This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regexec.c: Exchange 2 lines of code and add comments
[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{
a3b680e6 291 const int retval = PL_savestack_ix;
92da3157
DM
292 const int paren_elems_to_push =
293 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
e0fa7e2b
NC
294 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
295 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
87c0511b 296 I32 p;
40a82448 297 GET_RE_DEBUG_FLAGS_DECL;
a0d0e21e 298
b93070ed
DM
299 PERL_ARGS_ASSERT_REGCPPUSH;
300
e49a9654 301 if (paren_elems_to_push < 0)
e8a85d26
JH
302 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
303 (int)paren_elems_to_push, (int)maxopenparen,
304 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
e49a9654 305
e0fa7e2b
NC
306 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
307 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
5df417d0 308 " out of range (%lu-%ld)",
92da3157
DM
309 total_elems,
310 (unsigned long)maxopenparen,
311 (long)parenfloor);
e0fa7e2b 312
620d5b66 313 SSGROW(total_elems + REGCP_FRAME_ELEMS);
7f69552c 314
495f47a5 315 DEBUG_BUFFERS_r(
92da3157 316 if ((int)maxopenparen > (int)parenfloor)
495f47a5
DM
317 PerlIO_printf(Perl_debug_log,
318 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
319 PTR2UV(rex),
320 PTR2UV(rex->offs)
321 );
322 );
92da3157 323 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
b1ce53c5 324/* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
99a90e59
FC
325 SSPUSHIV(rex->offs[p].end);
326 SSPUSHIV(rex->offs[p].start);
1ca2007e 327 SSPUSHINT(rex->offs[p].start_tmp);
e7707071 328 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
495f47a5
DM
329 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
330 (UV)p,
331 (IV)rex->offs[p].start,
332 (IV)rex->offs[p].start_tmp,
333 (IV)rex->offs[p].end
40a82448 334 ));
a0d0e21e 335 }
b1ce53c5 336/* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
92da3157 337 SSPUSHINT(maxopenparen);
b93070ed
DM
338 SSPUSHINT(rex->lastparen);
339 SSPUSHINT(rex->lastcloseparen);
e0fa7e2b 340 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
41123dfd 341
a0d0e21e
LW
342 return retval;
343}
344
c277df42 345/* These are needed since we do not localize EVAL nodes: */
ab3bbdeb
YO
346#define REGCP_SET(cp) \
347 DEBUG_STATE_r( \
ab3bbdeb 348 PerlIO_printf(Perl_debug_log, \
e4f74956 349 " Setting an EVAL scope, savestack=%"IVdf"\n", \
ab3bbdeb
YO
350 (IV)PL_savestack_ix)); \
351 cp = PL_savestack_ix
c3464db5 352
ab3bbdeb 353#define REGCP_UNWIND(cp) \
e4f74956 354 DEBUG_STATE_r( \
ab3bbdeb 355 if (cp != PL_savestack_ix) \
e4f74956
YO
356 PerlIO_printf(Perl_debug_log, \
357 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
ab3bbdeb
YO
358 (IV)(cp), (IV)PL_savestack_ix)); \
359 regcpblow(cp)
c277df42 360
a8d1f4b4
DM
361#define UNWIND_PAREN(lp, lcp) \
362 for (n = rex->lastparen; n > lp; n--) \
363 rex->offs[n].end = -1; \
364 rex->lastparen = n; \
365 rex->lastcloseparen = lcp;
366
367
f067efbf 368STATIC void
92da3157 369S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
a0d0e21e 370{
e0fa7e2b 371 UV i;
87c0511b 372 U32 paren;
a3621e74
YO
373 GET_RE_DEBUG_FLAGS_DECL;
374
7918f24d
NC
375 PERL_ARGS_ASSERT_REGCPPOP;
376
b1ce53c5 377 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
c6bf6a65 378 i = SSPOPUV;
e0fa7e2b
NC
379 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
380 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
b93070ed
DM
381 rex->lastcloseparen = SSPOPINT;
382 rex->lastparen = SSPOPINT;
92da3157 383 *maxopenparen_p = SSPOPINT;
b1ce53c5 384
620d5b66 385 i -= REGCP_OTHER_ELEMS;
b1ce53c5 386 /* Now restore the parentheses context. */
495f47a5
DM
387 DEBUG_BUFFERS_r(
388 if (i || rex->lastparen + 1 <= rex->nparens)
389 PerlIO_printf(Perl_debug_log,
390 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
391 PTR2UV(rex),
392 PTR2UV(rex->offs)
393 );
394 );
92da3157 395 paren = *maxopenparen_p;
620d5b66 396 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
ea3daa5d 397 SSize_t tmps;
1ca2007e 398 rex->offs[paren].start_tmp = SSPOPINT;
99a90e59
FC
399 rex->offs[paren].start = SSPOPIV;
400 tmps = SSPOPIV;
b93070ed
DM
401 if (paren <= rex->lastparen)
402 rex->offs[paren].end = tmps;
495f47a5
DM
403 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
404 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
405 (UV)paren,
406 (IV)rex->offs[paren].start,
407 (IV)rex->offs[paren].start_tmp,
408 (IV)rex->offs[paren].end,
409 (paren > rex->lastparen ? "(skipped)" : ""));
c277df42 410 );
87c0511b 411 paren--;
a0d0e21e 412 }
daf18116 413#if 1
dafc8851
JH
414 /* It would seem that the similar code in regtry()
415 * already takes care of this, and in fact it is in
416 * a better location to since this code can #if 0-ed out
417 * but the code in regtry() is needed or otherwise tests
418 * requiring null fields (pat.t#187 and split.t#{13,14}
daf18116
JH
419 * (as of patchlevel 7877) will fail. Then again,
420 * this code seems to be necessary or otherwise
225593e1
DM
421 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
422 * --jhi updated by dapm */
b93070ed 423 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
92da3157 424 if (i > *maxopenparen_p)
b93070ed
DM
425 rex->offs[i].start = -1;
426 rex->offs[i].end = -1;
495f47a5
DM
427 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
428 " \\%"UVuf": %s ..-1 undeffing\n",
429 (UV)i,
92da3157 430 (i > *maxopenparen_p) ? "-1" : " "
495f47a5 431 ));
a0d0e21e 432 }
dafc8851 433#endif
a0d0e21e
LW
434}
435
74088413
DM
436/* restore the parens and associated vars at savestack position ix,
437 * but without popping the stack */
438
439STATIC void
92da3157 440S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
74088413
DM
441{
442 I32 tmpix = PL_savestack_ix;
443 PL_savestack_ix = ix;
92da3157 444 regcppop(rex, maxopenparen_p);
74088413
DM
445 PL_savestack_ix = tmpix;
446}
447
02db2b7b 448#define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
a0d0e21e 449
31c7f561
KW
450STATIC bool
451S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
452{
453 /* Returns a boolean as to whether or not 'character' is a member of the
454 * Posix character class given by 'classnum' that should be equivalent to a
455 * value in the typedef '_char_class_number'.
456 *
457 * Ideally this could be replaced by a just an array of function pointers
458 * to the C library functions that implement the macros this calls.
459 * However, to compile, the precise function signatures are required, and
460 * these may vary from platform to to platform. To avoid having to figure
461 * out what those all are on each platform, I (khw) am using this method,
7aee35ff
KW
462 * which adds an extra layer of function call overhead (unless the C
463 * optimizer strips it away). But we don't particularly care about
464 * performance with locales anyway. */
31c7f561
KW
465
466 switch ((_char_class_number) classnum) {
15861f94 467 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
31c7f561 468 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
e8d596e0
KW
469 case _CC_ENUM_ASCII: return isASCII_LC(character);
470 case _CC_ENUM_BLANK: return isBLANK_LC(character);
b0d691b2
KW
471 case _CC_ENUM_CASED: return isLOWER_LC(character)
472 || isUPPER_LC(character);
e8d596e0 473 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
31c7f561
KW
474 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
475 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
476 case _CC_ENUM_LOWER: return isLOWER_LC(character);
477 case _CC_ENUM_PRINT: return isPRINT_LC(character);
e8d596e0 478 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
31c7f561 479 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
e8d596e0 480 case _CC_ENUM_SPACE: return isSPACE_LC(character);
31c7f561
KW
481 case _CC_ENUM_UPPER: return isUPPER_LC(character);
482 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
31c7f561 483 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
31c7f561
KW
484 default: /* VERTSPACE should never occur in locales */
485 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
486 }
487
a25b5927 488 assert(0); /* NOTREACHED */
31c7f561
KW
489 return FALSE;
490}
491
3018b823
KW
492STATIC bool
493S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
494{
495 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
496 * 'character' is a member of the Posix character class given by 'classnum'
497 * that should be equivalent to a value in the typedef
498 * '_char_class_number'.
499 *
500 * This just calls isFOO_lc on the code point for the character if it is in
501 * the range 0-255. Outside that range, all characters avoid Unicode
502 * rules, ignoring any locale. So use the Unicode function if this class
503 * requires a swash, and use the Unicode macro otherwise. */
504
505 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
506
507 if (UTF8_IS_INVARIANT(*character)) {
508 return isFOO_lc(classnum, *character);
509 }
510 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
511 return isFOO_lc(classnum,
94bb8c36 512 TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1)));
3018b823
KW
513 }
514
515 if (classnum < _FIRST_NON_SWASH_CC) {
516
517 /* Initialize the swash unless done already */
518 if (! PL_utf8_swash_ptrs[classnum]) {
519 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2a16ac92
KW
520 PL_utf8_swash_ptrs[classnum] =
521 _core_swash_init("utf8",
522 "",
523 &PL_sv_undef, 1, 0,
524 PL_XPosix_ptrs[classnum], &flags);
3018b823
KW
525 }
526
92a2046b
KW
527 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
528 character,
529 TRUE /* is UTF */ ));
3018b823
KW
530 }
531
532 switch ((_char_class_number) classnum) {
533 case _CC_ENUM_SPACE:
534 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
535
536 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
537 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
538 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
539 default: return 0; /* Things like CNTRL are always
540 below 256 */
541 }
542
a25b5927 543 assert(0); /* NOTREACHED */
3018b823
KW
544 return FALSE;
545}
546
a687059c 547/*
e50aee73 548 * pregexec and friends
a687059c
LW
549 */
550
76234dfb 551#ifndef PERL_IN_XSUB_RE
a687059c 552/*
c277df42 553 - pregexec - match a regexp against a string
a687059c 554 */
c277df42 555I32
5aaab254 556Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
ea3daa5d 557 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
8fd1a950
DM
558/* stringarg: the point in the string at which to begin matching */
559/* strend: pointer to null at end of string */
560/* strbeg: real beginning of string */
561/* minend: end of match must be >= minend bytes after stringarg. */
562/* screamer: SV being matched: only used for utf8 flag, pos() etc; string
563 * itself is accessed via the pointers above */
564/* nosave: For optimizations. */
c277df42 565{
7918f24d
NC
566 PERL_ARGS_ASSERT_PREGEXEC;
567
c277df42 568 return
9041c2e3 569 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
c277df42
IZ
570 nosave ? 0 : REXEC_COPY_STR);
571}
76234dfb 572#endif
22e551b9 573
cad2e5aa 574
6eb5f6b9 575
1a4edc3c
DM
576/* re_intuit_start():
577 *
578 * Based on some optimiser hints, try to find the earliest position in the
579 * string where the regex could match.
580 *
581 * rx: the regex to match against
582 * sv: the SV being matched: only used for utf8 flag; the string
583 * itself is accessed via the pointers below. Note that on
584 * something like an overloaded SV, SvPOK(sv) may be false
585 * and the string pointers may point to something unrelated to
586 * the SV itself.
587 * strbeg: real beginning of string
588 * strpos: the point in the string at which to begin matching
589 * strend: pointer to the byte following the last char of the string
590 * flags currently unused; set to 0
591 * data: currently unused; set to NULL
592 *
593 * The basic idea of re_intuit_start() is to use some known information
594 * about the pattern, namely:
595 *
596 * a) the longest known anchored substring (i.e. one that's at a
597 * constant offset from the beginning of the pattern; but not
598 * necessarily at a fixed offset from the beginning of the
599 * string);
600 * b) the longest floating substring (i.e. one that's not at a constant
601 * offset from the beginning of the pattern);
602 * c) Whether the pattern is anchored to the string; either
603 * an absolute anchor: /^../, or anchored to \n: /^.../m,
604 * or anchored to pos(): /\G/;
605 * d) A start class: a real or synthetic character class which
606 * represents which characters are legal at the start of the pattern;
607 *
608 * to either quickly reject the match, or to find the earliest position
609 * within the string at which the pattern might match, thus avoiding
610 * running the full NFA engine at those earlier locations, only to
611 * eventually fail and retry further along.
612 *
613 * Returns NULL if the pattern can't match, or returns the address within
614 * the string which is the earliest place the match could occur.
615 *
616 * The longest of the anchored and floating substrings is called 'check'
617 * and is checked first. The other is called 'other' and is checked
618 * second. The 'other' substring may not be present. For example,
619 *
620 * /(abc|xyz)ABC\d{0,3}DEFG/
621 *
622 * will have
623 *
624 * check substr (float) = "DEFG", offset 6..9 chars
625 * other substr (anchored) = "ABC", offset 3..3 chars
626 * stclass = [ax]
627 *
628 * Be aware that during the course of this function, sometimes 'anchored'
629 * refers to a substring being anchored relative to the start of the
630 * pattern, and sometimes to the pattern itself being anchored relative to
631 * the string. For example:
632 *
633 * /\dabc/: "abc" is anchored to the pattern;
634 * /^\dabc/: "abc" is anchored to the pattern and the string;
635 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
636 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
637 * but the pattern is anchored to the string.
52a21eb3
DM
638 */
639
cad2e5aa 640char *
52a21eb3
DM
641Perl_re_intuit_start(pTHX_
642 REGEXP * const rx,
643 SV *sv,
644 const char * const strbeg,
645 char *strpos,
646 char *strend,
647 const U32 flags,
648 re_scream_pos_data *data)
cad2e5aa 649{
8d919b0a 650 struct regexp *const prog = ReANY(rx);
6b071d16 651 SSize_t start_shift = prog->check_offset_min;
cad2e5aa 652 /* Should be nonnegative! */
ea3daa5d 653 SSize_t end_shift = 0;
0fc004dd
DM
654 /* current lowest pos in string where the regex can start matching */
655 char *rx_origin = strpos;
eb578fdb 656 SV *check;
f2ed9b32 657 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
6480a6c4 658 U8 other_ix = 1 - prog->substrs->check_ix;
6ad5ffb3 659 bool ml_anch = 0;
8f4bf5fc 660 char *other_last = strpos;/* latest pos 'other' substr already checked to */
bd61b366 661 char *check_at = NULL; /* check substr found at this pos */
bbe252da 662 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
f8fc2ecf 663 RXi_GET_DECL(prog,progi);
02d5137b
DM
664 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
665 regmatch_info *const reginfo = &reginfo_buf;
a3621e74
YO
666 GET_RE_DEBUG_FLAGS_DECL;
667
7918f24d 668 PERL_ARGS_ASSERT_RE_INTUIT_START;
c33e64f0
FC
669 PERL_UNUSED_ARG(flags);
670 PERL_UNUSED_ARG(data);
7918f24d 671
1dc475d0
DM
672 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
673 "Intuit: trying to determine minimum start position...\n"));
674
fb9bbddb
DM
675 /* for now, assume that all substr offsets are positive. If at some point
676 * in the future someone wants to do clever things with look-behind and
677 * -ve offsets, they'll need to fix up any code in this function
678 * which uses these offsets. See the thread beginning
679 * <20140113145929.GF27210@iabyn.com>
680 */
681 assert(prog->substrs->data[0].min_offset >= 0);
682 assert(prog->substrs->data[0].max_offset >= 0);
683 assert(prog->substrs->data[1].min_offset >= 0);
684 assert(prog->substrs->data[1].max_offset >= 0);
685 assert(prog->substrs->data[2].min_offset >= 0);
686 assert(prog->substrs->data[2].max_offset >= 0);
687
f7022b5a 688 /* for now, assume that if both present, that the floating substring
83f2232d 689 * doesn't start before the anchored substring.
f7022b5a
DM
690 * If you break this assumption (e.g. doing better optimisations
691 * with lookahead/behind), then you'll need to audit the code in this
692 * function carefully first
693 */
694 assert(
695 ! ( (prog->anchored_utf8 || prog->anchored_substr)
696 && (prog->float_utf8 || prog->float_substr))
697 || (prog->float_min_offset >= prog->anchored_offset));
698
1a4edc3c
DM
699 /* byte rather than char calculation for efficiency. It fails
700 * to quickly reject some cases that can't match, but will reject
701 * them later after doing full char arithmetic */
c344f387 702 if (prog->minlen > strend - strpos) {
a3621e74 703 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 704 " String too short...\n"));
cad2e5aa 705 goto fail;
2c2d71f5 706 }
d8da0584 707
6c3fea77 708 reginfo->is_utf8_target = cBOOL(utf8_target);
bf2039a9 709 reginfo->info_aux = NULL;
9d9163fb 710 reginfo->strbeg = strbeg;
220db18a 711 reginfo->strend = strend;
aed7b151 712 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
02d5137b 713 reginfo->intuit = 1;
1cb48e53
DM
714 /* not actually used within intuit, but zero for safety anyway */
715 reginfo->poscache_maxiter = 0;
02d5137b 716
f2ed9b32 717 if (utf8_target) {
33b8afdf
JH
718 if (!prog->check_utf8 && prog->check_substr)
719 to_utf8_substr(prog);
720 check = prog->check_utf8;
721 } else {
7e0d5ad7
KW
722 if (!prog->check_substr && prog->check_utf8) {
723 if (! to_byte_substr(prog)) {
6b54ddc5 724 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
7e0d5ad7
KW
725 }
726 }
33b8afdf
JH
727 check = prog->check_substr;
728 }
274cd312 729
1dc475d0
DM
730 /* dump the various substring data */
731 DEBUG_OPTIMISE_MORE_r({
732 int i;
733 for (i=0; i<=2; i++) {
734 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
735 : prog->substrs->data[i].substr);
736 if (!sv)
737 continue;
738
739 PerlIO_printf(Perl_debug_log,
740 " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf
741 " useful=%"IVdf" utf8=%d [%s]\n",
742 i,
743 (IV)prog->substrs->data[i].min_offset,
744 (IV)prog->substrs->data[i].max_offset,
745 (IV)prog->substrs->data[i].end_shift,
746 BmUSEFUL(sv),
747 utf8_target ? 1 : 0,
748 SvPEEK(sv));
749 }
750 });
751
8e1490ee 752 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
9fc7410e
DM
753
754 /* ml_anch: check after \n?
755 *
756 * A note about IMPLICIT: on an un-anchored pattern beginning
757 * with /.*.../, these flags will have been added by the
758 * compiler:
759 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
760 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
761 */
7d2d37f5
DM
762 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
763 && !(prog->intflags & PREGf_IMPLICIT);
cad2e5aa 764
343c8a29 765 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
c889ccc8
DM
766 /* we are only allowed to match at BOS or \G */
767
57fcbfa7 768 /* trivially reject if there's a BOS anchor and we're not at BOS.
7bb3b9eb
DM
769 *
770 * Note that we don't try to do a similar quick reject for
771 * \G, since generally the caller will have calculated strpos
772 * based on pos() and gofs, so the string is already correctly
773 * anchored by definition; and handling the exceptions would
774 * be too fiddly (e.g. REXEC_IGNOREPOS).
57fcbfa7 775 */
7bb3b9eb 776 if ( strpos != strbeg
343c8a29 777 && (prog->intflags & (PREGf_ANCH_BOL|PREGf_ANCH_SBOL)))
c889ccc8
DM
778 {
779 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 780 " Not at start...\n"));
c889ccc8
DM
781 goto fail;
782 }
783
a5d12a4b
DM
784 /* in the presence of an anchor, the anchored (relative to the
785 * start of the regex) substr must also be anchored relative
66b7ec5c
DM
786 * to strpos. So quickly reject if substr isn't found there.
787 * This works for \G too, because the caller will already have
788 * subtracted gofs from pos, and gofs is the offset from the
789 * \G to the start of the regex. For example, in /.abc\Gdef/,
790 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
791 * caller will have set strpos=pos()-4; we look for the substr
792 * at position pos()-4+1, which lines up with the "a" */
a5d12a4b 793
c889ccc8 794 if (prog->check_offset_min == prog->check_offset_max
d0d44648 795 && !(prog->intflags & PREGf_CANY_SEEN))
c889ccc8
DM
796 {
797 /* Substring at constant offset from beg-of-str... */
d307bf57 798 SSize_t slen = SvCUR(check);
343c8a29 799 char *s = HOP3c(strpos, prog->check_offset_min, strend);
1de06328 800
1dc475d0
DM
801 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
802 " Looking for check substr at fixed offset %"IVdf"...\n",
803 (IV)prog->check_offset_min));
804
7742aa66
DM
805 if (SvTAIL(check)) {
806 /* In this case, the regex is anchored at the end too.
807 * Unless it's a multiline match, the lengths must match
808 * exactly, give or take a \n. NB: slen >= 1 since
809 * the last char of check is \n */
810 if (!multiline
811 && ( strend - s > slen
812 || strend - s < slen - 1
813 || (strend - s == slen && strend[-1] != '\n')))
c889ccc8
DM
814 {
815 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 816 " String too long...\n"));
c889ccc8
DM
817 goto fail_finish;
818 }
819 /* Now should match s[0..slen-2] */
820 slen--;
c889ccc8 821 }
d307bf57
DM
822 if (slen && (*SvPVX_const(check) != *s
823 || (slen > 1 && memNE(SvPVX_const(check), s, slen))))
824 {
825 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 826 " String not equal...\n"));
d307bf57
DM
827 goto fail_finish;
828 }
c889ccc8
DM
829
830 check_at = s;
831 goto success_at_start;
cad2e5aa 832 }
cad2e5aa 833 }
cad2e5aa 834 }
0fc004dd 835
c0e0ec46 836 end_shift = prog->check_end_shift;
cad2e5aa 837
19188028 838#ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
0033605d 839 if (end_shift < 0)
1de06328 840 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
220fc49f 841 (IV)end_shift, RX_PRECOMP(prog));
2c2d71f5
JH
842#endif
843
2c2d71f5 844 restart:
1de06328 845
66b7ec5c
DM
846 /* This is the (re)entry point of the main loop in this function.
847 * The goal of this loop is to:
848 * 1) find the "check" substring in the region rx_origin..strend
849 * (adjusted by start_shift / end_shift). If not found, reject
850 * immediately.
851 * 2) If it exists, look for the "other" substr too if defined; for
852 * example, if the check substr maps to the anchored substr, then
853 * check the floating substr, and vice-versa. If not found, go
854 * back to (1) with rx_origin suitably incremented.
855 * 3) If we find an rx_origin position that doesn't contradict
856 * either of the substrings, then check the possible additional
857 * constraints on rx_origin of /^.../m or a known start class.
858 * If these fail, then depending on which constraints fail, jump
859 * back to here, or to various other re-entry points further along
860 * that skip some of the first steps.
861 * 4) If we pass all those tests, update the BmUSEFUL() count on the
862 * substring. If the start position was determined to be at the
863 * beginning of the string - so, not rejected, but not optimised,
864 * since we have to run regmatch from position 0 - decrement the
865 * BmUSEFUL() count. Otherwise increment it.
866 */
867
1a4edc3c
DM
868
869 /* first, look for the 'check' substring */
870
1de06328 871 {
c33e64f0
FC
872 U8* start_point;
873 U8* end_point;
c889ccc8 874
c889ccc8 875 DEBUG_OPTIMISE_MORE_r({
1dc475d0 876 PerlIO_printf(Perl_debug_log,
ae5d4331 877 " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf
1dc475d0 878 " Start shift: %"IVdf" End shift %"IVdf
4d281893 879 " Real end Shift: %"IVdf"\n",
197641b6 880 (IV)(rx_origin - strpos),
c889ccc8 881 (IV)prog->check_offset_min,
12fbc530
DM
882 (IV)start_shift,
883 (IV)end_shift,
c889ccc8
DM
884 (IV)prog->check_end_shift);
885 });
1de06328 886
0d331aaf 887 if (prog->intflags & PREGf_CANY_SEEN) {
0fc004dd 888 start_point= (U8*)(rx_origin + start_shift);
12fbc530 889 end_point= (U8*)(strend - end_shift);
557f47af
DM
890 if (start_point > end_point)
891 goto fail_finish;
1de06328 892 } else {
557f47af
DM
893 end_point = HOP3(strend, -end_shift, strbeg);
894 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
895 if (!start_point)
896 goto fail_finish;
1de06328 897 }
c889ccc8 898
557f47af 899
e0362b86
DM
900 /* If the regex is absolutely anchored to either the start of the
901 * string (BOL,SBOL) or to pos() (ANCH_GPOS), then
902 * check_offset_max represents an upper bound on the string where
903 * the substr could start. For the ANCH_GPOS case, we assume that
904 * the caller of intuit will have already set strpos to
905 * pos()-gofs, so in this case strpos + offset_max will still be
906 * an upper bound on the substr.
907 */
c19c836a
DM
908 if (!ml_anch
909 && prog->intflags & PREGf_ANCH
e0362b86 910 && prog->check_offset_max != SSize_t_MAX)
c19c836a 911 {
1a08ba3a 912 SSize_t len = SvCUR(check) - !!SvTAIL(check);
e0362b86
DM
913 const char * const anchor =
914 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
915
916 /* do a bytes rather than chars comparison. It's conservative;
917 * so it skips doing the HOP if the result can't possibly end
918 * up earlier than the old value of end_point.
919 */
920 if ((char*)end_point - anchor > prog->check_offset_max) {
921 end_point = HOP3lim((U8*)anchor,
922 prog->check_offset_max,
923 end_point -len)
924 + len;
925 }
d6ef1678
DM
926 }
927
6bda09f9 928 DEBUG_OPTIMISE_MORE_r({
1dc475d0 929 PerlIO_printf(Perl_debug_log, " fbm_instr len=%d str=<%.*s>\n",
1de06328 930 (int)(end_point - start_point),
fc8cd66c 931 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
1de06328
YO
932 start_point);
933 });
934
ae5d4331 935 check_at = fbm_instr( start_point, end_point,
7fba1cd6 936 check, multiline ? FBMrf_MULTILINE : 0);
c889ccc8 937
8fd34720
DM
938 /* Update the count-of-usability, remove useless subpatterns,
939 unshift s. */
940
941 DEBUG_EXECUTE_r({
942 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
943 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
944 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s",
945 (check_at ? "Found" : "Did not find"),
946 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
947 ? "anchored" : "floating"),
948 quoted,
949 RE_SV_TAIL(check),
950 (check_at ? " at offset " : "...\n") );
951 });
2c2d71f5 952
8fd34720
DM
953 if (!check_at)
954 goto fail_finish;
955 /* Finish the diagnostic message */
956 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(check_at - strpos)) );
2c2d71f5 957
8fd34720
DM
958 /* set rx_origin to the minimum position where the regex could start
959 * matching, given the constraint of the just-matched check substring.
960 * But don't set it lower than previously.
961 */
fdc003fd 962
8fd34720
DM
963 if (check_at - rx_origin > prog->check_offset_max)
964 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
965 }
fdc003fd
DM
966
967
1a4edc3c 968 /* now look for the 'other' substring if defined */
2c2d71f5 969
6480a6c4
DM
970 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
971 : prog->substrs->data[other_ix].substr)
1de06328 972 {
30944b6d 973 /* Take into account the "other" substring. */
6c3343a6
DM
974 char *last, *last1;
975 char *s;
976 SV* must;
977 struct reg_substr_datum *other;
978
979 do_other_substr:
980 other = &prog->substrs->data[other_ix];
981
982 /* if "other" is anchored:
983 * we've previously found a floating substr starting at check_at.
984 * This means that the regex origin must lie somewhere
985 * between min (rx_origin): HOP3(check_at, -check_offset_max)
986 * and max: HOP3(check_at, -check_offset_min)
987 * (except that min will be >= strpos)
988 * So the fixed substr must lie somewhere between
989 * HOP3(min, anchored_offset)
990 * HOP3(max, anchored_offset) + SvCUR(substr)
991 */
992
993 /* if "other" is floating
994 * Calculate last1, the absolute latest point where the
995 * floating substr could start in the string, ignoring any
996 * constraints from the earlier fixed match. It is calculated
997 * as follows:
998 *
999 * strend - prog->minlen (in chars) is the absolute latest
1000 * position within the string where the origin of the regex
1001 * could appear. The latest start point for the floating
1002 * substr is float_min_offset(*) on from the start of the
1003 * regex. last1 simply combines thee two offsets.
1004 *
1005 * (*) You might think the latest start point should be
1006 * float_max_offset from the regex origin, and technically
1007 * you'd be correct. However, consider
1008 * /a\d{2,4}bcd\w/
1009 * Here, float min, max are 3,5 and minlen is 7.
1010 * This can match either
1011 * /a\d\dbcd\w/
1012 * /a\d\d\dbcd\w/
1013 * /a\d\d\d\dbcd\w/
1014 * In the first case, the regex matches minlen chars; in the
1015 * second, minlen+1, in the third, minlen+2.
1016 * In the first case, the floating offset is 3 (which equals
1017 * float_min), in the second, 4, and in the third, 5 (which
1018 * equals float_max). In all cases, the floating string bcd
1019 * can never start more than 4 chars from the end of the
1020 * string, which equals minlen - float_min. As the substring
1021 * starts to match more than float_min from the start of the
1022 * regex, it makes the regex match more than minlen chars,
1023 * and the two cancel each other out. So we can always use
1024 * float_min - minlen, rather than float_max - minlen for the
1025 * latest position in the string.
1026 *
1027 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1028 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1029 */
1030
e7a14a9c 1031 assert(prog->minlen >= other->min_offset);
6c3343a6
DM
1032 last1 = HOP3c(strend,
1033 other->min_offset - prog->minlen, strbeg);
1034
4d006249 1035 if (other_ix) {/* i.e. if (other-is-float) */
6c3343a6
DM
1036 /* last is the latest point where the floating substr could
1037 * start, *given* any constraints from the earlier fixed
1038 * match. This constraint is that the floating string starts
1039 * <= float_max_offset chars from the regex origin (rx_origin).
1040 * If this value is less than last1, use it instead.
eb3831ce 1041 */
6c3343a6
DM
1042 assert(rx_origin <= last1);
1043 last =
1044 /* this condition handles the offset==infinity case, and
1045 * is a short-cut otherwise. Although it's comparing a
1046 * byte offset to a char length, it does so in a safe way,
1047 * since 1 char always occupies 1 or more bytes,
1048 * so if a string range is (last1 - rx_origin) bytes,
1049 * it will be less than or equal to (last1 - rx_origin)
1050 * chars; meaning it errs towards doing the accurate HOP3
1051 * rather than just using last1 as a short-cut */
1052 (last1 - rx_origin) < other->max_offset
1053 ? last1
1054 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1055 }
1056 else {
1057 assert(strpos + start_shift <= check_at);
1058 last = HOP4c(check_at, other->min_offset - start_shift,
1059 strbeg, strend);
1060 }
ead917d0 1061
6c3343a6
DM
1062 s = HOP3c(rx_origin, other->min_offset, strend);
1063 if (s < other_last) /* These positions already checked */
1064 s = other_last;
1065
1066 must = utf8_target ? other->utf8_substr : other->substr;
1067 assert(SvPOK(must));
1068 s = fbm_instr(
1069 (unsigned char*)s,
1070 (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0),
1071 must,
1072 multiline ? FBMrf_MULTILINE : 0
1073 );
1074 DEBUG_EXECUTE_r({
1075 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1076 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1077 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s",
1078 s ? "Found" : "Contradicts",
1079 other_ix ? "floating" : "anchored",
1080 quoted, RE_SV_TAIL(must));
1081 });
ead917d0 1082
ead917d0 1083
6c3343a6
DM
1084 if (!s) {
1085 /* last1 is latest possible substr location. If we didn't
1086 * find it before there, we never will */
1087 if (last >= last1) {
1088 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1089 ", giving up...\n"));
1090 goto fail_finish;
ead917d0
DM
1091 }
1092
6c3343a6
DM
1093 /* try to find the check substr again at a later
1094 * position. Maybe next time we'll find the "other" substr
1095 * in range too */
1096 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1097 ", trying %s at offset %ld...\n",
1098 (other_ix ? "floating" : "anchored"),
197641b6 1099 (long)(HOP3c(check_at, 1, strend) - strpos)));
6c3343a6
DM
1100
1101 other_last = HOP3c(last, 1, strend) /* highest failure */;
1102 rx_origin =
4d006249 1103 other_ix /* i.e. if other-is-float */
6c3343a6
DM
1104 ? HOP3c(rx_origin, 1, strend)
1105 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1106 goto restart;
1107 }
1108 else {
1109 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
197641b6 1110 (long)(s - strpos)));
ead917d0 1111
4d006249 1112 if (other_ix) { /* if (other-is-float) */
6c3343a6
DM
1113 /* other_last is set to s, not s+1, since its possible for
1114 * a floating substr to fail first time, then succeed
1115 * second time at the same floating position; e.g.:
1116 * "-AB--AABZ" =~ /\wAB\d*Z/
1117 * The first time round, anchored and float match at
1118 * "-(AB)--AAB(Z)" then fail on the initial \w character
1119 * class. Second time round, they match at "-AB--A(AB)(Z)".
1120 */
1121 other_last = s;
ead917d0
DM
1122 }
1123 else {
6c3343a6
DM
1124 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1125 other_last = HOP3c(s, 1, strend);
ead917d0 1126 }
6c3343a6 1127 }
cad2e5aa 1128 }
acba93e8
DM
1129 else {
1130 DEBUG_OPTIMISE_MORE_r(
1131 PerlIO_printf(Perl_debug_log,
1132 " Check-only match: offset min:%"IVdf" max:%"IVdf
1c1c599d 1133 " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf
acba93e8
DM
1134 " strend-strpos:%"IVdf"\n",
1135 (IV)prog->check_offset_min,
1136 (IV)prog->check_offset_max,
1c1c599d 1137 (IV)(check_at-strpos),
acba93e8 1138 (IV)(rx_origin-strpos),
1c1c599d 1139 (IV)(rx_origin-check_at),
acba93e8
DM
1140 (IV)(strend-strpos)
1141 )
1142 );
1143 }
2c2d71f5 1144
acba93e8 1145 postprocess_substr_matches:
0991020e 1146
1a4edc3c 1147 /* handle the extra constraint of /^.../m if present */
e3c6feb0 1148
7d2d37f5 1149 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
4620cb61
DM
1150 char *s;
1151
a62659bd
DM
1152 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1153 " looking for /^/m anchor"));
d0880ea7
DM
1154
1155 /* we have failed the constraint of a \n before rx_origin.
2e759faa
DM
1156 * Find the next \n, if any, even if it's beyond the current
1157 * anchored and/or floating substrings. Whether we should be
1158 * scanning ahead for the next \n or the next substr is debatable.
1159 * On the one hand you'd expect rare substrings to appear less
1160 * often than \n's. On the other hand, searching for \n means
1161 * we're effectively flipping been check_substr and "\n" on each
1162 * iteration as the current "rarest" string candidate, which
1163 * means for example that we'll quickly reject the whole string if
1164 * hasn't got a \n, rather than trying every substr position
1165 * first
1166 */
d0880ea7 1167
4620cb61
DM
1168 s = HOP3c(strend, - prog->minlen, strpos);
1169 if (s <= rx_origin ||
1170 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1171 {
d0880ea7
DM
1172 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1173 " Did not find /%s^%s/m...\n",
1174 PL_colors[0], PL_colors[1]));
a62659bd
DM
1175 goto fail_finish;
1176 }
d0880ea7 1177
4ada1233
DM
1178 /* earliest possible origin is 1 char after the \n.
1179 * (since *rx_origin == '\n', it's safe to ++ here rather than
1180 * HOP(rx_origin, 1)) */
1181 rx_origin++;
d0880ea7 1182
f4f115de 1183 if (prog->substrs->check_ix == 0 /* check is anchored */
4ada1233 1184 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
f4f115de 1185 {
d0880ea7
DM
1186 /* Position contradicts check-string; either because
1187 * check was anchored (and thus has no wiggle room),
4ada1233 1188 * or check was float and rx_origin is above the float range */
d0880ea7
DM
1189 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1190 " Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
4ada1233 1191 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
d0880ea7
DM
1192 goto restart;
1193 }
1194
1195 /* if we get here, the check substr must have been float,
2e759faa 1196 * is in range, and we may or may not have had an anchored
d0880ea7
DM
1197 * "other" substr which still contradicts */
1198 assert(prog->substrs->check_ix); /* check is float */
1199
1200 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1201 /* whoops, the anchored "other" substr exists, so we still
1202 * contradict. On the other hand, the float "check" substr
1203 * didn't contradict, so just retry the anchored "other"
1204 * substr */
1205 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1206 " Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1207 PL_colors[0], PL_colors[1],
1208 (long)(rx_origin - strpos),
1209 (long)(rx_origin - strpos + prog->anchored_offset)));
1210 goto do_other_substr;
1211 }
1212
1213 /* success: we don't contradict the found floating substring
1214 * (and there's no anchored substr). */
d0880ea7
DM
1215 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1216 " Found /%s^%s/m at offset %ld...\n",
1217 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
e3c6feb0
DM
1218 }
1219 else {
2e759faa 1220 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
fe4f3442 1221 " (multiline anchor test skipped)\n"));
e3c6feb0
DM
1222 }
1223
ffad1e6a 1224 success_at_start:
e3c6feb0 1225
cad2e5aa 1226
dd170ff5
DM
1227 /* if we have a starting character class, then test that extra constraint.
1228 * (trie stclasses are too expensive to use here, we are better off to
1229 * leave it to regmatch itself) */
1230
f8fc2ecf 1231 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
f8fc2ecf 1232 const U8* const str = (U8*)STRING(progi->regstclass);
0991020e 1233
2c75e362 1234 /* XXX this value could be pre-computed */
f8fc2ecf 1235 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
2c75e362
DM
1236 ? (reginfo->is_utf8_pat
1237 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1238 : STR_LEN(progi->regstclass))
66e933ab 1239 : 1);
1de06328 1240 char * endpos;
fa3bb21d 1241 char *s;
000dfd2d
DM
1242 /* latest pos that a matching float substr constrains rx start to */
1243 char *rx_max_float = NULL;
1244
c75a3985
DM
1245 /* if the current rx_origin is anchored, either by satisfying an
1246 * anchored substring constraint, or a /^.../m constraint, then we
1247 * can reject the current origin if the start class isn't found
1248 * at the current position. If we have a float-only match, then
1249 * rx_origin is constrained to a range; so look for the start class
1250 * in that range. if neither, then look for the start class in the
1251 * whole rest of the string */
1252
dd170ff5
DM
1253 /* XXX DAPM it's not clear what the minlen test is for, and why
1254 * it's not used in the floating case. Nothing in the test suite
1255 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1256 * Here are some old comments, which may or may not be correct:
1257 *
1258 * minlen == 0 is possible if regstclass is \b or \B,
1259 * and the fixed substr is ''$.
1260 * Since minlen is already taken into account, rx_origin+1 is
1261 * before strend; accidentally, minlen >= 1 guaranties no false
1262 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1263 * 0) below assumes that regstclass does not come from lookahead...
1264 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1265 * This leaves EXACTF-ish only, which are dealt with in
1266 * find_byclass().
1267 */
1268
7d2d37f5 1269 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
fa3bb21d 1270 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
000dfd2d
DM
1271 else if (prog->float_substr || prog->float_utf8) {
1272 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1273 endpos= HOP3c(rx_max_float, cl_l, strend);
1274 }
1de06328
YO
1275 else
1276 endpos= strend;
1277
1dc475d0
DM
1278 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1279 " looking for class: start_shift: %"IVdf" check_at: %"IVdf
c43b5520 1280 " rx_origin: %"IVdf" endpos: %"IVdf"\n",
1dc475d0 1281 (IV)start_shift, (IV)(check_at - strbeg),
c43b5520 1282 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
d8080198 1283
c43b5520 1284 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
f9176b44 1285 reginfo);
be778b1a 1286 if (!s) {
6eb5f6b9 1287 if (endpos == strend) {
a3621e74 1288 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1dc475d0 1289 " Could not match STCLASS...\n") );
6eb5f6b9
JH
1290 goto fail;
1291 }
a3621e74 1292 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1dc475d0 1293 " This position contradicts STCLASS...\n") );
e0eb31e7
DM
1294 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1295 && !(prog->intflags & PREGf_IMPLICIT))
653099ff 1296 goto fail;
9fed8d02 1297
6eb5f6b9 1298 /* Contradict one of substrings */
97136c8a
DM
1299 if (prog->anchored_substr || prog->anchored_utf8) {
1300 if (prog->substrs->check_ix == 1) { /* check is float */
1301 /* Have both, check_string is floating */
1302 assert(rx_origin + start_shift <= check_at);
1303 if (rx_origin + start_shift != check_at) {
1304 /* not at latest position float substr could match:
c75a3985
DM
1305 * Recheck anchored substring, but not floating.
1306 * The condition above is in bytes rather than
1307 * chars for efficiency. It's conservative, in
1308 * that it errs on the side of doing 'goto
1309 * do_other_substr', where a more accurate
1310 * char-based calculation will be done */
97136c8a
DM
1311 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1312 " Looking for anchored substr starting at offset %ld...\n",
1313 (long)(other_last - strpos)) );
1314 goto do_other_substr;
3369914b 1315 }
3369914b
DM
1316 }
1317 }
97136c8a 1318 else {
9fed8d02
DM
1319 /* float-only */
1320
7d2d37f5 1321 if (ml_anch) {
c75a3985
DM
1322 /* In the presence of ml_anch, we might be able to
1323 * find another \n without breaking the current float
1324 * constraint. */
1325
1326 /* strictly speaking this should be HOP3c(..., 1, ...),
1327 * but since we goto a block of code that's going to
1328 * search for the next \n if any, its safe here */
9fed8d02 1329 rx_origin++;
9fed8d02
DM
1330 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1331 " Looking for /%s^%s/m starting at offset %ld...\n",
1332 PL_colors[0], PL_colors[1],
1333 (long)(rx_origin - strpos)) );
9fed8d02 1334 goto postprocess_substr_matches;
ab60c45a 1335 }
c75a3985
DM
1336
1337 /* strictly speaking this can never be true; but might
1338 * be if we ever allow intuit without substrings */
1339 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
9fed8d02 1340 goto fail;
c75a3985 1341
000dfd2d 1342 rx_origin = rx_max_float;
9fed8d02
DM
1343 }
1344
c75a3985
DM
1345 /* at this point, any matching substrings have been
1346 * contradicted. Start again... */
1347
9fed8d02 1348 rx_origin = HOP3c(rx_origin, 1, strend);
557f47af
DM
1349
1350 /* uses bytes rather than char calculations for efficiency.
1351 * It's conservative: it errs on the side of doing 'goto restart',
1352 * where there is code that does a proper char-based test */
9fed8d02 1353 if (rx_origin + start_shift + end_shift > strend) {
40268e5b 1354 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
9fed8d02
DM
1355 " Could not match STCLASS...\n") );
1356 goto fail;
1357 }
1358 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1359 " Looking for %s substr starting at offset %ld...\n",
1360 (prog->substrs->check_ix ? "floating" : "anchored"),
1361 (long)(rx_origin + start_shift - strpos)) );
1362 goto restart;
6eb5f6b9 1363 }
9fed8d02 1364
c75a3985
DM
1365 /* Success !!! */
1366
5f9c6575 1367 if (rx_origin != s) {
a3621e74 1368 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 1369 " By STCLASS: moving %ld --> %ld\n",
5f9c6575 1370 (long)(rx_origin - strpos), (long)(s - strpos))
b7953727
JH
1371 );
1372 }
1373 else {
a3621e74 1374 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1dc475d0 1375 " Does not contradict STCLASS...\n");
b7953727
JH
1376 );
1377 }
6eb5f6b9 1378 }
ffad1e6a
DM
1379
1380 /* Decide whether using the substrings helped */
1381
1382 if (rx_origin != strpos) {
1383 /* Fixed substring is found far enough so that the match
1384 cannot start at strpos. */
1385
1386 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n"));
1387 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1388 }
1389 else {
70563e16
DM
1390 /* The found rx_origin position does not prohibit matching at
1391 * strpos, so calling intuit didn't gain us anything. Decrement
1392 * the BmUSEFUL() count on the check substring, and if we reach
1393 * zero, free it. */
1394 if (!(prog->intflags & PREGf_NAUGHTY)
ffad1e6a
DM
1395 && (utf8_target ? (
1396 prog->check_utf8 /* Could be deleted already */
1397 && --BmUSEFUL(prog->check_utf8) < 0
1398 && (prog->check_utf8 == prog->float_utf8)
1399 ) : (
1400 prog->check_substr /* Could be deleted already */
1401 && --BmUSEFUL(prog->check_substr) < 0
1402 && (prog->check_substr == prog->float_substr)
1403 )))
1404 {
1405 /* If flags & SOMETHING - do not do it many times on the same match */
1406 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n"));
1407 /* XXX Does the destruction order has to change with utf8_target? */
1408 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1409 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1410 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1411 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1412 check = NULL; /* abort */
ffad1e6a
DM
1413 /* XXXX This is a remnant of the old implementation. It
1414 looks wasteful, since now INTUIT can use many
1415 other heuristics. */
1416 prog->extflags &= ~RXf_USE_INTUIT;
ffad1e6a
DM
1417 }
1418 }
1419
1420 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1421 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1422 PL_colors[4], PL_colors[5], (long)(rx_origin - strpos)) );
1423
c765d6e0 1424 return rx_origin;
2c2d71f5
JH
1425
1426 fail_finish: /* Substring not found */
33b8afdf 1427 if (prog->check_substr || prog->check_utf8) /* could be removed already */
f2ed9b32 1428 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
cad2e5aa 1429 fail:
a3621e74 1430 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
e4584336 1431 PL_colors[4], PL_colors[5]));
bd61b366 1432 return NULL;
cad2e5aa 1433}
9661b544 1434
70563e16 1435
a0a388a1 1436#define DECL_TRIE_TYPE(scan) \
098b07d5
KW
1437 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1438 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold } \
fab2782b
YO
1439 trie_type = ((scan->flags == EXACT) \
1440 ? (utf8_target ? trie_utf8 : trie_plain) \
098b07d5
KW
1441 : (scan->flags == EXACTFA) \
1442 ? (utf8_target ? trie_utf8_exactfa_fold : trie_latin_utf8_exactfa_fold) \
1443 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
fab2782b 1444
fd3249ee 1445#define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
baa60164 1446STMT_START { \
fab2782b 1447 STRLEN skiplen; \
baa60164 1448 U8 flags = FOLD_FLAGS_FULL; \
fab2782b 1449 switch (trie_type) { \
31f05a37 1450 case trie_utf8_exactfa_fold: \
baa60164 1451 flags |= FOLD_FLAGS_NOMIX_ASCII; \
924ba076 1452 /* FALLTHROUGH */ \
fab2782b
YO
1453 case trie_utf8_fold: \
1454 if ( foldlen>0 ) { \
c80e42f3 1455 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO
1456 foldlen -= len; \
1457 uscan += len; \
1458 len=0; \
1459 } else { \
445bf929 1460 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
fab2782b
YO
1461 len = UTF8SKIP(uc); \
1462 skiplen = UNISKIP( uvc ); \
1463 foldlen -= skiplen; \
1464 uscan = foldbuf + skiplen; \
1465 } \
1466 break; \
baa60164
KW
1467 case trie_latin_utf8_exactfa_fold: \
1468 flags |= FOLD_FLAGS_NOMIX_ASCII; \
924ba076 1469 /* FALLTHROUGH */ \
fab2782b
YO
1470 case trie_latin_utf8_fold: \
1471 if ( foldlen>0 ) { \
c80e42f3 1472 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO
1473 foldlen -= len; \
1474 uscan += len; \
1475 len=0; \
1476 } else { \
1477 len = 1; \
31f05a37 1478 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
fab2782b
YO
1479 skiplen = UNISKIP( uvc ); \
1480 foldlen -= skiplen; \
1481 uscan = foldbuf + skiplen; \
1482 } \
1483 break; \
1484 case trie_utf8: \
c80e42f3 1485 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
fab2782b
YO
1486 break; \
1487 case trie_plain: \
1488 uvc = (UV)*uc; \
1489 len = 1; \
1490 } \
1491 if (uvc < 256) { \
1492 charid = trie->charmap[ uvc ]; \
1493 } \
1494 else { \
1495 charid = 0; \
1496 if (widecharmap) { \
1497 SV** const svpp = hv_fetch(widecharmap, \
1498 (char*)&uvc, sizeof(UV), 0); \
1499 if (svpp) \
1500 charid = (U16)SvIV(*svpp); \
1501 } \
1502 } \
4cadc6a9
YO
1503} STMT_END
1504
c84a03c5 1505#define REXEC_FBC_EXACTISH_SCAN(COND) \
4cadc6a9
YO
1506STMT_START { \
1507 while (s <= e) { \
c84a03c5 1508 if ( (COND) \
fac1af77 1509 && (ln == 1 || folder(s, pat_string, ln)) \
02d5137b 1510 && (reginfo->intuit || regtry(reginfo, &s)) )\
4cadc6a9
YO
1511 goto got_it; \
1512 s++; \
1513 } \
1514} STMT_END
1515
c84a03c5 1516#define REXEC_FBC_UTF8_SCAN(CODE) \
4cadc6a9 1517STMT_START { \
9a902117 1518 while (s < strend) { \
c84a03c5 1519 CODE \
9a902117 1520 s += UTF8SKIP(s); \
4cadc6a9
YO
1521 } \
1522} STMT_END
1523
c84a03c5 1524#define REXEC_FBC_SCAN(CODE) \
4cadc6a9
YO
1525STMT_START { \
1526 while (s < strend) { \
c84a03c5 1527 CODE \
4cadc6a9
YO
1528 s++; \
1529 } \
1530} STMT_END
1531
c84a03c5 1532#define REXEC_FBC_UTF8_CLASS_SCAN(COND) \
4cadc6a9 1533REXEC_FBC_UTF8_SCAN( \
c84a03c5 1534 if (COND) { \
02d5137b 1535 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
4cadc6a9
YO
1536 goto got_it; \
1537 else \
1538 tmp = doevery; \
1539 } \
1540 else \
1541 tmp = 1; \
1542)
1543
c84a03c5 1544#define REXEC_FBC_CLASS_SCAN(COND) \
4cadc6a9 1545REXEC_FBC_SCAN( \
c84a03c5 1546 if (COND) { \
02d5137b 1547 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
4cadc6a9
YO
1548 goto got_it; \
1549 else \
1550 tmp = doevery; \
1551 } \
1552 else \
1553 tmp = 1; \
1554)
1555
baa60164 1556#define REXEC_FBC_TRYIT \
02d5137b 1557if ((reginfo->intuit || regtry(reginfo, &s))) \
4cadc6a9
YO
1558 goto got_it
1559
c84a03c5 1560#define REXEC_FBC_CSCAN(CONDUTF8,COND) \
baa60164 1561 if (utf8_target) { \
c84a03c5 1562 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \
e1d1eefb
YO
1563 } \
1564 else { \
c84a03c5 1565 REXEC_FBC_CLASS_SCAN(COND); \
d981ef24 1566 }
e1d1eefb 1567
baa60164 1568#define DUMP_EXEC_POS(li,s,doutf8) \
9d9163fb 1569 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
6d59b646 1570 startpos, doutf8)
786e8c11 1571
cfaf538b 1572
b2f4e957 1573#define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
9d9163fb 1574 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
cfaf538b
KW
1575 tmp = TEST_NON_UTF8(tmp); \
1576 REXEC_FBC_UTF8_SCAN( \
ce1d15d5 1577 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
cfaf538b
KW
1578 tmp = !tmp; \
1579 IF_SUCCESS; \
1580 } \
1581 else { \
1582 IF_FAIL; \
1583 } \
1584 ); \
1585
236d82fd 1586#define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
9d9163fb 1587 if (s == reginfo->strbeg) { \
cfaf538b
KW
1588 tmp = '\n'; \
1589 } \
1590 else { \
9d9163fb 1591 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
3db24e1e
KW
1592 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1593 0, UTF8_ALLOW_DEFAULT); \
cfaf538b 1594 } \
5b8bb145 1595 tmp = TEST_UV(tmp); \
3db24e1e 1596 LOAD_UTF8_CHARCLASS_ALNUM(); \
cfaf538b 1597 REXEC_FBC_UTF8_SCAN( \
236d82fd 1598 if (tmp == ! (TEST_UTF8((U8 *) s))) { \
cfaf538b
KW
1599 tmp = !tmp; \
1600 IF_SUCCESS; \
1601 } \
1602 else { \
1603 IF_FAIL; \
1604 } \
1605 ); \
1606
63ac0dad
KW
1607/* The only difference between the BOUND and NBOUND cases is that
1608 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1609 * NBOUND. This is accomplished by passing it in either the if or else clause,
1610 * with the other one being empty */
236d82fd
KW
1611#define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1612 FBC_BOUND_COMMON(FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
cfaf538b 1613
236d82fd 1614#define FBC_BOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
b2f4e957 1615 FBC_BOUND_COMMON(FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
63ac0dad 1616
236d82fd
KW
1617#define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1618 FBC_BOUND_COMMON(FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
cfaf538b 1619
236d82fd 1620#define FBC_NBOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
b2f4e957 1621 FBC_BOUND_COMMON(FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
cfaf538b 1622
236d82fd 1623/* Common to the BOUND and NBOUND cases. We
63ac0dad
KW
1624 * are looking for the boundary (or non-boundary between a word and non-word
1625 * character. The utf8 and non-utf8 cases have the same logic, but the details
1626 * must be different. Find the "wordness" of the character just prior to this
1627 * one, and compare it with the wordness of this one. If they differ, we have
1628 * a boundary. At the beginning of the string, pretend that the previous
760cfa8e
KW
1629 * character was a new-line.
1630 *
1631 * 'tmp' below in the REXEC_FBC_SCAN loop is a loop invariant, a bool giving
1632 * the return of TEST_NON_UTF8(s-1). To see this, note that that's what it is
1633 * defined to be at entry to the loop, and to get to the IF_FAIL branch, tmp
1634 * must equal TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is
1635 * that complement. But in that branch we complement tmp, meaning that at the
1636 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1637 * which means at the top of the loop in the next iteration, it is
1638 * TEST_NON_UTF8(s-1) */
baa60164 1639#define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
63ac0dad 1640 if (utf8_target) { \
baa60164 1641 UTF8_CODE \
63ac0dad
KW
1642 } \
1643 else { /* Not utf8 */ \
9d9163fb 1644 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
63ac0dad
KW
1645 tmp = TEST_NON_UTF8(tmp); \
1646 REXEC_FBC_SCAN( \
1647 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
63ac0dad 1648 IF_SUCCESS; \
760cfa8e 1649 tmp = !tmp; \
63ac0dad
KW
1650 } \
1651 else { \
1652 IF_FAIL; \
1653 } \
1654 ); \
1655 } \
baa60164 1656 if ((!prog->minlen && tmp) && (reginfo->intuit || regtry(reginfo, &s))) \
63ac0dad
KW
1657 goto got_it;
1658
786e8c11 1659/* We know what class REx starts with. Try to find this position... */
02d5137b 1660/* if reginfo->intuit, its a dryrun */
786e8c11
YO
1661/* annoyingly all the vars in this routine have different names from their counterparts
1662 in regmatch. /grrr */
1663
3c3eec57 1664STATIC char *
07be1b83 1665S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
f9176b44 1666 const char *strend, regmatch_info *reginfo)
a687059c 1667{
73104a1b
KW
1668 dVAR;
1669 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1670 char *pat_string; /* The pattern's exactish string */
1671 char *pat_end; /* ptr to end char of pat_string */
1672 re_fold_t folder; /* Function for computing non-utf8 folds */
1673 const U8 *fold_array; /* array for folding ords < 256 */
1674 STRLEN ln;
1675 STRLEN lnc;
73104a1b
KW
1676 U8 c1;
1677 U8 c2;
1678 char *e;
1679 I32 tmp = 1; /* Scratch variable? */
ba44c216 1680 const bool utf8_target = reginfo->is_utf8_target;
73104a1b 1681 UV utf8_fold_flags = 0;
f9176b44 1682 const bool is_utf8_pat = reginfo->is_utf8_pat;
3018b823
KW
1683 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1684 with a result inverts that result, as 0^1 =
1685 1 and 1^1 = 0 */
1686 _char_class_number classnum;
1687
73104a1b 1688 RXi_GET_DECL(prog,progi);
2f7f8cb1 1689
73104a1b 1690 PERL_ARGS_ASSERT_FIND_BYCLASS;
2f7f8cb1 1691
73104a1b
KW
1692 /* We know what class it must start with. */
1693 switch (OP(c)) {
1694 case ANYOF:
1695 if (utf8_target) {
1696 REXEC_FBC_UTF8_CLASS_SCAN(
3db24e1e 1697 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
73104a1b
KW
1698 }
1699 else {
1700 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1701 }
1702 break;
1703 case CANY:
1704 REXEC_FBC_SCAN(
02d5137b 1705 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
73104a1b
KW
1706 goto got_it;
1707 else
1708 tmp = doevery;
1709 );
1710 break;
1711
098b07d5
KW
1712 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1713 assert(! is_utf8_pat);
924ba076 1714 /* FALLTHROUGH */
73104a1b 1715 case EXACTFA:
984e6dd1 1716 if (is_utf8_pat || utf8_target) {
73104a1b
KW
1717 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1718 goto do_exactf_utf8;
1719 }
1720 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1721 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1722 goto do_exactf_non_utf8; /* isn't dealt with by these */
77a6d856 1723
2fdb7295
KW
1724 case EXACTF: /* This node only generated for non-utf8 patterns */
1725 assert(! is_utf8_pat);
73104a1b 1726 if (utf8_target) {
73104a1b
KW
1727 utf8_fold_flags = 0;
1728 goto do_exactf_utf8;
1729 }
1730 fold_array = PL_fold;
1731 folder = foldEQ;
1732 goto do_exactf_non_utf8;
1733
1734 case EXACTFL:
31f05a37 1735 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
cea315b6 1736 utf8_fold_flags = FOLDEQ_LOCALE;
73104a1b
KW
1737 goto do_exactf_utf8;
1738 }
1739 fold_array = PL_fold_locale;
1740 folder = foldEQ_locale;
1741 goto do_exactf_non_utf8;
3c760661 1742
73104a1b 1743 case EXACTFU_SS:
984e6dd1 1744 if (is_utf8_pat) {
73104a1b
KW
1745 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1746 }
1747 goto do_exactf_utf8;
16d951b7 1748
73104a1b 1749 case EXACTFU:
984e6dd1
DM
1750 if (is_utf8_pat || utf8_target) {
1751 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
73104a1b
KW
1752 goto do_exactf_utf8;
1753 }
fac1af77 1754
73104a1b
KW
1755 /* Any 'ss' in the pattern should have been replaced by regcomp,
1756 * so we don't have to worry here about this single special case
1757 * in the Latin1 range */
1758 fold_array = PL_fold_latin1;
1759 folder = foldEQ_latin1;
1760
924ba076 1761 /* FALLTHROUGH */
73104a1b
KW
1762
1763 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1764 are no glitches with fold-length differences
1765 between the target string and pattern */
1766
1767 /* The idea in the non-utf8 EXACTF* cases is to first find the
1768 * first character of the EXACTF* node and then, if necessary,
1769 * case-insensitively compare the full text of the node. c1 is the
1770 * first character. c2 is its fold. This logic will not work for
1771 * Unicode semantics and the german sharp ss, which hence should
1772 * not be compiled into a node that gets here. */
1773 pat_string = STRING(c);
1774 ln = STR_LEN(c); /* length to match in octets/bytes */
1775
1776 /* We know that we have to match at least 'ln' bytes (which is the
1777 * same as characters, since not utf8). If we have to match 3
1778 * characters, and there are only 2 availabe, we know without
1779 * trying that it will fail; so don't start a match past the
1780 * required minimum number from the far end */
ea3daa5d 1781 e = HOP3c(strend, -((SSize_t)ln), s);
73104a1b 1782
02d5137b 1783 if (reginfo->intuit && e < s) {
73104a1b
KW
1784 e = s; /* Due to minlen logic of intuit() */
1785 }
fac1af77 1786
73104a1b
KW
1787 c1 = *pat_string;
1788 c2 = fold_array[c1];
1789 if (c1 == c2) { /* If char and fold are the same */
1790 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1791 }
1792 else {
1793 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1794 }
1795 break;
fac1af77 1796
73104a1b
KW
1797 do_exactf_utf8:
1798 {
1799 unsigned expansion;
1800
1801 /* If one of the operands is in utf8, we can't use the simpler folding
1802 * above, due to the fact that many different characters can have the
1803 * same fold, or portion of a fold, or different- length fold */
1804 pat_string = STRING(c);
1805 ln = STR_LEN(c); /* length to match in octets/bytes */
1806 pat_end = pat_string + ln;
984e6dd1 1807 lnc = is_utf8_pat /* length to match in characters */
73104a1b
KW
1808 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1809 : ln;
1810
1811 /* We have 'lnc' characters to match in the pattern, but because of
1812 * multi-character folding, each character in the target can match
1813 * up to 3 characters (Unicode guarantees it will never exceed
1814 * this) if it is utf8-encoded; and up to 2 if not (based on the
1815 * fact that the Latin 1 folds are already determined, and the
1816 * only multi-char fold in that range is the sharp-s folding to
1817 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1818 * string character. Adjust lnc accordingly, rounding up, so that
1819 * if we need to match at least 4+1/3 chars, that really is 5. */
1820 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1821 lnc = (lnc + expansion - 1) / expansion;
1822
1823 /* As in the non-UTF8 case, if we have to match 3 characters, and
1824 * only 2 are left, it's guaranteed to fail, so don't start a
1825 * match that would require us to go beyond the end of the string
1826 */
ea3daa5d 1827 e = HOP3c(strend, -((SSize_t)lnc), s);
73104a1b 1828
02d5137b 1829 if (reginfo->intuit && e < s) {
73104a1b
KW
1830 e = s; /* Due to minlen logic of intuit() */
1831 }
0658cdde 1832
73104a1b
KW
1833 /* XXX Note that we could recalculate e to stop the loop earlier,
1834 * as the worst case expansion above will rarely be met, and as we
1835 * go along we would usually find that e moves further to the left.
1836 * This would happen only after we reached the point in the loop
1837 * where if there were no expansion we should fail. Unclear if
1838 * worth the expense */
1839
1840 while (s <= e) {
1841 char *my_strend= (char *)strend;
1842 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
984e6dd1 1843 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
02d5137b 1844 && (reginfo->intuit || regtry(reginfo, &s)) )
73104a1b
KW
1845 {
1846 goto got_it;
1847 }
1848 s += (utf8_target) ? UTF8SKIP(s) : 1;
1849 }
1850 break;
1851 }
236d82fd 1852
73104a1b 1853 case BOUNDL:
236d82fd 1854 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
73104a1b
KW
1855 break;
1856 case NBOUNDL:
236d82fd 1857 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
73104a1b
KW
1858 break;
1859 case BOUND:
236d82fd 1860 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
73104a1b
KW
1861 break;
1862 case BOUNDA:
236d82fd 1863 FBC_BOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
73104a1b
KW
1864 break;
1865 case NBOUND:
236d82fd 1866 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
73104a1b
KW
1867 break;
1868 case NBOUNDA:
236d82fd 1869 FBC_NBOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
73104a1b
KW
1870 break;
1871 case BOUNDU:
236d82fd 1872 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
73104a1b
KW
1873 break;
1874 case NBOUNDU:
236d82fd 1875 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
73104a1b 1876 break;
73104a1b
KW
1877 case LNBREAK:
1878 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1879 is_LNBREAK_latin1_safe(s, strend)
1880 );
1881 break;
3018b823
KW
1882
1883 /* The argument to all the POSIX node types is the class number to pass to
1884 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1885
1886 case NPOSIXL:
1887 to_complement = 1;
1888 /* FALLTHROUGH */
1889
1890 case POSIXL:
3018b823
KW
1891 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1892 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
73104a1b 1893 break;
3018b823
KW
1894
1895 case NPOSIXD:
1896 to_complement = 1;
1897 /* FALLTHROUGH */
1898
1899 case POSIXD:
1900 if (utf8_target) {
1901 goto posix_utf8;
1902 }
1903 goto posixa;
1904
1905 case NPOSIXA:
1906 if (utf8_target) {
1907 /* The complement of something that matches only ASCII matches all
837226c8
KW
1908 * non-ASCII, plus everything in ASCII that isn't in the class. */
1909 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
3018b823
KW
1910 || ! _generic_isCC_A(*s, FLAGS(c)));
1911 break;
1912 }
1913
1914 to_complement = 1;
1915 /* FALLTHROUGH */
1916
73104a1b 1917 case POSIXA:
3018b823 1918 posixa:
73104a1b 1919 /* Don't need to worry about utf8, as it can match only a single
3018b823
KW
1920 * byte invariant character. */
1921 REXEC_FBC_CLASS_SCAN(
1922 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
73104a1b 1923 break;
3018b823
KW
1924
1925 case NPOSIXU:
1926 to_complement = 1;
1927 /* FALLTHROUGH */
1928
1929 case POSIXU:
1930 if (! utf8_target) {
1931 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1932 FLAGS(c))));
1933 }
1934 else {
1935
1936 posix_utf8:
1937 classnum = (_char_class_number) FLAGS(c);
1938 if (classnum < _FIRST_NON_SWASH_CC) {
1939 while (s < strend) {
1940
1941 /* We avoid loading in the swash as long as possible, but
1942 * should we have to, we jump to a separate loop. This
1943 * extra 'if' statement is what keeps this code from being
1944 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1945 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1946 goto found_above_latin1;
1947 }
1948 if ((UTF8_IS_INVARIANT(*s)
1949 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
1950 classnum)))
1951 || (UTF8_IS_DOWNGRADEABLE_START(*s)
1952 && to_complement ^ cBOOL(
94bb8c36
KW
1953 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
1954 *(s + 1)),
3018b823
KW
1955 classnum))))
1956 {
02d5137b 1957 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
3018b823
KW
1958 goto got_it;
1959 else {
1960 tmp = doevery;
1961 }
1962 }
1963 else {
1964 tmp = 1;
1965 }
1966 s += UTF8SKIP(s);
1967 }
1968 }
1969 else switch (classnum) { /* These classes are implemented as
1970 macros */
1971 case _CC_ENUM_SPACE: /* XXX would require separate code if we
1972 revert the change of \v matching this */
924ba076 1973 /* FALLTHROUGH */
3018b823
KW
1974
1975 case _CC_ENUM_PSXSPC:
1976 REXEC_FBC_UTF8_CLASS_SCAN(
1977 to_complement ^ cBOOL(isSPACE_utf8(s)));
1978 break;
1979
1980 case _CC_ENUM_BLANK:
1981 REXEC_FBC_UTF8_CLASS_SCAN(
1982 to_complement ^ cBOOL(isBLANK_utf8(s)));
1983 break;
1984
1985 case _CC_ENUM_XDIGIT:
1986 REXEC_FBC_UTF8_CLASS_SCAN(
1987 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
1988 break;
1989
1990 case _CC_ENUM_VERTSPACE:
1991 REXEC_FBC_UTF8_CLASS_SCAN(
1992 to_complement ^ cBOOL(isVERTWS_utf8(s)));
1993 break;
1994
1995 case _CC_ENUM_CNTRL:
1996 REXEC_FBC_UTF8_CLASS_SCAN(
1997 to_complement ^ cBOOL(isCNTRL_utf8(s)));
1998 break;
1999
2000 default:
2001 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
a25b5927 2002 assert(0); /* NOTREACHED */
3018b823
KW
2003 }
2004 }
2005 break;
2006
2007 found_above_latin1: /* Here we have to load a swash to get the result
2008 for the current code point */
2009 if (! PL_utf8_swash_ptrs[classnum]) {
2010 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2011 PL_utf8_swash_ptrs[classnum] =
2a16ac92
KW
2012 _core_swash_init("utf8",
2013 "",
2014 &PL_sv_undef, 1, 0,
2015 PL_XPosix_ptrs[classnum], &flags);
3018b823
KW
2016 }
2017
2018 /* This is a copy of the loop above for swash classes, though using the
2019 * FBC macro instead of being expanded out. Since we've loaded the
2020 * swash, we don't have to check for that each time through the loop */
2021 REXEC_FBC_UTF8_CLASS_SCAN(
2022 to_complement ^ cBOOL(_generic_utf8(
2023 classnum,
2024 s,
2025 swash_fetch(PL_utf8_swash_ptrs[classnum],
2026 (U8 *) s, TRUE))));
73104a1b
KW
2027 break;
2028
2029 case AHOCORASICKC:
2030 case AHOCORASICK:
2031 {
2032 DECL_TRIE_TYPE(c);
2033 /* what trie are we using right now */
2034 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2035 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2036 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2037
2038 const char *last_start = strend - trie->minlen;
6148ee25 2039#ifdef DEBUGGING
73104a1b 2040 const char *real_start = s;
6148ee25 2041#endif
73104a1b
KW
2042 STRLEN maxlen = trie->maxlen;
2043 SV *sv_points;
2044 U8 **points; /* map of where we were in the input string
2045 when reading a given char. For ASCII this
2046 is unnecessary overhead as the relationship
2047 is always 1:1, but for Unicode, especially
2048 case folded Unicode this is not true. */
2049 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2050 U8 *bitmap=NULL;
2051
2052
2053 GET_RE_DEBUG_FLAGS_DECL;
2054
2055 /* We can't just allocate points here. We need to wrap it in
2056 * an SV so it gets freed properly if there is a croak while
2057 * running the match */
2058 ENTER;
2059 SAVETMPS;
2060 sv_points=newSV(maxlen * sizeof(U8 *));
2061 SvCUR_set(sv_points,
2062 maxlen * sizeof(U8 *));
2063 SvPOK_on(sv_points);
2064 sv_2mortal(sv_points);
2065 points=(U8**)SvPV_nolen(sv_points );
2066 if ( trie_type != trie_utf8_fold
2067 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2068 {
2069 if (trie->bitmap)
2070 bitmap=(U8*)trie->bitmap;
2071 else
2072 bitmap=(U8*)ANYOF_BITMAP(c);
2073 }
2074 /* this is the Aho-Corasick algorithm modified a touch
2075 to include special handling for long "unknown char" sequences.
2076 The basic idea being that we use AC as long as we are dealing
2077 with a possible matching char, when we encounter an unknown char
2078 (and we have not encountered an accepting state) we scan forward
2079 until we find a legal starting char.
2080 AC matching is basically that of trie matching, except that when
2081 we encounter a failing transition, we fall back to the current
2082 states "fail state", and try the current char again, a process
2083 we repeat until we reach the root state, state 1, or a legal
2084 transition. If we fail on the root state then we can either
2085 terminate if we have reached an accepting state previously, or
2086 restart the entire process from the beginning if we have not.
2087
2088 */
2089 while (s <= last_start) {
2090 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2091 U8 *uc = (U8*)s;
2092 U16 charid = 0;
2093 U32 base = 1;
2094 U32 state = 1;
2095 UV uvc = 0;
2096 STRLEN len = 0;
2097 STRLEN foldlen = 0;
2098 U8 *uscan = (U8*)NULL;
2099 U8 *leftmost = NULL;
2100#ifdef DEBUGGING
2101 U32 accepted_word= 0;
786e8c11 2102#endif
73104a1b
KW
2103 U32 pointpos = 0;
2104
2105 while ( state && uc <= (U8*)strend ) {
2106 int failed=0;
2107 U32 word = aho->states[ state ].wordnum;
2108
2109 if( state==1 ) {
2110 if ( bitmap ) {
2111 DEBUG_TRIE_EXECUTE_r(
2112 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2113 dump_exec_pos( (char *)uc, c, strend, real_start,
2114 (char *)uc, utf8_target );
2115 PerlIO_printf( Perl_debug_log,
2116 " Scanning for legal start char...\n");
2117 }
2118 );
2119 if (utf8_target) {
2120 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2121 uc += UTF8SKIP(uc);
2122 }
2123 } else {
2124 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2125 uc++;
2126 }
786e8c11 2127 }
73104a1b 2128 s= (char *)uc;
07be1b83 2129 }
73104a1b
KW
2130 if (uc >(U8*)last_start) break;
2131 }
2132
2133 if ( word ) {
2134 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2135 if (!leftmost || lpos < leftmost) {
2136 DEBUG_r(accepted_word=word);
2137 leftmost= lpos;
7016d6eb 2138 }
73104a1b 2139 if (base==0) break;
7016d6eb 2140
73104a1b
KW
2141 }
2142 points[pointpos++ % maxlen]= uc;
2143 if (foldlen || uc < (U8*)strend) {
2144 REXEC_TRIE_READ_CHAR(trie_type, trie,
2145 widecharmap, uc,
2146 uscan, len, uvc, charid, foldlen,
2147 foldbuf, uniflags);
2148 DEBUG_TRIE_EXECUTE_r({
2149 dump_exec_pos( (char *)uc, c, strend,
2150 real_start, s, utf8_target);
2151 PerlIO_printf(Perl_debug_log,
2152 " Charid:%3u CP:%4"UVxf" ",
2153 charid, uvc);
2154 });
2155 }
2156 else {
2157 len = 0;
2158 charid = 0;
2159 }
07be1b83 2160
73104a1b
KW
2161
2162 do {
6148ee25 2163#ifdef DEBUGGING
73104a1b 2164 word = aho->states[ state ].wordnum;
6148ee25 2165#endif
73104a1b
KW
2166 base = aho->states[ state ].trans.base;
2167
2168 DEBUG_TRIE_EXECUTE_r({
2169 if (failed)
2170 dump_exec_pos( (char *)uc, c, strend, real_start,
2171 s, utf8_target );
2172 PerlIO_printf( Perl_debug_log,
2173 "%sState: %4"UVxf", word=%"UVxf,
2174 failed ? " Fail transition to " : "",
2175 (UV)state, (UV)word);
2176 });
2177 if ( base ) {
2178 U32 tmp;
2179 I32 offset;
2180 if (charid &&
2181 ( ((offset = base + charid
2182 - 1 - trie->uniquecharcount)) >= 0)
2183 && ((U32)offset < trie->lasttrans)
2184 && trie->trans[offset].check == state
2185 && (tmp=trie->trans[offset].next))
2186 {
2187 DEBUG_TRIE_EXECUTE_r(
2188 PerlIO_printf( Perl_debug_log," - legal\n"));
2189 state = tmp;
2190 break;
07be1b83
YO
2191 }
2192 else {
786e8c11 2193 DEBUG_TRIE_EXECUTE_r(
73104a1b 2194 PerlIO_printf( Perl_debug_log," - fail\n"));
786e8c11 2195 failed = 1;
73104a1b 2196 state = aho->fail[state];
07be1b83 2197 }
07be1b83 2198 }
73104a1b
KW
2199 else {
2200 /* we must be accepting here */
2201 DEBUG_TRIE_EXECUTE_r(
2202 PerlIO_printf( Perl_debug_log," - accepting\n"));
2203 failed = 1;
2204 break;
786e8c11 2205 }
73104a1b
KW
2206 } while(state);
2207 uc += len;
2208 if (failed) {
2209 if (leftmost)
2210 break;
2211 if (!state) state = 1;
07be1b83 2212 }
73104a1b
KW
2213 }
2214 if ( aho->states[ state ].wordnum ) {
2215 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2216 if (!leftmost || lpos < leftmost) {
2217 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2218 leftmost = lpos;
07be1b83
YO
2219 }
2220 }
73104a1b
KW
2221 if (leftmost) {
2222 s = (char*)leftmost;
2223 DEBUG_TRIE_EXECUTE_r({
2224 PerlIO_printf(
2225 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2226 (UV)accepted_word, (IV)(s - real_start)
2227 );
2228 });
02d5137b 2229 if (reginfo->intuit || regtry(reginfo, &s)) {
73104a1b
KW
2230 FREETMPS;
2231 LEAVE;
2232 goto got_it;
2233 }
2234 s = HOPc(s,1);
2235 DEBUG_TRIE_EXECUTE_r({
2236 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2237 });
2238 } else {
2239 DEBUG_TRIE_EXECUTE_r(
2240 PerlIO_printf( Perl_debug_log,"No match.\n"));
2241 break;
2242 }
2243 }
2244 FREETMPS;
2245 LEAVE;
2246 }
2247 break;
2248 default:
2249 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
73104a1b
KW
2250 }
2251 return 0;
2252 got_it:
2253 return s;
6eb5f6b9
JH
2254}
2255
60165aa4
DM
2256/* set RX_SAVED_COPY, RX_SUBBEG etc.
2257 * flags have same meanings as with regexec_flags() */
2258
749f4950
DM
2259static void
2260S_reg_set_capture_string(pTHX_ REGEXP * const rx,
60165aa4
DM
2261 char *strbeg,
2262 char *strend,
2263 SV *sv,
2264 U32 flags,
2265 bool utf8_target)
2266{
2267 struct regexp *const prog = ReANY(rx);
2268
60165aa4
DM
2269 if (flags & REXEC_COPY_STR) {
2270#ifdef PERL_ANY_COW
2271 if (SvCANCOW(sv)) {
2272 if (DEBUG_C_TEST) {
2273 PerlIO_printf(Perl_debug_log,
2274 "Copy on write: regexp capture, type %d\n",
2275 (int) SvTYPE(sv));
2276 }
5411a0e5
DM
2277 /* Create a new COW SV to share the match string and store
2278 * in saved_copy, unless the current COW SV in saved_copy
2279 * is valid and suitable for our purpose */
2280 if (( prog->saved_copy
2281 && SvIsCOW(prog->saved_copy)
2282 && SvPOKp(prog->saved_copy)
2283 && SvIsCOW(sv)
2284 && SvPOKp(sv)
2285 && SvPVX(sv) == SvPVX(prog->saved_copy)))
a76b0e90 2286 {
5411a0e5
DM
2287 /* just reuse saved_copy SV */
2288 if (RXp_MATCH_COPIED(prog)) {
2289 Safefree(prog->subbeg);
2290 RXp_MATCH_COPIED_off(prog);
2291 }
2292 }
2293 else {
2294 /* create new COW SV to share string */
a76b0e90
DM
2295 RX_MATCH_COPY_FREE(rx);
2296 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
a76b0e90 2297 }
5411a0e5
DM
2298 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2299 assert (SvPOKp(prog->saved_copy));
60165aa4
DM
2300 prog->sublen = strend - strbeg;
2301 prog->suboffset = 0;
2302 prog->subcoffset = 0;
2303 } else
2304#endif
2305 {
99a90e59
FC
2306 SSize_t min = 0;
2307 SSize_t max = strend - strbeg;
ea3daa5d 2308 SSize_t sublen;
60165aa4
DM
2309
2310 if ( (flags & REXEC_COPY_SKIP_POST)
e322109a 2311 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM
2312 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2313 ) { /* don't copy $' part of string */
2314 U32 n = 0;
2315 max = -1;
2316 /* calculate the right-most part of the string covered
2317 * by a capture. Due to look-ahead, this may be to
2318 * the right of $&, so we have to scan all captures */
2319 while (n <= prog->lastparen) {
2320 if (prog->offs[n].end > max)
2321 max = prog->offs[n].end;
2322 n++;
2323 }
2324 if (max == -1)
2325 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2326 ? prog->offs[0].start
2327 : 0;
2328 assert(max >= 0 && max <= strend - strbeg);
2329 }
2330
2331 if ( (flags & REXEC_COPY_SKIP_PRE)
e322109a 2332 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM
2333 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2334 ) { /* don't copy $` part of string */
2335 U32 n = 0;
2336 min = max;
2337 /* calculate the left-most part of the string covered
2338 * by a capture. Due to look-behind, this may be to
2339 * the left of $&, so we have to scan all captures */
2340 while (min && n <= prog->lastparen) {
2341 if ( prog->offs[n].start != -1
2342 && prog->offs[n].start < min)
2343 {
2344 min = prog->offs[n].start;
2345 }
2346 n++;
2347 }
2348 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2349 && min > prog->offs[0].end
2350 )
2351 min = prog->offs[0].end;
2352
2353 }
2354
2355 assert(min >= 0 && min <= max && min <= strend - strbeg);
2356 sublen = max - min;
2357
2358 if (RX_MATCH_COPIED(rx)) {
2359 if (sublen > prog->sublen)
2360 prog->subbeg =
2361 (char*)saferealloc(prog->subbeg, sublen+1);
2362 }
2363 else
2364 prog->subbeg = (char*)safemalloc(sublen+1);
2365 Copy(strbeg + min, prog->subbeg, sublen, char);
2366 prog->subbeg[sublen] = '\0';
2367 prog->suboffset = min;
2368 prog->sublen = sublen;
2369 RX_MATCH_COPIED_on(rx);
2370 }
2371 prog->subcoffset = prog->suboffset;
2372 if (prog->suboffset && utf8_target) {
2373 /* Convert byte offset to chars.
2374 * XXX ideally should only compute this if @-/@+
2375 * has been seen, a la PL_sawampersand ??? */
2376
2377 /* If there's a direct correspondence between the
2378 * string which we're matching and the original SV,
2379 * then we can use the utf8 len cache associated with
2380 * the SV. In particular, it means that under //g,
2381 * sv_pos_b2u() will use the previously cached
2382 * position to speed up working out the new length of
2383 * subcoffset, rather than counting from the start of
2384 * the string each time. This stops
2385 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2386 * from going quadratic */
2387 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
ea3daa5d
FC
2388 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2389 SV_GMAGIC|SV_CONST_RETURN);
60165aa4
DM
2390 else
2391 prog->subcoffset = utf8_length((U8*)strbeg,
2392 (U8*)(strbeg+prog->suboffset));
2393 }
2394 }
2395 else {
2396 RX_MATCH_COPY_FREE(rx);
2397 prog->subbeg = strbeg;
2398 prog->suboffset = 0;
2399 prog->subcoffset = 0;
2400 prog->sublen = strend - strbeg;
2401 }
2402}
2403
2404
2405
fae667d5 2406
6eb5f6b9
JH
2407/*
2408 - regexec_flags - match a regexp against a string
2409 */
2410I32
5aaab254 2411Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
ea3daa5d 2412 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
8fd1a950
DM
2413/* stringarg: the point in the string at which to begin matching */
2414/* strend: pointer to null at end of string */
2415/* strbeg: real beginning of string */
2416/* minend: end of match must be >= minend bytes after stringarg. */
2417/* sv: SV being matched: only used for utf8 flag, pos() etc; string
2418 * itself is accessed via the pointers above */
2419/* data: May be used for some additional optimizations.
d058ec57 2420 Currently unused. */
a340edde 2421/* flags: For optimizations. See REXEC_* in regexp.h */
8fd1a950 2422
6eb5f6b9 2423{
8d919b0a 2424 struct regexp *const prog = ReANY(rx);
5aaab254 2425 char *s;
eb578fdb 2426 regnode *c;
03c83e26 2427 char *startpos;
ea3daa5d
FC
2428 SSize_t minlen; /* must match at least this many chars */
2429 SSize_t dontbother = 0; /* how many characters not to try at end */
f2ed9b32 2430 const bool utf8_target = cBOOL(DO_UTF8(sv));
2757e526 2431 I32 multiline;
f8fc2ecf 2432 RXi_GET_DECL(prog,progi);
02d5137b
DM
2433 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2434 regmatch_info *const reginfo = &reginfo_buf;
e9105d30 2435 regexp_paren_pair *swap = NULL;
006f26b2 2436 I32 oldsave;
a3621e74
YO
2437 GET_RE_DEBUG_FLAGS_DECL;
2438
7918f24d 2439 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
9d4ba2ae 2440 PERL_UNUSED_ARG(data);
6eb5f6b9
JH
2441
2442 /* Be paranoid... */
03c83e26 2443 if (prog == NULL || stringarg == NULL) {
6eb5f6b9 2444 Perl_croak(aTHX_ "NULL regexp parameter");
6eb5f6b9
JH
2445 }
2446
6c3fea77 2447 DEBUG_EXECUTE_r(
03c83e26 2448 debug_start_match(rx, utf8_target, stringarg, strend,
6c3fea77
DM
2449 "Matching");
2450 );
8adc0f72 2451
b342a604
DM
2452 startpos = stringarg;
2453
58430ea8 2454 if (prog->intflags & PREGf_GPOS_SEEN) {
d307c076
DM
2455 MAGIC *mg;
2456
fef7148b
DM
2457 /* set reginfo->ganch, the position where \G can match */
2458
2459 reginfo->ganch =
2460 (flags & REXEC_IGNOREPOS)
2461 ? stringarg /* use start pos rather than pos() */
2462 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
25fdce4a
FC
2463 /* Defined pos(): */
2464 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
fef7148b
DM
2465 : strbeg; /* pos() not defined; use start of string */
2466
2467 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
7b0eb0b8 2468 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg)));
fef7148b 2469
03c83e26
DM
2470 /* in the presence of \G, we may need to start looking earlier in
2471 * the string than the suggested start point of stringarg:
0b2c2a84 2472 * if prog->gofs is set, then that's a known, fixed minimum
03c83e26
DM
2473 * offset, such as
2474 * /..\G/: gofs = 2
2475 * /ab|c\G/: gofs = 1
2476 * or if the minimum offset isn't known, then we have to go back
2477 * to the start of the string, e.g. /w+\G/
2478 */
2bfbe302 2479
8e1490ee 2480 if (prog->intflags & PREGf_ANCH_GPOS) {
2bfbe302
DM
2481 startpos = reginfo->ganch - prog->gofs;
2482 if (startpos <
2483 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2484 {
2485 DEBUG_r(PerlIO_printf(Perl_debug_log,
2486 "fail: ganch-gofs before earliest possible start\n"));
2487 return 0;
2488 }
2489 }
2490 else if (prog->gofs) {
b342a604
DM
2491 if (startpos - prog->gofs < strbeg)
2492 startpos = strbeg;
2493 else
2494 startpos -= prog->gofs;
03c83e26 2495 }
58430ea8 2496 else if (prog->intflags & PREGf_GPOS_FLOAT)
b342a604 2497 startpos = strbeg;
03c83e26
DM
2498 }
2499
2500 minlen = prog->minlen;
b342a604 2501 if ((startpos + minlen) > strend || startpos < strbeg) {
03c83e26
DM
2502 DEBUG_r(PerlIO_printf(Perl_debug_log,
2503 "Regex match can't succeed, so not even tried\n"));
2504 return 0;
2505 }
2506
63a3746a
DM
2507 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2508 * which will call destuctors to reset PL_regmatch_state, free higher
2509 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2510 * regmatch_info_aux_eval */
2511
2512 oldsave = PL_savestack_ix;
2513
dfa77d06
DM
2514 s = startpos;
2515
e322109a 2516 if ((prog->extflags & RXf_USE_INTUIT)
7fadf4a7
DM
2517 && !(flags & REXEC_CHECKED))
2518 {
dfa77d06 2519 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
7fadf4a7 2520 flags, NULL);
dfa77d06 2521 if (!s)
7fadf4a7
DM
2522 return 0;
2523
e322109a 2524 if (prog->extflags & RXf_CHECK_ALL) {
7fadf4a7
DM
2525 /* we can match based purely on the result of INTUIT.
2526 * Set up captures etc just for $& and $-[0]
2527 * (an intuit-only match wont have $1,$2,..) */
2528 assert(!prog->nparens);
d5e7783a
DM
2529
2530 /* s/// doesn't like it if $& is earlier than where we asked it to
2531 * start searching (which can happen on something like /.\G/) */
2532 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2533 && (s < stringarg))
2534 {
2535 /* this should only be possible under \G */
58430ea8 2536 assert(prog->intflags & PREGf_GPOS_SEEN);
d5e7783a
DM
2537 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2538 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2539 goto phooey;
2540 }
2541
7fadf4a7
DM
2542 /* match via INTUIT shouldn't have any captures.
2543 * Let @-, @+, $^N know */
2544 prog->lastparen = prog->lastcloseparen = 0;
2545 RX_MATCH_UTF8_set(rx, utf8_target);
3ff69bd6
DM
2546 prog->offs[0].start = s - strbeg;
2547 prog->offs[0].end = utf8_target
2548 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2549 : s - strbeg + prog->minlenret;
7fadf4a7 2550 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 2551 S_reg_set_capture_string(aTHX_ rx,
7fadf4a7
DM
2552 strbeg, strend,
2553 sv, flags, utf8_target);
2554
7fadf4a7
DM
2555 return 1;
2556 }
2557 }
2558
6c3fea77 2559 multiline = prog->extflags & RXf_PMf_MULTILINE;
1de06328 2560
dfa77d06 2561 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
a3621e74 2562 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
a72c7584
JH
2563 "String too short [regexec_flags]...\n"));
2564 goto phooey;
1aa99e6b 2565 }
1de06328 2566
6eb5f6b9 2567 /* Check validity of program. */
f8fc2ecf 2568 if (UCHARAT(progi->program) != REG_MAGIC) {
6eb5f6b9
JH
2569 Perl_croak(aTHX_ "corrupted regexp program");
2570 }
2571
1738e041
DM
2572 RX_MATCH_TAINTED_off(rx);
2573
6c3fea77
DM
2574 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2575 reginfo->intuit = 0;
2576 reginfo->is_utf8_target = cBOOL(utf8_target);
02d5137b
DM
2577 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2578 reginfo->warned = FALSE;
9d9163fb 2579 reginfo->strbeg = strbeg;
02d5137b 2580 reginfo->sv = sv;
1cb48e53 2581 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
220db18a 2582 reginfo->strend = strend;
6eb5f6b9 2583 /* see how far we have to get to not match where we matched before */
fe3974be 2584 reginfo->till = stringarg + minend;
6eb5f6b9 2585
60779a30 2586 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
82c23608
FC
2587 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2588 S_cleanup_regmatch_info_aux has executed (registered by
2589 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2590 magic belonging to this SV.
2591 Not newSVsv, either, as it does not COW.
2592 */
60779a30 2593 assert(!IS_PADGV(sv));
82c23608 2594 reginfo->sv = newSV(0);
4cba5ac0 2595 SvSetSV_nosteal(reginfo->sv, sv);
82c23608
FC
2596 SAVEFREESV(reginfo->sv);
2597 }
2598
331b2dcc
DM
2599 /* reserve next 2 or 3 slots in PL_regmatch_state:
2600 * slot N+0: may currently be in use: skip it
2601 * slot N+1: use for regmatch_info_aux struct
2602 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2603 * slot N+3: ready for use by regmatch()
2604 */
bf2039a9 2605
331b2dcc
DM
2606 {
2607 regmatch_state *old_regmatch_state;
2608 regmatch_slab *old_regmatch_slab;
2609 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2610
2611 /* on first ever match, allocate first slab */
2612 if (!PL_regmatch_slab) {
2613 Newx(PL_regmatch_slab, 1, regmatch_slab);
2614 PL_regmatch_slab->prev = NULL;
2615 PL_regmatch_slab->next = NULL;
2616 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2617 }
bf2039a9 2618
331b2dcc
DM
2619 old_regmatch_state = PL_regmatch_state;
2620 old_regmatch_slab = PL_regmatch_slab;
bf2039a9 2621
331b2dcc
DM
2622 for (i=0; i <= max; i++) {
2623 if (i == 1)
2624 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2625 else if (i ==2)
2626 reginfo->info_aux_eval =
2627 reginfo->info_aux->info_aux_eval =
2628 &(PL_regmatch_state->u.info_aux_eval);
bf2039a9 2629
331b2dcc
DM
2630 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2631 PL_regmatch_state = S_push_slab(aTHX);
2632 }
bf2039a9 2633
331b2dcc
DM
2634 /* note initial PL_regmatch_state position; at end of match we'll
2635 * pop back to there and free any higher slabs */
bf2039a9 2636
331b2dcc
DM
2637 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2638 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2ac8ff4b 2639 reginfo->info_aux->poscache = NULL;
bf2039a9 2640
331b2dcc 2641 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
bf2039a9 2642
331b2dcc
DM
2643 if ((prog->extflags & RXf_EVAL_SEEN))
2644 S_setup_eval_state(aTHX_ reginfo);
2645 else
2646 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
bf2039a9 2647 }
d3aa529c 2648
6eb5f6b9 2649 /* If there is a "must appear" string, look for it. */
6eb5f6b9 2650
288b8c02 2651 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
e9105d30
GG
2652 /* We have to be careful. If the previous successful match
2653 was from this regex we don't want a subsequent partially
2654 successful match to clobber the old results.
2655 So when we detect this possibility we add a swap buffer
d8da0584
KW
2656 to the re, and switch the buffer each match. If we fail,
2657 we switch it back; otherwise we leave it swapped.
e9105d30
GG
2658 */
2659 swap = prog->offs;
2660 /* do we need a save destructor here for eval dies? */
2661 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
495f47a5
DM
2662 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2663 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2664 PTR2UV(prog),
2665 PTR2UV(swap),
2666 PTR2UV(prog->offs)
2667 ));
c74340f9 2668 }
6eb5f6b9
JH
2669
2670 /* Simplest case: anchored match need be tried only once. */
2671 /* [unless only anchor is BOL and multiline is set] */
8e1490ee 2672 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
3542935d 2673 if (s == startpos && regtry(reginfo, &s))
6eb5f6b9 2674 goto got_it;
8e1490ee 2675 else if (multiline || (prog->intflags & (PREGf_IMPLICIT | PREGf_ANCH_MBOL))) /* XXXX SBOL? */
6eb5f6b9
JH
2676 {
2677 char *end;
2678
2679 if (minlen)
2680 dontbother = minlen - 1;
1aa99e6b 2681 end = HOP3c(strend, -dontbother, strbeg) - 1;
6eb5f6b9 2682 /* for multiline we only have to try after newlines */
33b8afdf 2683 if (prog->check_substr || prog->check_utf8) {
92f3d482
YO
2684 /* because of the goto we can not easily reuse the macros for bifurcating the
2685 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2686 if (utf8_target) {
2687 if (s == startpos)
2688 goto after_try_utf8;
2689 while (1) {
02d5137b 2690 if (regtry(reginfo, &s)) {
92f3d482
YO
2691 goto got_it;
2692 }
2693 after_try_utf8:
2694 if (s > end) {
2695 goto phooey;
2696 }
2697 if (prog->extflags & RXf_USE_INTUIT) {
52a21eb3
DM
2698 s = re_intuit_start(rx, sv, strbeg,
2699 s + UTF8SKIP(s), strend, flags, NULL);
92f3d482
YO
2700 if (!s) {
2701 goto phooey;
2702 }
2703 }
2704 else {
2705 s += UTF8SKIP(s);
2706 }
2707 }
2708 } /* end search for check string in unicode */
2709 else {
2710 if (s == startpos) {
2711 goto after_try_latin;
2712 }
2713 while (1) {
02d5137b 2714 if (regtry(reginfo, &s)) {
92f3d482
YO
2715 goto got_it;
2716 }
2717 after_try_latin:
2718 if (s > end) {
2719 goto phooey;
2720 }
2721 if (prog->extflags & RXf_USE_INTUIT) {
52a21eb3
DM
2722 s = re_intuit_start(rx, sv, strbeg,
2723 s + 1, strend, flags, NULL);
92f3d482
YO
2724 if (!s) {
2725 goto phooey;
2726 }
2727 }
2728 else {
2729 s++;
2730 }
2731 }
2732 } /* end search for check string in latin*/
2733 } /* end search for check string */
2734 else { /* search for newline */
2735 if (s > startpos) {
2736 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
6eb5f6b9 2737 s--;
92f3d482 2738 }
21eede78
YO
2739 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2740 while (s <= end) { /* note it could be possible to match at the end of the string */
6eb5f6b9 2741 if (*s++ == '\n') { /* don't need PL_utf8skip here */
02d5137b 2742 if (regtry(reginfo, &s))
6eb5f6b9
JH
2743 goto got_it;
2744 }
92f3d482
YO
2745 }
2746 } /* end search for newline */
2747 } /* end anchored/multiline check string search */
6eb5f6b9 2748 goto phooey;
a8430a8b 2749 } else if (prog->intflags & PREGf_ANCH_GPOS)
f9f4320a 2750 {
a8430a8b
YO
2751 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
2752 assert(prog->intflags & PREGf_GPOS_SEEN);
2bfbe302
DM
2753 /* For anchored \G, the only position it can match from is
2754 * (ganch-gofs); we already set startpos to this above; if intuit
2755 * moved us on from there, we can't possibly succeed */
2756 assert(startpos == reginfo->ganch - prog->gofs);
2757 if (s == startpos && regtry(reginfo, &s))
6eb5f6b9
JH
2758 goto got_it;
2759 goto phooey;
2760 }
2761
2762 /* Messy cases: unanchored match. */
bbe252da 2763 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
6eb5f6b9 2764 /* we have /x+whatever/ */
984e6dd1 2765 /* it must be a one character string (XXXX Except is_utf8_pat?) */
33b8afdf 2766 char ch;
bf93d4cc
GS
2767#ifdef DEBUGGING
2768 int did_match = 0;
2769#endif
f2ed9b32 2770 if (utf8_target) {
7e0d5ad7
KW
2771 if (! prog->anchored_utf8) {
2772 to_utf8_substr(prog);
2773 }
2774 ch = SvPVX_const(prog->anchored_utf8)[0];
4cadc6a9 2775 REXEC_FBC_SCAN(
6eb5f6b9 2776 if (*s == ch) {
a3621e74 2777 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 2778 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH
2779 s += UTF8SKIP(s);
2780 while (s < strend && *s == ch)
2781 s += UTF8SKIP(s);
2782 }
4cadc6a9 2783 );
7e0d5ad7 2784
6eb5f6b9
JH
2785 }
2786 else {
7e0d5ad7
KW
2787 if (! prog->anchored_substr) {
2788 if (! to_byte_substr(prog)) {
6b54ddc5 2789 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2790 }
2791 }
2792 ch = SvPVX_const(prog->anchored_substr)[0];
4cadc6a9 2793 REXEC_FBC_SCAN(
6eb5f6b9 2794 if (*s == ch) {
a3621e74 2795 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 2796 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH
2797 s++;
2798 while (s < strend && *s == ch)
2799 s++;
2800 }
4cadc6a9 2801 );
6eb5f6b9 2802 }
a3621e74 2803 DEBUG_EXECUTE_r(if (!did_match)
bf93d4cc 2804 PerlIO_printf(Perl_debug_log,
b7953727
JH
2805 "Did not find anchored character...\n")
2806 );
6eb5f6b9 2807 }
a0714e2c
SS
2808 else if (prog->anchored_substr != NULL
2809 || prog->anchored_utf8 != NULL
2810 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
33b8afdf
JH
2811 && prog->float_max_offset < strend - s)) {
2812 SV *must;
ea3daa5d
FC
2813 SSize_t back_max;
2814 SSize_t back_min;
33b8afdf 2815 char *last;
6eb5f6b9 2816 char *last1; /* Last position checked before */
bf93d4cc
GS
2817#ifdef DEBUGGING
2818 int did_match = 0;
2819#endif
33b8afdf 2820 if (prog->anchored_substr || prog->anchored_utf8) {
7e0d5ad7
KW
2821 if (utf8_target) {
2822 if (! prog->anchored_utf8) {
2823 to_utf8_substr(prog);
2824 }
2825 must = prog->anchored_utf8;
2826 }
2827 else {
2828 if (! prog->anchored_substr) {
2829 if (! to_byte_substr(prog)) {
6b54ddc5 2830 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2831 }
2832 }
2833 must = prog->anchored_substr;
2834 }
33b8afdf
JH
2835 back_max = back_min = prog->anchored_offset;
2836 } else {
7e0d5ad7
KW
2837 if (utf8_target) {
2838 if (! prog->float_utf8) {
2839 to_utf8_substr(prog);
2840 }
2841 must = prog->float_utf8;
2842 }
2843 else {
2844 if (! prog->float_substr) {
2845 if (! to_byte_substr(prog)) {
6b54ddc5 2846 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2847 }
2848 }
2849 must = prog->float_substr;
2850 }
33b8afdf
JH
2851 back_max = prog->float_max_offset;
2852 back_min = prog->float_min_offset;
2853 }
1de06328 2854
1de06328
YO
2855 if (back_min<0) {
2856 last = strend;
2857 } else {
2858 last = HOP3c(strend, /* Cannot start after this */
ea3daa5d 2859 -(SSize_t)(CHR_SVLEN(must)
1de06328
YO
2860 - (SvTAIL(must) != 0) + back_min), strbeg);
2861 }
9d9163fb 2862 if (s > reginfo->strbeg)
6eb5f6b9
JH
2863 last1 = HOPc(s, -1);
2864 else
2865 last1 = s - 1; /* bogus */
2866
a0288114 2867 /* XXXX check_substr already used to find "s", can optimize if
6eb5f6b9 2868 check_substr==must. */
bf05793b 2869 dontbother = 0;
6eb5f6b9
JH
2870 strend = HOPc(strend, -dontbother);
2871 while ( (s <= last) &&
e50d57d4 2872 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
9041c2e3 2873 (unsigned char*)strend, must,
c33e64f0 2874 multiline ? FBMrf_MULTILINE : 0)) ) {
a3621e74 2875 DEBUG_EXECUTE_r( did_match = 1 );
6eb5f6b9
JH
2876 if (HOPc(s, -back_max) > last1) {
2877 last1 = HOPc(s, -back_min);
2878 s = HOPc(s, -back_max);
2879 }
2880 else {
9d9163fb
DM
2881 char * const t = (last1 >= reginfo->strbeg)
2882 ? HOPc(last1, 1) : last1 + 1;
6eb5f6b9
JH
2883
2884 last1 = HOPc(s, -back_min);
52657f30 2885 s = t;
6eb5f6b9 2886 }
f2ed9b32 2887 if (utf8_target) {
6eb5f6b9 2888 while (s <= last1) {
02d5137b 2889 if (regtry(reginfo, &s))
6eb5f6b9 2890 goto got_it;
7016d6eb
DM
2891 if (s >= last1) {
2892 s++; /* to break out of outer loop */
2893 break;
2894 }
2895 s += UTF8SKIP(s);
6eb5f6b9
JH
2896 }
2897 }
2898 else {
2899 while (s <= last1) {
02d5137b 2900 if (regtry(reginfo, &s))
6eb5f6b9
JH
2901 goto got_it;
2902 s++;
2903 }
2904 }
2905 }
ab3bbdeb 2906 DEBUG_EXECUTE_r(if (!did_match) {
f2ed9b32 2907 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
ab3bbdeb
YO
2908 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2909 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
33b8afdf 2910 ((must == prog->anchored_substr || must == prog->anchored_utf8)
bf93d4cc 2911 ? "anchored" : "floating"),
ab3bbdeb
YO
2912 quoted, RE_SV_TAIL(must));
2913 });
6eb5f6b9
JH
2914 goto phooey;
2915 }
f8fc2ecf 2916 else if ( (c = progi->regstclass) ) {
f14c76ed 2917 if (minlen) {
f8fc2ecf 2918 const OPCODE op = OP(progi->regstclass);
66e933ab 2919 /* don't bother with what can't match */
786e8c11 2920 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
f14c76ed
RGS
2921 strend = HOPc(strend, -(minlen - 1));
2922 }
a3621e74 2923 DEBUG_EXECUTE_r({
be8e71aa 2924 SV * const prop = sv_newmortal();
2395827c 2925 regprop(prog, prop, c, reginfo);
0df25f3d 2926 {
f2ed9b32 2927 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb 2928 s,strend-s,60);
0df25f3d 2929 PerlIO_printf(Perl_debug_log,
1c8f8eb1 2930 "Matching stclass %.*s against %s (%d bytes)\n",
e4f74956 2931 (int)SvCUR(prop), SvPVX_const(prop),
ab3bbdeb 2932 quoted, (int)(strend - s));
0df25f3d 2933 }
ffc61ed2 2934 });
f9176b44 2935 if (find_byclass(prog, c, s, strend, reginfo))
6eb5f6b9 2936 goto got_it;
07be1b83 2937 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
d6a28714
JH
2938 }
2939 else {
2940 dontbother = 0;
a0714e2c 2941 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
33b8afdf 2942 /* Trim the end. */
6af40bd7 2943 char *last= NULL;
33b8afdf 2944 SV* float_real;
c33e64f0
FC
2945 STRLEN len;
2946 const char *little;
33b8afdf 2947
7e0d5ad7
KW
2948 if (utf8_target) {
2949 if (! prog->float_utf8) {
2950 to_utf8_substr(prog);
2951 }
2952 float_real = prog->float_utf8;
2953 }
2954 else {
2955 if (! prog->float_substr) {
2956 if (! to_byte_substr(prog)) {
6b54ddc5 2957 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2958 }
2959 }
2960 float_real = prog->float_substr;
2961 }
d6a28714 2962
c33e64f0
FC
2963 little = SvPV_const(float_real, len);
2964 if (SvTAIL(float_real)) {
7f18ad16
KW
2965 /* This means that float_real contains an artificial \n on
2966 * the end due to the presence of something like this:
2967 * /foo$/ where we can match both "foo" and "foo\n" at the
2968 * end of the string. So we have to compare the end of the
2969 * string first against the float_real without the \n and
2970 * then against the full float_real with the string. We
2971 * have to watch out for cases where the string might be
2972 * smaller than the float_real or the float_real without
2973 * the \n. */
1a13b075
YO
2974 char *checkpos= strend - len;
2975 DEBUG_OPTIMISE_r(
2976 PerlIO_printf(Perl_debug_log,
2977 "%sChecking for float_real.%s\n",
2978 PL_colors[4], PL_colors[5]));
2979 if (checkpos + 1 < strbeg) {
7f18ad16
KW
2980 /* can't match, even if we remove the trailing \n
2981 * string is too short to match */
1a13b075
YO
2982 DEBUG_EXECUTE_r(
2983 PerlIO_printf(Perl_debug_log,
2984 "%sString shorter than required trailing substring, cannot match.%s\n",
2985 PL_colors[4], PL_colors[5]));
2986 goto phooey;
2987 } else if (memEQ(checkpos + 1, little, len - 1)) {
7f18ad16
KW
2988 /* can match, the end of the string matches without the
2989 * "\n" */
1a13b075
YO
2990 last = checkpos + 1;
2991 } else if (checkpos < strbeg) {
7f18ad16
KW
2992 /* cant match, string is too short when the "\n" is
2993 * included */
1a13b075
YO
2994 DEBUG_EXECUTE_r(
2995 PerlIO_printf(Perl_debug_log,
2996 "%sString does not contain required trailing substring, cannot match.%s\n",
2997 PL_colors[4], PL_colors[5]));
2998 goto phooey;
2999 } else if (!multiline) {
7f18ad16
KW
3000 /* non multiline match, so compare with the "\n" at the
3001 * end of the string */
1a13b075
YO
3002 if (memEQ(checkpos, little, len)) {
3003 last= checkpos;
3004 } else {
3005 DEBUG_EXECUTE_r(
3006 PerlIO_printf(Perl_debug_log,
3007 "%sString does not contain required trailing substring, cannot match.%s\n",
3008 PL_colors[4], PL_colors[5]));
3009 goto phooey;
3010 }
3011 } else {
7f18ad16
KW
3012 /* multiline match, so we have to search for a place
3013 * where the full string is located */
d6a28714 3014 goto find_last;
1a13b075 3015 }
c33e64f0 3016 } else {
d6a28714 3017 find_last:
9041c2e3 3018 if (len)
d6a28714 3019 last = rninstr(s, strend, little, little + len);
b8c5462f 3020 else
a0288114 3021 last = strend; /* matching "$" */
b8c5462f 3022 }
6af40bd7 3023 if (!last) {
7f18ad16
KW
3024 /* at one point this block contained a comment which was
3025 * probably incorrect, which said that this was a "should not
3026 * happen" case. Even if it was true when it was written I am
3027 * pretty sure it is not anymore, so I have removed the comment
3028 * and replaced it with this one. Yves */
6bda09f9
YO
3029 DEBUG_EXECUTE_r(
3030 PerlIO_printf(Perl_debug_log,
6af40bd7
YO
3031 "String does not contain required substring, cannot match.\n"
3032 ));
3033 goto phooey;
bf93d4cc 3034 }
d6a28714
JH
3035 dontbother = strend - last + prog->float_min_offset;
3036 }
3037 if (minlen && (dontbother < minlen))
3038 dontbother = minlen - 1;
3039 strend -= dontbother; /* this one's always in bytes! */
3040 /* We don't know much -- general case. */
f2ed9b32 3041 if (utf8_target) {
d6a28714 3042 for (;;) {
02d5137b 3043 if (regtry(reginfo, &s))
d6a28714
JH
3044 goto got_it;
3045 if (s >= strend)
3046 break;
b8c5462f 3047 s += UTF8SKIP(s);
d6a28714
JH
3048 };
3049 }
3050 else {
3051 do {
02d5137b 3052 if (regtry(reginfo, &s))
d6a28714
JH
3053 goto got_it;
3054 } while (s++ < strend);
3055 }
3056 }
3057
3058 /* Failure. */
3059 goto phooey;
3060
3061got_it:
d5e7783a
DM
3062 /* s/// doesn't like it if $& is earlier than where we asked it to
3063 * start searching (which can happen on something like /.\G/) */
3064 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3065 && (prog->offs[0].start < stringarg - strbeg))
3066 {
3067 /* this should only be possible under \G */
58430ea8 3068 assert(prog->intflags & PREGf_GPOS_SEEN);
d5e7783a
DM
3069 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3070 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3071 goto phooey;
3072 }
3073
495f47a5
DM
3074 DEBUG_BUFFERS_r(
3075 if (swap)
3076 PerlIO_printf(Perl_debug_log,
3077 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
3078 PTR2UV(prog),
3079 PTR2UV(swap)
3080 );
3081 );
e9105d30 3082 Safefree(swap);
d6a28714 3083
bf2039a9
DM
3084 /* clean up; this will trigger destructors that will free all slabs
3085 * above the current one, and cleanup the regmatch_info_aux
3086 * and regmatch_info_aux_eval sructs */
8adc0f72 3087
006f26b2
DM
3088 LEAVE_SCOPE(oldsave);
3089
5daac39c
NC
3090 if (RXp_PAREN_NAMES(prog))
3091 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
d6a28714 3092
0254aed9
DM
3093 RX_MATCH_UTF8_set(rx, utf8_target);
3094
d6a28714 3095 /* make sure $`, $&, $', and $digit will work later */
60165aa4 3096 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 3097 S_reg_set_capture_string(aTHX_ rx,
60165aa4
DM
3098 strbeg, reginfo->strend,
3099 sv, flags, utf8_target);
9041c2e3 3100
d6a28714
JH
3101 return 1;
3102
3103phooey:
a3621e74 3104 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
e4584336 3105 PL_colors[4], PL_colors[5]));
8adc0f72 3106
bf2039a9
DM
3107 /* clean up; this will trigger destructors that will free all slabs
3108 * above the current one, and cleanup the regmatch_info_aux
3109 * and regmatch_info_aux_eval sructs */
8adc0f72 3110
006f26b2
DM
3111 LEAVE_SCOPE(oldsave);
3112
e9105d30 3113 if (swap) {
c74340f9 3114 /* we failed :-( roll it back */
495f47a5
DM
3115 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
3116 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
3117 PTR2UV(prog),
3118 PTR2UV(prog->offs),
3119 PTR2UV(swap)
3120 ));
e9105d30
GG
3121 Safefree(prog->offs);
3122 prog->offs = swap;
3123 }
d6a28714
JH
3124 return 0;
3125}
3126
6bda09f9 3127
b3d298be 3128/* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
ec43f78b 3129 * Do inc before dec, in case old and new rex are the same */
baa60164 3130#define SET_reg_curpm(Re2) \
bf2039a9 3131 if (reginfo->info_aux_eval) { \
ec43f78b
DM
3132 (void)ReREFCNT_inc(Re2); \
3133 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3134 PM_SETRE((PL_reg_curpm), (Re2)); \
3135 }
3136
3137
d6a28714
JH
3138/*
3139 - regtry - try match at specific point
3140 */
3141STATIC I32 /* 0 failure, 1 success */
f73aaa43 3142S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
d6a28714 3143{
d6a28714 3144 CHECKPOINT lastcp;
288b8c02 3145 REGEXP *const rx = reginfo->prog;
8d919b0a 3146 regexp *const prog = ReANY(rx);
99a90e59 3147 SSize_t result;
f8fc2ecf 3148 RXi_GET_DECL(prog,progi);
a3621e74 3149 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
3150
3151 PERL_ARGS_ASSERT_REGTRY;
3152
24b23f37 3153 reginfo->cutpoint=NULL;
d6a28714 3154
9d9163fb 3155 prog->offs[0].start = *startposp - reginfo->strbeg;
d6a28714 3156 prog->lastparen = 0;
03994de8 3157 prog->lastcloseparen = 0;
d6a28714
JH
3158
3159 /* XXXX What this code is doing here?!!! There should be no need
b93070ed 3160 to do this again and again, prog->lastparen should take care of
3dd2943c 3161 this! --ilya*/
dafc8851
JH
3162
3163 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3164 * Actually, the code in regcppop() (which Ilya may be meaning by
b93070ed 3165 * prog->lastparen), is not needed at all by the test suite
225593e1
DM
3166 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3167 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3168 * Meanwhile, this code *is* needed for the
daf18116
JH
3169 * above-mentioned test suite tests to succeed. The common theme
3170 * on those tests seems to be returning null fields from matches.
225593e1 3171 * --jhi updated by dapm */
dafc8851 3172#if 1
d6a28714 3173 if (prog->nparens) {
b93070ed 3174 regexp_paren_pair *pp = prog->offs;
eb578fdb 3175 I32 i;
b93070ed 3176 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
f0ab9afb
NC
3177 ++pp;
3178 pp->start = -1;
3179 pp->end = -1;
d6a28714
JH
3180 }
3181 }
dafc8851 3182#endif
02db2b7b 3183 REGCP_SET(lastcp);
f73aaa43
DM
3184 result = regmatch(reginfo, *startposp, progi->program + 1);
3185 if (result != -1) {
3186 prog->offs[0].end = result;
d6a28714
JH
3187 return 1;
3188 }
24b23f37 3189 if (reginfo->cutpoint)
f73aaa43 3190 *startposp= reginfo->cutpoint;
02db2b7b 3191 REGCP_UNWIND(lastcp);
d6a28714
JH
3192 return 0;
3193}
3194
02db2b7b 3195
8ba1375e
MJD
3196#define sayYES goto yes
3197#define sayNO goto no
262b90c4 3198#define sayNO_SILENT goto no_silent
8ba1375e 3199
f9f4320a
YO
3200/* we dont use STMT_START/END here because it leads to
3201 "unreachable code" warnings, which are bogus, but distracting. */
3202#define CACHEsayNO \
c476f425 3203 if (ST.cache_mask) \
2ac8ff4b 3204 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
f9f4320a 3205 sayNO
3298f257 3206
a3621e74 3207/* this is used to determine how far from the left messages like
265c4333
YO
3208 'failed...' are printed. It should be set such that messages
3209 are inline with the regop output that created them.
a3621e74 3210*/
265c4333 3211#define REPORT_CODE_OFF 32
a3621e74
YO
3212
3213
40a82448
DM
3214#define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3215#define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
79a2a0e8
KW
3216#define CHRTEST_NOT_A_CP_1 -999
3217#define CHRTEST_NOT_A_CP_2 -998
9e137952 3218
5d9a96ca
DM
3219/* grab a new slab and return the first slot in it */
3220
3221STATIC regmatch_state *
3222S_push_slab(pTHX)
3223{
a35a87e7 3224#if PERL_VERSION < 9 && !defined(PERL_CORE)
54df2634
NC
3225 dMY_CXT;
3226#endif
5d9a96ca
DM
3227 regmatch_slab *s = PL_regmatch_slab->next;
3228 if (!s) {
3229 Newx(s, 1, regmatch_slab);
3230 s->prev = PL_regmatch_slab;
3231 s->next = NULL;
3232 PL_regmatch_slab->next = s;
3233 }
3234 PL_regmatch_slab = s;
86545054 3235 return SLAB_FIRST(s);
5d9a96ca 3236}
5b47454d 3237
95b24440 3238
40a82448
DM
3239/* push a new state then goto it */
3240
4d5016e5
DM
3241#define PUSH_STATE_GOTO(state, node, input) \
3242 pushinput = input; \
40a82448
DM
3243 scan = node; \
3244 st->resume_state = state; \
3245 goto push_state;
3246
3247/* push a new state with success backtracking, then goto it */
3248
4d5016e5
DM
3249#define PUSH_YES_STATE_GOTO(state, node, input) \
3250 pushinput = input; \
40a82448
DM
3251 scan = node; \
3252 st->resume_state = state; \
3253 goto push_yes_state;
3254
aa283a38 3255
aa283a38 3256
4d5016e5 3257
d6a28714 3258/*
95b24440 3259
bf1f174e
DM
3260regmatch() - main matching routine
3261
3262This is basically one big switch statement in a loop. We execute an op,
3263set 'next' to point the next op, and continue. If we come to a point which
3264we may need to backtrack to on failure such as (A|B|C), we push a
3265backtrack state onto the backtrack stack. On failure, we pop the top
3266state, and re-enter the loop at the state indicated. If there are no more
3267states to pop, we return failure.
3268
3269Sometimes we also need to backtrack on success; for example /A+/, where
3270after successfully matching one A, we need to go back and try to
3271match another one; similarly for lookahead assertions: if the assertion
3272completes successfully, we backtrack to the state just before the assertion
3273and then carry on. In these cases, the pushed state is marked as
3274'backtrack on success too'. This marking is in fact done by a chain of
3275pointers, each pointing to the previous 'yes' state. On success, we pop to
3276the nearest yes state, discarding any intermediate failure-only states.
3277Sometimes a yes state is pushed just to force some cleanup code to be
3278called at the end of a successful match or submatch; e.g. (??{$re}) uses
3279it to free the inner regex.
3280
3281Note that failure backtracking rewinds the cursor position, while
3282success backtracking leaves it alone.
3283
3284A pattern is complete when the END op is executed, while a subpattern
3285such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3286ops trigger the "pop to last yes state if any, otherwise return true"
3287behaviour.
3288
3289A common convention in this function is to use A and B to refer to the two
3290subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3291the subpattern to be matched possibly multiple times, while B is the entire
3292rest of the pattern. Variable and state names reflect this convention.
3293
3294The states in the main switch are the union of ops and failure/success of
3295substates associated with with that op. For example, IFMATCH is the op
3296that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3297'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3298successfully matched A and IFMATCH_A_fail is a state saying that we have
3299just failed to match A. Resume states always come in pairs. The backtrack
3300state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3301at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3302on success or failure.
3303
3304The struct that holds a backtracking state is actually a big union, with
3305one variant for each major type of op. The variable st points to the
3306top-most backtrack struct. To make the code clearer, within each
3307block of code we #define ST to alias the relevant union.
3308
3309Here's a concrete example of a (vastly oversimplified) IFMATCH
3310implementation:
3311
3312 switch (state) {
3313 ....
3314
3315#define ST st->u.ifmatch
3316
3317 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3318 ST.foo = ...; // some state we wish to save
95b24440 3319 ...
bf1f174e
DM
3320 // push a yes backtrack state with a resume value of
3321 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3322 // first node of A:
4d5016e5 3323 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
bf1f174e
DM
3324 // NOTREACHED
3325
3326 case IFMATCH_A: // we have successfully executed A; now continue with B
3327 next = B;
3328 bar = ST.foo; // do something with the preserved value
3329 break;
3330
3331 case IFMATCH_A_fail: // A failed, so the assertion failed
3332 ...; // do some housekeeping, then ...
3333 sayNO; // propagate the failure
3334
3335#undef ST
95b24440 3336
bf1f174e
DM
3337 ...
3338 }
95b24440 3339
bf1f174e
DM
3340For any old-timers reading this who are familiar with the old recursive
3341approach, the code above is equivalent to:
95b24440 3342
bf1f174e
DM
3343 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3344 {
3345 int foo = ...
95b24440 3346 ...
bf1f174e
DM
3347 if (regmatch(A)) {
3348 next = B;
3349 bar = foo;
3350 break;
95b24440 3351 }
bf1f174e
DM
3352 ...; // do some housekeeping, then ...
3353 sayNO; // propagate the failure
95b24440 3354 }
bf1f174e
DM
3355
3356The topmost backtrack state, pointed to by st, is usually free. If you
3357want to claim it, populate any ST.foo fields in it with values you wish to
3358save, then do one of
3359
4d5016e5
DM
3360 PUSH_STATE_GOTO(resume_state, node, newinput);
3361 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
bf1f174e
DM
3362
3363which sets that backtrack state's resume value to 'resume_state', pushes a
3364new free entry to the top of the backtrack stack, then goes to 'node'.
3365On backtracking, the free slot is popped, and the saved state becomes the
3366new free state. An ST.foo field in this new top state can be temporarily
3367accessed to retrieve values, but once the main loop is re-entered, it
3368becomes available for reuse.
3369
3370Note that the depth of the backtrack stack constantly increases during the
3371left-to-right execution of the pattern, rather than going up and down with
3372the pattern nesting. For example the stack is at its maximum at Z at the
3373end of the pattern, rather than at X in the following:
3374
3375 /(((X)+)+)+....(Y)+....Z/
3376
3377The only exceptions to this are lookahead/behind assertions and the cut,
3378(?>A), which pop all the backtrack states associated with A before
3379continuing.
3380
486ec47a 3381Backtrack state structs are allocated in slabs of about 4K in size.
bf1f174e
DM
3382PL_regmatch_state and st always point to the currently active state,
3383and PL_regmatch_slab points to the slab currently containing
3384PL_regmatch_state. The first time regmatch() is called, the first slab is
3385allocated, and is never freed until interpreter destruction. When the slab
3386is full, a new one is allocated and chained to the end. At exit from
3387regmatch(), slabs allocated since entry are freed.
3388
3389*/
95b24440 3390
40a82448 3391
5bc10b2c 3392#define DEBUG_STATE_pp(pp) \
265c4333 3393 DEBUG_STATE_r({ \
baa60164 3394 DUMP_EXEC_POS(locinput, scan, utf8_target); \
5bc10b2c 3395 PerlIO_printf(Perl_debug_log, \
5d458dd8 3396 " %*s"pp" %s%s%s%s%s\n", \
5bc10b2c 3397 depth*2, "", \
baa60164 3398 PL_reg_name[st->resume_state], \
5d458dd8
YO
3399 ((st==yes_state||st==mark_state) ? "[" : ""), \
3400 ((st==yes_state) ? "Y" : ""), \
3401 ((st==mark_state) ? "M" : ""), \
3402 ((st==yes_state||st==mark_state) ? "]" : "") \
3403 ); \
265c4333 3404 });
5bc10b2c 3405
40a82448 3406
3dab1dad 3407#define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
95b24440 3408
3df15adc 3409#ifdef DEBUGGING
5bc10b2c 3410
ab3bbdeb 3411STATIC void
f2ed9b32 3412S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
ab3bbdeb
YO
3413 const char *start, const char *end, const char *blurb)
3414{
efd26800 3415 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
7918f24d
NC
3416
3417 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3418
ab3bbdeb
YO
3419 if (!PL_colorset)
3420 reginitcolors();
3421 {
3422 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
d2c6dc5e 3423 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
ab3bbdeb 3424
f2ed9b32 3425 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb
YO
3426 start, end - start, 60);
3427
3428 PerlIO_printf(Perl_debug_log,
3429 "%s%s REx%s %s against %s\n",
3430 PL_colors[4], blurb, PL_colors[5], s0, s1);
3431
f2ed9b32 3432 if (utf8_target||utf8_pat)
1de06328
YO
3433 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3434 utf8_pat ? "pattern" : "",
f2ed9b32
KW
3435 utf8_pat && utf8_target ? " and " : "",
3436 utf8_target ? "string" : ""
ab3bbdeb
YO
3437 );
3438 }
3439}
3df15adc
YO
3440
3441STATIC void
786e8c11
YO
3442S_dump_exec_pos(pTHX_ const char *locinput,
3443 const regnode *scan,
3444 const char *loc_regeol,
3445 const char *loc_bostr,
3446 const char *loc_reg_starttry,
f2ed9b32 3447 const bool utf8_target)
07be1b83 3448{
786e8c11 3449 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
07be1b83 3450 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
786e8c11 3451 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
07be1b83
YO
3452 /* The part of the string before starttry has one color
3453 (pref0_len chars), between starttry and current
3454 position another one (pref_len - pref0_len chars),
3455 after the current position the third one.
3456 We assume that pref0_len <= pref_len, otherwise we
3457 decrease pref0_len. */
786e8c11
YO
3458 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3459 ? (5 + taill) - l : locinput - loc_bostr;
07be1b83
YO
3460 int pref0_len;
3461
7918f24d
NC
3462 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3463
f2ed9b32 3464 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
07be1b83 3465 pref_len++;
786e8c11
YO
3466 pref0_len = pref_len - (locinput - loc_reg_starttry);
3467 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3468 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3469 ? (5 + taill) - pref_len : loc_regeol - locinput);
f2ed9b32 3470 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
07be1b83
YO
3471 l--;
3472 if (pref0_len < 0)
3473 pref0_len = 0;
3474 if (pref0_len > pref_len)
3475 pref0_len = pref_len;
3476 {
f2ed9b32 3477 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
0df25f3d 3478
ab3bbdeb 3479 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
1de06328 3480 (locinput - pref_len),pref0_len, 60, 4, 5);
0df25f3d 3481
ab3bbdeb 3482 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3df15adc 3483 (locinput - pref_len + pref0_len),
1de06328 3484 pref_len - pref0_len, 60, 2, 3);
0df25f3d 3485
ab3bbdeb 3486 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
1de06328 3487 locinput, loc_regeol - locinput, 10, 0, 1);
0df25f3d 3488
1de06328 3489 const STRLEN tlen=len0+len1+len2;
3df15adc 3490 PerlIO_printf(Perl_debug_log,
ab3bbdeb 3491 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
786e8c11 3492 (IV)(locinput - loc_bostr),
07be1b83 3493 len0, s0,
07be1b83 3494 len1, s1,
07be1b83 3495 (docolor ? "" : "> <"),
07be1b83 3496 len2, s2,
f9f4320a 3497 (int)(tlen > 19 ? 0 : 19 - tlen),
07be1b83
YO
3498 "");
3499 }
3500}
3df15adc 3501
07be1b83
YO
3502#endif
3503
0a4db386
YO
3504/* reg_check_named_buff_matched()
3505 * Checks to see if a named buffer has matched. The data array of
3506 * buffer numbers corresponding to the buffer is expected to reside
3507 * in the regexp->data->data array in the slot stored in the ARG() of
3508 * node involved. Note that this routine doesn't actually care about the
3509 * name, that information is not preserved from compilation to execution.
3510 * Returns the index of the leftmost defined buffer with the given name
3511 * or 0 if non of the buffers matched.
3512 */
3513STATIC I32
dc3bf405 3514S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
7918f24d 3515{
0a4db386 3516 I32 n;
f8fc2ecf 3517 RXi_GET_DECL(rex,rexi);
ad64d0ec 3518 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
0a4db386 3519 I32 *nums=(I32*)SvPVX(sv_dat);
7918f24d
NC
3520
3521 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3522
0a4db386 3523 for ( n=0; n<SvIVX(sv_dat); n++ ) {
b93070ed
DM