This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regexec.c: Clean up macro/code interface slightly
[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
5b8bb145 1586#define FBC_UTF8(TEST_UV, TEST2_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( \
c84a03c5 1598 if (tmp == ! (TEST2_UTF8)) { \
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 */
5b8bb145
KW
1611#define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST2_UTF8) \
1612 FBC_BOUND_COMMON(FBC_UTF8(TEST_UV, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
cfaf538b 1613
5b8bb145 1614#define FBC_BOUND_A(TEST_NON_UTF8, TEST_UV, TEST2_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
5b8bb145
KW
1617#define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST2_UTF8) \
1618 FBC_BOUND_COMMON(FBC_UTF8(TEST_UV, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
cfaf538b 1619
5b8bb145 1620#define FBC_NBOUND_A(TEST_NON_UTF8, TEST_UV, TEST2_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
63ac0dad
KW
1623/* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1624 * be passed in completely with the variable name being tested, which isn't
1625 * such a clean interface, but this is easier to read than it was before. We
1626 * are looking for the boundary (or non-boundary between a word and non-word
1627 * character. The utf8 and non-utf8 cases have the same logic, but the details
1628 * must be different. Find the "wordness" of the character just prior to this
1629 * one, and compare it with the wordness of this one. If they differ, we have
1630 * a boundary. At the beginning of the string, pretend that the previous
1631 * character was a new-line */
baa60164 1632#define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
63ac0dad 1633 if (utf8_target) { \
baa60164 1634 UTF8_CODE \
63ac0dad
KW
1635 } \
1636 else { /* Not utf8 */ \
9d9163fb 1637 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
63ac0dad
KW
1638 tmp = TEST_NON_UTF8(tmp); \
1639 REXEC_FBC_SCAN( \
1640 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1641 tmp = !tmp; \
1642 IF_SUCCESS; \
1643 } \
1644 else { \
1645 IF_FAIL; \
1646 } \
1647 ); \
1648 } \
baa60164 1649 if ((!prog->minlen && tmp) && (reginfo->intuit || regtry(reginfo, &s))) \
63ac0dad
KW
1650 goto got_it;
1651
786e8c11 1652/* We know what class REx starts with. Try to find this position... */
02d5137b 1653/* if reginfo->intuit, its a dryrun */
786e8c11
YO
1654/* annoyingly all the vars in this routine have different names from their counterparts
1655 in regmatch. /grrr */
1656
3c3eec57 1657STATIC char *
07be1b83 1658S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
f9176b44 1659 const char *strend, regmatch_info *reginfo)
a687059c 1660{
73104a1b
KW
1661 dVAR;
1662 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1663 char *pat_string; /* The pattern's exactish string */
1664 char *pat_end; /* ptr to end char of pat_string */
1665 re_fold_t folder; /* Function for computing non-utf8 folds */
1666 const U8 *fold_array; /* array for folding ords < 256 */
1667 STRLEN ln;
1668 STRLEN lnc;
73104a1b
KW
1669 U8 c1;
1670 U8 c2;
1671 char *e;
1672 I32 tmp = 1; /* Scratch variable? */
ba44c216 1673 const bool utf8_target = reginfo->is_utf8_target;
73104a1b 1674 UV utf8_fold_flags = 0;
f9176b44 1675 const bool is_utf8_pat = reginfo->is_utf8_pat;
3018b823
KW
1676 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1677 with a result inverts that result, as 0^1 =
1678 1 and 1^1 = 0 */
1679 _char_class_number classnum;
1680
73104a1b 1681 RXi_GET_DECL(prog,progi);
2f7f8cb1 1682
73104a1b 1683 PERL_ARGS_ASSERT_FIND_BYCLASS;
2f7f8cb1 1684
73104a1b
KW
1685 /* We know what class it must start with. */
1686 switch (OP(c)) {
1687 case ANYOF:
1688 if (utf8_target) {
1689 REXEC_FBC_UTF8_CLASS_SCAN(
3db24e1e 1690 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
73104a1b
KW
1691 }
1692 else {
1693 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1694 }
1695 break;
1696 case CANY:
1697 REXEC_FBC_SCAN(
02d5137b 1698 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
73104a1b
KW
1699 goto got_it;
1700 else
1701 tmp = doevery;
1702 );
1703 break;
1704
098b07d5
KW
1705 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1706 assert(! is_utf8_pat);
924ba076 1707 /* FALLTHROUGH */
73104a1b 1708 case EXACTFA:
984e6dd1 1709 if (is_utf8_pat || utf8_target) {
73104a1b
KW
1710 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1711 goto do_exactf_utf8;
1712 }
1713 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1714 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1715 goto do_exactf_non_utf8; /* isn't dealt with by these */
77a6d856 1716
2fdb7295
KW
1717 case EXACTF: /* This node only generated for non-utf8 patterns */
1718 assert(! is_utf8_pat);
73104a1b 1719 if (utf8_target) {
73104a1b
KW
1720 utf8_fold_flags = 0;
1721 goto do_exactf_utf8;
1722 }
1723 fold_array = PL_fold;
1724 folder = foldEQ;
1725 goto do_exactf_non_utf8;
1726
1727 case EXACTFL:
31f05a37 1728 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
cea315b6 1729 utf8_fold_flags = FOLDEQ_LOCALE;
73104a1b
KW
1730 goto do_exactf_utf8;
1731 }
1732 fold_array = PL_fold_locale;
1733 folder = foldEQ_locale;
1734 goto do_exactf_non_utf8;
3c760661 1735
73104a1b 1736 case EXACTFU_SS:
984e6dd1 1737 if (is_utf8_pat) {
73104a1b
KW
1738 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1739 }
1740 goto do_exactf_utf8;
16d951b7 1741
73104a1b 1742 case EXACTFU:
984e6dd1
DM
1743 if (is_utf8_pat || utf8_target) {
1744 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
73104a1b
KW
1745 goto do_exactf_utf8;
1746 }
fac1af77 1747
73104a1b
KW
1748 /* Any 'ss' in the pattern should have been replaced by regcomp,
1749 * so we don't have to worry here about this single special case
1750 * in the Latin1 range */
1751 fold_array = PL_fold_latin1;
1752 folder = foldEQ_latin1;
1753
924ba076 1754 /* FALLTHROUGH */
73104a1b
KW
1755
1756 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1757 are no glitches with fold-length differences
1758 between the target string and pattern */
1759
1760 /* The idea in the non-utf8 EXACTF* cases is to first find the
1761 * first character of the EXACTF* node and then, if necessary,
1762 * case-insensitively compare the full text of the node. c1 is the
1763 * first character. c2 is its fold. This logic will not work for
1764 * Unicode semantics and the german sharp ss, which hence should
1765 * not be compiled into a node that gets here. */
1766 pat_string = STRING(c);
1767 ln = STR_LEN(c); /* length to match in octets/bytes */
1768
1769 /* We know that we have to match at least 'ln' bytes (which is the
1770 * same as characters, since not utf8). If we have to match 3
1771 * characters, and there are only 2 availabe, we know without
1772 * trying that it will fail; so don't start a match past the
1773 * required minimum number from the far end */
ea3daa5d 1774 e = HOP3c(strend, -((SSize_t)ln), s);
73104a1b 1775
02d5137b 1776 if (reginfo->intuit && e < s) {
73104a1b
KW
1777 e = s; /* Due to minlen logic of intuit() */
1778 }
fac1af77 1779
73104a1b
KW
1780 c1 = *pat_string;
1781 c2 = fold_array[c1];
1782 if (c1 == c2) { /* If char and fold are the same */
1783 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1784 }
1785 else {
1786 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1787 }
1788 break;
fac1af77 1789
73104a1b
KW
1790 do_exactf_utf8:
1791 {
1792 unsigned expansion;
1793
1794 /* If one of the operands is in utf8, we can't use the simpler folding
1795 * above, due to the fact that many different characters can have the
1796 * same fold, or portion of a fold, or different- length fold */
1797 pat_string = STRING(c);
1798 ln = STR_LEN(c); /* length to match in octets/bytes */
1799 pat_end = pat_string + ln;
984e6dd1 1800 lnc = is_utf8_pat /* length to match in characters */
73104a1b
KW
1801 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1802 : ln;
1803
1804 /* We have 'lnc' characters to match in the pattern, but because of
1805 * multi-character folding, each character in the target can match
1806 * up to 3 characters (Unicode guarantees it will never exceed
1807 * this) if it is utf8-encoded; and up to 2 if not (based on the
1808 * fact that the Latin 1 folds are already determined, and the
1809 * only multi-char fold in that range is the sharp-s folding to
1810 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1811 * string character. Adjust lnc accordingly, rounding up, so that
1812 * if we need to match at least 4+1/3 chars, that really is 5. */
1813 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1814 lnc = (lnc + expansion - 1) / expansion;
1815
1816 /* As in the non-UTF8 case, if we have to match 3 characters, and
1817 * only 2 are left, it's guaranteed to fail, so don't start a
1818 * match that would require us to go beyond the end of the string
1819 */
ea3daa5d 1820 e = HOP3c(strend, -((SSize_t)lnc), s);
73104a1b 1821
02d5137b 1822 if (reginfo->intuit && e < s) {
73104a1b
KW
1823 e = s; /* Due to minlen logic of intuit() */
1824 }
0658cdde 1825
73104a1b
KW
1826 /* XXX Note that we could recalculate e to stop the loop earlier,
1827 * as the worst case expansion above will rarely be met, and as we
1828 * go along we would usually find that e moves further to the left.
1829 * This would happen only after we reached the point in the loop
1830 * where if there were no expansion we should fail. Unclear if
1831 * worth the expense */
1832
1833 while (s <= e) {
1834 char *my_strend= (char *)strend;
1835 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
984e6dd1 1836 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
02d5137b 1837 && (reginfo->intuit || regtry(reginfo, &s)) )
73104a1b
KW
1838 {
1839 goto got_it;
1840 }
1841 s += (utf8_target) ? UTF8SKIP(s) : 1;
1842 }
1843 break;
1844 }
1845 case BOUNDL:
0eb30aeb 1846 FBC_BOUND(isWORDCHAR_LC,
5b8bb145 1847 isWORDCHAR_LC_uvchr,
0eb30aeb 1848 isWORDCHAR_LC_utf8((U8*)s));
73104a1b
KW
1849 break;
1850 case NBOUNDL:
0eb30aeb 1851 FBC_NBOUND(isWORDCHAR_LC,
5b8bb145 1852 isWORDCHAR_LC_uvchr,
0eb30aeb 1853 isWORDCHAR_LC_utf8((U8*)s));
73104a1b
KW
1854 break;
1855 case BOUND:
1856 FBC_BOUND(isWORDCHAR,
5b8bb145 1857 isWORDCHAR_uni,
03940dc2 1858 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
73104a1b
KW
1859 break;
1860 case BOUNDA:
b2f4e957 1861 FBC_BOUND_A(isWORDCHAR_A,
5b8bb145 1862 isWORDCHAR_A,
b2f4e957 1863 isWORDCHAR_A((U8*)s));
73104a1b
KW
1864 break;
1865 case NBOUND:
1866 FBC_NBOUND(isWORDCHAR,
5b8bb145 1867 isWORDCHAR_uni,
03940dc2 1868 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
73104a1b
KW
1869 break;
1870 case NBOUNDA:
b2f4e957 1871 FBC_NBOUND_A(isWORDCHAR_A,
5b8bb145 1872 isWORDCHAR_A,
b2f4e957 1873 isWORDCHAR_A((U8*)s));
73104a1b
KW
1874 break;
1875 case BOUNDU:
1876 FBC_BOUND(isWORDCHAR_L1,
5b8bb145 1877 isWORDCHAR_uni,
03940dc2 1878 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
73104a1b
KW
1879 break;
1880 case NBOUNDU:
1881 FBC_NBOUND(isWORDCHAR_L1,
5b8bb145 1882 isWORDCHAR_uni,
03940dc2 1883 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
73104a1b 1884 break;
73104a1b
KW
1885 case LNBREAK:
1886 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1887 is_LNBREAK_latin1_safe(s, strend)
1888 );
1889 break;
3018b823
KW
1890
1891 /* The argument to all the POSIX node types is the class number to pass to
1892 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1893
1894 case NPOSIXL:
1895 to_complement = 1;
1896 /* FALLTHROUGH */
1897
1898 case POSIXL:
3018b823
KW
1899 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1900 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
73104a1b 1901 break;
3018b823
KW
1902
1903 case NPOSIXD:
1904 to_complement = 1;
1905 /* FALLTHROUGH */
1906
1907 case POSIXD:
1908 if (utf8_target) {
1909 goto posix_utf8;
1910 }
1911 goto posixa;
1912
1913 case NPOSIXA:
1914 if (utf8_target) {
1915 /* The complement of something that matches only ASCII matches all
837226c8
KW
1916 * non-ASCII, plus everything in ASCII that isn't in the class. */
1917 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
3018b823
KW
1918 || ! _generic_isCC_A(*s, FLAGS(c)));
1919 break;
1920 }
1921
1922 to_complement = 1;
1923 /* FALLTHROUGH */
1924
73104a1b 1925 case POSIXA:
3018b823 1926 posixa:
73104a1b 1927 /* Don't need to worry about utf8, as it can match only a single
3018b823
KW
1928 * byte invariant character. */
1929 REXEC_FBC_CLASS_SCAN(
1930 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
73104a1b 1931 break;
3018b823
KW
1932
1933 case NPOSIXU:
1934 to_complement = 1;
1935 /* FALLTHROUGH */
1936
1937 case POSIXU:
1938 if (! utf8_target) {
1939 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1940 FLAGS(c))));
1941 }
1942 else {
1943
1944 posix_utf8:
1945 classnum = (_char_class_number) FLAGS(c);
1946 if (classnum < _FIRST_NON_SWASH_CC) {
1947 while (s < strend) {
1948
1949 /* We avoid loading in the swash as long as possible, but
1950 * should we have to, we jump to a separate loop. This
1951 * extra 'if' statement is what keeps this code from being
1952 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1953 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1954 goto found_above_latin1;
1955 }
1956 if ((UTF8_IS_INVARIANT(*s)
1957 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
1958 classnum)))
1959 || (UTF8_IS_DOWNGRADEABLE_START(*s)
1960 && to_complement ^ cBOOL(
94bb8c36
KW
1961 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
1962 *(s + 1)),
3018b823
KW
1963 classnum))))
1964 {
02d5137b 1965 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
3018b823
KW
1966 goto got_it;
1967 else {
1968 tmp = doevery;
1969 }
1970 }
1971 else {
1972 tmp = 1;
1973 }
1974 s += UTF8SKIP(s);
1975 }
1976 }
1977 else switch (classnum) { /* These classes are implemented as
1978 macros */
1979 case _CC_ENUM_SPACE: /* XXX would require separate code if we
1980 revert the change of \v matching this */
924ba076 1981 /* FALLTHROUGH */
3018b823
KW
1982
1983 case _CC_ENUM_PSXSPC:
1984 REXEC_FBC_UTF8_CLASS_SCAN(
1985 to_complement ^ cBOOL(isSPACE_utf8(s)));
1986 break;
1987
1988 case _CC_ENUM_BLANK:
1989 REXEC_FBC_UTF8_CLASS_SCAN(
1990 to_complement ^ cBOOL(isBLANK_utf8(s)));
1991 break;
1992
1993 case _CC_ENUM_XDIGIT:
1994 REXEC_FBC_UTF8_CLASS_SCAN(
1995 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
1996 break;
1997
1998 case _CC_ENUM_VERTSPACE:
1999 REXEC_FBC_UTF8_CLASS_SCAN(
2000 to_complement ^ cBOOL(isVERTWS_utf8(s)));
2001 break;
2002
2003 case _CC_ENUM_CNTRL:
2004 REXEC_FBC_UTF8_CLASS_SCAN(
2005 to_complement ^ cBOOL(isCNTRL_utf8(s)));
2006 break;
2007
2008 default:
2009 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
a25b5927 2010 assert(0); /* NOTREACHED */
3018b823
KW
2011 }
2012 }
2013 break;
2014
2015 found_above_latin1: /* Here we have to load a swash to get the result
2016 for the current code point */
2017 if (! PL_utf8_swash_ptrs[classnum]) {
2018 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2019 PL_utf8_swash_ptrs[classnum] =
2a16ac92
KW
2020 _core_swash_init("utf8",
2021 "",
2022 &PL_sv_undef, 1, 0,
2023 PL_XPosix_ptrs[classnum], &flags);
3018b823
KW
2024 }
2025
2026 /* This is a copy of the loop above for swash classes, though using the
2027 * FBC macro instead of being expanded out. Since we've loaded the
2028 * swash, we don't have to check for that each time through the loop */
2029 REXEC_FBC_UTF8_CLASS_SCAN(
2030 to_complement ^ cBOOL(_generic_utf8(
2031 classnum,
2032 s,
2033 swash_fetch(PL_utf8_swash_ptrs[classnum],
2034 (U8 *) s, TRUE))));
73104a1b
KW
2035 break;
2036
2037 case AHOCORASICKC:
2038 case AHOCORASICK:
2039 {
2040 DECL_TRIE_TYPE(c);
2041 /* what trie are we using right now */
2042 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2043 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2044 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2045
2046 const char *last_start = strend - trie->minlen;
6148ee25 2047#ifdef DEBUGGING
73104a1b 2048 const char *real_start = s;
6148ee25 2049#endif
73104a1b
KW
2050 STRLEN maxlen = trie->maxlen;
2051 SV *sv_points;
2052 U8 **points; /* map of where we were in the input string
2053 when reading a given char. For ASCII this
2054 is unnecessary overhead as the relationship
2055 is always 1:1, but for Unicode, especially
2056 case folded Unicode this is not true. */
2057 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2058 U8 *bitmap=NULL;
2059
2060
2061 GET_RE_DEBUG_FLAGS_DECL;
2062
2063 /* We can't just allocate points here. We need to wrap it in
2064 * an SV so it gets freed properly if there is a croak while
2065 * running the match */
2066 ENTER;
2067 SAVETMPS;
2068 sv_points=newSV(maxlen * sizeof(U8 *));
2069 SvCUR_set(sv_points,
2070 maxlen * sizeof(U8 *));
2071 SvPOK_on(sv_points);
2072 sv_2mortal(sv_points);
2073 points=(U8**)SvPV_nolen(sv_points );
2074 if ( trie_type != trie_utf8_fold
2075 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2076 {
2077 if (trie->bitmap)
2078 bitmap=(U8*)trie->bitmap;
2079 else
2080 bitmap=(U8*)ANYOF_BITMAP(c);
2081 }
2082 /* this is the Aho-Corasick algorithm modified a touch
2083 to include special handling for long "unknown char" sequences.
2084 The basic idea being that we use AC as long as we are dealing
2085 with a possible matching char, when we encounter an unknown char
2086 (and we have not encountered an accepting state) we scan forward
2087 until we find a legal starting char.
2088 AC matching is basically that of trie matching, except that when
2089 we encounter a failing transition, we fall back to the current
2090 states "fail state", and try the current char again, a process
2091 we repeat until we reach the root state, state 1, or a legal
2092 transition. If we fail on the root state then we can either
2093 terminate if we have reached an accepting state previously, or
2094 restart the entire process from the beginning if we have not.
2095
2096 */
2097 while (s <= last_start) {
2098 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2099 U8 *uc = (U8*)s;
2100 U16 charid = 0;
2101 U32 base = 1;
2102 U32 state = 1;
2103 UV uvc = 0;
2104 STRLEN len = 0;
2105 STRLEN foldlen = 0;
2106 U8 *uscan = (U8*)NULL;
2107 U8 *leftmost = NULL;
2108#ifdef DEBUGGING
2109 U32 accepted_word= 0;
786e8c11 2110#endif
73104a1b
KW
2111 U32 pointpos = 0;
2112
2113 while ( state && uc <= (U8*)strend ) {
2114 int failed=0;
2115 U32 word = aho->states[ state ].wordnum;
2116
2117 if( state==1 ) {
2118 if ( bitmap ) {
2119 DEBUG_TRIE_EXECUTE_r(
2120 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2121 dump_exec_pos( (char *)uc, c, strend, real_start,
2122 (char *)uc, utf8_target );
2123 PerlIO_printf( Perl_debug_log,
2124 " Scanning for legal start char...\n");
2125 }
2126 );
2127 if (utf8_target) {
2128 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2129 uc += UTF8SKIP(uc);
2130 }
2131 } else {
2132 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2133 uc++;
2134 }
786e8c11 2135 }
73104a1b 2136 s= (char *)uc;
07be1b83 2137 }
73104a1b
KW
2138 if (uc >(U8*)last_start) break;
2139 }
2140
2141 if ( word ) {
2142 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2143 if (!leftmost || lpos < leftmost) {
2144 DEBUG_r(accepted_word=word);
2145 leftmost= lpos;
7016d6eb 2146 }
73104a1b 2147 if (base==0) break;
7016d6eb 2148
73104a1b
KW
2149 }
2150 points[pointpos++ % maxlen]= uc;
2151 if (foldlen || uc < (U8*)strend) {
2152 REXEC_TRIE_READ_CHAR(trie_type, trie,
2153 widecharmap, uc,
2154 uscan, len, uvc, charid, foldlen,
2155 foldbuf, uniflags);
2156 DEBUG_TRIE_EXECUTE_r({
2157 dump_exec_pos( (char *)uc, c, strend,
2158 real_start, s, utf8_target);
2159 PerlIO_printf(Perl_debug_log,
2160 " Charid:%3u CP:%4"UVxf" ",
2161 charid, uvc);
2162 });
2163 }
2164 else {
2165 len = 0;
2166 charid = 0;
2167 }
07be1b83 2168
73104a1b
KW
2169
2170 do {
6148ee25 2171#ifdef DEBUGGING
73104a1b 2172 word = aho->states[ state ].wordnum;
6148ee25 2173#endif
73104a1b
KW
2174 base = aho->states[ state ].trans.base;
2175
2176 DEBUG_TRIE_EXECUTE_r({
2177 if (failed)
2178 dump_exec_pos( (char *)uc, c, strend, real_start,
2179 s, utf8_target );
2180 PerlIO_printf( Perl_debug_log,
2181 "%sState: %4"UVxf", word=%"UVxf,
2182 failed ? " Fail transition to " : "",
2183 (UV)state, (UV)word);
2184 });
2185 if ( base ) {
2186 U32 tmp;
2187 I32 offset;
2188 if (charid &&
2189 ( ((offset = base + charid
2190 - 1 - trie->uniquecharcount)) >= 0)
2191 && ((U32)offset < trie->lasttrans)
2192 && trie->trans[offset].check == state
2193 && (tmp=trie->trans[offset].next))
2194 {
2195 DEBUG_TRIE_EXECUTE_r(
2196 PerlIO_printf( Perl_debug_log," - legal\n"));
2197 state = tmp;
2198 break;
07be1b83
YO
2199 }
2200 else {
786e8c11 2201 DEBUG_TRIE_EXECUTE_r(
73104a1b 2202 PerlIO_printf( Perl_debug_log," - fail\n"));
786e8c11 2203 failed = 1;
73104a1b 2204 state = aho->fail[state];
07be1b83 2205 }
07be1b83 2206 }
73104a1b
KW
2207 else {
2208 /* we must be accepting here */
2209 DEBUG_TRIE_EXECUTE_r(
2210 PerlIO_printf( Perl_debug_log," - accepting\n"));
2211 failed = 1;
2212 break;
786e8c11 2213 }
73104a1b
KW
2214 } while(state);
2215 uc += len;
2216 if (failed) {
2217 if (leftmost)
2218 break;
2219 if (!state) state = 1;
07be1b83 2220 }
73104a1b
KW
2221 }
2222 if ( aho->states[ state ].wordnum ) {
2223 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2224 if (!leftmost || lpos < leftmost) {
2225 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2226 leftmost = lpos;
07be1b83
YO
2227 }
2228 }
73104a1b
KW
2229 if (leftmost) {
2230 s = (char*)leftmost;
2231 DEBUG_TRIE_EXECUTE_r({
2232 PerlIO_printf(
2233 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2234 (UV)accepted_word, (IV)(s - real_start)
2235 );
2236 });
02d5137b 2237 if (reginfo->intuit || regtry(reginfo, &s)) {
73104a1b
KW
2238 FREETMPS;
2239 LEAVE;
2240 goto got_it;
2241 }
2242 s = HOPc(s,1);
2243 DEBUG_TRIE_EXECUTE_r({
2244 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2245 });
2246 } else {
2247 DEBUG_TRIE_EXECUTE_r(
2248 PerlIO_printf( Perl_debug_log,"No match.\n"));
2249 break;
2250 }
2251 }
2252 FREETMPS;
2253 LEAVE;
2254 }
2255 break;
2256 default:
2257 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
73104a1b
KW
2258 }
2259 return 0;
2260 got_it:
2261 return s;
6eb5f6b9
JH
2262}
2263
60165aa4
DM
2264/* set RX_SAVED_COPY, RX_SUBBEG etc.
2265 * flags have same meanings as with regexec_flags() */
2266
749f4950
DM
2267static void
2268S_reg_set_capture_string(pTHX_ REGEXP * const rx,
60165aa4
DM
2269 char *strbeg,
2270 char *strend,
2271 SV *sv,
2272 U32 flags,
2273 bool utf8_target)
2274{
2275 struct regexp *const prog = ReANY(rx);
2276
60165aa4
DM
2277 if (flags & REXEC_COPY_STR) {
2278#ifdef PERL_ANY_COW
2279 if (SvCANCOW(sv)) {
2280 if (DEBUG_C_TEST) {
2281 PerlIO_printf(Perl_debug_log,
2282 "Copy on write: regexp capture, type %d\n",
2283 (int) SvTYPE(sv));
2284 }
5411a0e5
DM
2285 /* Create a new COW SV to share the match string and store
2286 * in saved_copy, unless the current COW SV in saved_copy
2287 * is valid and suitable for our purpose */
2288 if (( prog->saved_copy
2289 && SvIsCOW(prog->saved_copy)
2290 && SvPOKp(prog->saved_copy)
2291 && SvIsCOW(sv)
2292 && SvPOKp(sv)
2293 && SvPVX(sv) == SvPVX(prog->saved_copy)))
a76b0e90 2294 {
5411a0e5
DM
2295 /* just reuse saved_copy SV */
2296 if (RXp_MATCH_COPIED(prog)) {
2297 Safefree(prog->subbeg);
2298 RXp_MATCH_COPIED_off(prog);
2299 }
2300 }
2301 else {
2302 /* create new COW SV to share string */
a76b0e90
DM
2303 RX_MATCH_COPY_FREE(rx);
2304 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
a76b0e90 2305 }
5411a0e5
DM
2306 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2307 assert (SvPOKp(prog->saved_copy));
60165aa4
DM
2308 prog->sublen = strend - strbeg;
2309 prog->suboffset = 0;
2310 prog->subcoffset = 0;
2311 } else
2312#endif
2313 {
99a90e59
FC
2314 SSize_t min = 0;
2315 SSize_t max = strend - strbeg;
ea3daa5d 2316 SSize_t sublen;
60165aa4
DM
2317
2318 if ( (flags & REXEC_COPY_SKIP_POST)
e322109a 2319 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM
2320 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2321 ) { /* don't copy $' part of string */
2322 U32 n = 0;
2323 max = -1;
2324 /* calculate the right-most part of the string covered
2325 * by a capture. Due to look-ahead, this may be to
2326 * the right of $&, so we have to scan all captures */
2327 while (n <= prog->lastparen) {
2328 if (prog->offs[n].end > max)
2329 max = prog->offs[n].end;
2330 n++;
2331 }
2332 if (max == -1)
2333 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2334 ? prog->offs[0].start
2335 : 0;
2336 assert(max >= 0 && max <= strend - strbeg);
2337 }
2338
2339 if ( (flags & REXEC_COPY_SKIP_PRE)
e322109a 2340 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM
2341 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2342 ) { /* don't copy $` part of string */
2343 U32 n = 0;
2344 min = max;
2345 /* calculate the left-most part of the string covered
2346 * by a capture. Due to look-behind, this may be to
2347 * the left of $&, so we have to scan all captures */
2348 while (min && n <= prog->lastparen) {
2349 if ( prog->offs[n].start != -1
2350 && prog->offs[n].start < min)
2351 {
2352 min = prog->offs[n].start;
2353 }
2354 n++;
2355 }
2356 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2357 && min > prog->offs[0].end
2358 )
2359 min = prog->offs[0].end;
2360
2361 }
2362
2363 assert(min >= 0 && min <= max && min <= strend - strbeg);
2364 sublen = max - min;
2365
2366 if (RX_MATCH_COPIED(rx)) {
2367 if (sublen > prog->sublen)
2368 prog->subbeg =
2369 (char*)saferealloc(prog->subbeg, sublen+1);
2370 }
2371 else
2372 prog->subbeg = (char*)safemalloc(sublen+1);
2373 Copy(strbeg + min, prog->subbeg, sublen, char);
2374 prog->subbeg[sublen] = '\0';
2375 prog->suboffset = min;
2376 prog->sublen = sublen;
2377 RX_MATCH_COPIED_on(rx);
2378 }
2379 prog->subcoffset = prog->suboffset;
2380 if (prog->suboffset && utf8_target) {
2381 /* Convert byte offset to chars.
2382 * XXX ideally should only compute this if @-/@+
2383 * has been seen, a la PL_sawampersand ??? */
2384
2385 /* If there's a direct correspondence between the
2386 * string which we're matching and the original SV,
2387 * then we can use the utf8 len cache associated with
2388 * the SV. In particular, it means that under //g,
2389 * sv_pos_b2u() will use the previously cached
2390 * position to speed up working out the new length of
2391 * subcoffset, rather than counting from the start of
2392 * the string each time. This stops
2393 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2394 * from going quadratic */
2395 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
ea3daa5d
FC
2396 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2397 SV_GMAGIC|SV_CONST_RETURN);
60165aa4
DM
2398 else
2399 prog->subcoffset = utf8_length((U8*)strbeg,
2400 (U8*)(strbeg+prog->suboffset));
2401 }
2402 }
2403 else {
2404 RX_MATCH_COPY_FREE(rx);
2405 prog->subbeg = strbeg;
2406 prog->suboffset = 0;
2407 prog->subcoffset = 0;
2408 prog->sublen = strend - strbeg;
2409 }
2410}
2411
2412
2413
fae667d5 2414
6eb5f6b9
JH
2415/*
2416 - regexec_flags - match a regexp against a string
2417 */
2418I32
5aaab254 2419Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
ea3daa5d 2420 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
8fd1a950
DM
2421/* stringarg: the point in the string at which to begin matching */
2422/* strend: pointer to null at end of string */
2423/* strbeg: real beginning of string */
2424/* minend: end of match must be >= minend bytes after stringarg. */
2425/* sv: SV being matched: only used for utf8 flag, pos() etc; string
2426 * itself is accessed via the pointers above */
2427/* data: May be used for some additional optimizations.
d058ec57 2428 Currently unused. */
a340edde 2429/* flags: For optimizations. See REXEC_* in regexp.h */
8fd1a950 2430
6eb5f6b9 2431{
8d919b0a 2432 struct regexp *const prog = ReANY(rx);
5aaab254 2433 char *s;
eb578fdb 2434 regnode *c;
03c83e26 2435 char *startpos;
ea3daa5d
FC
2436 SSize_t minlen; /* must match at least this many chars */
2437 SSize_t dontbother = 0; /* how many characters not to try at end */
f2ed9b32 2438 const bool utf8_target = cBOOL(DO_UTF8(sv));
2757e526 2439 I32 multiline;
f8fc2ecf 2440 RXi_GET_DECL(prog,progi);
02d5137b
DM
2441 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2442 regmatch_info *const reginfo = &reginfo_buf;
e9105d30 2443 regexp_paren_pair *swap = NULL;
006f26b2 2444 I32 oldsave;
a3621e74
YO
2445 GET_RE_DEBUG_FLAGS_DECL;
2446
7918f24d 2447 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
9d4ba2ae 2448 PERL_UNUSED_ARG(data);
6eb5f6b9
JH
2449
2450 /* Be paranoid... */
03c83e26 2451 if (prog == NULL || stringarg == NULL) {
6eb5f6b9 2452 Perl_croak(aTHX_ "NULL regexp parameter");
6eb5f6b9
JH
2453 }
2454
6c3fea77 2455 DEBUG_EXECUTE_r(
03c83e26 2456 debug_start_match(rx, utf8_target, stringarg, strend,
6c3fea77
DM
2457 "Matching");
2458 );
8adc0f72 2459
b342a604
DM
2460 startpos = stringarg;
2461
58430ea8 2462 if (prog->intflags & PREGf_GPOS_SEEN) {
d307c076
DM
2463 MAGIC *mg;
2464
fef7148b
DM
2465 /* set reginfo->ganch, the position where \G can match */
2466
2467 reginfo->ganch =
2468 (flags & REXEC_IGNOREPOS)
2469 ? stringarg /* use start pos rather than pos() */
2470 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
25fdce4a
FC
2471 /* Defined pos(): */
2472 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
fef7148b
DM
2473 : strbeg; /* pos() not defined; use start of string */
2474
2475 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
7b0eb0b8 2476 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg)));
fef7148b 2477
03c83e26
DM
2478 /* in the presence of \G, we may need to start looking earlier in
2479 * the string than the suggested start point of stringarg:
0b2c2a84 2480 * if prog->gofs is set, then that's a known, fixed minimum
03c83e26
DM
2481 * offset, such as
2482 * /..\G/: gofs = 2
2483 * /ab|c\G/: gofs = 1
2484 * or if the minimum offset isn't known, then we have to go back
2485 * to the start of the string, e.g. /w+\G/
2486 */
2bfbe302 2487
8e1490ee 2488 if (prog->intflags & PREGf_ANCH_GPOS) {
2bfbe302
DM
2489 startpos = reginfo->ganch - prog->gofs;
2490 if (startpos <
2491 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2492 {
2493 DEBUG_r(PerlIO_printf(Perl_debug_log,
2494 "fail: ganch-gofs before earliest possible start\n"));
2495 return 0;
2496 }
2497 }
2498 else if (prog->gofs) {
b342a604
DM
2499 if (startpos - prog->gofs < strbeg)
2500 startpos = strbeg;
2501 else
2502 startpos -= prog->gofs;
03c83e26 2503 }
58430ea8 2504 else if (prog->intflags & PREGf_GPOS_FLOAT)
b342a604 2505 startpos = strbeg;
03c83e26
DM
2506 }
2507
2508 minlen = prog->minlen;
b342a604 2509 if ((startpos + minlen) > strend || startpos < strbeg) {
03c83e26
DM
2510 DEBUG_r(PerlIO_printf(Perl_debug_log,
2511 "Regex match can't succeed, so not even tried\n"));
2512 return 0;
2513 }
2514
63a3746a
DM
2515 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2516 * which will call destuctors to reset PL_regmatch_state, free higher
2517 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2518 * regmatch_info_aux_eval */
2519
2520 oldsave = PL_savestack_ix;
2521
dfa77d06
DM
2522 s = startpos;
2523
e322109a 2524 if ((prog->extflags & RXf_USE_INTUIT)
7fadf4a7
DM
2525 && !(flags & REXEC_CHECKED))
2526 {
dfa77d06 2527 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
7fadf4a7 2528 flags, NULL);
dfa77d06 2529 if (!s)
7fadf4a7
DM
2530 return 0;
2531
e322109a 2532 if (prog->extflags & RXf_CHECK_ALL) {
7fadf4a7
DM
2533 /* we can match based purely on the result of INTUIT.
2534 * Set up captures etc just for $& and $-[0]
2535 * (an intuit-only match wont have $1,$2,..) */
2536 assert(!prog->nparens);
d5e7783a
DM
2537
2538 /* s/// doesn't like it if $& is earlier than where we asked it to
2539 * start searching (which can happen on something like /.\G/) */
2540 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2541 && (s < stringarg))
2542 {
2543 /* this should only be possible under \G */
58430ea8 2544 assert(prog->intflags & PREGf_GPOS_SEEN);
d5e7783a
DM
2545 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2546 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2547 goto phooey;
2548 }
2549
7fadf4a7
DM
2550 /* match via INTUIT shouldn't have any captures.
2551 * Let @-, @+, $^N know */
2552 prog->lastparen = prog->lastcloseparen = 0;
2553 RX_MATCH_UTF8_set(rx, utf8_target);
3ff69bd6
DM
2554 prog->offs[0].start = s - strbeg;
2555 prog->offs[0].end = utf8_target
2556 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2557 : s - strbeg + prog->minlenret;
7fadf4a7 2558 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 2559 S_reg_set_capture_string(aTHX_ rx,
7fadf4a7
DM
2560 strbeg, strend,
2561 sv, flags, utf8_target);
2562
7fadf4a7
DM
2563 return 1;
2564 }
2565 }
2566
6c3fea77 2567 multiline = prog->extflags & RXf_PMf_MULTILINE;
1de06328 2568
dfa77d06 2569 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
a3621e74 2570 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
a72c7584
JH
2571 "String too short [regexec_flags]...\n"));
2572 goto phooey;
1aa99e6b 2573 }
1de06328 2574
6eb5f6b9 2575 /* Check validity of program. */
f8fc2ecf 2576 if (UCHARAT(progi->program) != REG_MAGIC) {
6eb5f6b9
JH
2577 Perl_croak(aTHX_ "corrupted regexp program");
2578 }
2579
1738e041
DM
2580 RX_MATCH_TAINTED_off(rx);
2581
6c3fea77
DM
2582 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2583 reginfo->intuit = 0;
2584 reginfo->is_utf8_target = cBOOL(utf8_target);
02d5137b
DM
2585 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2586 reginfo->warned = FALSE;
9d9163fb 2587 reginfo->strbeg = strbeg;
02d5137b 2588 reginfo->sv = sv;
1cb48e53 2589 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
220db18a 2590 reginfo->strend = strend;
6eb5f6b9 2591 /* see how far we have to get to not match where we matched before */
fe3974be 2592 reginfo->till = stringarg + minend;
6eb5f6b9 2593
60779a30 2594 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
82c23608
FC
2595 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2596 S_cleanup_regmatch_info_aux has executed (registered by
2597 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2598 magic belonging to this SV.
2599 Not newSVsv, either, as it does not COW.
2600 */
60779a30 2601 assert(!IS_PADGV(sv));
82c23608 2602 reginfo->sv = newSV(0);
4cba5ac0 2603 SvSetSV_nosteal(reginfo->sv, sv);
82c23608
FC
2604 SAVEFREESV(reginfo->sv);
2605 }
2606
331b2dcc
DM
2607 /* reserve next 2 or 3 slots in PL_regmatch_state:
2608 * slot N+0: may currently be in use: skip it
2609 * slot N+1: use for regmatch_info_aux struct
2610 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2611 * slot N+3: ready for use by regmatch()
2612 */
bf2039a9 2613
331b2dcc
DM
2614 {
2615 regmatch_state *old_regmatch_state;
2616 regmatch_slab *old_regmatch_slab;
2617 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2618
2619 /* on first ever match, allocate first slab */
2620 if (!PL_regmatch_slab) {
2621 Newx(PL_regmatch_slab, 1, regmatch_slab);
2622 PL_regmatch_slab->prev = NULL;
2623 PL_regmatch_slab->next = NULL;
2624 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2625 }
bf2039a9 2626
331b2dcc
DM
2627 old_regmatch_state = PL_regmatch_state;
2628 old_regmatch_slab = PL_regmatch_slab;
bf2039a9 2629
331b2dcc
DM
2630 for (i=0; i <= max; i++) {
2631 if (i == 1)
2632 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2633 else if (i ==2)
2634 reginfo->info_aux_eval =
2635 reginfo->info_aux->info_aux_eval =
2636 &(PL_regmatch_state->u.info_aux_eval);
bf2039a9 2637
331b2dcc
DM
2638 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2639 PL_regmatch_state = S_push_slab(aTHX);
2640 }
bf2039a9 2641
331b2dcc
DM
2642 /* note initial PL_regmatch_state position; at end of match we'll
2643 * pop back to there and free any higher slabs */
bf2039a9 2644
331b2dcc
DM
2645 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2646 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2ac8ff4b 2647 reginfo->info_aux->poscache = NULL;
bf2039a9 2648
331b2dcc 2649 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
bf2039a9 2650
331b2dcc
DM
2651 if ((prog->extflags & RXf_EVAL_SEEN))
2652 S_setup_eval_state(aTHX_ reginfo);
2653 else
2654 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
bf2039a9 2655 }
d3aa529c 2656
6eb5f6b9 2657 /* If there is a "must appear" string, look for it. */
6eb5f6b9 2658
288b8c02 2659 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
e9105d30
GG
2660 /* We have to be careful. If the previous successful match
2661 was from this regex we don't want a subsequent partially
2662 successful match to clobber the old results.
2663 So when we detect this possibility we add a swap buffer
d8da0584
KW
2664 to the re, and switch the buffer each match. If we fail,
2665 we switch it back; otherwise we leave it swapped.
e9105d30
GG
2666 */
2667 swap = prog->offs;
2668 /* do we need a save destructor here for eval dies? */
2669 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
495f47a5
DM
2670 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2671 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2672 PTR2UV(prog),
2673 PTR2UV(swap),
2674 PTR2UV(prog->offs)
2675 ));
c74340f9 2676 }
6eb5f6b9
JH
2677
2678 /* Simplest case: anchored match need be tried only once. */
2679 /* [unless only anchor is BOL and multiline is set] */
8e1490ee 2680 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
3542935d 2681 if (s == startpos && regtry(reginfo, &s))
6eb5f6b9 2682 goto got_it;
8e1490ee 2683 else if (multiline || (prog->intflags & (PREGf_IMPLICIT | PREGf_ANCH_MBOL))) /* XXXX SBOL? */
6eb5f6b9
JH
2684 {
2685 char *end;
2686
2687 if (minlen)
2688 dontbother = minlen - 1;
1aa99e6b 2689 end = HOP3c(strend, -dontbother, strbeg) - 1;
6eb5f6b9 2690 /* for multiline we only have to try after newlines */
33b8afdf 2691 if (prog->check_substr || prog->check_utf8) {
92f3d482
YO
2692 /* because of the goto we can not easily reuse the macros for bifurcating the
2693 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2694 if (utf8_target) {
2695 if (s == startpos)
2696 goto after_try_utf8;
2697 while (1) {
02d5137b 2698 if (regtry(reginfo, &s)) {
92f3d482
YO
2699 goto got_it;
2700 }
2701 after_try_utf8:
2702 if (s > end) {
2703 goto phooey;
2704 }
2705 if (prog->extflags & RXf_USE_INTUIT) {
52a21eb3
DM
2706 s = re_intuit_start(rx, sv, strbeg,
2707 s + UTF8SKIP(s), strend, flags, NULL);
92f3d482
YO
2708 if (!s) {
2709 goto phooey;
2710 }
2711 }
2712 else {
2713 s += UTF8SKIP(s);
2714 }
2715 }
2716 } /* end search for check string in unicode */
2717 else {
2718 if (s == startpos) {
2719 goto after_try_latin;
2720 }
2721 while (1) {
02d5137b 2722 if (regtry(reginfo, &s)) {
92f3d482
YO
2723 goto got_it;
2724 }
2725 after_try_latin:
2726 if (s > end) {
2727 goto phooey;
2728 }
2729 if (prog->extflags & RXf_USE_INTUIT) {
52a21eb3
DM
2730 s = re_intuit_start(rx, sv, strbeg,
2731 s + 1, strend, flags, NULL);
92f3d482
YO
2732 if (!s) {
2733 goto phooey;
2734 }
2735 }
2736 else {
2737 s++;
2738 }
2739 }
2740 } /* end search for check string in latin*/
2741 } /* end search for check string */
2742 else { /* search for newline */
2743 if (s > startpos) {
2744 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
6eb5f6b9 2745 s--;
92f3d482 2746 }
21eede78
YO
2747 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2748 while (s <= end) { /* note it could be possible to match at the end of the string */
6eb5f6b9 2749 if (*s++ == '\n') { /* don't need PL_utf8skip here */
02d5137b 2750 if (regtry(reginfo, &s))
6eb5f6b9
JH
2751 goto got_it;
2752 }
92f3d482
YO
2753 }
2754 } /* end search for newline */
2755 } /* end anchored/multiline check string search */
6eb5f6b9 2756 goto phooey;
a8430a8b 2757 } else if (prog->intflags & PREGf_ANCH_GPOS)
f9f4320a 2758 {
a8430a8b
YO
2759 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
2760 assert(prog->intflags & PREGf_GPOS_SEEN);
2bfbe302
DM
2761 /* For anchored \G, the only position it can match from is
2762 * (ganch-gofs); we already set startpos to this above; if intuit
2763 * moved us on from there, we can't possibly succeed */
2764 assert(startpos == reginfo->ganch - prog->gofs);
2765 if (s == startpos && regtry(reginfo, &s))
6eb5f6b9
JH
2766 goto got_it;
2767 goto phooey;
2768 }
2769
2770 /* Messy cases: unanchored match. */
bbe252da 2771 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
6eb5f6b9 2772 /* we have /x+whatever/ */
984e6dd1 2773 /* it must be a one character string (XXXX Except is_utf8_pat?) */
33b8afdf 2774 char ch;
bf93d4cc
GS
2775#ifdef DEBUGGING
2776 int did_match = 0;
2777#endif
f2ed9b32 2778 if (utf8_target) {
7e0d5ad7
KW
2779 if (! prog->anchored_utf8) {
2780 to_utf8_substr(prog);
2781 }
2782 ch = SvPVX_const(prog->anchored_utf8)[0];
4cadc6a9 2783 REXEC_FBC_SCAN(
6eb5f6b9 2784 if (*s == ch) {
a3621e74 2785 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 2786 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH
2787 s += UTF8SKIP(s);
2788 while (s < strend && *s == ch)
2789 s += UTF8SKIP(s);
2790 }
4cadc6a9 2791 );
7e0d5ad7 2792
6eb5f6b9
JH
2793 }
2794 else {
7e0d5ad7
KW
2795 if (! prog->anchored_substr) {
2796 if (! to_byte_substr(prog)) {
6b54ddc5 2797 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2798 }
2799 }
2800 ch = SvPVX_const(prog->anchored_substr)[0];
4cadc6a9 2801 REXEC_FBC_SCAN(
6eb5f6b9 2802 if (*s == ch) {
a3621e74 2803 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 2804 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH
2805 s++;
2806 while (s < strend && *s == ch)
2807 s++;
2808 }
4cadc6a9 2809 );
6eb5f6b9 2810 }
a3621e74 2811 DEBUG_EXECUTE_r(if (!did_match)
bf93d4cc 2812 PerlIO_printf(Perl_debug_log,
b7953727
JH
2813 "Did not find anchored character...\n")
2814 );
6eb5f6b9 2815 }
a0714e2c
SS
2816 else if (prog->anchored_substr != NULL
2817 || prog->anchored_utf8 != NULL
2818 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
33b8afdf
JH
2819 && prog->float_max_offset < strend - s)) {
2820 SV *must;
ea3daa5d
FC
2821 SSize_t back_max;
2822 SSize_t back_min;
33b8afdf 2823 char *last;
6eb5f6b9 2824 char *last1; /* Last position checked before */
bf93d4cc
GS
2825#ifdef DEBUGGING
2826 int did_match = 0;
2827#endif
33b8afdf 2828 if (prog->anchored_substr || prog->anchored_utf8) {
7e0d5ad7
KW
2829 if (utf8_target) {
2830 if (! prog->anchored_utf8) {
2831 to_utf8_substr(prog);
2832 }
2833 must = prog->anchored_utf8;
2834 }
2835 else {
2836 if (! prog->anchored_substr) {
2837 if (! to_byte_substr(prog)) {
6b54ddc5 2838 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2839 }
2840 }
2841 must = prog->anchored_substr;
2842 }
33b8afdf
JH
2843 back_max = back_min = prog->anchored_offset;
2844 } else {
7e0d5ad7
KW
2845 if (utf8_target) {
2846 if (! prog->float_utf8) {
2847 to_utf8_substr(prog);
2848 }
2849 must = prog->float_utf8;
2850 }
2851 else {
2852 if (! prog->float_substr) {
2853 if (! to_byte_substr(prog)) {
6b54ddc5 2854 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2855 }
2856 }
2857 must = prog->float_substr;
2858 }
33b8afdf
JH
2859 back_max = prog->float_max_offset;
2860 back_min = prog->float_min_offset;
2861 }
1de06328 2862
1de06328
YO
2863 if (back_min<0) {
2864 last = strend;
2865 } else {
2866 last = HOP3c(strend, /* Cannot start after this */
ea3daa5d 2867 -(SSize_t)(CHR_SVLEN(must)
1de06328
YO
2868 - (SvTAIL(must) != 0) + back_min), strbeg);
2869 }
9d9163fb 2870 if (s > reginfo->strbeg)
6eb5f6b9
JH
2871 last1 = HOPc(s, -1);
2872 else
2873 last1 = s - 1; /* bogus */
2874
a0288114 2875 /* XXXX check_substr already used to find "s", can optimize if
6eb5f6b9 2876 check_substr==must. */
bf05793b 2877 dontbother = 0;
6eb5f6b9
JH
2878 strend = HOPc(strend, -dontbother);
2879 while ( (s <= last) &&
e50d57d4 2880 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
9041c2e3 2881 (unsigned char*)strend, must,
c33e64f0 2882 multiline ? FBMrf_MULTILINE : 0)) ) {
a3621e74 2883 DEBUG_EXECUTE_r( did_match = 1 );
6eb5f6b9
JH
2884 if (HOPc(s, -back_max) > last1) {
2885 last1 = HOPc(s, -back_min);
2886 s = HOPc(s, -back_max);
2887 }
2888 else {
9d9163fb
DM
2889 char * const t = (last1 >= reginfo->strbeg)
2890 ? HOPc(last1, 1) : last1 + 1;
6eb5f6b9
JH
2891
2892 last1 = HOPc(s, -back_min);
52657f30 2893 s = t;
6eb5f6b9 2894 }
f2ed9b32 2895 if (utf8_target) {
6eb5f6b9 2896 while (s <= last1) {
02d5137b 2897 if (regtry(reginfo, &s))
6eb5f6b9 2898 goto got_it;
7016d6eb
DM
2899 if (s >= last1) {
2900 s++; /* to break out of outer loop */
2901 break;
2902 }
2903 s += UTF8SKIP(s);
6eb5f6b9
JH
2904 }
2905 }
2906 else {
2907 while (s <= last1) {
02d5137b 2908 if (regtry(reginfo, &s))
6eb5f6b9
JH
2909 goto got_it;
2910 s++;
2911 }
2912 }
2913 }
ab3bbdeb 2914 DEBUG_EXECUTE_r(if (!did_match) {
f2ed9b32 2915 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
ab3bbdeb
YO
2916 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2917 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
33b8afdf 2918 ((must == prog->anchored_substr || must == prog->anchored_utf8)
bf93d4cc 2919 ? "anchored" : "floating"),
ab3bbdeb
YO
2920 quoted, RE_SV_TAIL(must));
2921 });
6eb5f6b9
JH
2922 goto phooey;
2923 }
f8fc2ecf 2924 else if ( (c = progi->regstclass) ) {
f14c76ed 2925 if (minlen) {
f8fc2ecf 2926 const OPCODE op = OP(progi->regstclass);
66e933ab 2927 /* don't bother with what can't match */
786e8c11 2928 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
f14c76ed
RGS
2929 strend = HOPc(strend, -(minlen - 1));
2930 }
a3621e74 2931 DEBUG_EXECUTE_r({
be8e71aa 2932 SV * const prop = sv_newmortal();
2395827c 2933 regprop(prog, prop, c, reginfo);
0df25f3d 2934 {
f2ed9b32 2935 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb 2936 s,strend-s,60);
0df25f3d 2937 PerlIO_printf(Perl_debug_log,
1c8f8eb1 2938 "Matching stclass %.*s against %s (%d bytes)\n",
e4f74956 2939 (int)SvCUR(prop), SvPVX_const(prop),
ab3bbdeb 2940 quoted, (int)(strend - s));
0df25f3d 2941 }
ffc61ed2 2942 });
f9176b44 2943 if (find_byclass(prog, c, s, strend, reginfo))
6eb5f6b9 2944 goto got_it;
07be1b83 2945 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
d6a28714
JH
2946 }
2947 else {
2948 dontbother = 0;
a0714e2c 2949 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
33b8afdf 2950 /* Trim the end. */
6af40bd7 2951 char *last= NULL;
33b8afdf 2952 SV* float_real;
c33e64f0
FC
2953 STRLEN len;
2954 const char *little;
33b8afdf 2955
7e0d5ad7
KW
2956 if (utf8_target) {
2957 if (! prog->float_utf8) {
2958 to_utf8_substr(prog);
2959 }
2960 float_real = prog->float_utf8;
2961 }
2962 else {
2963 if (! prog->float_substr) {
2964 if (! to_byte_substr(prog)) {
6b54ddc5 2965 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2966 }
2967 }
2968 float_real = prog->float_substr;
2969 }
d6a28714 2970
c33e64f0
FC
2971 little = SvPV_const(float_real, len);
2972 if (SvTAIL(float_real)) {
7f18ad16
KW
2973 /* This means that float_real contains an artificial \n on
2974 * the end due to the presence of something like this:
2975 * /foo$/ where we can match both "foo" and "foo\n" at the
2976 * end of the string. So we have to compare the end of the
2977 * string first against the float_real without the \n and
2978 * then against the full float_real with the string. We
2979 * have to watch out for cases where the string might be
2980 * smaller than the float_real or the float_real without
2981 * the \n. */
1a13b075
YO
2982 char *checkpos= strend - len;
2983 DEBUG_OPTIMISE_r(
2984 PerlIO_printf(Perl_debug_log,
2985 "%sChecking for float_real.%s\n",
2986 PL_colors[4], PL_colors[5]));
2987 if (checkpos + 1 < strbeg) {
7f18ad16
KW
2988 /* can't match, even if we remove the trailing \n
2989 * string is too short to match */
1a13b075
YO
2990 DEBUG_EXECUTE_r(
2991 PerlIO_printf(Perl_debug_log,
2992 "%sString shorter than required trailing substring, cannot match.%s\n",
2993 PL_colors[4], PL_colors[5]));
2994 goto phooey;
2995 } else if (memEQ(checkpos + 1, little, len - 1)) {
7f18ad16
KW
2996 /* can match, the end of the string matches without the
2997 * "\n" */
1a13b075
YO
2998 last = checkpos + 1;
2999 } else if (checkpos < strbeg) {
7f18ad16
KW
3000 /* cant match, string is too short when the "\n" is
3001 * included */
1a13b075
YO
3002 DEBUG_EXECUTE_r(
3003 PerlIO_printf(Perl_debug_log,
3004 "%sString does not contain required trailing substring, cannot match.%s\n",
3005 PL_colors[4], PL_colors[5]));
3006 goto phooey;
3007 } else if (!multiline) {
7f18ad16
KW
3008 /* non multiline match, so compare with the "\n" at the
3009 * end of the string */
1a13b075
YO
3010 if (memEQ(checkpos, little, len)) {
3011 last= checkpos;
3012 } else {
3013 DEBUG_EXECUTE_r(
3014 PerlIO_printf(Perl_debug_log,
3015 "%sString does not contain required trailing substring, cannot match.%s\n",
3016 PL_colors[4], PL_colors[5]));
3017 goto phooey;
3018 }
3019 } else {
7f18ad16
KW
3020 /* multiline match, so we have to search for a place
3021 * where the full string is located */
d6a28714 3022 goto find_last;
1a13b075 3023 }
c33e64f0 3024 } else {
d6a28714 3025 find_last:
9041c2e3 3026 if (len)
d6a28714 3027 last = rninstr(s, strend, little, little + len);
b8c5462f 3028 else
a0288114 3029 last = strend; /* matching "$" */
b8c5462f 3030 }
6af40bd7 3031 if (!last) {
7f18ad16
KW
3032 /* at one point this block contained a comment which was
3033 * probably incorrect, which said that this was a "should not
3034 * happen" case. Even if it was true when it was written I am
3035 * pretty sure it is not anymore, so I have removed the comment
3036 * and replaced it with this one. Yves */
6bda09f9
YO
3037 DEBUG_EXECUTE_r(
3038 PerlIO_printf(Perl_debug_log,
6af40bd7
YO
3039 "String does not contain required substring, cannot match.\n"
3040 ));
3041 goto phooey;
bf93d4cc 3042 }
d6a28714
JH
3043 dontbother = strend - last + prog->float_min_offset;
3044 }
3045 if (minlen && (dontbother < minlen))
3046 dontbother = minlen - 1;
3047 strend -= dontbother; /* this one's always in bytes! */
3048 /* We don't know much -- general case. */
f2ed9b32 3049 if (utf8_target) {
d6a28714 3050 for (;;) {
02d5137b 3051 if (regtry(reginfo, &s))
d6a28714
JH
3052 goto got_it;
3053 if (s >= strend)
3054 break;
b8c5462f 3055 s += UTF8SKIP(s);
d6a28714
JH
3056 };
3057 }
3058 else {
3059 do {
02d5137b 3060 if (regtry(reginfo, &s))
d6a28714
JH
3061 goto got_it;
3062 } while (s++ < strend);
3063 }
3064 }
3065
3066 /* Failure. */
3067 goto phooey;
3068
3069got_it:
d5e7783a
DM
3070 /* s/// doesn't like it if $& is earlier than where we asked it to
3071 * start searching (which can happen on something like /.\G/) */
3072 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3073 && (prog->offs[0].start < stringarg - strbeg))
3074 {
3075 /* this should only be possible under \G */
58430ea8 3076 assert(prog->intflags & PREGf_GPOS_SEEN);
d5e7783a
DM
3077 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3078 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3079 goto phooey;
3080 }
3081
495f47a5
DM
3082 DEBUG_BUFFERS_r(
3083 if (swap)
3084 PerlIO_printf(Perl_debug_log,
3085 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
3086 PTR2UV(prog),
3087 PTR2UV(swap)
3088 );
3089 );
e9105d30 3090 Safefree(swap);
d6a28714 3091
bf2039a9
DM
3092 /* clean up; this will trigger destructors that will free all slabs
3093 * above the current one, and cleanup the regmatch_info_aux
3094 * and regmatch_info_aux_eval sructs */
8adc0f72 3095
006f26b2
DM
3096 LEAVE_SCOPE(oldsave);
3097
5daac39c
NC
3098 if (RXp_PAREN_NAMES(prog))
3099 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
d6a28714 3100
0254aed9
DM
3101 RX_MATCH_UTF8_set(rx, utf8_target);
3102
d6a28714 3103 /* make sure $`, $&, $', and $digit will work later */
60165aa4 3104 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 3105 S_reg_set_capture_string(aTHX_ rx,
60165aa4
DM
3106 strbeg, reginfo->strend,
3107 sv, flags, utf8_target);
9041c2e3 3108
d6a28714
JH
3109 return 1;
3110
3111phooey:
a3621e74 3112 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
e4584336 3113 PL_colors[4], PL_colors[5]));
8adc0f72 3114
bf2039a9
DM
3115 /* clean up; this will trigger destructors that will free all slabs
3116 * above the current one, and cleanup the regmatch_info_aux
3117 * and regmatch_info_aux_eval sructs */
8adc0f72 3118
006f26b2
DM
3119 LEAVE_SCOPE(oldsave);
3120
e9105d30 3121 if (swap) {
c74340f9 3122 /* we failed :-( roll it back */
495f47a5
DM
3123 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
3124 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
3125 PTR2UV(prog),
3126 PTR2UV(prog->offs),
3127 PTR2UV(swap)
3128 ));
e9105d30
GG
3129 Safefree(prog->offs);
3130 prog->offs = swap;
3131 }
d6a28714
JH
3132 return 0;
3133}
3134
6bda09f9 3135
b3d298be 3136/* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
ec43f78b 3137 * Do inc before dec, in case old and new rex are the same */
baa60164 3138#define SET_reg_curpm(Re2) \
bf2039a9 3139 if (reginfo->info_aux_eval) { \
ec43f78b
DM
3140 (void)ReREFCNT_inc(Re2); \
3141 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3142 PM_SETRE((PL_reg_curpm), (Re2)); \
3143 }
3144
3145
d6a28714
JH
3146/*
3147 - regtry - try match at specific point
3148 */
3149STATIC I32 /* 0 failure, 1 success */
f73aaa43 3150S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
d6a28714 3151{
d6a28714 3152 CHECKPOINT lastcp;
288b8c02 3153 REGEXP *const rx = reginfo->prog;
8d919b0a 3154 regexp *const prog = ReANY(rx);
99a90e59 3155 SSize_t result;
f8fc2ecf 3156 RXi_GET_DECL(prog,progi);
a3621e74 3157 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
3158
3159 PERL_ARGS_ASSERT_REGTRY;
3160
24b23f37 3161 reginfo->cutpoint=NULL;
d6a28714 3162
9d9163fb 3163 prog->offs[0].start = *startposp - reginfo->strbeg;
d6a28714 3164 prog->lastparen = 0;
03994de8 3165 prog->lastcloseparen = 0;
d6a28714
JH
3166
3167 /* XXXX What this code is doing here?!!! There should be no need
b93070ed 3168 to do this again and again, prog->lastparen should take care of
3dd2943c 3169 this! --ilya*/
dafc8851
JH
3170
3171 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3172 * Actually, the code in regcppop() (which Ilya may be meaning by
b93070ed 3173 * prog->lastparen), is not needed at all by the test suite
225593e1
DM
3174 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3175 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3176 * Meanwhile, this code *is* needed for the
daf18116
JH
3177 * above-mentioned test suite tests to succeed. The common theme
3178 * on those tests seems to be returning null fields from matches.
225593e1 3179 * --jhi updated by dapm */
dafc8851 3180#if 1
d6a28714 3181 if (prog->nparens) {
b93070ed 3182 regexp_paren_pair *pp = prog->offs;
eb578fdb 3183 I32 i;
b93070ed 3184 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
f0ab9afb
NC
3185 ++pp;
3186 pp->start = -1;
3187 pp->end = -1;
d6a28714
JH
3188 }
3189 }
dafc8851 3190#endif
02db2b7b 3191 REGCP_SET(lastcp);
f73aaa43
DM
3192 result = regmatch(reginfo, *startposp, progi->program + 1);
3193 if (result != -1) {
3194 prog->offs[0].end = result;
d6a28714
JH
3195 return 1;
3196 }
24b23f37 3197 if (reginfo->cutpoint)
f73aaa43 3198 *startposp= reginfo->cutpoint;
02db2b7b 3199 REGCP_UNWIND(lastcp);
d6a28714
JH
3200 return 0;
3201}
3202
02db2b7b 3203
8ba1375e
MJD
3204#define sayYES goto yes
3205#define sayNO goto no
262b90c4 3206#define sayNO_SILENT goto no_silent
8ba1375e 3207
f9f4320a
YO
3208/* we dont use STMT_START/END here because it leads to
3209 "unreachable code" warnings, which are bogus, but distracting. */
3210#define CACHEsayNO \
c476f425 3211 if (ST.cache_mask) \
2ac8ff4b 3212 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
f9f4320a 3213 sayNO
3298f257 3214
a3621e74 3215/* this is used to determine how far from the left messages like
265c4333
YO
3216 'failed...' are printed. It should be set such that messages
3217 are inline with the regop output that created them.
a3621e74 3218*/
265c4333 3219#define REPORT_CODE_OFF 32
a3621e74
YO
3220
3221
40a82448
DM
3222#define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3223#define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
79a2a0e8
KW
3224#define CHRTEST_NOT_A_CP_1 -999
3225#define CHRTEST_NOT_A_CP_2 -998
9e137952 3226
5d9a96ca
DM
3227/* grab a new slab and return the first slot in it */
3228
3229STATIC regmatch_state *
3230S_push_slab(pTHX)
3231{
a35a87e7 3232#if PERL_VERSION < 9 && !defined(PERL_CORE)
54df2634
NC
3233 dMY_CXT;
3234#endif
5d9a96ca
DM
3235 regmatch_slab *s = PL_regmatch_slab->next;
3236 if (!s) {
3237 Newx(s, 1, regmatch_slab);
3238 s->prev = PL_regmatch_slab;
3239 s->next = NULL;
3240 PL_regmatch_slab->next = s;
3241 }
3242 PL_regmatch_slab = s;
86545054 3243 return SLAB_FIRST(s);
5d9a96ca 3244}
5b47454d 3245
95b24440 3246
40a82448
DM
3247/* push a new state then goto it */
3248
4d5016e5
DM
3249#define PUSH_STATE_GOTO(state, node, input) \
3250 pushinput = input; \
40a82448
DM
3251 scan = node; \
3252 st->resume_state = state; \
3253 goto push_state;
3254
3255/* push a new state with success backtracking, then goto it */
3256
4d5016e5
DM
3257#define PUSH_YES_STATE_GOTO(state, node, input) \
3258 pushinput = input; \
40a82448
DM
3259 scan = node; \
3260 st->resume_state = state; \
3261 goto push_yes_state;
3262
aa283a38 3263
aa283a38 3264
4d5016e5 3265
d6a28714 3266/*
95b24440 3267
bf1f174e
DM
3268regmatch() - main matching routine
3269
3270This is basically one big switch statement in a loop. We execute an op,
3271set 'next' to point the next op, and continue. If we come to a point which
3272we may need to backtrack to on failure such as (A|B|C), we push a
3273backtrack state onto the backtrack stack. On failure, we pop the top
3274state, and re-enter the loop at the state indicated. If there are no more
3275states to pop, we return failure.
3276
3277Sometimes we also need to backtrack on success; for example /A+/, where
3278after successfully matching one A, we need to go back and try to
3279match another one; similarly for lookahead assertions: if the assertion
3280completes successfully, we backtrack to the state just before the assertion
3281and then carry on. In these cases, the pushed state is marked as
3282'backtrack on success too'. This marking is in fact done by a chain of
3283pointers, each pointing to the previous 'yes' state. On success, we pop to
3284the nearest yes state, discarding any intermediate failure-only states.
3285Sometimes a yes state is pushed just to force some cleanup code to be
3286called at the end of a successful match or submatch; e.g. (??{$re}) uses
3287it to free the inner regex.
3288
3289Note that failure backtracking rewinds the cursor position, while
3290success backtracking leaves it alone.
3291
3292A pattern is complete when the END op is executed, while a subpattern
3293such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3294ops trigger the "pop to last yes state if any, otherwise return true"
3295behaviour.
3296
3297A common convention in this function is to use A and B to refer to the two
3298subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3299the subpattern to be matched possibly multiple times, while B is the entire
3300rest of the pattern. Variable and state names reflect this convention.
3301
3302The states in the main switch are the union of ops and failure/success of
3303substates associated with with that op. For example, IFMATCH is the op
3304that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3305'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3306successfully matched A and IFMATCH_A_fail is a state saying that we have
3307just failed to match A. Resume states always come in pairs. The backtrack
3308state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3309at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3310on success or failure.
3311
3312The struct that holds a backtracking state is actually a big union, with
3313one variant for each major type of op. The variable st points to the
3314top-most backtrack struct. To make the code clearer, within each
3315block of code we #define ST to alias the relevant union.
3316
3317Here's a concrete example of a (vastly oversimplified) IFMATCH
3318implementation:
3319
3320 switch (state) {
3321 ....
3322
3323#define ST st->u.ifmatch
3324
3325 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3326 ST.foo = ...; // some state we wish to save
95b24440 3327 ...
bf1f174e
DM
3328 // push a yes backtrack state with a resume value of
3329 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3330 // first node of A:
4d5016e5 3331 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
bf1f174e
DM
3332 // NOTREACHED
3333
3334 case IFMATCH_A: // we have successfully executed A; now continue with B
3335 next = B;
3336 bar = ST.foo; // do something with the preserved value
3337 break;
3338
3339 case IFMATCH_A_fail: // A failed, so the assertion failed
3340 ...; // do some housekeeping, then ...
3341 sayNO; // propagate the failure
3342
3343#undef ST
95b24440 3344
bf1f174e
DM
3345 ...
3346 }
95b24440 3347
bf1f174e
DM
3348For any old-timers reading this who are familiar with the old recursive
3349approach, the code above is equivalent to:
95b24440 3350
bf1f174e
DM
3351 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3352 {
3353 int foo = ...
95b24440 3354 ...
bf1f174e
DM
3355 if (regmatch(A)) {
3356 next = B;
3357 bar = foo;
3358 break;
95b24440 3359 }
bf1f174e
DM
3360 ...; // do some housekeeping, then ...
3361 sayNO; // propagate the failure
95b24440 3362 }
bf1f174e
DM
3363
3364The topmost backtrack state, pointed to by st, is usually free. If you
3365want to claim it, populate any ST.foo fields in it with values you wish to
3366save, then do one of
3367
4d5016e5
DM
3368 PUSH_STATE_GOTO(resume_state, node, newinput);
3369 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
bf1f174e
DM
3370
3371which sets that backtrack state's resume value to 'resume_state', pushes a
3372new free entry to the top of the backtrack stack, then goes to 'node'.
3373On backtracking, the free slot is popped, and the saved state becomes the
3374new free state. An ST.foo field in this new top state can be temporarily
3375accessed to retrieve values, but once the main loop is re-entered, it
3376becomes available for reuse.
3377
3378Note that the depth of the backtrack stack constantly increases during the
3379left-to-right execution of the pattern, rather than going up and down with
3380the pattern nesting. For example the stack is at its maximum at Z at the
3381end of the pattern, rather than at X in the following:
3382
3383 /(((X)+)+)+....(Y)+....Z/
3384
3385The only exceptions to this are lookahead/behind assertions and the cut,
3386(?>A), which pop all the backtrack states associated with A before
3387continuing.
3388
486ec47a 3389Backtrack state structs are allocated in slabs of about 4K in size.
bf1f174e
DM
3390PL_regmatch_state and st always point to the currently active state,
3391and PL_regmatch_slab points to the slab currently containing
3392PL_regmatch_state. The first time regmatch() is called, the first slab is
3393allocated, and is never freed until interpreter destruction. When the slab
3394is full, a new one is allocated and chained to the end. At exit from
3395regmatch(), slabs allocated since entry are freed.
3396
3397*/
95b24440 3398
40a82448 3399
5bc10b2c 3400#define DEBUG_STATE_pp(pp) \
265c4333 3401 DEBUG_STATE_r({ \
baa60164 3402 DUMP_EXEC_POS(locinput, scan, utf8_target); \
5bc10b2c 3403 PerlIO_printf(Perl_debug_log, \
5d458dd8 3404 " %*s"pp" %s%s%s%s%s\n", \
5bc10b2c 3405 depth*2, "", \
baa60164 3406 PL_reg_name[st->resume_state], \
5d458dd8
YO
3407 ((st==yes_state||st==mark_state) ? "[" : ""), \
3408 ((st==yes_state) ? "Y" : ""), \
3409 ((st==mark_state) ? "M" : ""), \
3410 ((st==yes_state||st==mark_state) ? "]" : "") \
3411 ); \
265c4333 3412 });
5bc10b2c 3413
40a82448 3414
3dab1dad 3415#define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
95b24440 3416
3df15adc 3417#ifdef DEBUGGING
5bc10b2c 3418
ab3bbdeb 3419STATIC void
f2ed9b32 3420S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
ab3bbdeb
YO
3421 const char *start, const char *end, const char *blurb)
3422{
efd26800 3423 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
7918f24d
NC
3424
3425 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3426
ab3bbdeb
YO
3427 if (!PL_colorset)
3428 reginitcolors();
3429 {
3430 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
d2c6dc5e 3431 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
ab3bbdeb 3432
f2ed9b32 3433 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb
YO
3434 start, end - start, 60);
3435
3436 PerlIO_printf(Perl_debug_log,
3437 "%s%s REx%s %s against %s\n",
3438 PL_colors[4], blurb, PL_colors[5], s0, s1);
3439
f2ed9b32 3440 if (utf8_target||utf8_pat)
1de06328
YO
3441 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3442 utf8_pat ? "pattern" : "",
f2ed9b32
KW
3443 utf8_pat && utf8_target ? " and " : "",
3444 utf8_target ? "string" : ""
ab3bbdeb
YO
3445 );
3446 }
3447}
3df15adc
YO
3448
3449STATIC void
786e8c11
YO
3450S_dump_exec_pos(pTHX_ const char *locinput,
3451 const regnode *scan,
3452 const char *loc_regeol,
3453 const char *loc_bostr,
3454 const char *loc_reg_starttry,
f2ed9b32 3455 const bool utf8_target)
07be1b83 3456{
786e8c11 3457 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
07be1b83 3458 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
786e8c11 3459 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
07be1b83
YO
3460 /* The part of the string before starttry has one color
3461 (pref0_len chars), between starttry and current
3462 position another one (pref_len - pref0_len chars),
3463 after the current position the third one.
3464 We assume that pref0_len <= pref_len, otherwise we
3465 decrease pref0_len. */
786e8c11
YO
3466 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3467 ? (5 + taill) - l : locinput - loc_bostr;
07be1b83
YO
3468 int pref0_len;
3469
7918f24d
NC
3470 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3471
f2ed9b32 3472 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
07be1b83 3473 pref_len++;
786e8c11
YO
3474 pref0_len = pref_len - (locinput - loc_reg_starttry);
3475 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3476 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3477 ? (5 + taill) - pref_len : loc_regeol - locinput);
f2ed9b32 3478 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
07be1b83
YO
3479 l--;
3480 if (pref0_len < 0)
3481 pref0_len = 0;
3482 if (pref0_len > pref_len)
3483 pref0_len = pref_len;
3484 {
f2ed9b32 3485 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
0df25f3d 3486
ab3bbdeb 3487 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
1de06328 3488 (locinput - pref_len),pref0_len, 60, 4, 5);
0df25f3d 3489
ab3bbdeb 3490 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3df15adc 3491 (locinput - pref_len + pref0_len),
1de06328 3492 pref_len - pref0_len, 60, 2, 3);
0df25f3d 3493
ab3bbdeb 3494 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
1de06328 3495 locinput, loc_regeol - locinput, 10, 0, 1);
0df25f3d 3496
1de06328 3497 const STRLEN tlen=len0+len1+len2;
3df15adc 3498 PerlIO_printf(Perl_debug_log,
ab3bbdeb 3499 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
786e8c11 3500 (IV)(locinput - loc_bostr),
07be1b83 3501 len0, s0,
07be1b83 3502 len1, s1,
07be1b83 3503 (docolor ? "" : "> <"),
07be1b83 3504 len2, s2,
f9f4320a 3505 (int)(tlen > 19 ? 0 : 19 - tlen),
07be1b83
YO
3506 "");
3507 }
3508}
3df15adc 3509
07be1b83
YO
3510#endif
3511
0a4db386
YO
3512/* reg_check_named_buff_matched()
3513 * Checks to see if a named buffer has matched. The data array of
3514 * buffer numbers corresponding to the buffer is expected to reside
3515 * in the regexp->data->data array in the slot stored in the ARG() of
3516 * node involved. Note that this routine doesn't actually care about the
3517 * name, that information is not preserved from compilation to execution.
3518 * Returns the index of the leftmost defined buffer with the given name
3519 * or 0 if non of the buffers matched.
3520 */
3521STATIC I32
dc3bf405 3522S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
7918f24d 3523{
0a4db386 3524 I32 n;
f8fc2ecf