This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regexec.c: Fix some EBCDIC problems
[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
234 we don't need this definition. */
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. */
241#define IS_TEXT(rn) ( OP(rn)==EXACT )
098b07d5 242#define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || 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{
97aff369 291 dVAR;
a3b680e6 292 const int retval = PL_savestack_ix;
92da3157
DM
293 const int paren_elems_to_push =
294 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
e0fa7e2b
NC
295 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
296 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
87c0511b 297 I32 p;
40a82448 298 GET_RE_DEBUG_FLAGS_DECL;
a0d0e21e 299
b93070ed
DM
300 PERL_ARGS_ASSERT_REGCPPUSH;
301
e49a9654 302 if (paren_elems_to_push < 0)
d1b2014a
YO
303 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %i",
304 paren_elems_to_push, maxopenparen, parenfloor, 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{
97aff369 371 dVAR;
e0fa7e2b 372 UV i;
87c0511b 373 U32 paren;
a3621e74
YO
374 GET_RE_DEBUG_FLAGS_DECL;
375
7918f24d
NC
376 PERL_ARGS_ASSERT_REGCPPOP;
377
b1ce53c5 378 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
c6bf6a65 379 i = SSPOPUV;
e0fa7e2b
NC
380 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
381 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
b93070ed
DM
382 rex->lastcloseparen = SSPOPINT;
383 rex->lastparen = SSPOPINT;
92da3157 384 *maxopenparen_p = SSPOPINT;
b1ce53c5 385
620d5b66 386 i -= REGCP_OTHER_ELEMS;
b1ce53c5 387 /* Now restore the parentheses context. */
495f47a5
DM
388 DEBUG_BUFFERS_r(
389 if (i || rex->lastparen + 1 <= rex->nparens)
390 PerlIO_printf(Perl_debug_log,
391 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
392 PTR2UV(rex),
393 PTR2UV(rex->offs)
394 );
395 );
92da3157 396 paren = *maxopenparen_p;
620d5b66 397 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
ea3daa5d 398 SSize_t tmps;
1ca2007e 399 rex->offs[paren].start_tmp = SSPOPINT;
99a90e59
FC
400 rex->offs[paren].start = SSPOPIV;
401 tmps = SSPOPIV;
b93070ed
DM
402 if (paren <= rex->lastparen)
403 rex->offs[paren].end = tmps;
495f47a5
DM
404 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
405 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
406 (UV)paren,
407 (IV)rex->offs[paren].start,
408 (IV)rex->offs[paren].start_tmp,
409 (IV)rex->offs[paren].end,
410 (paren > rex->lastparen ? "(skipped)" : ""));
c277df42 411 );
87c0511b 412 paren--;
a0d0e21e 413 }
daf18116 414#if 1
dafc8851
JH
415 /* It would seem that the similar code in regtry()
416 * already takes care of this, and in fact it is in
417 * a better location to since this code can #if 0-ed out
418 * but the code in regtry() is needed or otherwise tests
419 * requiring null fields (pat.t#187 and split.t#{13,14}
daf18116
JH
420 * (as of patchlevel 7877) will fail. Then again,
421 * this code seems to be necessary or otherwise
225593e1
DM
422 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
423 * --jhi updated by dapm */
b93070ed 424 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
92da3157 425 if (i > *maxopenparen_p)
b93070ed
DM
426 rex->offs[i].start = -1;
427 rex->offs[i].end = -1;
495f47a5
DM
428 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
429 " \\%"UVuf": %s ..-1 undeffing\n",
430 (UV)i,
92da3157 431 (i > *maxopenparen_p) ? "-1" : " "
495f47a5 432 ));
a0d0e21e 433 }
dafc8851 434#endif
a0d0e21e
LW
435}
436
74088413
DM
437/* restore the parens and associated vars at savestack position ix,
438 * but without popping the stack */
439
440STATIC void
92da3157 441S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
74088413
DM
442{
443 I32 tmpix = PL_savestack_ix;
444 PL_savestack_ix = ix;
92da3157 445 regcppop(rex, maxopenparen_p);
74088413
DM
446 PL_savestack_ix = tmpix;
447}
448
02db2b7b 449#define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
a0d0e21e 450
31c7f561
KW
451STATIC bool
452S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
453{
454 /* Returns a boolean as to whether or not 'character' is a member of the
455 * Posix character class given by 'classnum' that should be equivalent to a
456 * value in the typedef '_char_class_number'.
457 *
458 * Ideally this could be replaced by a just an array of function pointers
459 * to the C library functions that implement the macros this calls.
460 * However, to compile, the precise function signatures are required, and
461 * these may vary from platform to to platform. To avoid having to figure
462 * out what those all are on each platform, I (khw) am using this method,
7aee35ff
KW
463 * which adds an extra layer of function call overhead (unless the C
464 * optimizer strips it away). But we don't particularly care about
465 * performance with locales anyway. */
31c7f561
KW
466
467 switch ((_char_class_number) classnum) {
15861f94 468 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
31c7f561 469 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
e8d596e0
KW
470 case _CC_ENUM_ASCII: return isASCII_LC(character);
471 case _CC_ENUM_BLANK: return isBLANK_LC(character);
b0d691b2
KW
472 case _CC_ENUM_CASED: return isLOWER_LC(character)
473 || isUPPER_LC(character);
e8d596e0 474 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
31c7f561
KW
475 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
476 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
477 case _CC_ENUM_LOWER: return isLOWER_LC(character);
478 case _CC_ENUM_PRINT: return isPRINT_LC(character);
e8d596e0 479 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
31c7f561 480 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
e8d596e0 481 case _CC_ENUM_SPACE: return isSPACE_LC(character);
31c7f561
KW
482 case _CC_ENUM_UPPER: return isUPPER_LC(character);
483 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
31c7f561 484 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
31c7f561
KW
485 default: /* VERTSPACE should never occur in locales */
486 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
487 }
488
489 assert(0); /* NOTREACHED */
490 return FALSE;
491}
492
3018b823
KW
493STATIC bool
494S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
495{
496 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
497 * 'character' is a member of the Posix character class given by 'classnum'
498 * that should be equivalent to a value in the typedef
499 * '_char_class_number'.
500 *
501 * This just calls isFOO_lc on the code point for the character if it is in
502 * the range 0-255. Outside that range, all characters avoid Unicode
503 * rules, ignoring any locale. So use the Unicode function if this class
504 * requires a swash, and use the Unicode macro otherwise. */
505
506 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
507
508 if (UTF8_IS_INVARIANT(*character)) {
509 return isFOO_lc(classnum, *character);
510 }
511 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
512 return isFOO_lc(classnum,
94bb8c36 513 TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1)));
3018b823
KW
514 }
515
516 if (classnum < _FIRST_NON_SWASH_CC) {
517
518 /* Initialize the swash unless done already */
519 if (! PL_utf8_swash_ptrs[classnum]) {
520 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2a16ac92
KW
521 PL_utf8_swash_ptrs[classnum] =
522 _core_swash_init("utf8",
523 "",
524 &PL_sv_undef, 1, 0,
525 PL_XPosix_ptrs[classnum], &flags);
3018b823
KW
526 }
527
92a2046b
KW
528 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
529 character,
530 TRUE /* is UTF */ ));
3018b823
KW
531 }
532
533 switch ((_char_class_number) classnum) {
534 case _CC_ENUM_SPACE:
535 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
536
537 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
538 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
539 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
540 default: return 0; /* Things like CNTRL are always
541 below 256 */
542 }
543
544 assert(0); /* NOTREACHED */
545 return FALSE;
546}
547
a687059c 548/*
e50aee73 549 * pregexec and friends
a687059c
LW
550 */
551
76234dfb 552#ifndef PERL_IN_XSUB_RE
a687059c 553/*
c277df42 554 - pregexec - match a regexp against a string
a687059c 555 */
c277df42 556I32
5aaab254 557Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
ea3daa5d 558 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
8fd1a950
DM
559/* stringarg: the point in the string at which to begin matching */
560/* strend: pointer to null at end of string */
561/* strbeg: real beginning of string */
562/* minend: end of match must be >= minend bytes after stringarg. */
563/* screamer: SV being matched: only used for utf8 flag, pos() etc; string
564 * itself is accessed via the pointers above */
565/* nosave: For optimizations. */
c277df42 566{
7918f24d
NC
567 PERL_ARGS_ASSERT_PREGEXEC;
568
c277df42 569 return
9041c2e3 570 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
c277df42
IZ
571 nosave ? 0 : REXEC_COPY_STR);
572}
76234dfb 573#endif
22e551b9 574
cad2e5aa 575
6eb5f6b9 576
1a4edc3c
DM
577/* re_intuit_start():
578 *
579 * Based on some optimiser hints, try to find the earliest position in the
580 * string where the regex could match.
581 *
582 * rx: the regex to match against
583 * sv: the SV being matched: only used for utf8 flag; the string
584 * itself is accessed via the pointers below. Note that on
585 * something like an overloaded SV, SvPOK(sv) may be false
586 * and the string pointers may point to something unrelated to
587 * the SV itself.
588 * strbeg: real beginning of string
589 * strpos: the point in the string at which to begin matching
590 * strend: pointer to the byte following the last char of the string
591 * flags currently unused; set to 0
592 * data: currently unused; set to NULL
593 *
594 * The basic idea of re_intuit_start() is to use some known information
595 * about the pattern, namely:
596 *
597 * a) the longest known anchored substring (i.e. one that's at a
598 * constant offset from the beginning of the pattern; but not
599 * necessarily at a fixed offset from the beginning of the
600 * string);
601 * b) the longest floating substring (i.e. one that's not at a constant
602 * offset from the beginning of the pattern);
603 * c) Whether the pattern is anchored to the string; either
604 * an absolute anchor: /^../, or anchored to \n: /^.../m,
605 * or anchored to pos(): /\G/;
606 * d) A start class: a real or synthetic character class which
607 * represents which characters are legal at the start of the pattern;
608 *
609 * to either quickly reject the match, or to find the earliest position
610 * within the string at which the pattern might match, thus avoiding
611 * running the full NFA engine at those earlier locations, only to
612 * eventually fail and retry further along.
613 *
614 * Returns NULL if the pattern can't match, or returns the address within
615 * the string which is the earliest place the match could occur.
616 *
617 * The longest of the anchored and floating substrings is called 'check'
618 * and is checked first. The other is called 'other' and is checked
619 * second. The 'other' substring may not be present. For example,
620 *
621 * /(abc|xyz)ABC\d{0,3}DEFG/
622 *
623 * will have
624 *
625 * check substr (float) = "DEFG", offset 6..9 chars
626 * other substr (anchored) = "ABC", offset 3..3 chars
627 * stclass = [ax]
628 *
629 * Be aware that during the course of this function, sometimes 'anchored'
630 * refers to a substring being anchored relative to the start of the
631 * pattern, and sometimes to the pattern itself being anchored relative to
632 * the string. For example:
633 *
634 * /\dabc/: "abc" is anchored to the pattern;
635 * /^\dabc/: "abc" is anchored to the pattern and the string;
636 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
637 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
638 * but the pattern is anchored to the string.
52a21eb3
DM
639 */
640
cad2e5aa 641char *
52a21eb3
DM
642Perl_re_intuit_start(pTHX_
643 REGEXP * const rx,
644 SV *sv,
645 const char * const strbeg,
646 char *strpos,
647 char *strend,
648 const U32 flags,
649 re_scream_pos_data *data)
cad2e5aa 650{
97aff369 651 dVAR;
8d919b0a 652 struct regexp *const prog = ReANY(rx);
6b071d16 653 SSize_t start_shift = prog->check_offset_min;
cad2e5aa 654 /* Should be nonnegative! */
ea3daa5d 655 SSize_t end_shift = 0;
0fc004dd
DM
656 /* current lowest pos in string where the regex can start matching */
657 char *rx_origin = strpos;
eb578fdb 658 SV *check;
f2ed9b32 659 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
6480a6c4 660 U8 other_ix = 1 - prog->substrs->check_ix;
6ad5ffb3 661 bool ml_anch = 0;
8f4bf5fc 662 char *other_last = strpos;/* latest pos 'other' substr already checked to */
bd61b366 663 char *check_at = NULL; /* check substr found at this pos */
bbe252da 664 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
f8fc2ecf 665 RXi_GET_DECL(prog,progi);
02d5137b
DM
666 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
667 regmatch_info *const reginfo = &reginfo_buf;
a3621e74
YO
668 GET_RE_DEBUG_FLAGS_DECL;
669
7918f24d 670 PERL_ARGS_ASSERT_RE_INTUIT_START;
c33e64f0
FC
671 PERL_UNUSED_ARG(flags);
672 PERL_UNUSED_ARG(data);
7918f24d 673
1dc475d0
DM
674 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
675 "Intuit: trying to determine minimum start position...\n"));
676
fb9bbddb
DM
677 /* for now, assume that all substr offsets are positive. If at some point
678 * in the future someone wants to do clever things with look-behind and
679 * -ve offsets, they'll need to fix up any code in this function
680 * which uses these offsets. See the thread beginning
681 * <20140113145929.GF27210@iabyn.com>
682 */
683 assert(prog->substrs->data[0].min_offset >= 0);
684 assert(prog->substrs->data[0].max_offset >= 0);
685 assert(prog->substrs->data[1].min_offset >= 0);
686 assert(prog->substrs->data[1].max_offset >= 0);
687 assert(prog->substrs->data[2].min_offset >= 0);
688 assert(prog->substrs->data[2].max_offset >= 0);
689
f7022b5a 690 /* for now, assume that if both present, that the floating substring
83f2232d 691 * doesn't start before the anchored substring.
f7022b5a
DM
692 * If you break this assumption (e.g. doing better optimisations
693 * with lookahead/behind), then you'll need to audit the code in this
694 * function carefully first
695 */
696 assert(
697 ! ( (prog->anchored_utf8 || prog->anchored_substr)
698 && (prog->float_utf8 || prog->float_substr))
699 || (prog->float_min_offset >= prog->anchored_offset));
700
1a4edc3c
DM
701 /* byte rather than char calculation for efficiency. It fails
702 * to quickly reject some cases that can't match, but will reject
703 * them later after doing full char arithmetic */
c344f387 704 if (prog->minlen > strend - strpos) {
a3621e74 705 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 706 " String too short...\n"));
cad2e5aa 707 goto fail;
2c2d71f5 708 }
d8da0584 709
6c3fea77 710 reginfo->is_utf8_target = cBOOL(utf8_target);
bf2039a9 711 reginfo->info_aux = NULL;
9d9163fb 712 reginfo->strbeg = strbeg;
220db18a 713 reginfo->strend = strend;
aed7b151 714 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
02d5137b 715 reginfo->intuit = 1;
1cb48e53
DM
716 /* not actually used within intuit, but zero for safety anyway */
717 reginfo->poscache_maxiter = 0;
02d5137b 718
f2ed9b32 719 if (utf8_target) {
33b8afdf
JH
720 if (!prog->check_utf8 && prog->check_substr)
721 to_utf8_substr(prog);
722 check = prog->check_utf8;
723 } else {
7e0d5ad7
KW
724 if (!prog->check_substr && prog->check_utf8) {
725 if (! to_byte_substr(prog)) {
6b54ddc5 726 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
7e0d5ad7
KW
727 }
728 }
33b8afdf
JH
729 check = prog->check_substr;
730 }
274cd312 731
1dc475d0
DM
732 /* dump the various substring data */
733 DEBUG_OPTIMISE_MORE_r({
734 int i;
735 for (i=0; i<=2; i++) {
736 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
737 : prog->substrs->data[i].substr);
738 if (!sv)
739 continue;
740
741 PerlIO_printf(Perl_debug_log,
742 " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf
743 " useful=%"IVdf" utf8=%d [%s]\n",
744 i,
745 (IV)prog->substrs->data[i].min_offset,
746 (IV)prog->substrs->data[i].max_offset,
747 (IV)prog->substrs->data[i].end_shift,
748 BmUSEFUL(sv),
749 utf8_target ? 1 : 0,
750 SvPEEK(sv));
751 }
752 });
753
8e1490ee 754 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
9fc7410e
DM
755
756 /* ml_anch: check after \n?
757 *
758 * A note about IMPLICIT: on an un-anchored pattern beginning
759 * with /.*.../, these flags will have been added by the
760 * compiler:
761 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
762 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
763 */
7d2d37f5
DM
764 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
765 && !(prog->intflags & PREGf_IMPLICIT);
cad2e5aa 766
343c8a29 767 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
c889ccc8
DM
768 /* we are only allowed to match at BOS or \G */
769
57fcbfa7 770 /* trivially reject if there's a BOS anchor and we're not at BOS.
7bb3b9eb
DM
771 *
772 * Note that we don't try to do a similar quick reject for
773 * \G, since generally the caller will have calculated strpos
774 * based on pos() and gofs, so the string is already correctly
775 * anchored by definition; and handling the exceptions would
776 * be too fiddly (e.g. REXEC_IGNOREPOS).
57fcbfa7 777 */
7bb3b9eb 778 if ( strpos != strbeg
343c8a29 779 && (prog->intflags & (PREGf_ANCH_BOL|PREGf_ANCH_SBOL)))
c889ccc8
DM
780 {
781 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 782 " Not at start...\n"));
c889ccc8
DM
783 goto fail;
784 }
785
a5d12a4b
DM
786 /* in the presence of an anchor, the anchored (relative to the
787 * start of the regex) substr must also be anchored relative
66b7ec5c
DM
788 * to strpos. So quickly reject if substr isn't found there.
789 * This works for \G too, because the caller will already have
790 * subtracted gofs from pos, and gofs is the offset from the
791 * \G to the start of the regex. For example, in /.abc\Gdef/,
792 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
793 * caller will have set strpos=pos()-4; we look for the substr
794 * at position pos()-4+1, which lines up with the "a" */
a5d12a4b 795
c889ccc8 796 if (prog->check_offset_min == prog->check_offset_max
d0d44648 797 && !(prog->intflags & PREGf_CANY_SEEN))
c889ccc8
DM
798 {
799 /* Substring at constant offset from beg-of-str... */
d307bf57 800 SSize_t slen = SvCUR(check);
343c8a29 801 char *s = HOP3c(strpos, prog->check_offset_min, strend);
1de06328 802
1dc475d0
DM
803 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
804 " Looking for check substr at fixed offset %"IVdf"...\n",
805 (IV)prog->check_offset_min));
806
7742aa66
DM
807 if (SvTAIL(check)) {
808 /* In this case, the regex is anchored at the end too.
809 * Unless it's a multiline match, the lengths must match
810 * exactly, give or take a \n. NB: slen >= 1 since
811 * the last char of check is \n */
812 if (!multiline
813 && ( strend - s > slen
814 || strend - s < slen - 1
815 || (strend - s == slen && strend[-1] != '\n')))
c889ccc8
DM
816 {
817 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 818 " String too long...\n"));
c889ccc8
DM
819 goto fail_finish;
820 }
821 /* Now should match s[0..slen-2] */
822 slen--;
c889ccc8 823 }
d307bf57
DM
824 if (slen && (*SvPVX_const(check) != *s
825 || (slen > 1 && memNE(SvPVX_const(check), s, slen))))
826 {
827 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 828 " String not equal...\n"));
d307bf57
DM
829 goto fail_finish;
830 }
c889ccc8
DM
831
832 check_at = s;
833 goto success_at_start;
cad2e5aa 834 }
cad2e5aa 835 }
cad2e5aa 836 }
0fc004dd 837
c0e0ec46 838 end_shift = prog->check_end_shift;
cad2e5aa 839
19188028 840#ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
0033605d 841 if (end_shift < 0)
1de06328 842 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
220fc49f 843 (IV)end_shift, RX_PRECOMP(prog));
2c2d71f5
JH
844#endif
845
2c2d71f5 846 restart:
1de06328 847
66b7ec5c
DM
848 /* This is the (re)entry point of the main loop in this function.
849 * The goal of this loop is to:
850 * 1) find the "check" substring in the region rx_origin..strend
851 * (adjusted by start_shift / end_shift). If not found, reject
852 * immediately.
853 * 2) If it exists, look for the "other" substr too if defined; for
854 * example, if the check substr maps to the anchored substr, then
855 * check the floating substr, and vice-versa. If not found, go
856 * back to (1) with rx_origin suitably incremented.
857 * 3) If we find an rx_origin position that doesn't contradict
858 * either of the substrings, then check the possible additional
859 * constraints on rx_origin of /^.../m or a known start class.
860 * If these fail, then depending on which constraints fail, jump
861 * back to here, or to various other re-entry points further along
862 * that skip some of the first steps.
863 * 4) If we pass all those tests, update the BmUSEFUL() count on the
864 * substring. If the start position was determined to be at the
865 * beginning of the string - so, not rejected, but not optimised,
866 * since we have to run regmatch from position 0 - decrement the
867 * BmUSEFUL() count. Otherwise increment it.
868 */
869
1a4edc3c
DM
870
871 /* first, look for the 'check' substring */
872
1de06328 873 {
c33e64f0
FC
874 U8* start_point;
875 U8* end_point;
c889ccc8 876
c889ccc8 877 DEBUG_OPTIMISE_MORE_r({
1dc475d0 878 PerlIO_printf(Perl_debug_log,
ae5d4331 879 " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf
1dc475d0 880 " Start shift: %"IVdf" End shift %"IVdf
4d281893 881 " Real end Shift: %"IVdf"\n",
197641b6 882 (IV)(rx_origin - strpos),
c889ccc8 883 (IV)prog->check_offset_min,
12fbc530
DM
884 (IV)start_shift,
885 (IV)end_shift,
c889ccc8
DM
886 (IV)prog->check_end_shift);
887 });
1de06328 888
0d331aaf 889 if (prog->intflags & PREGf_CANY_SEEN) {
0fc004dd 890 start_point= (U8*)(rx_origin + start_shift);
12fbc530 891 end_point= (U8*)(strend - end_shift);
557f47af
DM
892 if (start_point > end_point)
893 goto fail_finish;
1de06328 894 } else {
557f47af
DM
895 end_point = HOP3(strend, -end_shift, strbeg);
896 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
897 if (!start_point)
898 goto fail_finish;
1de06328 899 }
c889ccc8 900
557f47af 901
e0362b86
DM
902 /* If the regex is absolutely anchored to either the start of the
903 * string (BOL,SBOL) or to pos() (ANCH_GPOS), then
904 * check_offset_max represents an upper bound on the string where
905 * the substr could start. For the ANCH_GPOS case, we assume that
906 * the caller of intuit will have already set strpos to
907 * pos()-gofs, so in this case strpos + offset_max will still be
908 * an upper bound on the substr.
909 */
c19c836a
DM
910 if (!ml_anch
911 && prog->intflags & PREGf_ANCH
e0362b86 912 && prog->check_offset_max != SSize_t_MAX)
c19c836a 913 {
1a08ba3a 914 SSize_t len = SvCUR(check) - !!SvTAIL(check);
e0362b86
DM
915 const char * const anchor =
916 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
917
918 /* do a bytes rather than chars comparison. It's conservative;
919 * so it skips doing the HOP if the result can't possibly end
920 * up earlier than the old value of end_point.
921 */
922 if ((char*)end_point - anchor > prog->check_offset_max) {
923 end_point = HOP3lim((U8*)anchor,
924 prog->check_offset_max,
925 end_point -len)
926 + len;
927 }
d6ef1678
DM
928 }
929
6bda09f9 930 DEBUG_OPTIMISE_MORE_r({
1dc475d0 931 PerlIO_printf(Perl_debug_log, " fbm_instr len=%d str=<%.*s>\n",
1de06328 932 (int)(end_point - start_point),
fc8cd66c 933 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
1de06328
YO
934 start_point);
935 });
936
ae5d4331 937 check_at = fbm_instr( start_point, end_point,
7fba1cd6 938 check, multiline ? FBMrf_MULTILINE : 0);
c889ccc8 939
8fd34720
DM
940 /* Update the count-of-usability, remove useless subpatterns,
941 unshift s. */
942
943 DEBUG_EXECUTE_r({
944 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
945 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
946 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s",
947 (check_at ? "Found" : "Did not find"),
948 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
949 ? "anchored" : "floating"),
950 quoted,
951 RE_SV_TAIL(check),
952 (check_at ? " at offset " : "...\n") );
953 });
2c2d71f5 954
8fd34720
DM
955 if (!check_at)
956 goto fail_finish;
957 /* Finish the diagnostic message */
958 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(check_at - strpos)) );
2c2d71f5 959
8fd34720
DM
960 /* set rx_origin to the minimum position where the regex could start
961 * matching, given the constraint of the just-matched check substring.
962 * But don't set it lower than previously.
963 */
fdc003fd 964
8fd34720
DM
965 if (check_at - rx_origin > prog->check_offset_max)
966 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
967 }
fdc003fd
DM
968
969
1a4edc3c 970 /* now look for the 'other' substring if defined */
2c2d71f5 971
6480a6c4
DM
972 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
973 : prog->substrs->data[other_ix].substr)
1de06328 974 {
30944b6d 975 /* Take into account the "other" substring. */
6c3343a6
DM
976 char *last, *last1;
977 char *s;
978 SV* must;
979 struct reg_substr_datum *other;
980
981 do_other_substr:
982 other = &prog->substrs->data[other_ix];
983
984 /* if "other" is anchored:
985 * we've previously found a floating substr starting at check_at.
986 * This means that the regex origin must lie somewhere
987 * between min (rx_origin): HOP3(check_at, -check_offset_max)
988 * and max: HOP3(check_at, -check_offset_min)
989 * (except that min will be >= strpos)
990 * So the fixed substr must lie somewhere between
991 * HOP3(min, anchored_offset)
992 * HOP3(max, anchored_offset) + SvCUR(substr)
993 */
994
995 /* if "other" is floating
996 * Calculate last1, the absolute latest point where the
997 * floating substr could start in the string, ignoring any
998 * constraints from the earlier fixed match. It is calculated
999 * as follows:
1000 *
1001 * strend - prog->minlen (in chars) is the absolute latest
1002 * position within the string where the origin of the regex
1003 * could appear. The latest start point for the floating
1004 * substr is float_min_offset(*) on from the start of the
1005 * regex. last1 simply combines thee two offsets.
1006 *
1007 * (*) You might think the latest start point should be
1008 * float_max_offset from the regex origin, and technically
1009 * you'd be correct. However, consider
1010 * /a\d{2,4}bcd\w/
1011 * Here, float min, max are 3,5 and minlen is 7.
1012 * This can match either
1013 * /a\d\dbcd\w/
1014 * /a\d\d\dbcd\w/
1015 * /a\d\d\d\dbcd\w/
1016 * In the first case, the regex matches minlen chars; in the
1017 * second, minlen+1, in the third, minlen+2.
1018 * In the first case, the floating offset is 3 (which equals
1019 * float_min), in the second, 4, and in the third, 5 (which
1020 * equals float_max). In all cases, the floating string bcd
1021 * can never start more than 4 chars from the end of the
1022 * string, which equals minlen - float_min. As the substring
1023 * starts to match more than float_min from the start of the
1024 * regex, it makes the regex match more than minlen chars,
1025 * and the two cancel each other out. So we can always use
1026 * float_min - minlen, rather than float_max - minlen for the
1027 * latest position in the string.
1028 *
1029 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1030 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1031 */
1032
e7a14a9c 1033 assert(prog->minlen >= other->min_offset);
6c3343a6
DM
1034 last1 = HOP3c(strend,
1035 other->min_offset - prog->minlen, strbeg);
1036
4d006249 1037 if (other_ix) {/* i.e. if (other-is-float) */
6c3343a6
DM
1038 /* last is the latest point where the floating substr could
1039 * start, *given* any constraints from the earlier fixed
1040 * match. This constraint is that the floating string starts
1041 * <= float_max_offset chars from the regex origin (rx_origin).
1042 * If this value is less than last1, use it instead.
eb3831ce 1043 */
6c3343a6
DM
1044 assert(rx_origin <= last1);
1045 last =
1046 /* this condition handles the offset==infinity case, and
1047 * is a short-cut otherwise. Although it's comparing a
1048 * byte offset to a char length, it does so in a safe way,
1049 * since 1 char always occupies 1 or more bytes,
1050 * so if a string range is (last1 - rx_origin) bytes,
1051 * it will be less than or equal to (last1 - rx_origin)
1052 * chars; meaning it errs towards doing the accurate HOP3
1053 * rather than just using last1 as a short-cut */
1054 (last1 - rx_origin) < other->max_offset
1055 ? last1
1056 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1057 }
1058 else {
1059 assert(strpos + start_shift <= check_at);
1060 last = HOP4c(check_at, other->min_offset - start_shift,
1061 strbeg, strend);
1062 }
ead917d0 1063
6c3343a6
DM
1064 s = HOP3c(rx_origin, other->min_offset, strend);
1065 if (s < other_last) /* These positions already checked */
1066 s = other_last;
1067
1068 must = utf8_target ? other->utf8_substr : other->substr;
1069 assert(SvPOK(must));
1070 s = fbm_instr(
1071 (unsigned char*)s,
1072 (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0),
1073 must,
1074 multiline ? FBMrf_MULTILINE : 0
1075 );
1076 DEBUG_EXECUTE_r({
1077 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1078 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1079 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s",
1080 s ? "Found" : "Contradicts",
1081 other_ix ? "floating" : "anchored",
1082 quoted, RE_SV_TAIL(must));
1083 });
ead917d0 1084
ead917d0 1085
6c3343a6
DM
1086 if (!s) {
1087 /* last1 is latest possible substr location. If we didn't
1088 * find it before there, we never will */
1089 if (last >= last1) {
1090 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1091 ", giving up...\n"));
1092 goto fail_finish;
ead917d0
DM
1093 }
1094
6c3343a6
DM
1095 /* try to find the check substr again at a later
1096 * position. Maybe next time we'll find the "other" substr
1097 * in range too */
1098 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1099 ", trying %s at offset %ld...\n",
1100 (other_ix ? "floating" : "anchored"),
197641b6 1101 (long)(HOP3c(check_at, 1, strend) - strpos)));
6c3343a6
DM
1102
1103 other_last = HOP3c(last, 1, strend) /* highest failure */;
1104 rx_origin =
4d006249 1105 other_ix /* i.e. if other-is-float */
6c3343a6
DM
1106 ? HOP3c(rx_origin, 1, strend)
1107 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1108 goto restart;
1109 }
1110 else {
1111 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
197641b6 1112 (long)(s - strpos)));
ead917d0 1113
4d006249 1114 if (other_ix) { /* if (other-is-float) */
6c3343a6
DM
1115 /* other_last is set to s, not s+1, since its possible for
1116 * a floating substr to fail first time, then succeed
1117 * second time at the same floating position; e.g.:
1118 * "-AB--AABZ" =~ /\wAB\d*Z/
1119 * The first time round, anchored and float match at
1120 * "-(AB)--AAB(Z)" then fail on the initial \w character
1121 * class. Second time round, they match at "-AB--A(AB)(Z)".
1122 */
1123 other_last = s;
ead917d0
DM
1124 }
1125 else {
6c3343a6
DM
1126 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1127 other_last = HOP3c(s, 1, strend);
ead917d0 1128 }
6c3343a6 1129 }
cad2e5aa 1130 }
acba93e8
DM
1131 else {
1132 DEBUG_OPTIMISE_MORE_r(
1133 PerlIO_printf(Perl_debug_log,
1134 " Check-only match: offset min:%"IVdf" max:%"IVdf
1c1c599d 1135 " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf
acba93e8
DM
1136 " strend-strpos:%"IVdf"\n",
1137 (IV)prog->check_offset_min,
1138 (IV)prog->check_offset_max,
1c1c599d 1139 (IV)(check_at-strpos),
acba93e8 1140 (IV)(rx_origin-strpos),
1c1c599d 1141 (IV)(rx_origin-check_at),
acba93e8
DM
1142 (IV)(strend-strpos)
1143 )
1144 );
1145 }
2c2d71f5 1146
acba93e8 1147 postprocess_substr_matches:
0991020e 1148
1a4edc3c 1149 /* handle the extra constraint of /^.../m if present */
e3c6feb0 1150
7d2d37f5 1151 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
4620cb61
DM
1152 char *s;
1153
a62659bd
DM
1154 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1155 " looking for /^/m anchor"));
d0880ea7
DM
1156
1157 /* we have failed the constraint of a \n before rx_origin.
2e759faa
DM
1158 * Find the next \n, if any, even if it's beyond the current
1159 * anchored and/or floating substrings. Whether we should be
1160 * scanning ahead for the next \n or the next substr is debatable.
1161 * On the one hand you'd expect rare substrings to appear less
1162 * often than \n's. On the other hand, searching for \n means
1163 * we're effectively flipping been check_substr and "\n" on each
1164 * iteration as the current "rarest" string candidate, which
1165 * means for example that we'll quickly reject the whole string if
1166 * hasn't got a \n, rather than trying every substr position
1167 * first
1168 */
d0880ea7 1169
4620cb61
DM
1170 s = HOP3c(strend, - prog->minlen, strpos);
1171 if (s <= rx_origin ||
1172 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1173 {
d0880ea7
DM
1174 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1175 " Did not find /%s^%s/m...\n",
1176 PL_colors[0], PL_colors[1]));
a62659bd
DM
1177 goto fail_finish;
1178 }
d0880ea7 1179
4ada1233
DM
1180 /* earliest possible origin is 1 char after the \n.
1181 * (since *rx_origin == '\n', it's safe to ++ here rather than
1182 * HOP(rx_origin, 1)) */
1183 rx_origin++;
d0880ea7 1184
f4f115de 1185 if (prog->substrs->check_ix == 0 /* check is anchored */
4ada1233 1186 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
f4f115de 1187 {
d0880ea7
DM
1188 /* Position contradicts check-string; either because
1189 * check was anchored (and thus has no wiggle room),
4ada1233 1190 * or check was float and rx_origin is above the float range */
d0880ea7
DM
1191 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1192 " Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
4ada1233 1193 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
d0880ea7
DM
1194 goto restart;
1195 }
1196
1197 /* if we get here, the check substr must have been float,
2e759faa 1198 * is in range, and we may or may not have had an anchored
d0880ea7
DM
1199 * "other" substr which still contradicts */
1200 assert(prog->substrs->check_ix); /* check is float */
1201
1202 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1203 /* whoops, the anchored "other" substr exists, so we still
1204 * contradict. On the other hand, the float "check" substr
1205 * didn't contradict, so just retry the anchored "other"
1206 * substr */
1207 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1208 " Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1209 PL_colors[0], PL_colors[1],
1210 (long)(rx_origin - strpos),
1211 (long)(rx_origin - strpos + prog->anchored_offset)));
1212 goto do_other_substr;
1213 }
1214
1215 /* success: we don't contradict the found floating substring
1216 * (and there's no anchored substr). */
d0880ea7
DM
1217 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1218 " Found /%s^%s/m at offset %ld...\n",
1219 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
e3c6feb0
DM
1220 }
1221 else {
2e759faa 1222 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
fe4f3442 1223 " (multiline anchor test skipped)\n"));
e3c6feb0
DM
1224 }
1225
ffad1e6a 1226 success_at_start:
e3c6feb0 1227
cad2e5aa 1228
dd170ff5
DM
1229 /* if we have a starting character class, then test that extra constraint.
1230 * (trie stclasses are too expensive to use here, we are better off to
1231 * leave it to regmatch itself) */
1232
f8fc2ecf 1233 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
f8fc2ecf 1234 const U8* const str = (U8*)STRING(progi->regstclass);
0991020e 1235
2c75e362 1236 /* XXX this value could be pre-computed */
f8fc2ecf 1237 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
2c75e362
DM
1238 ? (reginfo->is_utf8_pat
1239 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1240 : STR_LEN(progi->regstclass))
66e933ab 1241 : 1);
1de06328 1242 char * endpos;
fa3bb21d 1243 char *s;
000dfd2d
DM
1244 /* latest pos that a matching float substr constrains rx start to */
1245 char *rx_max_float = NULL;
1246
c75a3985
DM
1247 /* if the current rx_origin is anchored, either by satisfying an
1248 * anchored substring constraint, or a /^.../m constraint, then we
1249 * can reject the current origin if the start class isn't found
1250 * at the current position. If we have a float-only match, then
1251 * rx_origin is constrained to a range; so look for the start class
1252 * in that range. if neither, then look for the start class in the
1253 * whole rest of the string */
1254
dd170ff5
DM
1255 /* XXX DAPM it's not clear what the minlen test is for, and why
1256 * it's not used in the floating case. Nothing in the test suite
1257 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1258 * Here are some old comments, which may or may not be correct:
1259 *
1260 * minlen == 0 is possible if regstclass is \b or \B,
1261 * and the fixed substr is ''$.
1262 * Since minlen is already taken into account, rx_origin+1 is
1263 * before strend; accidentally, minlen >= 1 guaranties no false
1264 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1265 * 0) below assumes that regstclass does not come from lookahead...
1266 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1267 * This leaves EXACTF-ish only, which are dealt with in
1268 * find_byclass().
1269 */
1270
7d2d37f5 1271 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
fa3bb21d 1272 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
000dfd2d
DM
1273 else if (prog->float_substr || prog->float_utf8) {
1274 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1275 endpos= HOP3c(rx_max_float, cl_l, strend);
1276 }
1de06328
YO
1277 else
1278 endpos= strend;
1279
1dc475d0
DM
1280 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1281 " looking for class: start_shift: %"IVdf" check_at: %"IVdf
c43b5520 1282 " rx_origin: %"IVdf" endpos: %"IVdf"\n",
1dc475d0 1283 (IV)start_shift, (IV)(check_at - strbeg),
c43b5520 1284 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
d8080198 1285
c43b5520 1286 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
f9176b44 1287 reginfo);
be778b1a 1288 if (!s) {
6eb5f6b9 1289 if (endpos == strend) {
a3621e74 1290 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1dc475d0 1291 " Could not match STCLASS...\n") );
6eb5f6b9
JH
1292 goto fail;
1293 }
a3621e74 1294 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1dc475d0 1295 " This position contradicts STCLASS...\n") );
e0eb31e7
DM
1296 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1297 && !(prog->intflags & PREGf_IMPLICIT))
653099ff 1298 goto fail;
9fed8d02 1299
6eb5f6b9 1300 /* Contradict one of substrings */
97136c8a
DM
1301 if (prog->anchored_substr || prog->anchored_utf8) {
1302 if (prog->substrs->check_ix == 1) { /* check is float */
1303 /* Have both, check_string is floating */
1304 assert(rx_origin + start_shift <= check_at);
1305 if (rx_origin + start_shift != check_at) {
1306 /* not at latest position float substr could match:
c75a3985
DM
1307 * Recheck anchored substring, but not floating.
1308 * The condition above is in bytes rather than
1309 * chars for efficiency. It's conservative, in
1310 * that it errs on the side of doing 'goto
1311 * do_other_substr', where a more accurate
1312 * char-based calculation will be done */
97136c8a
DM
1313 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1314 " Looking for anchored substr starting at offset %ld...\n",
1315 (long)(other_last - strpos)) );
1316 goto do_other_substr;
3369914b 1317 }
3369914b
DM
1318 }
1319 }
97136c8a 1320 else {
9fed8d02
DM
1321 /* float-only */
1322
7d2d37f5 1323 if (ml_anch) {
c75a3985
DM
1324 /* In the presence of ml_anch, we might be able to
1325 * find another \n without breaking the current float
1326 * constraint. */
1327
1328 /* strictly speaking this should be HOP3c(..., 1, ...),
1329 * but since we goto a block of code that's going to
1330 * search for the next \n if any, its safe here */
9fed8d02 1331 rx_origin++;
9fed8d02
DM
1332 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1333 " Looking for /%s^%s/m starting at offset %ld...\n",
1334 PL_colors[0], PL_colors[1],
1335 (long)(rx_origin - strpos)) );
9fed8d02 1336 goto postprocess_substr_matches;
ab60c45a 1337 }
c75a3985
DM
1338
1339 /* strictly speaking this can never be true; but might
1340 * be if we ever allow intuit without substrings */
1341 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
9fed8d02 1342 goto fail;
c75a3985 1343
000dfd2d 1344 rx_origin = rx_max_float;
9fed8d02
DM
1345 }
1346
c75a3985
DM
1347 /* at this point, any matching substrings have been
1348 * contradicted. Start again... */
1349
9fed8d02 1350 rx_origin = HOP3c(rx_origin, 1, strend);
557f47af
DM
1351
1352 /* uses bytes rather than char calculations for efficiency.
1353 * It's conservative: it errs on the side of doing 'goto restart',
1354 * where there is code that does a proper char-based test */
9fed8d02 1355 if (rx_origin + start_shift + end_shift > strend) {
40268e5b 1356 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
9fed8d02
DM
1357 " Could not match STCLASS...\n") );
1358 goto fail;
1359 }
1360 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1361 " Looking for %s substr starting at offset %ld...\n",
1362 (prog->substrs->check_ix ? "floating" : "anchored"),
1363 (long)(rx_origin + start_shift - strpos)) );
1364 goto restart;
6eb5f6b9 1365 }
9fed8d02 1366
c75a3985
DM
1367 /* Success !!! */
1368
5f9c6575 1369 if (rx_origin != s) {
a3621e74 1370 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 1371 " By STCLASS: moving %ld --> %ld\n",
5f9c6575 1372 (long)(rx_origin - strpos), (long)(s - strpos))
b7953727
JH
1373 );
1374 }
1375 else {
a3621e74 1376 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 1377 " Does not contradict STCLASS...\n");
b7953727
JH
1378 );
1379 }
6eb5f6b9 1380 }
ffad1e6a
DM
1381
1382 /* Decide whether using the substrings helped */
1383
1384 if (rx_origin != strpos) {
1385 /* Fixed substring is found far enough so that the match
1386 cannot start at strpos. */
1387
1388 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n"));
1389 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1390 }
1391 else {
70563e16
DM
1392 /* The found rx_origin position does not prohibit matching at
1393 * strpos, so calling intuit didn't gain us anything. Decrement
1394 * the BmUSEFUL() count on the check substring, and if we reach
1395 * zero, free it. */
1396 if (!(prog->intflags & PREGf_NAUGHTY)
ffad1e6a
DM
1397 && (utf8_target ? (
1398 prog->check_utf8 /* Could be deleted already */
1399 && --BmUSEFUL(prog->check_utf8) < 0
1400 && (prog->check_utf8 == prog->float_utf8)
1401 ) : (
1402 prog->check_substr /* Could be deleted already */
1403 && --BmUSEFUL(prog->check_substr) < 0
1404 && (prog->check_substr == prog->float_substr)
1405 )))
1406 {
1407 /* If flags & SOMETHING - do not do it many times on the same match */
1408 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n"));
1409 /* XXX Does the destruction order has to change with utf8_target? */
1410 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1411 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1412 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1413 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1414 check = NULL; /* abort */
ffad1e6a
DM
1415 /* XXXX This is a remnant of the old implementation. It
1416 looks wasteful, since now INTUIT can use many
1417 other heuristics. */
1418 prog->extflags &= ~RXf_USE_INTUIT;
ffad1e6a
DM
1419 }
1420 }
1421
1422 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1423 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1424 PL_colors[4], PL_colors[5], (long)(rx_origin - strpos)) );
1425
c765d6e0 1426 return rx_origin;
2c2d71f5
JH
1427
1428 fail_finish: /* Substring not found */
33b8afdf 1429 if (prog->check_substr || prog->check_utf8) /* could be removed already */
f2ed9b32 1430 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
cad2e5aa 1431 fail:
a3621e74 1432 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
e4584336 1433 PL_colors[4], PL_colors[5]));
bd61b366 1434 return NULL;
cad2e5aa 1435}
9661b544 1436
70563e16 1437
a0a388a1 1438#define DECL_TRIE_TYPE(scan) \
098b07d5
KW
1439 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1440 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold } \
fab2782b
YO
1441 trie_type = ((scan->flags == EXACT) \
1442 ? (utf8_target ? trie_utf8 : trie_plain) \
098b07d5
KW
1443 : (scan->flags == EXACTFA) \
1444 ? (utf8_target ? trie_utf8_exactfa_fold : trie_latin_utf8_exactfa_fold) \
1445 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
fab2782b 1446
fd3249ee 1447#define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
baa60164 1448STMT_START { \
fab2782b 1449 STRLEN skiplen; \
baa60164 1450 U8 flags = FOLD_FLAGS_FULL; \
fab2782b 1451 switch (trie_type) { \
31f05a37 1452 case trie_utf8_exactfa_fold: \
baa60164 1453 flags |= FOLD_FLAGS_NOMIX_ASCII; \
924ba076 1454 /* FALLTHROUGH */ \
fab2782b
YO
1455 case trie_utf8_fold: \
1456 if ( foldlen>0 ) { \
c80e42f3 1457 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO
1458 foldlen -= len; \
1459 uscan += len; \
1460 len=0; \
1461 } else { \
445bf929 1462 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
fab2782b
YO
1463 len = UTF8SKIP(uc); \
1464 skiplen = UNISKIP( uvc ); \
1465 foldlen -= skiplen; \
1466 uscan = foldbuf + skiplen; \
1467 } \
1468 break; \
baa60164
KW
1469 case trie_latin_utf8_exactfa_fold: \
1470 flags |= FOLD_FLAGS_NOMIX_ASCII; \
924ba076 1471 /* FALLTHROUGH */ \
fab2782b
YO
1472 case trie_latin_utf8_fold: \
1473 if ( foldlen>0 ) { \
c80e42f3 1474 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO
1475 foldlen -= len; \
1476 uscan += len; \
1477 len=0; \
1478 } else { \
1479 len = 1; \
31f05a37 1480 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
fab2782b
YO
1481 skiplen = UNISKIP( uvc ); \
1482 foldlen -= skiplen; \
1483 uscan = foldbuf + skiplen; \
1484 } \
1485 break; \
1486 case trie_utf8: \
c80e42f3 1487 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO
1488 break; \
1489 case trie_plain: \
1490 uvc = (UV)*uc; \
1491 len = 1; \
1492 } \
1493 if (uvc < 256) { \
1494 charid = trie->charmap[ uvc ]; \
1495 } \
1496 else { \
1497 charid = 0; \
1498 if (widecharmap) { \
1499 SV** const svpp = hv_fetch(widecharmap, \
1500 (char*)&uvc, sizeof(UV), 0); \
1501 if (svpp) \
1502 charid = (U16)SvIV(*svpp); \
1503 } \
1504 } \
4cadc6a9
YO
1505} STMT_END
1506
4cadc6a9
YO
1507#define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1508STMT_START { \
1509 while (s <= e) { \
1510 if ( (CoNd) \
fac1af77 1511 && (ln == 1 || folder(s, pat_string, ln)) \
02d5137b 1512 && (reginfo->intuit || regtry(reginfo, &s)) )\
4cadc6a9
YO
1513 goto got_it; \
1514 s++; \
1515 } \
1516} STMT_END
1517
1518#define REXEC_FBC_UTF8_SCAN(CoDe) \
1519STMT_START { \
9a902117 1520 while (s < strend) { \
4cadc6a9 1521 CoDe \
9a902117 1522 s += UTF8SKIP(s); \
4cadc6a9
YO
1523 } \
1524} STMT_END
1525
1526#define REXEC_FBC_SCAN(CoDe) \
1527STMT_START { \
1528 while (s < strend) { \
1529 CoDe \
1530 s++; \
1531 } \
1532} STMT_END
1533
1534#define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1535REXEC_FBC_UTF8_SCAN( \
1536 if (CoNd) { \
02d5137b 1537 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
4cadc6a9
YO
1538 goto got_it; \
1539 else \
1540 tmp = doevery; \
1541 } \
1542 else \
1543 tmp = 1; \
1544)
1545
1546#define REXEC_FBC_CLASS_SCAN(CoNd) \
1547REXEC_FBC_SCAN( \
1548 if (CoNd) { \
02d5137b 1549 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
4cadc6a9
YO
1550 goto got_it; \
1551 else \
1552 tmp = doevery; \
1553 } \
1554 else \
1555 tmp = 1; \
1556)
1557
baa60164 1558#define REXEC_FBC_TRYIT \
02d5137b 1559if ((reginfo->intuit || regtry(reginfo, &s))) \
4cadc6a9
YO
1560 goto got_it
1561
e1d1eefb 1562#define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
baa60164 1563 if (utf8_target) { \
e1d1eefb
YO
1564 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1565 } \
1566 else { \
1567 REXEC_FBC_CLASS_SCAN(CoNd); \
d981ef24 1568 }
e1d1eefb 1569
baa60164 1570#define DUMP_EXEC_POS(li,s,doutf8) \
9d9163fb 1571 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
6d59b646 1572 startpos, doutf8)
786e8c11 1573
cfaf538b 1574
baa60164 1575#define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
9d9163fb 1576 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
cfaf538b
KW
1577 tmp = TEST_NON_UTF8(tmp); \
1578 REXEC_FBC_UTF8_SCAN( \
ce1d15d5 1579 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
cfaf538b
KW
1580 tmp = !tmp; \
1581 IF_SUCCESS; \
1582 } \
1583 else { \
1584 IF_FAIL; \
1585 } \
1586 ); \
1587
baa60164 1588#define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
9d9163fb 1589 if (s == reginfo->strbeg) { \
cfaf538b
KW
1590 tmp = '\n'; \
1591 } \
1592 else { \
9d9163fb 1593 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
3db24e1e
KW
1594 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1595 0, UTF8_ALLOW_DEFAULT); \
cfaf538b
KW
1596 } \
1597 tmp = TeSt1_UtF8; \
3db24e1e 1598 LOAD_UTF8_CHARCLASS_ALNUM(); \
cfaf538b 1599 REXEC_FBC_UTF8_SCAN( \
3db24e1e 1600 if (tmp == ! (TeSt2_UtF8)) { \
cfaf538b
KW
1601 tmp = !tmp; \
1602 IF_SUCCESS; \
1603 } \
1604 else { \
1605 IF_FAIL; \
1606 } \
1607 ); \
1608
63ac0dad
KW
1609/* The only difference between the BOUND and NBOUND cases is that
1610 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1611 * NBOUND. This is accomplished by passing it in either the if or else clause,
1612 * with the other one being empty */
1613#define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1dcf4a1b 1614 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
cfaf538b
KW
1615
1616#define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1dcf4a1b 1617 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
63ac0dad
KW
1618
1619#define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1dcf4a1b 1620 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
cfaf538b
KW
1621
1622#define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1dcf4a1b 1623 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
cfaf538b 1624
63ac0dad
KW
1625
1626/* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1627 * be passed in completely with the variable name being tested, which isn't
1628 * such a clean interface, but this is easier to read than it was before. We
1629 * are looking for the boundary (or non-boundary between a word and non-word
1630 * character. The utf8 and non-utf8 cases have the same logic, but the details
1631 * must be different. Find the "wordness" of the character just prior to this
1632 * one, and compare it with the wordness of this one. If they differ, we have
1633 * a boundary. At the beginning of the string, pretend that the previous
1634 * character was a new-line */
baa60164 1635#define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
63ac0dad 1636 if (utf8_target) { \
baa60164 1637 UTF8_CODE \
63ac0dad
KW
1638 } \
1639 else { /* Not utf8 */ \
9d9163fb 1640 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
63ac0dad
KW
1641 tmp = TEST_NON_UTF8(tmp); \
1642 REXEC_FBC_SCAN( \
1643 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1644 tmp = !tmp; \
1645 IF_SUCCESS; \
1646 } \
1647 else { \
1648 IF_FAIL; \
1649 } \
1650 ); \
1651 } \
baa60164 1652 if ((!prog->minlen && tmp) && (reginfo->intuit || regtry(reginfo, &s))) \
63ac0dad
KW
1653 goto got_it;
1654
786e8c11 1655/* We know what class REx starts with. Try to find this position... */
02d5137b 1656/* if reginfo->intuit, its a dryrun */
786e8c11
YO
1657/* annoyingly all the vars in this routine have different names from their counterparts
1658 in regmatch. /grrr */
1659
3c3eec57 1660STATIC char *
07be1b83 1661S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
f9176b44 1662 const char *strend, regmatch_info *reginfo)
a687059c 1663{
73104a1b
KW
1664 dVAR;
1665 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1666 char *pat_string; /* The pattern's exactish string */
1667 char *pat_end; /* ptr to end char of pat_string */
1668 re_fold_t folder; /* Function for computing non-utf8 folds */
1669 const U8 *fold_array; /* array for folding ords < 256 */
1670 STRLEN ln;
1671 STRLEN lnc;
73104a1b
KW
1672 U8 c1;
1673 U8 c2;
1674 char *e;
1675 I32 tmp = 1; /* Scratch variable? */
ba44c216 1676 const bool utf8_target = reginfo->is_utf8_target;
73104a1b 1677 UV utf8_fold_flags = 0;
f9176b44 1678 const bool is_utf8_pat = reginfo->is_utf8_pat;
3018b823
KW
1679 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1680 with a result inverts that result, as 0^1 =
1681 1 and 1^1 = 0 */
1682 _char_class_number classnum;
1683
73104a1b 1684 RXi_GET_DECL(prog,progi);
2f7f8cb1 1685
73104a1b 1686 PERL_ARGS_ASSERT_FIND_BYCLASS;
2f7f8cb1 1687
73104a1b
KW
1688 /* We know what class it must start with. */
1689 switch (OP(c)) {
1690 case ANYOF:
1691 if (utf8_target) {
1692 REXEC_FBC_UTF8_CLASS_SCAN(
3db24e1e 1693 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
73104a1b
KW
1694 }
1695 else {
1696 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1697 }
1698 break;
1699 case CANY:
1700 REXEC_FBC_SCAN(
02d5137b 1701 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
73104a1b
KW
1702 goto got_it;
1703 else
1704 tmp = doevery;
1705 );
1706 break;
1707
098b07d5
KW
1708 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1709 assert(! is_utf8_pat);
924ba076 1710 /* FALLTHROUGH */
73104a1b 1711 case EXACTFA:
984e6dd1 1712 if (is_utf8_pat || utf8_target) {
73104a1b
KW
1713 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1714 goto do_exactf_utf8;
1715 }
1716 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1717 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1718 goto do_exactf_non_utf8; /* isn't dealt with by these */
77a6d856 1719
2fdb7295
KW
1720 case EXACTF: /* This node only generated for non-utf8 patterns */
1721 assert(! is_utf8_pat);
73104a1b 1722 if (utf8_target) {
73104a1b
KW
1723 utf8_fold_flags = 0;
1724 goto do_exactf_utf8;
1725 }
1726 fold_array = PL_fold;
1727 folder = foldEQ;
1728 goto do_exactf_non_utf8;
1729
1730 case EXACTFL:
31f05a37 1731 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
cea315b6 1732 utf8_fold_flags = FOLDEQ_LOCALE;
73104a1b
KW
1733 goto do_exactf_utf8;
1734 }
1735 fold_array = PL_fold_locale;
1736 folder = foldEQ_locale;
1737 goto do_exactf_non_utf8;
3c760661 1738
73104a1b 1739 case EXACTFU_SS:
984e6dd1 1740 if (is_utf8_pat) {
73104a1b
KW
1741 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1742 }
1743 goto do_exactf_utf8;
16d951b7 1744
73104a1b 1745 case EXACTFU:
984e6dd1
DM
1746 if (is_utf8_pat || utf8_target) {
1747 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
73104a1b
KW
1748 goto do_exactf_utf8;
1749 }
fac1af77 1750
73104a1b
KW
1751 /* Any 'ss' in the pattern should have been replaced by regcomp,
1752 * so we don't have to worry here about this single special case
1753 * in the Latin1 range */
1754 fold_array = PL_fold_latin1;
1755 folder = foldEQ_latin1;
1756
924ba076 1757 /* FALLTHROUGH */
73104a1b
KW
1758
1759 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1760 are no glitches with fold-length differences
1761 between the target string and pattern */
1762
1763 /* The idea in the non-utf8 EXACTF* cases is to first find the
1764 * first character of the EXACTF* node and then, if necessary,
1765 * case-insensitively compare the full text of the node. c1 is the
1766 * first character. c2 is its fold. This logic will not work for
1767 * Unicode semantics and the german sharp ss, which hence should
1768 * not be compiled into a node that gets here. */
1769 pat_string = STRING(c);
1770 ln = STR_LEN(c); /* length to match in octets/bytes */
1771
1772 /* We know that we have to match at least 'ln' bytes (which is the
1773 * same as characters, since not utf8). If we have to match 3
1774 * characters, and there are only 2 availabe, we know without
1775 * trying that it will fail; so don't start a match past the
1776 * required minimum number from the far end */
ea3daa5d 1777 e = HOP3c(strend, -((SSize_t)ln), s);
73104a1b 1778
02d5137b 1779 if (reginfo->intuit && e < s) {
73104a1b
KW
1780 e = s; /* Due to minlen logic of intuit() */
1781 }
fac1af77 1782
73104a1b
KW
1783 c1 = *pat_string;
1784 c2 = fold_array[c1];
1785 if (c1 == c2) { /* If char and fold are the same */
1786 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1787 }
1788 else {
1789 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1790 }
1791 break;
fac1af77 1792
73104a1b
KW
1793 do_exactf_utf8:
1794 {
1795 unsigned expansion;
1796
1797 /* If one of the operands is in utf8, we can't use the simpler folding
1798 * above, due to the fact that many different characters can have the
1799 * same fold, or portion of a fold, or different- length fold */
1800 pat_string = STRING(c);
1801 ln = STR_LEN(c); /* length to match in octets/bytes */
1802 pat_end = pat_string + ln;
984e6dd1 1803 lnc = is_utf8_pat /* length to match in characters */
73104a1b
KW
1804 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1805 : ln;
1806
1807 /* We have 'lnc' characters to match in the pattern, but because of
1808 * multi-character folding, each character in the target can match
1809 * up to 3 characters (Unicode guarantees it will never exceed
1810 * this) if it is utf8-encoded; and up to 2 if not (based on the
1811 * fact that the Latin 1 folds are already determined, and the
1812 * only multi-char fold in that range is the sharp-s folding to
1813 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1814 * string character. Adjust lnc accordingly, rounding up, so that
1815 * if we need to match at least 4+1/3 chars, that really is 5. */
1816 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1817 lnc = (lnc + expansion - 1) / expansion;
1818
1819 /* As in the non-UTF8 case, if we have to match 3 characters, and
1820 * only 2 are left, it's guaranteed to fail, so don't start a
1821 * match that would require us to go beyond the end of the string
1822 */
ea3daa5d 1823 e = HOP3c(strend, -((SSize_t)lnc), s);
73104a1b 1824
02d5137b 1825 if (reginfo->intuit && e < s) {
73104a1b
KW
1826 e = s; /* Due to minlen logic of intuit() */
1827 }
0658cdde 1828
73104a1b
KW
1829 /* XXX Note that we could recalculate e to stop the loop earlier,
1830 * as the worst case expansion above will rarely be met, and as we
1831 * go along we would usually find that e moves further to the left.
1832 * This would happen only after we reached the point in the loop
1833 * where if there were no expansion we should fail. Unclear if
1834 * worth the expense */
1835
1836 while (s <= e) {
1837 char *my_strend= (char *)strend;
1838 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
984e6dd1 1839 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
02d5137b 1840 && (reginfo->intuit || regtry(reginfo, &s)) )
73104a1b
KW
1841 {
1842 goto got_it;
1843 }
1844 s += (utf8_target) ? UTF8SKIP(s) : 1;
1845 }
1846 break;
1847 }
1848 case BOUNDL:
0eb30aeb 1849 FBC_BOUND(isWORDCHAR_LC,
f4cd282c 1850 isWORDCHAR_LC_uvchr(tmp),
0eb30aeb 1851 isWORDCHAR_LC_utf8((U8*)s));
73104a1b
KW
1852 break;
1853 case NBOUNDL:
0eb30aeb 1854 FBC_NBOUND(isWORDCHAR_LC,
f4cd282c 1855 isWORDCHAR_LC_uvchr(tmp),
0eb30aeb 1856 isWORDCHAR_LC_utf8((U8*)s));
73104a1b
KW
1857 break;
1858 case BOUND:
1859 FBC_BOUND(isWORDCHAR,
0eb30aeb 1860 isWORDCHAR_uni(tmp),
03940dc2 1861 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
73104a1b
KW
1862 break;
1863 case BOUNDA:
1864 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1865 isWORDCHAR_A(tmp),
1866 isWORDCHAR_A((U8*)s));
1867 break;
1868 case NBOUND:
1869 FBC_NBOUND(isWORDCHAR,
0eb30aeb 1870 isWORDCHAR_uni(tmp),
03940dc2 1871 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
73104a1b
KW
1872 break;
1873 case NBOUNDA:
1874 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1875 isWORDCHAR_A(tmp),
1876 isWORDCHAR_A((U8*)s));
1877 break;
1878 case BOUNDU:
1879 FBC_BOUND(isWORDCHAR_L1,
0eb30aeb 1880 isWORDCHAR_uni(tmp),
03940dc2 1881 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
73104a1b
KW
1882 break;
1883 case NBOUNDU:
1884 FBC_NBOUND(isWORDCHAR_L1,
0eb30aeb 1885 isWORDCHAR_uni(tmp),
03940dc2 1886 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
73104a1b 1887 break;
73104a1b
KW
1888 case LNBREAK:
1889 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1890 is_LNBREAK_latin1_safe(s, strend)
1891 );
1892 break;
3018b823
KW
1893
1894 /* The argument to all the POSIX node types is the class number to pass to
1895 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1896
1897 case NPOSIXL:
1898 to_complement = 1;
1899 /* FALLTHROUGH */
1900
1901 case POSIXL:
3018b823
KW
1902 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1903 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
73104a1b 1904 break;
3018b823
KW
1905
1906 case NPOSIXD:
1907 to_complement = 1;
1908 /* FALLTHROUGH */
1909
1910 case POSIXD:
1911 if (utf8_target) {
1912 goto posix_utf8;
1913 }
1914 goto posixa;
1915
1916 case NPOSIXA:
1917 if (utf8_target) {
1918 /* The complement of something that matches only ASCII matches all
837226c8
KW
1919 * non-ASCII, plus everything in ASCII that isn't in the class. */
1920 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
3018b823
KW
1921 || ! _generic_isCC_A(*s, FLAGS(c)));
1922 break;
1923 }
1924
1925 to_complement = 1;
1926 /* FALLTHROUGH */
1927
73104a1b 1928 case POSIXA:
3018b823 1929 posixa:
73104a1b 1930 /* Don't need to worry about utf8, as it can match only a single
3018b823
KW
1931 * byte invariant character. */
1932 REXEC_FBC_CLASS_SCAN(
1933 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
73104a1b 1934 break;
3018b823
KW
1935
1936 case NPOSIXU:
1937 to_complement = 1;
1938 /* FALLTHROUGH */
1939
1940 case POSIXU:
1941 if (! utf8_target) {
1942 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1943 FLAGS(c))));
1944 }
1945 else {
1946
1947 posix_utf8:
1948 classnum = (_char_class_number) FLAGS(c);
1949 if (classnum < _FIRST_NON_SWASH_CC) {
1950 while (s < strend) {
1951
1952 /* We avoid loading in the swash as long as possible, but
1953 * should we have to, we jump to a separate loop. This
1954 * extra 'if' statement is what keeps this code from being
1955 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1956 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1957 goto found_above_latin1;
1958 }
1959 if ((UTF8_IS_INVARIANT(*s)
1960 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
1961 classnum)))
1962 || (UTF8_IS_DOWNGRADEABLE_START(*s)
1963 && to_complement ^ cBOOL(
94bb8c36
KW
1964 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
1965 *(s + 1)),
3018b823
KW
1966 classnum))))
1967 {
02d5137b 1968 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
3018b823
KW
1969 goto got_it;
1970 else {
1971 tmp = doevery;
1972 }
1973 }
1974 else {
1975 tmp = 1;
1976 }
1977 s += UTF8SKIP(s);
1978 }
1979 }
1980 else switch (classnum) { /* These classes are implemented as
1981 macros */
1982 case _CC_ENUM_SPACE: /* XXX would require separate code if we
1983 revert the change of \v matching this */
924ba076 1984 /* FALLTHROUGH */
3018b823
KW
1985
1986 case _CC_ENUM_PSXSPC:
1987 REXEC_FBC_UTF8_CLASS_SCAN(
1988 to_complement ^ cBOOL(isSPACE_utf8(s)));
1989 break;
1990
1991 case _CC_ENUM_BLANK:
1992 REXEC_FBC_UTF8_CLASS_SCAN(
1993 to_complement ^ cBOOL(isBLANK_utf8(s)));
1994 break;
1995
1996 case _CC_ENUM_XDIGIT:
1997 REXEC_FBC_UTF8_CLASS_SCAN(
1998 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
1999 break;
2000
2001 case _CC_ENUM_VERTSPACE:
2002 REXEC_FBC_UTF8_CLASS_SCAN(
2003 to_complement ^ cBOOL(isVERTWS_utf8(s)));
2004 break;
2005
2006 case _CC_ENUM_CNTRL:
2007 REXEC_FBC_UTF8_CLASS_SCAN(
2008 to_complement ^ cBOOL(isCNTRL_utf8(s)));
2009 break;
2010
2011 default:
2012 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
2013 assert(0); /* NOTREACHED */
2014 }
2015 }
2016 break;
2017
2018 found_above_latin1: /* Here we have to load a swash to get the result
2019 for the current code point */
2020 if (! PL_utf8_swash_ptrs[classnum]) {
2021 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2022 PL_utf8_swash_ptrs[classnum] =
2a16ac92
KW
2023 _core_swash_init("utf8",
2024 "",
2025 &PL_sv_undef, 1, 0,
2026 PL_XPosix_ptrs[classnum], &flags);
3018b823
KW
2027 }
2028
2029 /* This is a copy of the loop above for swash classes, though using the
2030 * FBC macro instead of being expanded out. Since we've loaded the
2031 * swash, we don't have to check for that each time through the loop */
2032 REXEC_FBC_UTF8_CLASS_SCAN(
2033 to_complement ^ cBOOL(_generic_utf8(
2034 classnum,
2035 s,
2036 swash_fetch(PL_utf8_swash_ptrs[classnum],
2037 (U8 *) s, TRUE))));
73104a1b
KW
2038 break;
2039
2040 case AHOCORASICKC:
2041 case AHOCORASICK:
2042 {
2043 DECL_TRIE_TYPE(c);
2044 /* what trie are we using right now */
2045 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2046 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2047 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2048
2049 const char *last_start = strend - trie->minlen;
6148ee25 2050#ifdef DEBUGGING
73104a1b 2051 const char *real_start = s;
6148ee25 2052#endif
73104a1b
KW
2053 STRLEN maxlen = trie->maxlen;
2054 SV *sv_points;
2055 U8 **points; /* map of where we were in the input string
2056 when reading a given char. For ASCII this
2057 is unnecessary overhead as the relationship
2058 is always 1:1, but for Unicode, especially
2059 case folded Unicode this is not true. */
2060 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2061 U8 *bitmap=NULL;
2062
2063
2064 GET_RE_DEBUG_FLAGS_DECL;
2065
2066 /* We can't just allocate points here. We need to wrap it in
2067 * an SV so it gets freed properly if there is a croak while
2068 * running the match */
2069 ENTER;
2070 SAVETMPS;
2071 sv_points=newSV(maxlen * sizeof(U8 *));
2072 SvCUR_set(sv_points,
2073 maxlen * sizeof(U8 *));
2074 SvPOK_on(sv_points);
2075 sv_2mortal(sv_points);
2076 points=(U8**)SvPV_nolen(sv_points );
2077 if ( trie_type != trie_utf8_fold
2078 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2079 {
2080 if (trie->bitmap)
2081 bitmap=(U8*)trie->bitmap;
2082 else
2083 bitmap=(U8*)ANYOF_BITMAP(c);
2084 }
2085 /* this is the Aho-Corasick algorithm modified a touch
2086 to include special handling for long "unknown char" sequences.
2087 The basic idea being that we use AC as long as we are dealing
2088 with a possible matching char, when we encounter an unknown char
2089 (and we have not encountered an accepting state) we scan forward
2090 until we find a legal starting char.
2091 AC matching is basically that of trie matching, except that when
2092 we encounter a failing transition, we fall back to the current
2093 states "fail state", and try the current char again, a process
2094 we repeat until we reach the root state, state 1, or a legal
2095 transition. If we fail on the root state then we can either
2096 terminate if we have reached an accepting state previously, or
2097 restart the entire process from the beginning if we have not.
2098
2099 */
2100 while (s <= last_start) {
2101 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2102 U8 *uc = (U8*)s;
2103 U16 charid = 0;
2104 U32 base = 1;
2105 U32 state = 1;
2106 UV uvc = 0;
2107 STRLEN len = 0;
2108 STRLEN foldlen = 0;
2109 U8 *uscan = (U8*)NULL;
2110 U8 *leftmost = NULL;
2111#ifdef DEBUGGING
2112 U32 accepted_word= 0;
786e8c11 2113#endif
73104a1b
KW
2114 U32 pointpos = 0;
2115
2116 while ( state && uc <= (U8*)strend ) {
2117 int failed=0;
2118 U32 word = aho->states[ state ].wordnum;
2119
2120 if( state==1 ) {
2121 if ( bitmap ) {
2122 DEBUG_TRIE_EXECUTE_r(
2123 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2124 dump_exec_pos( (char *)uc, c, strend, real_start,
2125 (char *)uc, utf8_target );
2126 PerlIO_printf( Perl_debug_log,
2127 " Scanning for legal start char...\n");
2128 }
2129 );
2130 if (utf8_target) {
2131 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2132 uc += UTF8SKIP(uc);
2133 }
2134 } else {
2135 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2136 uc++;
2137 }
786e8c11 2138 }
73104a1b 2139 s= (char *)uc;
07be1b83 2140 }
73104a1b
KW
2141 if (uc >(U8*)last_start) break;
2142 }
2143
2144 if ( word ) {
2145 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2146 if (!leftmost || lpos < leftmost) {
2147 DEBUG_r(accepted_word=word);
2148 leftmost= lpos;
7016d6eb 2149 }
73104a1b 2150 if (base==0) break;
7016d6eb 2151
73104a1b
KW
2152 }
2153 points[pointpos++ % maxlen]= uc;
2154 if (foldlen || uc < (U8*)strend) {
2155 REXEC_TRIE_READ_CHAR(trie_type, trie,
2156 widecharmap, uc,
2157 uscan, len, uvc, charid, foldlen,
2158 foldbuf, uniflags);
2159 DEBUG_TRIE_EXECUTE_r({
2160 dump_exec_pos( (char *)uc, c, strend,
2161 real_start, s, utf8_target);
2162 PerlIO_printf(Perl_debug_log,
2163 " Charid:%3u CP:%4"UVxf" ",
2164 charid, uvc);
2165 });
2166 }
2167 else {
2168 len = 0;
2169 charid = 0;
2170 }
07be1b83 2171
73104a1b
KW
2172
2173 do {
6148ee25 2174#ifdef DEBUGGING
73104a1b 2175 word = aho->states[ state ].wordnum;
6148ee25 2176#endif
73104a1b
KW
2177 base = aho->states[ state ].trans.base;
2178
2179 DEBUG_TRIE_EXECUTE_r({
2180 if (failed)
2181 dump_exec_pos( (char *)uc, c, strend, real_start,
2182 s, utf8_target );
2183 PerlIO_printf( Perl_debug_log,
2184 "%sState: %4"UVxf", word=%"UVxf,
2185 failed ? " Fail transition to " : "",
2186 (UV)state, (UV)word);
2187 });
2188 if ( base ) {
2189 U32 tmp;
2190 I32 offset;
2191 if (charid &&
2192 ( ((offset = base + charid
2193 - 1 - trie->uniquecharcount)) >= 0)
2194 && ((U32)offset < trie->lasttrans)
2195 && trie->trans[offset].check == state
2196 && (tmp=trie->trans[offset].next))
2197 {
2198 DEBUG_TRIE_EXECUTE_r(
2199 PerlIO_printf( Perl_debug_log," - legal\n"));
2200 state = tmp;
2201 break;
07be1b83
YO
2202 }
2203 else {
786e8c11 2204 DEBUG_TRIE_EXECUTE_r(
73104a1b 2205 PerlIO_printf( Perl_debug_log," - fail\n"));
786e8c11 2206 failed = 1;
73104a1b 2207 state = aho->fail[state];
07be1b83 2208 }
07be1b83 2209 }
73104a1b
KW
2210 else {
2211 /* we must be accepting here */
2212 DEBUG_TRIE_EXECUTE_r(
2213 PerlIO_printf( Perl_debug_log," - accepting\n"));
2214 failed = 1;
2215 break;
786e8c11 2216 }
73104a1b
KW
2217 } while(state);
2218 uc += len;
2219 if (failed) {
2220 if (leftmost)
2221 break;
2222 if (!state) state = 1;
07be1b83 2223 }
73104a1b
KW
2224 }
2225 if ( aho->states[ state ].wordnum ) {
2226 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2227 if (!leftmost || lpos < leftmost) {
2228 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2229 leftmost = lpos;
07be1b83
YO
2230 }
2231 }
73104a1b
KW
2232 if (leftmost) {
2233 s = (char*)leftmost;
2234 DEBUG_TRIE_EXECUTE_r({
2235 PerlIO_printf(
2236 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2237 (UV)accepted_word, (IV)(s - real_start)
2238 );
2239 });
02d5137b 2240 if (reginfo->intuit || regtry(reginfo, &s)) {
73104a1b
KW
2241 FREETMPS;
2242 LEAVE;
2243 goto got_it;
2244 }
2245 s = HOPc(s,1);
2246 DEBUG_TRIE_EXECUTE_r({
2247 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2248 });
2249 } else {
2250 DEBUG_TRIE_EXECUTE_r(
2251 PerlIO_printf( Perl_debug_log,"No match.\n"));
2252 break;
2253 }
2254 }
2255 FREETMPS;
2256 LEAVE;
2257 }
2258 break;
2259 default:
2260 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2261 break;
2262 }
2263 return 0;
2264 got_it:
2265 return s;
6eb5f6b9
JH
2266}
2267
60165aa4
DM
2268/* set RX_SAVED_COPY, RX_SUBBEG etc.
2269 * flags have same meanings as with regexec_flags() */
2270
749f4950
DM
2271static void
2272S_reg_set_capture_string(pTHX_ REGEXP * const rx,
60165aa4
DM
2273 char *strbeg,
2274 char *strend,
2275 SV *sv,
2276 U32 flags,
2277 bool utf8_target)
2278{
2279 struct regexp *const prog = ReANY(rx);
2280
60165aa4
DM
2281 if (flags & REXEC_COPY_STR) {
2282#ifdef PERL_ANY_COW
2283 if (SvCANCOW(sv)) {
2284 if (DEBUG_C_TEST) {
2285 PerlIO_printf(Perl_debug_log,
2286 "Copy on write: regexp capture, type %d\n",
2287 (int) SvTYPE(sv));
2288 }
5411a0e5
DM
2289 /* Create a new COW SV to share the match string and store
2290 * in saved_copy, unless the current COW SV in saved_copy
2291 * is valid and suitable for our purpose */
2292 if (( prog->saved_copy
2293 && SvIsCOW(prog->saved_copy)
2294 && SvPOKp(prog->saved_copy)
2295 && SvIsCOW(sv)
2296 && SvPOKp(sv)
2297 && SvPVX(sv) == SvPVX(prog->saved_copy)))
a76b0e90 2298 {
5411a0e5
DM
2299 /* just reuse saved_copy SV */
2300 if (RXp_MATCH_COPIED(prog)) {
2301 Safefree(prog->subbeg);
2302 RXp_MATCH_COPIED_off(prog);
2303 }
2304 }
2305 else {
2306 /* create new COW SV to share string */
a76b0e90
DM
2307 RX_MATCH_COPY_FREE(rx);
2308 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
a76b0e90 2309 }
5411a0e5
DM
2310 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2311 assert (SvPOKp(prog->saved_copy));
60165aa4
DM
2312 prog->sublen = strend - strbeg;
2313 prog->suboffset = 0;
2314 prog->subcoffset = 0;
2315 } else
2316#endif
2317 {
99a90e59
FC
2318 SSize_t min = 0;
2319 SSize_t max = strend - strbeg;
ea3daa5d 2320 SSize_t sublen;
60165aa4
DM
2321
2322 if ( (flags & REXEC_COPY_SKIP_POST)
e322109a 2323 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM
2324 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2325 ) { /* don't copy $' part of string */
2326 U32 n = 0;
2327 max = -1;
2328 /* calculate the right-most part of the string covered
2329 * by a capture. Due to look-ahead, this may be to
2330 * the right of $&, so we have to scan all captures */
2331 while (n <= prog->lastparen) {
2332 if (prog->offs[n].end > max)
2333 max = prog->offs[n].end;
2334 n++;
2335 }
2336 if (max == -1)
2337 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2338 ? prog->offs[0].start
2339 : 0;
2340 assert(max >= 0 && max <= strend - strbeg);
2341 }
2342
2343 if ( (flags & REXEC_COPY_SKIP_PRE)
e322109a 2344 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM
2345 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2346 ) { /* don't copy $` part of string */
2347 U32 n = 0;
2348 min = max;
2349 /* calculate the left-most part of the string covered
2350 * by a capture. Due to look-behind, this may be to
2351 * the left of $&, so we have to scan all captures */
2352 while (min && n <= prog->lastparen) {
2353 if ( prog->offs[n].start != -1
2354 && prog->offs[n].start < min)
2355 {
2356 min = prog->offs[n].start;
2357 }
2358 n++;
2359 }
2360 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2361 && min > prog->offs[0].end
2362 )
2363 min = prog->offs[0].end;
2364
2365 }
2366
2367 assert(min >= 0 && min <= max && min <= strend - strbeg);
2368 sublen = max - min;
2369
2370 if (RX_MATCH_COPIED(rx)) {
2371 if (sublen > prog->sublen)
2372 prog->subbeg =
2373 (char*)saferealloc(prog->subbeg, sublen+1);
2374 }
2375 else
2376 prog->subbeg = (char*)safemalloc(sublen+1);
2377 Copy(strbeg + min, prog->subbeg, sublen, char);
2378 prog->subbeg[sublen] = '\0';
2379 prog->suboffset = min;
2380 prog->sublen = sublen;
2381 RX_MATCH_COPIED_on(rx);
2382 }
2383 prog->subcoffset = prog->suboffset;
2384 if (prog->suboffset && utf8_target) {
2385 /* Convert byte offset to chars.
2386 * XXX ideally should only compute this if @-/@+
2387 * has been seen, a la PL_sawampersand ??? */
2388
2389 /* If there's a direct correspondence between the
2390 * string which we're matching and the original SV,
2391 * then we can use the utf8 len cache associated with
2392 * the SV. In particular, it means that under //g,
2393 * sv_pos_b2u() will use the previously cached
2394 * position to speed up working out the new length of
2395 * subcoffset, rather than counting from the start of
2396 * the string each time. This stops
2397 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2398 * from going quadratic */
2399 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
ea3daa5d
FC
2400 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2401 SV_GMAGIC|SV_CONST_RETURN);
60165aa4
DM
2402 else
2403 prog->subcoffset = utf8_length((U8*)strbeg,
2404 (U8*)(strbeg+prog->suboffset));
2405 }
2406 }
2407 else {
2408 RX_MATCH_COPY_FREE(rx);
2409 prog->subbeg = strbeg;
2410 prog->suboffset = 0;
2411 prog->subcoffset = 0;
2412 prog->sublen = strend - strbeg;
2413 }
2414}
2415
2416
2417
fae667d5 2418
6eb5f6b9
JH
2419/*
2420 - regexec_flags - match a regexp against a string
2421 */
2422I32
5aaab254 2423Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
ea3daa5d 2424 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
8fd1a950
DM
2425/* stringarg: the point in the string at which to begin matching */
2426/* strend: pointer to null at end of string */
2427/* strbeg: real beginning of string */
2428/* minend: end of match must be >= minend bytes after stringarg. */
2429/* sv: SV being matched: only used for utf8 flag, pos() etc; string
2430 * itself is accessed via the pointers above */
2431/* data: May be used for some additional optimizations.
d058ec57 2432 Currently unused. */
a340edde 2433/* flags: For optimizations. See REXEC_* in regexp.h */
8fd1a950 2434
6eb5f6b9 2435{
97aff369 2436 dVAR;
8d919b0a 2437 struct regexp *const prog = ReANY(rx);
5aaab254 2438 char *s;
eb578fdb 2439 regnode *c;
03c83e26 2440 char *startpos;
ea3daa5d
FC
2441 SSize_t minlen; /* must match at least this many chars */
2442 SSize_t dontbother = 0; /* how many characters not to try at end */
f2ed9b32 2443 const bool utf8_target = cBOOL(DO_UTF8(sv));
2757e526 2444 I32 multiline;
f8fc2ecf 2445 RXi_GET_DECL(prog,progi);
02d5137b
DM
2446 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2447 regmatch_info *const reginfo = &reginfo_buf;
e9105d30 2448 regexp_paren_pair *swap = NULL;
006f26b2 2449 I32 oldsave;
a3621e74
YO
2450 GET_RE_DEBUG_FLAGS_DECL;
2451
7918f24d 2452 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
9d4ba2ae 2453 PERL_UNUSED_ARG(data);
6eb5f6b9
JH
2454
2455 /* Be paranoid... */
03c83e26 2456 if (prog == NULL || stringarg == NULL) {
6eb5f6b9
JH
2457 Perl_croak(aTHX_ "NULL regexp parameter");
2458 return 0;
2459 }
2460
6c3fea77 2461 DEBUG_EXECUTE_r(
03c83e26 2462 debug_start_match(rx, utf8_target, stringarg, strend,
6c3fea77
DM
2463 "Matching");
2464 );
8adc0f72 2465
b342a604
DM
2466 startpos = stringarg;
2467
58430ea8 2468 if (prog->intflags & PREGf_GPOS_SEEN) {
d307c076
DM
2469 MAGIC *mg;
2470
fef7148b
DM
2471 /* set reginfo->ganch, the position where \G can match */
2472
2473 reginfo->ganch =
2474 (flags & REXEC_IGNOREPOS)
2475 ? stringarg /* use start pos rather than pos() */
2476 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
25fdce4a
FC
2477 /* Defined pos(): */
2478 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
fef7148b
DM
2479 : strbeg; /* pos() not defined; use start of string */
2480
2481 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
7b0eb0b8 2482 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg)));
fef7148b 2483
03c83e26
DM
2484 /* in the presence of \G, we may need to start looking earlier in
2485 * the string than the suggested start point of stringarg:
0b2c2a84 2486 * if prog->gofs is set, then that's a known, fixed minimum
03c83e26
DM
2487 * offset, such as
2488 * /..\G/: gofs = 2
2489 * /ab|c\G/: gofs = 1
2490 * or if the minimum offset isn't known, then we have to go back
2491 * to the start of the string, e.g. /w+\G/
2492 */
2bfbe302 2493
8e1490ee 2494 if (prog->intflags & PREGf_ANCH_GPOS) {
2bfbe302
DM
2495 startpos = reginfo->ganch - prog->gofs;
2496 if (startpos <
2497 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2498 {
2499 DEBUG_r(PerlIO_printf(Perl_debug_log,
2500 "fail: ganch-gofs before earliest possible start\n"));
2501 return 0;
2502 }
2503 }
2504 else if (prog->gofs) {
b342a604
DM
2505 if (startpos - prog->gofs < strbeg)
2506 startpos = strbeg;
2507 else
2508 startpos -= prog->gofs;
03c83e26 2509 }
58430ea8 2510 else if (prog->intflags & PREGf_GPOS_FLOAT)
b342a604 2511 startpos = strbeg;
03c83e26
DM
2512 }
2513
2514 minlen = prog->minlen;
b342a604 2515 if ((startpos + minlen) > strend || startpos < strbeg) {
03c83e26
DM
2516 DEBUG_r(PerlIO_printf(Perl_debug_log,
2517 "Regex match can't succeed, so not even tried\n"));
2518 return 0;
2519 }
2520
63a3746a
DM
2521 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2522 * which will call destuctors to reset PL_regmatch_state, free higher
2523 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2524 * regmatch_info_aux_eval */
2525
2526 oldsave = PL_savestack_ix;
2527
dfa77d06
DM
2528 s = startpos;
2529
e322109a 2530 if ((prog->extflags & RXf_USE_INTUIT)
7fadf4a7
DM
2531 && !(flags & REXEC_CHECKED))
2532 {
dfa77d06 2533 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
7fadf4a7 2534 flags, NULL);
dfa77d06 2535 if (!s)
7fadf4a7
DM
2536 return 0;
2537
e322109a 2538 if (prog->extflags & RXf_CHECK_ALL) {
7fadf4a7
DM
2539 /* we can match based purely on the result of INTUIT.
2540 * Set up captures etc just for $& and $-[0]
2541 * (an intuit-only match wont have $1,$2,..) */
2542 assert(!prog->nparens);
d5e7783a
DM
2543
2544 /* s/// doesn't like it if $& is earlier than where we asked it to
2545 * start searching (which can happen on something like /.\G/) */
2546 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2547 && (s < stringarg))
2548 {
2549 /* this should only be possible under \G */
58430ea8 2550 assert(prog->intflags & PREGf_GPOS_SEEN);
d5e7783a
DM
2551 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2552 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2553 goto phooey;
2554 }
2555
7fadf4a7
DM
2556 /* match via INTUIT shouldn't have any captures.
2557 * Let @-, @+, $^N know */
2558 prog->lastparen = prog->lastcloseparen = 0;
2559 RX_MATCH_UTF8_set(rx, utf8_target);
3ff69bd6
DM
2560 prog->offs[0].start = s - strbeg;
2561 prog->offs[0].end = utf8_target
2562 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2563 : s - strbeg + prog->minlenret;
7fadf4a7 2564 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 2565 S_reg_set_capture_string(aTHX_ rx,
7fadf4a7
DM
2566 strbeg, strend,
2567 sv, flags, utf8_target);
2568
7fadf4a7
DM
2569 return 1;
2570 }
2571 }
2572
6c3fea77 2573 multiline = prog->extflags & RXf_PMf_MULTILINE;
1de06328 2574
dfa77d06 2575 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
a3621e74 2576 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
a72c7584
JH
2577 "String too short [regexec_flags]...\n"));
2578 goto phooey;
1aa99e6b 2579 }
1de06328 2580
6eb5f6b9 2581 /* Check validity of program. */
f8fc2ecf 2582 if (UCHARAT(progi->program) != REG_MAGIC) {
6eb5f6b9
JH
2583 Perl_croak(aTHX_ "corrupted regexp program");
2584 }
2585
1738e041
DM
2586 RX_MATCH_TAINTED_off(rx);
2587
6c3fea77
DM
2588 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2589 reginfo->intuit = 0;
2590 reginfo->is_utf8_target = cBOOL(utf8_target);
02d5137b
DM
2591 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2592 reginfo->warned = FALSE;
9d9163fb 2593 reginfo->strbeg = strbeg;
02d5137b 2594 reginfo->sv = sv;
1cb48e53 2595 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
220db18a 2596 reginfo->strend = strend;
6eb5f6b9 2597 /* see how far we have to get to not match where we matched before */
fe3974be 2598 reginfo->till = stringarg + minend;
6eb5f6b9 2599
60779a30 2600 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
82c23608
FC
2601 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2602 S_cleanup_regmatch_info_aux has executed (registered by
2603 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2604 magic belonging to this SV.
2605 Not newSVsv, either, as it does not COW.
2606 */
60779a30 2607 assert(!IS_PADGV(sv));
82c23608 2608 reginfo->sv = newSV(0);
4cba5ac0 2609 SvSetSV_nosteal(reginfo->sv, sv);
82c23608
FC
2610 SAVEFREESV(reginfo->sv);
2611 }
2612
331b2dcc
DM
2613 /* reserve next 2 or 3 slots in PL_regmatch_state:
2614 * slot N+0: may currently be in use: skip it
2615 * slot N+1: use for regmatch_info_aux struct
2616 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2617 * slot N+3: ready for use by regmatch()
2618 */
bf2039a9 2619
331b2dcc
DM
2620 {
2621 regmatch_state *old_regmatch_state;
2622 regmatch_slab *old_regmatch_slab;
2623 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2624
2625 /* on first ever match, allocate first slab */
2626 if (!PL_regmatch_slab) {
2627 Newx(PL_regmatch_slab, 1, regmatch_slab);
2628 PL_regmatch_slab->prev = NULL;
2629 PL_regmatch_slab->next = NULL;
2630 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2631 }
bf2039a9 2632
331b2dcc
DM
2633 old_regmatch_state = PL_regmatch_state;
2634 old_regmatch_slab = PL_regmatch_slab;
bf2039a9 2635
331b2dcc
DM
2636 for (i=0; i <= max; i++) {
2637 if (i == 1)
2638 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2639 else if (i ==2)
2640 reginfo->info_aux_eval =
2641 reginfo->info_aux->info_aux_eval =
2642 &(PL_regmatch_state->u.info_aux_eval);
bf2039a9 2643
331b2dcc
DM
2644 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2645 PL_regmatch_state = S_push_slab(aTHX);
2646 }
bf2039a9 2647
331b2dcc
DM
2648 /* note initial PL_regmatch_state position; at end of match we'll
2649 * pop back to there and free any higher slabs */
bf2039a9 2650
331b2dcc
DM
2651 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2652 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2ac8ff4b 2653 reginfo->info_aux->poscache = NULL;
bf2039a9 2654
331b2dcc 2655 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
bf2039a9 2656
331b2dcc
DM
2657 if ((prog->extflags & RXf_EVAL_SEEN))
2658 S_setup_eval_state(aTHX_ reginfo);
2659 else
2660 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
bf2039a9 2661 }
d3aa529c 2662
6eb5f6b9 2663 /* If there is a "must appear" string, look for it. */
6eb5f6b9 2664
288b8c02 2665 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
e9105d30
GG
2666 /* We have to be careful. If the previous successful match
2667 was from this regex we don't want a subsequent partially
2668 successful match to clobber the old results.
2669 So when we detect this possibility we add a swap buffer
d8da0584
KW
2670 to the re, and switch the buffer each match. If we fail,
2671 we switch it back; otherwise we leave it swapped.
e9105d30
GG
2672 */
2673 swap = prog->offs;
2674 /* do we need a save destructor here for eval dies? */
2675 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
495f47a5
DM
2676 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2677 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2678 PTR2UV(prog),
2679 PTR2UV(swap),
2680 PTR2UV(prog->offs)
2681 ));
c74340f9 2682 }
6eb5f6b9
JH
2683
2684 /* Simplest case: anchored match need be tried only once. */
2685 /* [unless only anchor is BOL and multiline is set] */
8e1490ee 2686 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
3542935d 2687 if (s == startpos && regtry(reginfo, &s))
6eb5f6b9 2688 goto got_it;
8e1490ee 2689 else if (multiline || (prog->intflags & (PREGf_IMPLICIT | PREGf_ANCH_MBOL))) /* XXXX SBOL? */
6eb5f6b9
JH
2690 {
2691 char *end;
2692
2693 if (minlen)
2694 dontbother = minlen - 1;
1aa99e6b 2695 end = HOP3c(strend, -dontbother, strbeg) - 1;
6eb5f6b9 2696 /* for multiline we only have to try after newlines */
33b8afdf 2697 if (prog->check_substr || prog->check_utf8) {
92f3d482
YO
2698 /* because of the goto we can not easily reuse the macros for bifurcating the
2699 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2700 if (utf8_target) {
2701 if (s == startpos)
2702 goto after_try_utf8;
2703 while (1) {
02d5137b 2704 if (regtry(reginfo, &s)) {
92f3d482
YO
2705 goto got_it;
2706 }
2707 after_try_utf8:
2708 if (s > end) {
2709 goto phooey;
2710 }
2711 if (prog->extflags & RXf_USE_INTUIT) {
52a21eb3
DM
2712 s = re_intuit_start(rx, sv, strbeg,
2713 s + UTF8SKIP(s), strend, flags, NULL);
92f3d482
YO
2714 if (!s) {
2715 goto phooey;
2716 }
2717 }
2718 else {
2719 s += UTF8SKIP(s);
2720 }
2721 }
2722 } /* end search for check string in unicode */
2723 else {
2724 if (s == startpos) {
2725 goto after_try_latin;
2726 }
2727 while (1) {
02d5137b 2728 if (regtry(reginfo, &s)) {
92f3d482
YO
2729 goto got_it;
2730 }
2731 after_try_latin:
2732 if (s > end) {
2733 goto phooey;
2734 }
2735 if (prog->extflags & RXf_USE_INTUIT) {
52a21eb3
DM
2736 s = re_intuit_start(rx, sv, strbeg,
2737 s + 1, strend, flags, NULL);
92f3d482
YO
2738 if (!s) {
2739 goto phooey;
2740 }
2741 }
2742 else {
2743 s++;
2744 }
2745 }
2746 } /* end search for check string in latin*/
2747 } /* end search for check string */
2748 else { /* search for newline */
2749 if (s > startpos) {
2750 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
6eb5f6b9 2751 s--;
92f3d482 2752 }
21eede78
YO
2753 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2754 while (s <= end) { /* note it could be possible to match at the end of the string */
6eb5f6b9 2755 if (*s++ == '\n') { /* don't need PL_utf8skip here */
02d5137b 2756 if (regtry(reginfo, &s))
6eb5f6b9
JH
2757 goto got_it;
2758 }
92f3d482
YO
2759 }
2760 } /* end search for newline */
2761 } /* end anchored/multiline check string search */
6eb5f6b9 2762 goto phooey;
a8430a8b 2763 } else if (prog->intflags & PREGf_ANCH_GPOS)
f9f4320a 2764 {
a8430a8b
YO
2765 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
2766 assert(prog->intflags & PREGf_GPOS_SEEN);
2bfbe302
DM
2767 /* For anchored \G, the only position it can match from is
2768 * (ganch-gofs); we already set startpos to this above; if intuit
2769 * moved us on from there, we can't possibly succeed */
2770 assert(startpos == reginfo->ganch - prog->gofs);
2771 if (s == startpos && regtry(reginfo, &s))
6eb5f6b9
JH
2772 goto got_it;
2773 goto phooey;
2774 }
2775
2776 /* Messy cases: unanchored match. */
bbe252da 2777 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
6eb5f6b9 2778 /* we have /x+whatever/ */
984e6dd1 2779 /* it must be a one character string (XXXX Except is_utf8_pat?) */
33b8afdf 2780 char ch;
bf93d4cc
GS
2781#ifdef DEBUGGING
2782 int did_match = 0;
2783#endif
f2ed9b32 2784 if (utf8_target) {
7e0d5ad7
KW
2785 if (! prog->anchored_utf8) {
2786 to_utf8_substr(prog);
2787 }
2788 ch = SvPVX_const(prog->anchored_utf8)[0];
4cadc6a9 2789 REXEC_FBC_SCAN(
6eb5f6b9 2790 if (*s == ch) {
a3621e74 2791 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 2792 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH
2793 s += UTF8SKIP(s);
2794 while (s < strend && *s == ch)
2795 s += UTF8SKIP(s);
2796 }
4cadc6a9 2797 );
7e0d5ad7 2798
6eb5f6b9
JH
2799 }
2800 else {
7e0d5ad7
KW
2801 if (! prog->anchored_substr) {
2802 if (! to_byte_substr(prog)) {
6b54ddc5 2803 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2804 }
2805 }
2806 ch = SvPVX_const(prog->anchored_substr)[0];
4cadc6a9 2807 REXEC_FBC_SCAN(
6eb5f6b9 2808 if (*s == ch) {
a3621e74 2809 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 2810 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH
2811 s++;
2812 while (s < strend && *s == ch)
2813 s++;
2814 }
4cadc6a9 2815 );
6eb5f6b9 2816 }
a3621e74 2817 DEBUG_EXECUTE_r(if (!did_match)
bf93d4cc 2818 PerlIO_printf(Perl_debug_log,
b7953727
JH
2819 "Did not find anchored character...\n")
2820 );
6eb5f6b9 2821 }
a0714e2c
SS
2822 else if (prog->anchored_substr != NULL
2823 || prog->anchored_utf8 != NULL
2824 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
33b8afdf
JH
2825 && prog->float_max_offset < strend - s)) {
2826 SV *must;
ea3daa5d
FC
2827 SSize_t back_max;
2828 SSize_t back_min;
33b8afdf 2829 char *last;
6eb5f6b9 2830 char *last1; /* Last position checked before */
bf93d4cc
GS
2831#ifdef DEBUGGING
2832 int did_match = 0;
2833#endif
33b8afdf 2834 if (prog->anchored_substr || prog->anchored_utf8) {
7e0d5ad7
KW
2835 if (utf8_target) {
2836 if (! prog->anchored_utf8) {
2837 to_utf8_substr(prog);
2838 }
2839 must = prog->anchored_utf8;
2840 }
2841 else {
2842 if (! prog->anchored_substr) {
2843 if (! to_byte_substr(prog)) {
6b54ddc5 2844 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2845 }
2846 }
2847 must = prog->anchored_substr;
2848 }
33b8afdf
JH
2849 back_max = back_min = prog->anchored_offset;
2850 } else {
7e0d5ad7
KW
2851 if (utf8_target) {
2852 if (! prog->float_utf8) {
2853 to_utf8_substr(prog);
2854 }
2855 must = prog->float_utf8;
2856 }
2857 else {
2858 if (! prog->float_substr) {
2859 if (! to_byte_substr(prog)) {
6b54ddc5 2860 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2861 }
2862 }
2863 must = prog->float_substr;
2864 }
33b8afdf
JH
2865 back_max = prog->float_max_offset;
2866 back_min = prog->float_min_offset;
2867 }
1de06328 2868
1de06328
YO
2869 if (back_min<0) {
2870 last = strend;
2871 } else {
2872 last = HOP3c(strend, /* Cannot start after this */
ea3daa5d 2873 -(SSize_t)(CHR_SVLEN(must)
1de06328
YO
2874 - (SvTAIL(must) != 0) + back_min), strbeg);
2875 }
9d9163fb 2876 if (s > reginfo->strbeg)
6eb5f6b9
JH
2877 last1 = HOPc(s, -1);
2878 else
2879 last1 = s - 1; /* bogus */
2880
a0288114 2881 /* XXXX check_substr already used to find "s", can optimize if
6eb5f6b9 2882 check_substr==must. */
bf05793b 2883 dontbother = 0;
6eb5f6b9
JH
2884 strend = HOPc(strend, -dontbother);
2885 while ( (s <= last) &&
e50d57d4 2886 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
9041c2e3 2887 (unsigned char*)strend, must,
c33e64f0 2888 multiline ? FBMrf_MULTILINE : 0)) ) {
a3621e74 2889 DEBUG_EXECUTE_r( did_match = 1 );
6eb5f6b9
JH
2890 if (HOPc(s, -back_max) > last1) {
2891 last1 = HOPc(s, -back_min);
2892 s = HOPc(s, -back_max);
2893 }
2894 else {
9d9163fb
DM
2895 char * const t = (last1 >= reginfo->strbeg)
2896 ? HOPc(last1, 1) : last1 + 1;
6eb5f6b9
JH
2897
2898 last1 = HOPc(s, -back_min);
52657f30 2899 s = t;
6eb5f6b9 2900 }
f2ed9b32 2901 if (utf8_target) {
6eb5f6b9 2902 while (s <= last1) {
02d5137b 2903 if (regtry(reginfo, &s))
6eb5f6b9 2904 goto got_it;
7016d6eb
DM
2905 if (s >= last1) {
2906 s++; /* to break out of outer loop */
2907 break;
2908 }
2909 s += UTF8SKIP(s);
6eb5f6b9
JH
2910 }
2911 }
2912 else {
2913 while (s <= last1) {
02d5137b 2914 if (regtry(reginfo, &s))
6eb5f6b9
JH
2915 goto got_it;
2916 s++;
2917 }
2918 }
2919 }
ab3bbdeb 2920 DEBUG_EXECUTE_r(if (!did_match) {
f2ed9b32 2921 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
ab3bbdeb
YO
2922 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2923 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
33b8afdf 2924 ((must == prog->anchored_substr || must == prog->anchored_utf8)
bf93d4cc 2925 ? "anchored" : "floating"),
ab3bbdeb
YO
2926 quoted, RE_SV_TAIL(must));
2927 });
6eb5f6b9
JH
2928 goto phooey;
2929 }
f8fc2ecf 2930 else if ( (c = progi->regstclass) ) {
f14c76ed 2931 if (minlen) {
f8fc2ecf 2932 const OPCODE op = OP(progi->regstclass);
66e933ab 2933 /* don't bother with what can't match */
786e8c11 2934 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
f14c76ed
RGS
2935 strend = HOPc(strend, -(minlen - 1));
2936 }
a3621e74 2937 DEBUG_EXECUTE_r({
be8e71aa 2938 SV * const prop = sv_newmortal();
2395827c 2939 regprop(prog, prop, c, reginfo);
0df25f3d 2940 {
f2ed9b32 2941 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb 2942 s,strend-s,60);
0df25f3d 2943 PerlIO_printf(Perl_debug_log,
1c8f8eb1 2944 "Matching stclass %.*s against %s (%d bytes)\n",
e4f74956 2945 (int)SvCUR(prop), SvPVX_const(prop),
ab3bbdeb 2946 quoted, (int)(strend - s));
0df25f3d 2947 }
ffc61ed2 2948 });
f9176b44 2949 if (find_byclass(prog, c, s, strend, reginfo))
6eb5f6b9 2950 goto got_it;
07be1b83 2951 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
d6a28714
JH
2952 }
2953 else {
2954 dontbother = 0;
a0714e2c 2955 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
33b8afdf 2956 /* Trim the end. */
6af40bd7 2957 char *last= NULL;
33b8afdf 2958 SV* float_real;
c33e64f0
FC
2959 STRLEN len;
2960 const char *little;
33b8afdf 2961
7e0d5ad7
KW
2962 if (utf8_target) {
2963 if (! prog->float_utf8) {
2964 to_utf8_substr(prog);
2965 }
2966 float_real = prog->float_utf8;
2967 }
2968 else {
2969 if (! prog->float_substr) {
2970 if (! to_byte_substr(prog)) {
6b54ddc5 2971 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2972 }
2973 }
2974 float_real = prog->float_substr;
2975 }
d6a28714 2976
c33e64f0
FC
2977 little = SvPV_const(float_real, len);
2978 if (SvTAIL(float_real)) {
7f18ad16
KW
2979 /* This means that float_real contains an artificial \n on
2980 * the end due to the presence of something like this:
2981 * /foo$/ where we can match both "foo" and "foo\n" at the
2982 * end of the string. So we have to compare the end of the
2983 * string first against the float_real without the \n and
2984 * then against the full float_real with the string. We
2985 * have to watch out for cases where the string might be
2986 * smaller than the float_real or the float_real without
2987 * the \n. */
1a13b075
YO
2988 char *checkpos= strend - len;
2989 DEBUG_OPTIMISE_r(
2990 PerlIO_printf(Perl_debug_log,
2991 "%sChecking for float_real.%s\n",
2992 PL_colors[4], PL_colors[5]));
2993 if (checkpos + 1 < strbeg) {
7f18ad16
KW
2994 /* can't match, even if we remove the trailing \n
2995 * string is too short to match */
1a13b075
YO
2996 DEBUG_EXECUTE_r(
2997 PerlIO_printf(Perl_debug_log,
2998 "%sString shorter than required trailing substring, cannot match.%s\n",
2999 PL_colors[4], PL_colors[5]));
3000 goto phooey;
3001 } else if (memEQ(checkpos + 1, little, len - 1)) {
7f18ad16
KW
3002 /* can match, the end of the string matches without the
3003 * "\n" */
1a13b075
YO
3004 last = checkpos + 1;
3005 } else if (checkpos < strbeg) {
7f18ad16
KW
3006 /* cant match, string is too short when the "\n" is
3007 * included */
1a13b075
YO
3008 DEBUG_EXECUTE_r(
3009 PerlIO_printf(Perl_debug_log,
3010 "%sString does not contain required trailing substring, cannot match.%s\n",
3011 PL_colors[4], PL_colors[5]));
3012 goto phooey;
3013 } else if (!multiline) {
7f18ad16
KW
3014 /* non multiline match, so compare with the "\n" at the
3015 * end of the string */
1a13b075
YO
3016 if (memEQ(checkpos, little, len)) {
3017 last= checkpos;
3018 } else {
3019 DEBUG_EXECUTE_r(
3020 PerlIO_printf(Perl_debug_log,
3021 "%sString does not contain required trailing substring, cannot match.%s\n",
3022 PL_colors[4], PL_colors[5]));
3023 goto phooey;
3024 }
3025 } else {
7f18ad16
KW
3026 /* multiline match, so we have to search for a place
3027 * where the full string is located */
d6a28714 3028 goto find_last;
1a13b075 3029 }
c33e64f0 3030 } else {
d6a28714 3031 find_last:
9041c2e3 3032 if (len)
d6a28714 3033 last = rninstr(s, strend, little, little + len);
b8c5462f 3034 else
a0288114 3035 last = strend; /* matching "$" */
b8c5462f 3036 }
6af40bd7 3037 if (!last) {
7f18ad16
KW
3038 /* at one point this block contained a comment which was
3039 * probably incorrect, which said that this was a "should not
3040 * happen" case. Even if it was true when it was written I am
3041 * pretty sure it is not anymore, so I have removed the comment
3042 * and replaced it with this one. Yves */
6bda09f9
YO
3043 DEBUG_EXECUTE_r(
3044 PerlIO_printf(Perl_debug_log,
6af40bd7
YO
3045 "String does not contain required substring, cannot match.\n"
3046 ));
3047 goto phooey;
bf93d4cc 3048 }
d6a28714
JH
3049 dontbother = strend - last + prog->float_min_offset;
3050 }
3051 if (minlen && (dontbother < minlen))
3052 dontbother = minlen - 1;
3053 strend -= dontbother; /* this one's always in bytes! */
3054 /* We don't know much -- general case. */
f2ed9b32 3055 if (utf8_target) {
d6a28714 3056 for (;;) {
02d5137b 3057 if (regtry(reginfo, &s))
d6a28714
JH
3058 goto got_it;
3059 if (s >= strend)
3060 break;
b8c5462f 3061 s += UTF8SKIP(s);
d6a28714
JH
3062 };
3063 }
3064 else {
3065 do {
02d5137b 3066 if (regtry(reginfo, &s))
d6a28714
JH
3067 goto got_it;
3068 } while (s++ < strend);
3069 }
3070 }
3071
3072 /* Failure. */
3073 goto phooey;
3074
3075got_it:
d5e7783a
DM
3076 /* s/// doesn't like it if $& is earlier than where we asked it to
3077 * start searching (which can happen on something like /.\G/) */
3078 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3079 && (prog->offs[0].start < stringarg - strbeg))
3080 {
3081 /* this should only be possible under \G */
58430ea8 3082 assert(prog->intflags & PREGf_GPOS_SEEN);
d5e7783a
DM
3083 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3084 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3085 goto phooey;
3086 }
3087
495f47a5
DM
3088 DEBUG_BUFFERS_r(
3089 if (swap)
3090 PerlIO_printf(Perl_debug_log,
3091 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
3092 PTR2UV(prog),
3093 PTR2UV(swap)
3094 );
3095 );
e9105d30 3096 Safefree(swap);
d6a28714 3097
bf2039a9
DM
3098 /* clean up; this will trigger destructors that will free all slabs
3099 * above the current one, and cleanup the regmatch_info_aux
3100 * and regmatch_info_aux_eval sructs */
8adc0f72 3101
006f26b2
DM
3102 LEAVE_SCOPE(oldsave);
3103
5daac39c
NC
3104 if (RXp_PAREN_NAMES(prog))
3105 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
d6a28714 3106
0254aed9
DM
3107 RX_MATCH_UTF8_set(rx, utf8_target);
3108
d6a28714 3109 /* make sure $`, $&, $', and $digit will work later */
60165aa4 3110 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 3111 S_reg_set_capture_string(aTHX_ rx,
60165aa4
DM
3112 strbeg, reginfo->strend,
3113 sv, flags, utf8_target);
9041c2e3 3114
d6a28714
JH
3115 return 1;
3116
3117phooey:
a3621e74 3118 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
e4584336 3119 PL_colors[4], PL_colors[5]));
8adc0f72 3120
bf2039a9
DM
3121 /* clean up; this will trigger destructors that will free all slabs
3122 * above the current one, and cleanup the regmatch_info_aux
3123 * and regmatch_info_aux_eval sructs */
8adc0f72 3124
006f26b2
DM
3125 LEAVE_SCOPE(oldsave);
3126
e9105d30 3127 if (swap) {
c74340f9 3128 /* we failed :-( roll it back */
495f47a5
DM
3129 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
3130 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
3131 PTR2UV(prog),
3132 PTR2UV(prog->offs),
3133 PTR2UV(swap)
3134 ));
e9105d30
GG
3135 Safefree(prog->offs);
3136 prog->offs = swap;
3137 }
d6a28714
JH
3138 return 0;
3139}
3140
6bda09f9 3141
b3d298be 3142/* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
ec43f78b 3143 * Do inc before dec, in case old and new rex are the same */
baa60164 3144#define SET_reg_curpm(Re2) \
bf2039a9 3145 if (reginfo->info_aux_eval) { \
ec43f78b
DM
3146 (void)ReREFCNT_inc(Re2); \
3147 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3148 PM_SETRE((PL_reg_curpm), (Re2)); \
3149 }
3150
3151
d6a28714
JH
3152/*
3153 - regtry - try match at specific point
3154 */
3155STATIC I32 /* 0 failure, 1 success */
f73aaa43 3156S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
d6a28714 3157{
97aff369 3158 dVAR;
d6a28714 3159 CHECKPOINT lastcp;
288b8c02 3160 REGEXP *const rx = reginfo->prog;
8d919b0a 3161 regexp *const prog = ReANY(rx);
99a90e59 3162 SSize_t result;
f8fc2ecf 3163 RXi_GET_DECL(prog,progi);
a3621e74 3164 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
3165
3166 PERL_ARGS_ASSERT_REGTRY;
3167
24b23f37 3168 reginfo->cutpoint=NULL;
d6a28714 3169
9d9163fb 3170 prog->offs[0].start = *startposp - reginfo->strbeg;
d6a28714 3171 prog->lastparen = 0;
03994de8 3172 prog->lastcloseparen = 0;
d6a28714
JH
3173
3174 /* XXXX What this code is doing here?!!! There should be no need
b93070ed 3175 to do this again and again, prog->lastparen should take care of
3dd2943c 3176 this! --ilya*/
dafc8851
JH
3177
3178 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3179 * Actually, the code in regcppop() (which Ilya may be meaning by
b93070ed 3180 * prog->lastparen), is not needed at all by the test suite
225593e1
DM
3181 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3182 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3183 * Meanwhile, this code *is* needed for the
daf18116
JH
3184 * above-mentioned test suite tests to succeed. The common theme
3185 * on those tests seems to be returning null fields from matches.
225593e1 3186 * --jhi updated by dapm */
dafc8851 3187#if 1
d6a28714 3188 if (prog->nparens) {
b93070ed 3189 regexp_paren_pair *pp = prog->offs;
eb578fdb 3190 I32 i;
b93070ed 3191 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
f0ab9afb
NC
3192 ++pp;
3193 pp->start = -1;
3194 pp->end = -1;
d6a28714
JH
3195 }
3196 }
dafc8851 3197#endif
02db2b7b 3198 REGCP_SET(lastcp);
f73aaa43
DM
3199 result = regmatch(reginfo, *startposp, progi->program + 1);
3200 if (result != -1) {
3201 prog->offs[0].end = result;
d6a28714
JH
3202 return 1;
3203 }
24b23f37 3204 if (reginfo->cutpoint)
f73aaa43 3205 *startposp= reginfo->cutpoint;
02db2b7b 3206 REGCP_UNWIND(lastcp);
d6a28714
JH
3207 return 0;
3208}
3209
02db2b7b 3210
8ba1375e
MJD
3211#define sayYES goto yes
3212#define sayNO goto no
262b90c4 3213#define sayNO_SILENT goto no_silent
8ba1375e 3214
f9f4320a
YO
3215/* we dont use STMT_START/END here because it leads to
3216 "unreachable code" warnings, which are bogus, but distracting. */
3217#define CACHEsayNO \
c476f425 3218 if (ST.cache_mask) \
2ac8ff4b 3219 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
f9f4320a 3220 sayNO
3298f257 3221
a3621e74 3222/* this is used to determine how far from the left messages like
265c4333
YO
3223 'failed...' are printed. It should be set such that messages
3224 are inline with the regop output that created them.
a3621e74 3225*/
265c4333 3226#define REPORT_CODE_OFF 32
a3621e74
YO
3227
3228
40a82448
DM
3229#define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3230#define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
79a2a0e8
KW
3231#define CHRTEST_NOT_A_CP_1 -999
3232#define CHRTEST_NOT_A_CP_2 -998
9e137952 3233
5d9a96ca
DM
3234/* grab a new slab and return the first slot in it */
3235
3236STATIC regmatch_state *
3237S_push_slab(pTHX)
3238{
a35a87e7 3239#if PERL_VERSION < 9 && !defined(PERL_CORE)
54df2634
NC
3240 dMY_CXT;
3241#endif
5d9a96ca
DM
3242 regmatch_slab *s = PL_regmatch_slab->next;
3243 if (!s) {
3244 Newx(s, 1, regmatch_slab);
3245 s->prev = PL_regmatch_slab;
3246 s->next = NULL;
3247 PL_regmatch_slab->next = s;
3248 }
3249 PL_regmatch_slab = s;
86545054 3250 return SLAB_FIRST(s);
5d9a96ca 3251}
5b47454d 3252
95b24440 3253
40a82448
DM
3254/* push a new state then goto it */
3255
4d5016e5
DM
3256#define PUSH_STATE_GOTO(state, node, input) \
3257 pushinput = input; \
40a82448
DM
3258 scan = node; \
3259 st->resume_state = state; \
3260 goto push_state;
3261
3262/* push a new state with success backtracking, then goto it */
3263
4d5016e5
DM
3264#define PUSH_YES_STATE_GOTO(state, node, input) \
3265 pushinput = input; \
40a82448
DM
3266 scan = node; \
3267 st->resume_state = state; \
3268 goto push_yes_state;
3269
aa283a38 3270
aa283a38 3271
4d5016e5 3272
d6a28714 3273/*
95b24440 3274
bf1f174e
DM
3275regmatch() - main matching routine
3276
3277This is basically one big switch statement in a loop. We execute an op,
3278set 'next' to point the next op, and continue. If we come to a point which
3279we may need to backtrack to on failure such as (A|B|C), we push a
3280backtrack state onto the backtrack stack. On failure, we pop the top
3281state, and re-enter the loop at the state indicated. If there are no more
3282states to pop, we return failure.
3283
3284Sometimes we also need to backtrack on success; for example /A+/, where
3285after successfully matching one A, we need to go back and try to
3286match another one; similarly for lookahead assertions: if the assertion
3287completes successfully, we backtrack to the state just before the assertion
3288and then carry on. In these cases, the pushed state is marked as
3289'backtrack on success too'. This marking is in fact done by a chain of
3290pointers, each pointing to the previous 'yes' state. On success, we pop to
3291the nearest yes state, discarding any intermediate failure-only states.
3292Sometimes a yes state is pushed just to force some cleanup code to be
3293called at the end of a successful match or submatch; e.g. (??{$re}) uses
3294it to free the inner regex.
3295
3296Note that failure backtracking rewinds the cursor position, while
3297success backtracking leaves it alone.
3298
3299A pattern is complete when the END op is executed, while a subpattern
3300such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3301ops trigger the "pop to last yes state if any, otherwise return true"
3302behaviour.
3303
3304A common convention in this function is to use A and B to refer to the two
3305subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3306the subpattern to be matched possibly multiple times, while B is the entire
3307rest of the pattern. Variable and state names reflect this convention.
3308
3309The states in the main switch are the union of ops and failure/success of
3310substates associated with with that op. For example, IFMATCH is the op
3311that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3312'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3313successfully matched A and IFMATCH_A_fail is a state saying that we have
3314just failed to match A. Resume states always come in pairs. The backtrack
3315state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3316at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3317on success or failure.
3318
3319The struct that holds a backtracking state is actually a big union, with
3320one variant for each major type of op. The variable st points to the
3321top-most backtrack struct. To make the code clearer, within each
3322block of code we #define ST to alias the relevant union.
3323
3324Here's a concrete example of a (vastly oversimplified) IFMATCH
3325implementation:
3326
3327 switch (state) {
3328 ....
3329
3330#define ST st->u.ifmatch
3331
3332 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3333 ST.foo = ...; // some state we wish to save
95b24440 3334 ...
bf1f174e
DM
3335 // push a yes backtrack state with a resume value of
3336 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3337 // first node of A:
4d5016e5 3338 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
bf1f174e
DM
3339 // NOTREACHED
3340
3341 case IFMATCH_A: // we have successfully executed A; now continue with B
3342 next = B;
3343 bar = ST.foo; // do something with the preserved value
3344 break;
3345
3346 case IFMATCH_A_fail: // A failed, so the assertion failed
3347 ...; // do some housekeeping, then ...
3348 sayNO; // propagate the failure
3349
3350#undef ST
95b24440 3351
bf1f174e
DM
3352 ...
3353 }
95b24440 3354
bf1f174e
DM
3355For any old-timers reading this who are familiar with the old recursive
3356approach, the code above is equivalent to:
95b24440 3357
bf1f174e
DM
3358 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3359 {
3360 int foo = ...
95b24440 3361 ...
bf1f174e
DM
3362 if (regmatch(A)) {
3363 next = B;
3364 bar = foo;
3365 break;
95b24440 3366 }
bf1f174e
DM
3367 ...; // do some housekeeping, then ...
3368 sayNO; // propagate the failure
95b24440 3369 }
bf1f174e
DM
3370
3371The topmost backtrack state, pointed to by st, is usually free. If you
3372want to claim it, populate any ST.foo fields in it with values you wish to
3373save, then do one of
3374
4d5016e5
DM
3375 PUSH_STATE_GOTO(resume_state, node, newinput);
3376 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
bf1f174e
DM
3377
3378which sets that backtrack state's resume value to 'resume_state', pushes a
3379new free entry to the top of the backtrack stack, then goes to 'node'.
3380On backtracking, the free slot is popped, and the saved state becomes the
3381new free state. An ST.foo field in this new top state can be temporarily
3382accessed to retrieve values, but once the main loop is re-entered, it
3383becomes available for reuse.
3384
3385Note that the depth of the backtrack stack constantly increases during the
3386left-to-right execution of the pattern, rather than going up and down with
3387the pattern nesting. For example the stack is at its maximum at Z at the
3388end of the pattern, rather than at X in the following:
3389
3390 /(((X)+)+)+....(Y)+....Z/
3391
3392The only exceptions to this are lookahead/behind assertions and the cut,
3393(?>A), which pop all the backtrack states associated with A before
3394continuing.
3395
486ec47a 3396Backtrack state structs are allocated in slabs of about 4K in size.
bf1f174e
DM
3397PL_regmatch_state and st always point to the currently active state,
3398and PL_regmatch_slab points to the slab currently containing
3399PL_regmatch_state. The first time regmatch() is called, the first slab is
3400allocated, and is never freed until interpreter destruction. When the slab
3401is full, a new one is allocated and chained to the end. At exit from
3402regmatch(), slabs allocated since entry are freed.
3403
3404*/
95b24440 3405
40a82448 3406
5bc10b2c 3407#define DEBUG_STATE_pp(pp) \
265c4333 3408 DEBUG_STATE_r({ \
baa60164 3409 DUMP_EXEC_POS(locinput, scan, utf8_target); \
5bc10b2c 3410 PerlIO_printf(Perl_debug_log, \
5d458dd8 3411 " %*s"pp" %s%s%s%s%s\n", \
5bc10b2c 3412 depth*2, "", \
baa60164 3413 PL_reg_name[st->resume_state], \
5d458dd8
YO
3414 ((st==yes_state||st==mark_state) ? "[" : ""), \
3415 ((st==yes_state) ? "Y" : ""), \
3416 ((st==mark_state) ? "M" : ""), \
3417 ((st==yes_state||st==mark_state) ? "]" : "") \
3418 ); \
265c4333 3419 });
5bc10b2c 3420
40a82448 3421
3dab1dad 3422#define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
95b24440 3423
3df15adc 3424#ifdef DEBUGGING
5bc10b2c 3425
ab3bbdeb 3426STATIC void
f2ed9b32 3427S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
ab3bbdeb
YO
3428 const char *start, const char *end, const char *blurb)
3429{
efd26800 3430 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
7918f24d
NC
3431
3432 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3433
ab3bbdeb
YO
3434 if (!PL_colorset)
3435 reginitcolors();
3436 {
3437 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
d2c6dc5e 3438 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
ab3bbdeb 3439
f2ed9b32 3440 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb
YO
3441 start, end - start, 60);
3442
3443 PerlIO_printf(Perl_debug_log,
3444 "%s%s REx%s %s against %s\n",
3445 PL_colors[4], blurb, PL_colors[5], s0, s1);
3446
f2ed9b32 3447 if (utf8_target||utf8_pat)
1de06328
YO
3448 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3449 utf8_pat ? "pattern" : "",
f2ed9b32
KW
3450 utf8_pat && utf8_target ? " and " : "",
3451 utf8_target ? "string" : ""
ab3bbdeb
YO
3452 );
3453 }
3454}
3df15adc
YO
3455
3456STATIC void
786e8c11
YO
3457S_dump_exec_pos(pTHX_ const char *locinput,
3458 const regnode *scan,
3459 const char *loc_regeol,
3460 const char *loc_bostr,
3461 const char *loc_reg_starttry,
f2ed9b32 3462 const bool utf8_target)
07be1b83 3463{
786e8c11 3464 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
07be1b83 3465 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
786e8c11 3466 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
07be1b83
YO
3467 /* The part of the string before starttry has one color
3468 (pref0_len chars), between starttry and current
3469 position another one (pref_len - pref0_len chars),
3470 after the current position the third one.
3471 We assume that pref0_len <= pref_len, otherwise we
3472 decrease pref0_len. */
786e8c11
YO
3473 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3474 ? (5 + taill) - l : locinput - loc_bostr;
07be1b83
YO
3475 int pref0_len;
3476
7918f24d
NC
3477 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3478
f2ed9b32 3479 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
07be1b83 3480 pref_len++;
786e8c11
YO
3481 pref0_len = pref_len - (locinput - loc_reg_starttry);
3482 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3483 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3484 ? (5 + taill) - pref_len : loc_regeol - locinput);
f2ed9b32 3485 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
07be1b83
YO
3486 l--;
3487 if (pref0_len < 0)
3488 pref0_len = 0;
3489 if (pref0_len > pref_len)
3490 pref0_len = pref_len;
3491 {
f2ed9b32 3492 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
0df25f3d 3493
ab3bbdeb 3494 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
1de06328 3495 (locinput - pref_len),pref0_len, 60, 4, 5);
0df25f3d 3496
ab3bbdeb 3497 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3df15adc 3498 (locinput - pref_len + pref0_len),
1de06328 3499 pref_len - pref0_len, 60, 2, 3);
0df25f3d 3500
ab3bbdeb 3501 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
1de06328 3502 locinput, loc_regeol - locinput, 10, 0, 1);
0df25f3d 3503
1de06328 3504 const STRLEN tlen=len0+len1+len2;
3df15adc 3505 PerlIO_printf(Perl_debug_log,
ab3bbdeb 3506 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
786e8c11 3507 (IV)(locinput - loc_bostr),
07be1b83 3508 len0, s0,
07be1b83 3509 len1, s1,
07be1b83 3510 (docolor ? "" : "> <"),
07be1b83 3511 len2, s2,
f9f4320a 3512 (int)(tlen > 19 ? 0 : 19 - tlen),
07be1b83
YO
3513 "");
3514 }
3515}
3df15adc 3516
07be1b83
YO
3517#endif
3518
0a4db386
YO
3519/* reg_check_named_buff_matched()
3520 * Checks to see if a named buffer has matched. The data array of
3521 * buffer numbers corresponding to the buffer is expected to reside
3522 * in the regexp->data->data array in the slot stored in the ARG() of
3523 * node involved. Note that this routine doesn't actually care about the
3524 * name, that information is not preserved from compilation to execution.
3525 * Returns the index of the leftmost defined buffer with the given name
3526 * or 0 if non of the buffers matched.
3527 */
3528STATIC I32
7918f24d
NC
3529S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3530{
0a4db386 3531 I32 n;
f8fc2ecf 3532 RXi_GET_DECL(rex,rexi);
ad64d0ec 3533 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
0a4db386 3534 I32 *nums=(I32*)SvPVX(sv_dat);
7918f24d
NC
3535
3536 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3537
0a4db386 3538 for ( n=0; n<SvIVX(sv_dat); n++ ) {
b93070ed
DM
3539 if ((I32)rex->lastparen >= nums[n] &&
3540 rex->offs[nums[n]].end != -1)
0a4db386
YO
3541 {
3542 return nums[n];
3543 }
3544 }