ST
95b24440 3389
bf1f174e
DM		 3390 ...
3391 }
95b24440 3392
bf1f174e
DM		 3393For any old-timers reading this who are familiar with the old recursive
3394approach, the code above is equivalent to:
95b24440 3395
bf1f174e
DM		 3396 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3397 {
3398 int foo = ...
95b24440 3399 ...
bf1f174e
DM		 3400 if (regmatch(A)) {
3401 next = B;
3402 bar = foo;
3403 break;
95b24440 3404 }
bf1f174e
DM		 3405 ...; // do some housekeeping, then ...
3406 sayNO; // propagate the failure
95b24440 3407 }
bf1f174e
DM		 3408
3409The topmost backtrack state, pointed to by st, is usually free. If you
3410want to claim it, populate any ST.foo fields in it with values you wish to
3411save, then do one of
3412
4d5016e5
DM		 3413 PUSH_STATE_GOTO(resume_state, node, newinput);
3414 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
bf1f174e
DM		 3415
3416which sets that backtrack state's resume value to 'resume_state', pushes a
3417new free entry to the top of the backtrack stack, then goes to 'node'.
3418On backtracking, the free slot is popped, and the saved state becomes the
3419new free state. An ST.foo field in this new top state can be temporarily
3420accessed to retrieve values, but once the main loop is re-entered, it
3421becomes available for reuse.
3422
3423Note that the depth of the backtrack stack constantly increases during the
3424left-to-right execution of the pattern, rather than going up and down with
3425the pattern nesting. For example the stack is at its maximum at Z at the
3426end of the pattern, rather than at X in the following:
3427
3428 /(((X)+)+)+....(Y)+....Z/
3429
3430The only exceptions to this are lookahead/behind assertions and the cut,
3431(?>A), which pop all the backtrack states associated with A before
3432continuing.
3433
486ec47a 3434Backtrack state structs are allocated in slabs of about 4K in size.
bf1f174e
DM		 3435PL_regmatch_state and st always point to the currently active state,
3436and PL_regmatch_slab points to the slab currently containing
3437PL_regmatch_state. The first time regmatch() is called, the first slab is
3438allocated, and is never freed until interpreter destruction. When the slab
3439is full, a new one is allocated and chained to the end. At exit from
3440regmatch(), slabs allocated since entry are freed.
3441
3442*/
95b24440 3443
40a82448 3444
5bc10b2c 3445#define DEBUG_STATE_pp(pp) \
265c4333 3446 DEBUG_STATE_r({ \
baa60164 3447 DUMP_EXEC_POS(locinput, scan, utf8_target); \
5bc10b2c 3448 PerlIO_printf(Perl_debug_log, \
5d458dd8 3449 " %*s"pp" %s%s%s%s%s\n", \
5bc10b2c 3450 depth*2, "", \
baa60164 3451 PL_reg_name[st->resume_state], \
5d458dd8
YO		 3452 ((st==yes_state||st==mark_state) ? "[" : ""), \
3453 ((st==yes_state) ? "Y" : ""), \
3454 ((st==mark_state) ? "M" : ""), \
3455 ((st==yes_state||st==mark_state) ? "]" : "") \
3456 ); \
265c4333 3457 });
5bc10b2c 3458
40a82448 3459
3dab1dad 3460#define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
95b24440 3461
3df15adc 3462#ifdef DEBUGGING
5bc10b2c 3463
ab3bbdeb 3464STATIC void
f2ed9b32 3465S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
ab3bbdeb
YO		 3466 const char *start, const char *end, const char *blurb)
3467{
efd26800 3468 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
7918f24d
NC		 3469
3470 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3471
ab3bbdeb
YO		 3472 if (!PL_colorset)
3473 reginitcolors();
3474 {
3475 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
d2c6dc5e 3476 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
ab3bbdeb 3477
f2ed9b32 3478 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb
YO		 3479 start, end - start, 60);
3480
3481 PerlIO_printf(Perl_debug_log,
3482 "%s%s REx%s %s against %s\n",
3483 PL_colors[4], blurb, PL_colors[5], s0, s1);
3484
f2ed9b32 3485 if (utf8_target||utf8_pat)
1de06328
YO		 3486 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3487 utf8_pat ? "pattern" : "",
f2ed9b32
KW		 3488 utf8_pat && utf8_target ? " and " : "",
3489 utf8_target ? "string" : ""
ab3bbdeb
YO		 3490 );
3491 }
3492}
3df15adc
YO		 3493
3494STATIC void
786e8c11
YO		 3495S_dump_exec_pos(pTHX_ const char *locinput,
3496 const regnode *scan,
3497 const char *loc_regeol,
3498 const char *loc_bostr,
3499 const char *loc_reg_starttry,
f2ed9b32 3500 const bool utf8_target)
07be1b83 3501{
786e8c11 3502 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
07be1b83 3503 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
786e8c11 3504 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
07be1b83
YO		 3505 /* The part of the string before starttry has one color
3506 (pref0_len chars), between starttry and current
3507 position another one (pref_len - pref0_len chars),
3508 after the current position the third one.
3509 We assume that pref0_len <= pref_len, otherwise we
3510 decrease pref0_len. */
786e8c11
YO		 3511 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3512 ? (5 + taill) - l : locinput - loc_bostr;
07be1b83
YO		 3513 int pref0_len;
3514
7918f24d
NC		 3515 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3516
f2ed9b32 3517 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
07be1b83 3518 pref_len++;
786e8c11
YO		 3519 pref0_len = pref_len - (locinput - loc_reg_starttry);
3520 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3521 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3522 ? (5 + taill) - pref_len : loc_regeol - locinput);
f2ed9b32 3523 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
07be1b83
YO		 3524 l--;
3525 if (pref0_len < 0)
3526 pref0_len = 0;
3527 if (pref0_len > pref_len)
3528 pref0_len = pref_len;
3529 {
f2ed9b32 3530 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
0df25f3d 3531
ab3bbdeb 3532 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
1de06328 3533 (locinput - pref_len),pref0_len, 60, 4, 5);
0df25f3d 3534
ab3bbdeb 3535 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3df15adc