This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regexec(): access extflags directly
[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
7e0d5ad7
KW
40/* At least one required character in the target string is expressible only in
41 * UTF-8. */
991fc03a 42static const char* const non_utf8_target_but_utf8_required
7e0d5ad7
KW
43 = "Can't match, because target string needs to be in UTF-8\n";
44
6b54ddc5
YO
45#define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
46 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\
47 goto target; \
48} STMT_END
49
a687059c 50/*
e50aee73 51 * pregcomp and pregexec -- regsub and regerror are not used in perl
a687059c
LW
52 *
53 * Copyright (c) 1986 by University of Toronto.
54 * Written by Henry Spencer. Not derived from licensed software.
55 *
56 * Permission is granted to anyone to use this software for any
57 * purpose on any computer system, and to redistribute it freely,
58 * subject to the following restrictions:
59 *
60 * 1. The author is not responsible for the consequences of use of
61 * this software, no matter how awful, even if they arise
62 * from defects in it.
63 *
64 * 2. The origin of this software must not be misrepresented, either
65 * by explicit claim or by omission.
66 *
67 * 3. Altered versions must be plainly marked as such, and must not
68 * be misrepresented as being the original software.
69 *
70 **** Alterations to Henry's code are...
71 ****
4bb101f2 72 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
1129b882
NC
73 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
74 **** by Larry Wall and others
a687059c 75 ****
9ef589d8
LW
76 **** You may distribute under the terms of either the GNU General Public
77 **** License or the Artistic License, as specified in the README file.
a687059c
LW
78 *
79 * Beware that some of this code is subtly aware of the way operator
80 * precedence is structured in regular expressions. Serious changes in
81 * regular-expression syntax might require a total rethink.
82 */
83#include "EXTERN.h"
864dbfa3 84#define PERL_IN_REGEXEC_C
a687059c 85#include "perl.h"
0f5d15d6 86
54df2634
NC
87#ifdef PERL_IN_XSUB_RE
88# include "re_comp.h"
89#else
90# include "regcomp.h"
91#endif
a687059c 92
81e983c1 93#include "inline_invlist.c"
1b0f46bf 94#include "unicode_constants.h"
81e983c1 95
c74f6de9
KW
96#define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
97
a687059c
LW
98#ifndef STATIC
99#define STATIC static
100#endif
101
e0193e47 102/* Valid for non-utf8 strings: avoids the reginclass
7e2509c1
KW
103 * call if there are no complications: i.e., if everything matchable is
104 * straight forward in the bitmap */
635cd5d4 105#define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0) \
af364d03 106 : ANYOF_BITMAP_TEST(p,*(c)))
7d3e948e 107
c277df42
IZ
108/*
109 * Forwards.
110 */
111
f2ed9b32 112#define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
ba44c216 113#define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b)
a0ed51b3 114
3dab1dad 115#define HOPc(pos,off) \
ba44c216 116 (char *)(reginfo->is_utf8_target \
220db18a 117 ? reghop3((U8*)pos, off, \
9d9163fb 118 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
3dab1dad
YO
119 : (U8*)(pos + off))
120#define HOPBACKc(pos, off) \
ba44c216 121 (char*)(reginfo->is_utf8_target \
9d9163fb
DM
122 ? reghopmaybe3((U8*)pos, -off, (U8*)(reginfo->strbeg)) \
123 : (pos - off >= reginfo->strbeg) \
8e11feef 124 ? (U8*)pos - off \
3dab1dad 125 : NULL)
efb30f32 126
ba44c216 127#define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
1aa99e6b 128#define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
1aa99e6b 129
7016d6eb
DM
130
131#define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
132#define NEXTCHR_IS_EOS (nextchr < 0)
133
134#define SET_nextchr \
220db18a 135 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
7016d6eb
DM
136
137#define SET_locinput(p) \
138 locinput = (p); \
139 SET_nextchr
140
141
c7304fe2
KW
142#define LOAD_UTF8_CHARCLASS(swash_ptr, property_name) STMT_START { \
143 if (!swash_ptr) { \
144 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
c7304fe2
KW
145 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
146 1, 0, NULL, &flags); \
147 assert(swash_ptr); \
148 } \
149 } STMT_END
150
151/* If in debug mode, we test that a known character properly matches */
152#ifdef DEBUGGING
153# define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
154 property_name, \
155 utf8_char_in_property) \
156 LOAD_UTF8_CHARCLASS(swash_ptr, property_name); \
157 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
158#else
159# define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
160 property_name, \
161 utf8_char_in_property) \
162 LOAD_UTF8_CHARCLASS(swash_ptr, property_name)
163#endif
d1eb3177 164
c7304fe2
KW
165#define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
166 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
167 swash_property_names[_CC_WORDCHAR], \
168 GREEK_SMALL_LETTER_IOTA_UTF8)
169
170#define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
171 STMT_START { \
172 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \
173 "_X_regular_begin", \
174 GREEK_SMALL_LETTER_IOTA_UTF8); \
175 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \
176 "_X_extend", \
177 COMBINING_GRAVE_ACCENT_UTF8); \
178 } STMT_END
d1eb3177 179
c7304fe2 180#define PLACEHOLDER /* Something for the preprocessor to grab onto */
3dab1dad
YO
181/* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
182
5f80c4cf 183/* for use after a quantifier and before an EXACT-like node -- japhy */
c35dcbe2
YO
184/* it would be nice to rework regcomp.sym to generate this stuff. sigh
185 *
186 * NOTE that *nothing* that affects backtracking should be in here, specifically
187 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
188 * node that is in between two EXACT like nodes when ascertaining what the required
189 * "follow" character is. This should probably be moved to regex compile time
190 * although it may be done at run time beause of the REF possibility - more
191 * investigation required. -- demerphq
192*/
3e901dc0
YO
193#define JUMPABLE(rn) ( \
194 OP(rn) == OPEN || \
195 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
196 OP(rn) == EVAL || \
cca55fe3
JP
197 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
198 OP(rn) == PLUS || OP(rn) == MINMOD || \
d1c771f5 199 OP(rn) == KEEPS || \
3dab1dad 200 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
e2d8ce26 201)
ee9b8eae 202#define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
e2d8ce26 203
ee9b8eae
YO
204#define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
205
206#if 0
207/* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
208 we don't need this definition. */
209#define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
fab2782b 210#define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn)==EXACTFA || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
ee9b8eae
YO
211#define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
212
213#else
214/* ... so we use this as its faster. */
215#define IS_TEXT(rn) ( OP(rn)==EXACT )
fab2782b 216#define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
ee9b8eae
YO
217#define IS_TEXTF(rn) ( OP(rn)==EXACTF )
218#define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
219
220#endif
e2d8ce26 221
a84d97b6
HS
222/*
223 Search for mandatory following text node; for lookahead, the text must
224 follow but for lookbehind (rn->flags != 0) we skip to the next step.
225*/
cca55fe3 226#define FIND_NEXT_IMPT(rn) STMT_START { \
3dab1dad
YO
227 while (JUMPABLE(rn)) { \
228 const OPCODE type = OP(rn); \
229 if (type == SUSPEND || PL_regkind[type] == CURLY) \
e2d8ce26 230 rn = NEXTOPER(NEXTOPER(rn)); \
3dab1dad 231 else if (type == PLUS) \
cca55fe3 232 rn = NEXTOPER(rn); \
3dab1dad 233 else if (type == IFMATCH) \
a84d97b6 234 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
e2d8ce26 235 else rn += NEXT_OFF(rn); \
3dab1dad 236 } \
5f80c4cf 237} STMT_END
74750237 238
22913b96
KW
239/* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode.
240 * These are for the pre-composed Hangul syllables, which are all in a
241 * contiguous block and arranged there in such a way so as to facilitate
242 * alorithmic determination of their characteristics. As such, they don't need
243 * a swash, but can be determined by simple arithmetic. Almost all are
244 * GCB=LVT, but every 28th one is a GCB=LV */
245#define SBASE 0xAC00 /* Start of block */
246#define SCount 11172 /* Length of block */
247#define TCount 28
c476f425 248
006f26b2
DM
249#define SLAB_FIRST(s) (&(s)->states[0])
250#define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
251
a75351a1 252static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
bf2039a9 253static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
bf2039a9 254static regmatch_state * S_push_slab(pTHX);
51371543 255
87c0511b 256#define REGCP_PAREN_ELEMS 3
f067efbf 257#define REGCP_OTHER_ELEMS 3
e0fa7e2b 258#define REGCP_FRAME_ELEMS 1
620d5b66
NC
259/* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
260 * are needed for the regexp context stack bookkeeping. */
261
76e3520e 262STATIC CHECKPOINT
92da3157 263S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
a0d0e21e 264{
97aff369 265 dVAR;
a3b680e6 266 const int retval = PL_savestack_ix;
92da3157
DM
267 const int paren_elems_to_push =
268 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
e0fa7e2b
NC
269 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
270 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
87c0511b 271 I32 p;
40a82448 272 GET_RE_DEBUG_FLAGS_DECL;
a0d0e21e 273
b93070ed
DM
274 PERL_ARGS_ASSERT_REGCPPUSH;
275
e49a9654 276 if (paren_elems_to_push < 0)
5637ef5b
NC
277 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
278 paren_elems_to_push);
e49a9654 279
e0fa7e2b
NC
280 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
281 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
5df417d0 282 " out of range (%lu-%ld)",
92da3157
DM
283 total_elems,
284 (unsigned long)maxopenparen,
285 (long)parenfloor);
e0fa7e2b 286
620d5b66 287 SSGROW(total_elems + REGCP_FRAME_ELEMS);
7f69552c 288
495f47a5 289 DEBUG_BUFFERS_r(
92da3157 290 if ((int)maxopenparen > (int)parenfloor)
495f47a5
DM
291 PerlIO_printf(Perl_debug_log,
292 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
293 PTR2UV(rex),
294 PTR2UV(rex->offs)
295 );
296 );
92da3157 297 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
b1ce53c5 298/* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
b93070ed
DM
299 SSPUSHINT(rex->offs[p].end);
300 SSPUSHINT(rex->offs[p].start);
1ca2007e 301 SSPUSHINT(rex->offs[p].start_tmp);
e7707071 302 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
495f47a5
DM
303 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
304 (UV)p,
305 (IV)rex->offs[p].start,
306 (IV)rex->offs[p].start_tmp,
307 (IV)rex->offs[p].end
40a82448 308 ));
a0d0e21e 309 }
b1ce53c5 310/* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
92da3157 311 SSPUSHINT(maxopenparen);
b93070ed
DM
312 SSPUSHINT(rex->lastparen);
313 SSPUSHINT(rex->lastcloseparen);
e0fa7e2b 314 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
41123dfd 315
a0d0e21e
LW
316 return retval;
317}
318
c277df42 319/* These are needed since we do not localize EVAL nodes: */
ab3bbdeb
YO
320#define REGCP_SET(cp) \
321 DEBUG_STATE_r( \
ab3bbdeb 322 PerlIO_printf(Perl_debug_log, \
e4f74956 323 " Setting an EVAL scope, savestack=%"IVdf"\n", \
ab3bbdeb
YO
324 (IV)PL_savestack_ix)); \
325 cp = PL_savestack_ix
c3464db5 326
ab3bbdeb 327#define REGCP_UNWIND(cp) \
e4f74956 328 DEBUG_STATE_r( \
ab3bbdeb 329 if (cp != PL_savestack_ix) \
e4f74956
YO
330 PerlIO_printf(Perl_debug_log, \
331 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
ab3bbdeb
YO
332 (IV)(cp), (IV)PL_savestack_ix)); \
333 regcpblow(cp)
c277df42 334
a8d1f4b4
DM
335#define UNWIND_PAREN(lp, lcp) \
336 for (n = rex->lastparen; n > lp; n--) \
337 rex->offs[n].end = -1; \
338 rex->lastparen = n; \
339 rex->lastcloseparen = lcp;
340
341
f067efbf 342STATIC void
92da3157 343S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
a0d0e21e 344{
97aff369 345 dVAR;
e0fa7e2b 346 UV i;
87c0511b 347 U32 paren;
a3621e74
YO
348 GET_RE_DEBUG_FLAGS_DECL;
349
7918f24d
NC
350 PERL_ARGS_ASSERT_REGCPPOP;
351
b1ce53c5 352 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
c6bf6a65 353 i = SSPOPUV;
e0fa7e2b
NC
354 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
355 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
b93070ed
DM
356 rex->lastcloseparen = SSPOPINT;
357 rex->lastparen = SSPOPINT;
92da3157 358 *maxopenparen_p = SSPOPINT;
b1ce53c5 359
620d5b66 360 i -= REGCP_OTHER_ELEMS;
b1ce53c5 361 /* Now restore the parentheses context. */
495f47a5
DM
362 DEBUG_BUFFERS_r(
363 if (i || rex->lastparen + 1 <= rex->nparens)
364 PerlIO_printf(Perl_debug_log,
365 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
366 PTR2UV(rex),
367 PTR2UV(rex->offs)
368 );
369 );
92da3157 370 paren = *maxopenparen_p;
620d5b66 371 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
1df70142 372 I32 tmps;
1ca2007e 373 rex->offs[paren].start_tmp = SSPOPINT;
b93070ed 374 rex->offs[paren].start = SSPOPINT;
cf93c79d 375 tmps = SSPOPINT;
b93070ed
DM
376 if (paren <= rex->lastparen)
377 rex->offs[paren].end = tmps;
495f47a5
DM
378 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
379 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
380 (UV)paren,
381 (IV)rex->offs[paren].start,
382 (IV)rex->offs[paren].start_tmp,
383 (IV)rex->offs[paren].end,
384 (paren > rex->lastparen ? "(skipped)" : ""));
c277df42 385 );
87c0511b 386 paren--;
a0d0e21e 387 }
daf18116 388#if 1
dafc8851
JH
389 /* It would seem that the similar code in regtry()
390 * already takes care of this, and in fact it is in
391 * a better location to since this code can #if 0-ed out
392 * but the code in regtry() is needed or otherwise tests
393 * requiring null fields (pat.t#187 and split.t#{13,14}
daf18116
JH
394 * (as of patchlevel 7877) will fail. Then again,
395 * this code seems to be necessary or otherwise
225593e1
DM
396 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
397 * --jhi updated by dapm */
b93070ed 398 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
92da3157 399 if (i > *maxopenparen_p)
b93070ed
DM
400 rex->offs[i].start = -1;
401 rex->offs[i].end = -1;
495f47a5
DM
402 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
403 " \\%"UVuf": %s ..-1 undeffing\n",
404 (UV)i,
92da3157 405 (i > *maxopenparen_p) ? "-1" : " "
495f47a5 406 ));
a0d0e21e 407 }
dafc8851 408#endif
a0d0e21e
LW
409}
410
74088413
DM
411/* restore the parens and associated vars at savestack position ix,
412 * but without popping the stack */
413
414STATIC void
92da3157 415S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
74088413
DM
416{
417 I32 tmpix = PL_savestack_ix;
418 PL_savestack_ix = ix;
92da3157 419 regcppop(rex, maxopenparen_p);
74088413
DM
420 PL_savestack_ix = tmpix;
421}
422
02db2b7b 423#define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
a0d0e21e 424
31c7f561
KW
425STATIC bool
426S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
427{
428 /* Returns a boolean as to whether or not 'character' is a member of the
429 * Posix character class given by 'classnum' that should be equivalent to a
430 * value in the typedef '_char_class_number'.
431 *
432 * Ideally this could be replaced by a just an array of function pointers
433 * to the C library functions that implement the macros this calls.
434 * However, to compile, the precise function signatures are required, and
435 * these may vary from platform to to platform. To avoid having to figure
436 * out what those all are on each platform, I (khw) am using this method,
7aee35ff
KW
437 * which adds an extra layer of function call overhead (unless the C
438 * optimizer strips it away). But we don't particularly care about
439 * performance with locales anyway. */
31c7f561
KW
440
441 switch ((_char_class_number) classnum) {
15861f94 442 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
31c7f561 443 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
e8d596e0
KW
444 case _CC_ENUM_ASCII: return isASCII_LC(character);
445 case _CC_ENUM_BLANK: return isBLANK_LC(character);
b0d691b2
KW
446 case _CC_ENUM_CASED: return isLOWER_LC(character)
447 || isUPPER_LC(character);
e8d596e0 448 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
31c7f561
KW
449 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
450 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
451 case _CC_ENUM_LOWER: return isLOWER_LC(character);
452 case _CC_ENUM_PRINT: return isPRINT_LC(character);
e8d596e0 453 case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character);
31c7f561 454 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
e8d596e0 455 case _CC_ENUM_SPACE: return isSPACE_LC(character);
31c7f561
KW
456 case _CC_ENUM_UPPER: return isUPPER_LC(character);
457 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
31c7f561 458 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
31c7f561
KW
459 default: /* VERTSPACE should never occur in locales */
460 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
461 }
462
463 assert(0); /* NOTREACHED */
464 return FALSE;
465}
466
3018b823
KW
467STATIC bool
468S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
469{
470 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
471 * 'character' is a member of the Posix character class given by 'classnum'
472 * that should be equivalent to a value in the typedef
473 * '_char_class_number'.
474 *
475 * This just calls isFOO_lc on the code point for the character if it is in
476 * the range 0-255. Outside that range, all characters avoid Unicode
477 * rules, ignoring any locale. So use the Unicode function if this class
478 * requires a swash, and use the Unicode macro otherwise. */
479
480 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
481
482 if (UTF8_IS_INVARIANT(*character)) {
483 return isFOO_lc(classnum, *character);
484 }
485 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
486 return isFOO_lc(classnum,
487 TWO_BYTE_UTF8_TO_UNI(*character, *(character + 1)));
488 }
489
490 if (classnum < _FIRST_NON_SWASH_CC) {
491
492 /* Initialize the swash unless done already */
493 if (! PL_utf8_swash_ptrs[classnum]) {
494 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
495 PL_utf8_swash_ptrs[classnum] = _core_swash_init("utf8",
496 swash_property_names[classnum], &PL_sv_undef, 1, 0, NULL, &flags);
497 }
498
92a2046b
KW
499 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
500 character,
501 TRUE /* is UTF */ ));
3018b823
KW
502 }
503
504 switch ((_char_class_number) classnum) {
505 case _CC_ENUM_SPACE:
506 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
507
508 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
509 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
510 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
511 default: return 0; /* Things like CNTRL are always
512 below 256 */
513 }
514
515 assert(0); /* NOTREACHED */
516 return FALSE;
517}
518
a687059c 519/*
e50aee73 520 * pregexec and friends
a687059c
LW
521 */
522
76234dfb 523#ifndef PERL_IN_XSUB_RE
a687059c 524/*
c277df42 525 - pregexec - match a regexp against a string
a687059c 526 */
c277df42 527I32
5aaab254 528Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
c3464db5 529 char *strbeg, I32 minend, SV *screamer, U32 nosave)
8fd1a950
DM
530/* stringarg: the point in the string at which to begin matching */
531/* strend: pointer to null at end of string */
532/* strbeg: real beginning of string */
533/* minend: end of match must be >= minend bytes after stringarg. */
534/* screamer: SV being matched: only used for utf8 flag, pos() etc; string
535 * itself is accessed via the pointers above */
536/* nosave: For optimizations. */
c277df42 537{
7918f24d
NC
538 PERL_ARGS_ASSERT_PREGEXEC;
539
c277df42 540 return
9041c2e3 541 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
c277df42
IZ
542 nosave ? 0 : REXEC_COPY_STR);
543}
76234dfb 544#endif
22e551b9 545
9041c2e3 546/*
cad2e5aa
JH
547 * Need to implement the following flags for reg_anch:
548 *
549 * USE_INTUIT_NOML - Useful to call re_intuit_start() first
550 * USE_INTUIT_ML
551 * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
552 * INTUIT_AUTORITATIVE_ML
553 * INTUIT_ONCE_NOML - Intuit can match in one location only.
554 * INTUIT_ONCE_ML
555 *
556 * Another flag for this function: SECOND_TIME (so that float substrs
557 * with giant delta may be not rechecked).
558 */
559
3f7c398e 560/* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
cad2e5aa
JH
561 Otherwise, only SvCUR(sv) is used to get strbeg. */
562
6eb5f6b9
JH
563/* XXXX Some places assume that there is a fixed substring.
564 An update may be needed if optimizer marks as "INTUITable"
565 RExen without fixed substrings. Similarly, it is assumed that
566 lengths of all the strings are no more than minlen, thus they
567 cannot come from lookahead.
40d049e4
YO
568 (Or minlen should take into account lookahead.)
569 NOTE: Some of this comment is not correct. minlen does now take account
570 of lookahead/behind. Further research is required. -- demerphq
571
572*/
6eb5f6b9 573
2c2d71f5
JH
574/* A failure to find a constant substring means that there is no need to make
575 an expensive call to REx engine, thus we celebrate a failure. Similarly,
d8da0584 576 finding a substring too deep into the string means that fewer calls to
30944b6d
IZ
577 regtry() should be needed.
578
579 REx compiler's optimizer found 4 possible hints:
580 a) Anchored substring;
581 b) Fixed substring;
582 c) Whether we are anchored (beginning-of-line or \G);
486ec47a 583 d) First node (of those at offset 0) which may distinguish positions;
6eb5f6b9 584 We use a)b)d) and multiline-part of c), and try to find a position in the
30944b6d
IZ
585 string which does not contradict any of them.
586 */
2c2d71f5 587
6eb5f6b9
JH
588/* Most of decisions we do here should have been done at compile time.
589 The nodes of the REx which we used for the search should have been
590 deleted from the finite automaton. */
591
52a21eb3
DM
592/* args:
593 * rx: the regex to match against
594 * sv: the SV being matched: only used for utf8 flag; the string
595 * itself is accessed via the pointers below. Note that on
596 * something like an overloaded SV, SvPOK(sv) may be false
597 * and the string pointers may point to something unrelated to
598 * the SV itself.
599 * strbeg: real beginning of string
600 * strpos: the point in the string at which to begin matching
601 * strend: pointer to the byte following the last char of the string
602 * flags currently unused; set to 0
603 * data: currently unused; set to NULL
604 */
605
cad2e5aa 606char *
52a21eb3
DM
607Perl_re_intuit_start(pTHX_
608 REGEXP * const rx,
609 SV *sv,
610 const char * const strbeg,
611 char *strpos,
612 char *strend,
613 const U32 flags,
614 re_scream_pos_data *data)
cad2e5aa 615{
97aff369 616 dVAR;
8d919b0a 617 struct regexp *const prog = ReANY(rx);
eb578fdb 618 I32 start_shift = 0;
cad2e5aa 619 /* Should be nonnegative! */
eb578fdb
KW
620 I32 end_shift = 0;
621 char *s;
622 SV *check;
cad2e5aa 623 char *t;
f2ed9b32 624 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
cad2e5aa 625 I32 ml_anch;
eb578fdb 626 char *other_last = NULL; /* other substr checked before this */
bd61b366 627 char *check_at = NULL; /* check substr found at this pos */
d8080198 628 char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
bbe252da 629 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
f8fc2ecf 630 RXi_GET_DECL(prog,progi);
02d5137b
DM
631 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
632 regmatch_info *const reginfo = &reginfo_buf;
30944b6d 633#ifdef DEBUGGING
890ce7af 634 const char * const i_strpos = strpos;
30944b6d 635#endif
a3621e74
YO
636 GET_RE_DEBUG_FLAGS_DECL;
637
7918f24d 638 PERL_ARGS_ASSERT_RE_INTUIT_START;
c33e64f0
FC
639 PERL_UNUSED_ARG(flags);
640 PERL_UNUSED_ARG(data);
7918f24d 641
c344f387
JH
642 /* CHR_DIST() would be more correct here but it makes things slow. */
643 if (prog->minlen > strend - strpos) {
a3621e74 644 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
a72c7584 645 "String too short... [re_intuit_start]\n"));
cad2e5aa 646 goto fail;
2c2d71f5 647 }
d8da0584 648
6c3fea77 649 reginfo->is_utf8_target = cBOOL(utf8_target);
bf2039a9 650 reginfo->info_aux = NULL;
9d9163fb 651 reginfo->strbeg = strbeg;
220db18a 652 reginfo->strend = strend;
aed7b151 653 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
02d5137b 654 reginfo->intuit = 1;
1cb48e53
DM
655 /* not actually used within intuit, but zero for safety anyway */
656 reginfo->poscache_maxiter = 0;
02d5137b 657
f2ed9b32 658 if (utf8_target) {
33b8afdf
JH
659 if (!prog->check_utf8 && prog->check_substr)
660 to_utf8_substr(prog);
661 check = prog->check_utf8;
662 } else {
7e0d5ad7
KW
663 if (!prog->check_substr && prog->check_utf8) {
664 if (! to_byte_substr(prog)) {
6b54ddc5 665 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
7e0d5ad7
KW
666 }
667 }
33b8afdf
JH
668 check = prog->check_substr;
669 }
cf44e600
DM
670 if ((prog->extflags & RXf_ANCH) /* Match at beg-of-str or after \n */
671 && !(prog->extflags & RXf_ANCH_GPOS)) /* \G isn't a BOS or \n */
672 {
673 ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
bbe252da 674 || ( (prog->extflags & RXf_ANCH_BOL)
7fba1cd6 675 && !multiline ) ); /* Check after \n? */
cad2e5aa 676
7e25d62c 677 if (!ml_anch) {
cf44e600 678 if ( !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
7e25d62c 679 && (strpos != strbeg)) {
a3621e74 680 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
7e25d62c
JH
681 goto fail;
682 }
d46b78c6
KW
683 if (prog->check_offset_min == prog->check_offset_max
684 && !(prog->extflags & RXf_CANY_SEEN)
685 && ! multiline) /* /m can cause \n's to match that aren't
686 accounted for in the string max length.
687 See [perl #115242] */
688 {
2c2d71f5 689 /* Substring at constant offset from beg-of-str... */
cad2e5aa
JH
690 I32 slen;
691
1aa99e6b 692 s = HOP3c(strpos, prog->check_offset_min, strend);
1de06328 693
653099ff
GS
694 if (SvTAIL(check)) {
695 slen = SvCUR(check); /* >= 1 */
cad2e5aa 696
9041c2e3 697 if ( strend - s > slen || strend - s < slen - 1
2c2d71f5 698 || (strend - s == slen && strend[-1] != '\n')) {
a3621e74 699 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
2c2d71f5 700 goto fail_finish;
cad2e5aa
JH
701 }
702 /* Now should match s[0..slen-2] */
703 slen--;
3f7c398e 704 if (slen && (*SvPVX_const(check) != *s
cad2e5aa 705 || (slen > 1
3f7c398e 706 && memNE(SvPVX_const(check), s, slen)))) {
2c2d71f5 707 report_neq:
a3621e74 708 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
2c2d71f5
JH
709 goto fail_finish;
710 }
cad2e5aa 711 }
3f7c398e 712 else if (*SvPVX_const(check) != *s
653099ff 713 || ((slen = SvCUR(check)) > 1
3f7c398e 714 && memNE(SvPVX_const(check), s, slen)))
2c2d71f5 715 goto report_neq;
c315bfe8 716 check_at = s;
2c2d71f5 717 goto success_at_start;
7e25d62c 718 }
cad2e5aa 719 }
2c2d71f5 720 /* Match is anchored, but substr is not anchored wrt beg-of-str. */
cad2e5aa 721 s = strpos;
2c2d71f5 722 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
1de06328
YO
723 end_shift = prog->check_end_shift;
724
2c2d71f5 725 if (!ml_anch) {
a3b680e6 726 const I32 end = prog->check_offset_max + CHR_SVLEN(check)
653099ff 727 - (SvTAIL(check) != 0);
a3b680e6 728 const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
2c2d71f5
JH
729
730 if (end_shift < eshift)
731 end_shift = eshift;
732 }
cad2e5aa 733 }
2c2d71f5 734 else { /* Can match at random position */
cad2e5aa
JH
735 ml_anch = 0;
736 s = strpos;
1de06328
YO
737 start_shift = prog->check_offset_min; /* okay to underestimate on CC */
738 end_shift = prog->check_end_shift;
739
740 /* end shift should be non negative here */
cad2e5aa
JH
741 }
742
bcdf7404 743#ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
0033605d 744 if (end_shift < 0)
1de06328 745 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
220fc49f 746 (IV)end_shift, RX_PRECOMP(prog));
2c2d71f5
JH
747#endif
748
2c2d71f5
JH
749 restart:
750 /* Find a possible match in the region s..strend by looking for
751 the "check" substring in the region corrected by start/end_shift. */
1de06328
YO
752
753 {
754 I32 srch_start_shift = start_shift;
755 I32 srch_end_shift = end_shift;
c33e64f0
FC
756 U8* start_point;
757 U8* end_point;
1de06328
YO
758 if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
759 srch_end_shift -= ((strbeg - s) - srch_start_shift);
760 srch_start_shift = strbeg - s;
761 }
6bda09f9 762 DEBUG_OPTIMISE_MORE_r({
1de06328
YO
763 PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
764 (IV)prog->check_offset_min,
765 (IV)srch_start_shift,
766 (IV)srch_end_shift,
767 (IV)prog->check_end_shift);
768 });
769
bbe252da 770 if (prog->extflags & RXf_CANY_SEEN) {
1de06328
YO
771 start_point= (U8*)(s + srch_start_shift);
772 end_point= (U8*)(strend - srch_end_shift);
773 } else {
774 start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
775 end_point= HOP3(strend, -srch_end_shift, strbeg);
776 }
6bda09f9 777 DEBUG_OPTIMISE_MORE_r({
56570a2c 778 PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
1de06328 779 (int)(end_point - start_point),
fc8cd66c 780 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
1de06328
YO
781 start_point);
782 });
783
784 s = fbm_instr( start_point, end_point,
7fba1cd6 785 check, multiline ? FBMrf_MULTILINE : 0);
1de06328 786 }
cad2e5aa
JH
787 /* Update the count-of-usability, remove useless subpatterns,
788 unshift s. */
2c2d71f5 789
ab3bbdeb 790 DEBUG_EXECUTE_r({
f2ed9b32 791 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
ab3bbdeb
YO
792 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
793 PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
2c2d71f5 794 (s ? "Found" : "Did not find"),
f2ed9b32 795 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
ab3bbdeb
YO
796 ? "anchored" : "floating"),
797 quoted,
798 RE_SV_TAIL(check),
799 (s ? " at offset " : "...\n") );
800 });
2c2d71f5
JH
801
802 if (!s)
803 goto fail_finish;
2c2d71f5 804 /* Finish the diagnostic message */
a3621e74 805 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
2c2d71f5 806
1de06328
YO
807 /* XXX dmq: first branch is for positive lookbehind...
808 Our check string is offset from the beginning of the pattern.
809 So we need to do any stclass tests offset forward from that
810 point. I think. :-(
811 */
812
813
814
815 check_at=s;
816
817
2c2d71f5
JH
818 /* Got a candidate. Check MBOL anchoring, and the *other* substr.
819 Start with the other substr.
820 XXXX no SCREAM optimization yet - and a very coarse implementation
a0288114 821 XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
2c2d71f5
JH
822 *always* match. Probably should be marked during compile...
823 Probably it is right to do no SCREAM here...
824 */
825
f2ed9b32 826 if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
1de06328
YO
827 : (prog->float_substr && prog->anchored_substr))
828 {
30944b6d 829 /* Take into account the "other" substring. */
2c2d71f5
JH
830 /* XXXX May be hopelessly wrong for UTF... */
831 if (!other_last)
6eb5f6b9 832 other_last = strpos;
f2ed9b32 833 if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
30944b6d
IZ
834 do_other_anchored:
835 {
890ce7af
AL
836 char * const last = HOP3c(s, -start_shift, strbeg);
837 char *last1, *last2;
be8e71aa 838 char * const saved_s = s;
33b8afdf 839 SV* must;
2c2d71f5 840
2c2d71f5
JH
841 t = s - prog->check_offset_max;
842 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
f2ed9b32 843 && (!utf8_target
0ce71af7 844 || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
2c2d71f5 845 && t > strpos)))
6f207bd3 846 NOOP;
2c2d71f5
JH
847 else
848 t = strpos;
1aa99e6b 849 t = HOP3c(t, prog->anchored_offset, strend);
6eb5f6b9
JH
850 if (t < other_last) /* These positions already checked */
851 t = other_last;
1aa99e6b 852 last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
2c2d71f5
JH
853 if (last < last1)
854 last1 = last;
1de06328
YO
855 /* XXXX It is not documented what units *_offsets are in.
856 We assume bytes, but this is clearly wrong.
857 Meaning this code needs to be carefully reviewed for errors.
858 dmq.
859 */
860
2c2d71f5 861 /* On end-of-str: see comment below. */
f2ed9b32 862 must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
33b8afdf
JH
863 if (must == &PL_sv_undef) {
864 s = (char*)NULL;
1de06328 865 DEBUG_r(must = prog->anchored_utf8); /* for debug */
33b8afdf
JH
866 }
867 else
868 s = fbm_instr(
869 (unsigned char*)t,
870 HOP3(HOP3(last1, prog->anchored_offset, strend)
871 + SvCUR(must), -(SvTAIL(must)!=0), strbeg),
872 must,
7fba1cd6 873 multiline ? FBMrf_MULTILINE : 0
33b8afdf 874 );
ab3bbdeb 875 DEBUG_EXECUTE_r({
f2ed9b32 876 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
ab3bbdeb
YO
877 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
878 PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
2c2d71f5 879 (s ? "Found" : "Contradicts"),
ab3bbdeb
YO
880 quoted, RE_SV_TAIL(must));
881 });
882
883
2c2d71f5
JH
884 if (!s) {
885 if (last1 >= last2) {
a3621e74 886 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2c2d71f5
JH
887 ", giving up...\n"));
888 goto fail_finish;
889 }
a3621e74 890 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2c2d71f5 891 ", trying floating at offset %ld...\n",
be8e71aa 892 (long)(HOP3c(saved_s, 1, strend) - i_strpos)));
1aa99e6b
IH
893 other_last = HOP3c(last1, prog->anchored_offset+1, strend);
894 s = HOP3c(last, 1, strend);
2c2d71f5
JH
895 goto restart;
896 }
897 else {
a3621e74 898 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
30944b6d 899 (long)(s - i_strpos)));
1aa99e6b
IH
900 t = HOP3c(s, -prog->anchored_offset, strbeg);
901 other_last = HOP3c(s, 1, strend);
be8e71aa 902 s = saved_s;
2c2d71f5
JH
903 if (t == strpos)
904 goto try_at_start;
2c2d71f5
JH
905 goto try_at_offset;
906 }
30944b6d 907 }
2c2d71f5
JH
908 }
909 else { /* Take into account the floating substring. */
33b8afdf 910 char *last, *last1;
be8e71aa 911 char * const saved_s = s;
33b8afdf
JH
912 SV* must;
913
914 t = HOP3c(s, -start_shift, strbeg);
915 last1 = last =
916 HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
917 if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
918 last = HOP3c(t, prog->float_max_offset, strend);
919 s = HOP3c(t, prog->float_min_offset, strend);
920 if (s < other_last)
921 s = other_last;
2c2d71f5 922 /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
f2ed9b32 923 must = utf8_target ? prog->float_utf8 : prog->float_substr;
33b8afdf
JH
924 /* fbm_instr() takes into account exact value of end-of-str
925 if the check is SvTAIL(ed). Since false positives are OK,
926 and end-of-str is not later than strend we are OK. */
927 if (must == &PL_sv_undef) {
928 s = (char*)NULL;
1de06328 929 DEBUG_r(must = prog->float_utf8); /* for debug message */
33b8afdf
JH
930 }
931 else
2c2d71f5 932 s = fbm_instr((unsigned char*)s,
33b8afdf
JH
933 (unsigned char*)last + SvCUR(must)
934 - (SvTAIL(must)!=0),
7fba1cd6 935 must, multiline ? FBMrf_MULTILINE : 0);
ab3bbdeb 936 DEBUG_EXECUTE_r({
f2ed9b32 937 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
ab3bbdeb
YO
938 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
939 PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
33b8afdf 940 (s ? "Found" : "Contradicts"),
ab3bbdeb
YO
941 quoted, RE_SV_TAIL(must));
942 });
33b8afdf
JH
943 if (!s) {
944 if (last1 == last) {
a3621e74 945 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
33b8afdf
JH
946 ", giving up...\n"));
947 goto fail_finish;
2c2d71f5 948 }
a3621e74 949 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
33b8afdf 950 ", trying anchored starting at offset %ld...\n",
be8e71aa 951 (long)(saved_s + 1 - i_strpos)));
33b8afdf
JH
952 other_last = last;
953 s = HOP3c(t, 1, strend);
954 goto restart;
955 }
956 else {
a3621e74 957 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
33b8afdf
JH
958 (long)(s - i_strpos)));
959 other_last = s; /* Fix this later. --Hugo */
be8e71aa 960 s = saved_s;
33b8afdf
JH
961 if (t == strpos)
962 goto try_at_start;
963 goto try_at_offset;
964 }
2c2d71f5 965 }
cad2e5aa 966 }
2c2d71f5 967
1de06328 968
9ef43ace 969 t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
1de06328 970
6bda09f9 971 DEBUG_OPTIMISE_MORE_r(
1de06328
YO
972 PerlIO_printf(Perl_debug_log,
973 "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
974 (IV)prog->check_offset_min,
975 (IV)prog->check_offset_max,
976 (IV)(s-strpos),
977 (IV)(t-strpos),
978 (IV)(t-s),
979 (IV)(strend-strpos)
980 )
981 );
982
2c2d71f5 983 if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
f2ed9b32 984 && (!utf8_target
9ef43ace 985 || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
1de06328
YO
986 && t > strpos)))
987 {
2c2d71f5
JH
988 /* Fixed substring is found far enough so that the match
989 cannot start at strpos. */
990 try_at_offset:
cad2e5aa 991 if (ml_anch && t[-1] != '\n') {
30944b6d
IZ
992 /* Eventually fbm_*() should handle this, but often
993 anchored_offset is not 0, so this check will not be wasted. */
994 /* XXXX In the code below we prefer to look for "^" even in
995 presence of anchored substrings. And we search even
996 beyond the found float position. These pessimizations
997 are historical artefacts only. */
998 find_anchor:
2c2d71f5 999 while (t < strend - prog->minlen) {
cad2e5aa 1000 if (*t == '\n') {
4ee3650e 1001 if (t < check_at - prog->check_offset_min) {
f2ed9b32 1002 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
4ee3650e
GS
1003 /* Since we moved from the found position,
1004 we definitely contradict the found anchored
30944b6d
IZ
1005 substr. Due to the above check we do not
1006 contradict "check" substr.
1007 Thus we can arrive here only if check substr
1008 is float. Redo checking for "other"=="fixed".
1009 */
9041c2e3 1010 strpos = t + 1;
a3621e74 1011 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
e4584336 1012 PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
30944b6d
IZ
1013 goto do_other_anchored;
1014 }
4ee3650e
GS
1015 /* We don't contradict the found floating substring. */
1016 /* XXXX Why not check for STCLASS? */
cad2e5aa 1017 s = t + 1;
a3621e74 1018 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
e4584336 1019 PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
cad2e5aa
JH
1020 goto set_useful;
1021 }
4ee3650e
GS
1022 /* Position contradicts check-string */
1023 /* XXXX probably better to look for check-string
1024 than for "\n", so one should lower the limit for t? */
a3621e74 1025 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
e4584336 1026 PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
0e41cd87 1027 other_last = strpos = s = t + 1;
cad2e5aa
JH
1028 goto restart;
1029 }
1030 t++;
1031 }
a3621e74 1032 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
e4584336 1033 PL_colors[0], PL_colors[1]));
2c2d71f5 1034 goto fail_finish;
cad2e5aa 1035 }
f5952150 1036 else {
a3621e74 1037 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
e4584336 1038 PL_colors[0], PL_colors[1]));
f5952150 1039 }
cad2e5aa
JH
1040 s = t;
1041 set_useful:
f2ed9b32 1042 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
cad2e5aa
JH
1043 }
1044 else {
f5952150 1045 /* The found string does not prohibit matching at strpos,
2c2d71f5 1046 - no optimization of calling REx engine can be performed,
f5952150
GS
1047 unless it was an MBOL and we are not after MBOL,
1048 or a future STCLASS check will fail this. */
2c2d71f5
JH
1049 try_at_start:
1050 /* Even in this situation we may use MBOL flag if strpos is offset
1051 wrt the start of the string. */
52a21eb3 1052 if (ml_anch && (strpos != strbeg) && strpos[-1] != '\n'
d506a20d 1053 /* May be due to an implicit anchor of m{.*foo} */
bbe252da 1054 && !(prog->intflags & PREGf_IMPLICIT))
d506a20d 1055 {
cad2e5aa
JH
1056 t = strpos;
1057 goto find_anchor;
1058 }
a3621e74 1059 DEBUG_EXECUTE_r( if (ml_anch)
f5952150 1060 PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
70685ca0 1061 (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
30944b6d 1062 );
2c2d71f5 1063 success_at_start:
bbe252da 1064 if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
f2ed9b32 1065 && (utf8_target ? (
33b8afdf
JH
1066 prog->check_utf8 /* Could be deleted already */
1067 && --BmUSEFUL(prog->check_utf8) < 0
1068 && (prog->check_utf8 == prog->float_utf8)
1069 ) : (
1070 prog->check_substr /* Could be deleted already */
1071 && --BmUSEFUL(prog->check_substr) < 0
1072 && (prog->check_substr == prog->float_substr)
1073 )))
66e933ab 1074 {
cad2e5aa 1075 /* If flags & SOMETHING - do not do it many times on the same match */
a3621e74 1076 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
f2ed9b32
KW
1077 /* XXX Does the destruction order has to change with utf8_target? */
1078 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1079 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
a0714e2c
SS
1080 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1081 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1082 check = NULL; /* abort */
cad2e5aa 1083 s = strpos;
486ec47a 1084 /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
c9415951
YO
1085 see http://bugs.activestate.com/show_bug.cgi?id=87173 */
1086 if (prog->intflags & PREGf_IMPLICIT)
1087 prog->extflags &= ~RXf_ANCH_MBOL;
3cf5c195
IZ
1088 /* XXXX This is a remnant of the old implementation. It
1089 looks wasteful, since now INTUIT can use many
6eb5f6b9 1090 other heuristics. */
bbe252da 1091 prog->extflags &= ~RXf_USE_INTUIT;
c9415951 1092 /* XXXX What other flags might need to be cleared in this branch? */
cad2e5aa
JH
1093 }
1094 else
1095 s = strpos;
1096 }
1097
6eb5f6b9
JH
1098 /* Last resort... */
1099 /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
1de06328
YO
1100 /* trie stclasses are too expensive to use here, we are better off to
1101 leave it to regmatch itself */
f8fc2ecf 1102 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
6eb5f6b9
JH
1103 /* minlen == 0 is possible if regstclass is \b or \B,
1104 and the fixed substr is ''$.
1105 Since minlen is already taken into account, s+1 is before strend;
1106 accidentally, minlen >= 1 guaranties no false positives at s + 1
1107 even for \b or \B. But (minlen? 1 : 0) below assumes that
1108 regstclass does not come from lookahead... */
1109 /* If regstclass takes bytelength more than 1: If charlength==1, OK.
af944926 1110 This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
f8fc2ecf 1111 const U8* const str = (U8*)STRING(progi->regstclass);
2c75e362 1112 /* XXX this value could be pre-computed */
f8fc2ecf 1113 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
2c75e362
DM
1114 ? (reginfo->is_utf8_pat
1115 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1116 : STR_LEN(progi->regstclass))
66e933ab 1117 : 1);
1de06328
YO
1118 char * endpos;
1119 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1120 endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
1121 else if (prog->float_substr || prog->float_utf8)
1122 endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
1123 else
1124 endpos= strend;
1125
d8080198
YO
1126 if (checked_upto < s)
1127 checked_upto = s;
1128 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1129 (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
1130
6eb5f6b9 1131 t = s;
984e6dd1 1132 s = find_byclass(prog, progi->regstclass, checked_upto, endpos,
f9176b44 1133 reginfo);
d8080198
YO
1134 if (s) {
1135 checked_upto = s;
1136 } else {
6eb5f6b9 1137#ifdef DEBUGGING
cbbf8932 1138 const char *what = NULL;
6eb5f6b9
JH
1139#endif
1140 if (endpos == strend) {
a3621e74 1141 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
6eb5f6b9
JH
1142 "Could not match STCLASS...\n") );
1143 goto fail;
1144 }
a3621e74 1145 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
66e933ab 1146 "This position contradicts STCLASS...\n") );
bbe252da 1147 if ((prog->extflags & RXf_ANCH) && !ml_anch)
653099ff 1148 goto fail;
d8080198
YO
1149 checked_upto = HOPBACKc(endpos, start_shift);
1150 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
1151 (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
6eb5f6b9 1152 /* Contradict one of substrings */
33b8afdf 1153 if (prog->anchored_substr || prog->anchored_utf8) {
f2ed9b32 1154 if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
a3621e74 1155 DEBUG_EXECUTE_r( what = "anchored" );
6eb5f6b9 1156 hop_and_restart:
1aa99e6b 1157 s = HOP3c(t, 1, strend);
66e933ab
GS
1158 if (s + start_shift + end_shift > strend) {
1159 /* XXXX Should be taken into account earlier? */
a3621e74 1160 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
66e933ab
GS
1161 "Could not match STCLASS...\n") );
1162 goto fail;
1163 }
5e39e1e5
HS
1164 if (!check)
1165 goto giveup;
a3621e74 1166 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
f5952150 1167 "Looking for %s substr starting at offset %ld...\n",
6eb5f6b9
JH
1168 what, (long)(s + start_shift - i_strpos)) );
1169 goto restart;
1170 }
66e933ab 1171 /* Have both, check_string is floating */
6eb5f6b9
JH
1172 if (t + start_shift >= check_at) /* Contradicts floating=check */
1173 goto retry_floating_check;
1174 /* Recheck anchored substring, but not floating... */
9041c2e3 1175 s = check_at;
5e39e1e5
HS
1176 if (!check)
1177 goto giveup;
a3621e74 1178 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
f5952150 1179 "Looking for anchored substr starting at offset %ld...\n",
6eb5f6b9
JH
1180 (long)(other_last - i_strpos)) );
1181 goto do_other_anchored;
1182 }
60e71179
GS
1183 /* Another way we could have checked stclass at the
1184 current position only: */
1185 if (ml_anch) {
1186 s = t = t + 1;
5e39e1e5
HS
1187 if (!check)
1188 goto giveup;
a3621e74 1189 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
f5952150 1190 "Looking for /%s^%s/m starting at offset %ld...\n",
e4584336 1191 PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
60e71179 1192 goto try_at_offset;
66e933ab 1193 }
f2ed9b32 1194 if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
60e71179 1195 goto fail;
486ec47a 1196 /* Check is floating substring. */
6eb5f6b9
JH
1197 retry_floating_check:
1198 t = check_at - start_shift;
a3621e74 1199 DEBUG_EXECUTE_r( what = "floating" );
6eb5f6b9
JH
1200 goto hop_and_restart;
1201 }
b7953727 1202 if (t != s) {
a3621e74 1203 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6eb5f6b9 1204 "By STCLASS: moving %ld --> %ld\n",
b7953727
JH
1205 (long)(t - i_strpos), (long)(s - i_strpos))
1206 );
1207 }
1208 else {
a3621e74 1209 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
b7953727
JH
1210 "Does not contradict STCLASS...\n");
1211 );
1212 }
6eb5f6b9 1213 }
5e39e1e5 1214 giveup:
a3621e74 1215 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
5e39e1e5
HS
1216 PL_colors[4], (check ? "Guessed" : "Giving up"),
1217 PL_colors[5], (long)(s - i_strpos)) );
cad2e5aa 1218 return s;
2c2d71f5
JH
1219
1220 fail_finish: /* Substring not found */
33b8afdf 1221 if (prog->check_substr || prog->check_utf8) /* could be removed already */
f2ed9b32 1222 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
cad2e5aa 1223 fail:
a3621e74 1224 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
e4584336 1225 PL_colors[4], PL_colors[5]));
bd61b366 1226 return NULL;
cad2e5aa 1227}
9661b544 1228
a0a388a1
YO
1229#define DECL_TRIE_TYPE(scan) \
1230 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
fab2782b
YO
1231 trie_type = ((scan->flags == EXACT) \
1232 ? (utf8_target ? trie_utf8 : trie_plain) \
1233 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1234
fd3249ee
YO
1235#define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1236STMT_START { \
fab2782b
YO
1237 STRLEN skiplen; \
1238 switch (trie_type) { \
1239 case trie_utf8_fold: \
1240 if ( foldlen>0 ) { \
1241 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1242 foldlen -= len; \
1243 uscan += len; \
1244 len=0; \
1245 } else { \
1246 uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
1247 len = UTF8SKIP(uc); \
1248 skiplen = UNISKIP( uvc ); \
1249 foldlen -= skiplen; \
1250 uscan = foldbuf + skiplen; \
1251 } \
1252 break; \
1253 case trie_latin_utf8_fold: \
1254 if ( foldlen>0 ) { \
1255 uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1256 foldlen -= len; \
1257 uscan += len; \
1258 len=0; \
1259 } else { \
1260 len = 1; \
51910141 1261 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, FOLD_FLAGS_FULL); \
fab2782b
YO
1262 skiplen = UNISKIP( uvc ); \
1263 foldlen -= skiplen; \
1264 uscan = foldbuf + skiplen; \
1265 } \
1266 break; \
1267 case trie_utf8: \
1268 uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1269 break; \
1270 case trie_plain: \
1271 uvc = (UV)*uc; \
1272 len = 1; \
1273 } \
1274 if (uvc < 256) { \
1275 charid = trie->charmap[ uvc ]; \
1276 } \
1277 else { \
1278 charid = 0; \
1279 if (widecharmap) { \
1280 SV** const svpp = hv_fetch(widecharmap, \
1281 (char*)&uvc, sizeof(UV), 0); \
1282 if (svpp) \
1283 charid = (U16)SvIV(*svpp); \
1284 } \
1285 } \
4cadc6a9
YO
1286} STMT_END
1287
4cadc6a9
YO
1288#define REXEC_FBC_EXACTISH_SCAN(CoNd) \
1289STMT_START { \
1290 while (s <= e) { \
1291 if ( (CoNd) \
fac1af77 1292 && (ln == 1 || folder(s, pat_string, ln)) \
02d5137b 1293 && (reginfo->intuit || regtry(reginfo, &s)) )\
4cadc6a9
YO
1294 goto got_it; \
1295 s++; \
1296 } \
1297} STMT_END
1298
1299#define REXEC_FBC_UTF8_SCAN(CoDe) \
1300STMT_START { \
9a902117 1301 while (s < strend) { \
4cadc6a9 1302 CoDe \
9a902117 1303 s += UTF8SKIP(s); \
4cadc6a9
YO
1304 } \
1305} STMT_END
1306
1307#define REXEC_FBC_SCAN(CoDe) \
1308STMT_START { \
1309 while (s < strend) { \
1310 CoDe \
1311 s++; \
1312 } \
1313} STMT_END
1314
1315#define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
1316REXEC_FBC_UTF8_SCAN( \
1317 if (CoNd) { \
02d5137b 1318 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
4cadc6a9
YO
1319 goto got_it; \
1320 else \
1321 tmp = doevery; \
1322 } \
1323 else \
1324 tmp = 1; \
1325)
1326
1327#define REXEC_FBC_CLASS_SCAN(CoNd) \
1328REXEC_FBC_SCAN( \
1329 if (CoNd) { \
02d5137b 1330 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
4cadc6a9
YO
1331 goto got_it; \
1332 else \
1333 tmp = doevery; \
1334 } \
1335 else \
1336 tmp = 1; \
1337)
1338
1339#define REXEC_FBC_TRYIT \
02d5137b 1340if ((reginfo->intuit || regtry(reginfo, &s))) \
4cadc6a9
YO
1341 goto got_it
1342
e1d1eefb 1343#define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
f2ed9b32 1344 if (utf8_target) { \
e1d1eefb
YO
1345 REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
1346 } \
1347 else { \
1348 REXEC_FBC_CLASS_SCAN(CoNd); \
d981ef24 1349 }
e1d1eefb 1350
786e8c11 1351#define DUMP_EXEC_POS(li,s,doutf8) \
9d9163fb 1352 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
6d59b646 1353 startpos, doutf8)
786e8c11 1354
cfaf538b
KW
1355
1356#define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
9d9163fb 1357 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
cfaf538b
KW
1358 tmp = TEST_NON_UTF8(tmp); \
1359 REXEC_FBC_UTF8_SCAN( \
1360 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1361 tmp = !tmp; \
1362 IF_SUCCESS; \
1363 } \
1364 else { \
1365 IF_FAIL; \
1366 } \
1367 ); \
1368
1369#define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
9d9163fb 1370 if (s == reginfo->strbeg) { \
cfaf538b
KW
1371 tmp = '\n'; \
1372 } \
1373 else { \
9d9163fb 1374 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
cfaf538b
KW
1375 tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
1376 } \
1377 tmp = TeSt1_UtF8; \
1378 LOAD_UTF8_CHARCLASS_ALNUM(); \
1379 REXEC_FBC_UTF8_SCAN( \
1380 if (tmp == ! (TeSt2_UtF8)) { \
1381 tmp = !tmp; \
1382 IF_SUCCESS; \
1383 } \
1384 else { \
1385 IF_FAIL; \
1386 } \
1387 ); \
1388
63ac0dad
KW
1389/* The only difference between the BOUND and NBOUND cases is that
1390 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1391 * NBOUND. This is accomplished by passing it in either the if or else clause,
1392 * with the other one being empty */
1393#define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1dcf4a1b 1394 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
cfaf538b
KW
1395
1396#define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1dcf4a1b 1397 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
63ac0dad
KW
1398
1399#define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1dcf4a1b 1400 FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
cfaf538b
KW
1401
1402#define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
1dcf4a1b 1403 FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
cfaf538b 1404
63ac0dad
KW
1405
1406/* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
1407 * be passed in completely with the variable name being tested, which isn't
1408 * such a clean interface, but this is easier to read than it was before. We
1409 * are looking for the boundary (or non-boundary between a word and non-word
1410 * character. The utf8 and non-utf8 cases have the same logic, but the details
1411 * must be different. Find the "wordness" of the character just prior to this
1412 * one, and compare it with the wordness of this one. If they differ, we have
1413 * a boundary. At the beginning of the string, pretend that the previous
1414 * character was a new-line */
cfaf538b 1415#define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
63ac0dad 1416 if (utf8_target) { \
cfaf538b 1417 UTF8_CODE \
63ac0dad
KW
1418 } \
1419 else { /* Not utf8 */ \
9d9163fb 1420 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
63ac0dad
KW
1421 tmp = TEST_NON_UTF8(tmp); \
1422 REXEC_FBC_SCAN( \
1423 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1424 tmp = !tmp; \
1425 IF_SUCCESS; \
1426 } \
1427 else { \
1428 IF_FAIL; \
1429 } \
1430 ); \
1431 } \
02d5137b 1432 if ((!prog->minlen && tmp) && (reginfo->intuit || regtry(reginfo, &s))) \
63ac0dad
KW
1433 goto got_it;
1434
786e8c11 1435/* We know what class REx starts with. Try to find this position... */
02d5137b 1436/* if reginfo->intuit, its a dryrun */
786e8c11
YO
1437/* annoyingly all the vars in this routine have different names from their counterparts
1438 in regmatch. /grrr */
1439
3c3eec57 1440STATIC char *
07be1b83 1441S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
f9176b44 1442 const char *strend, regmatch_info *reginfo)
a687059c 1443{
73104a1b
KW
1444 dVAR;
1445 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1446 char *pat_string; /* The pattern's exactish string */
1447 char *pat_end; /* ptr to end char of pat_string */
1448 re_fold_t folder; /* Function for computing non-utf8 folds */
1449 const U8 *fold_array; /* array for folding ords < 256 */
1450 STRLEN ln;
1451 STRLEN lnc;
73104a1b
KW
1452 U8 c1;
1453 U8 c2;
1454 char *e;
1455 I32 tmp = 1; /* Scratch variable? */
ba44c216 1456 const bool utf8_target = reginfo->is_utf8_target;
73104a1b 1457 UV utf8_fold_flags = 0;
f9176b44 1458 const bool is_utf8_pat = reginfo->is_utf8_pat;
3018b823
KW
1459 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1460 with a result inverts that result, as 0^1 =
1461 1 and 1^1 = 0 */
1462 _char_class_number classnum;
1463
73104a1b 1464 RXi_GET_DECL(prog,progi);
2f7f8cb1 1465
73104a1b 1466 PERL_ARGS_ASSERT_FIND_BYCLASS;
2f7f8cb1 1467
73104a1b
KW
1468 /* We know what class it must start with. */
1469 switch (OP(c)) {
1470 case ANYOF:
9aa1e39f 1471 case ANYOF_SYNTHETIC:
954a2af6 1472 case ANYOF_WARN_SUPER:
73104a1b
KW
1473 if (utf8_target) {
1474 REXEC_FBC_UTF8_CLASS_SCAN(
1475 reginclass(prog, c, (U8*)s, utf8_target));
1476 }
1477 else {
1478 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1479 }
1480 break;
1481 case CANY:
1482 REXEC_FBC_SCAN(
02d5137b 1483 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
73104a1b
KW
1484 goto got_it;
1485 else
1486 tmp = doevery;
1487 );
1488 break;
1489
1490 case EXACTFA:
984e6dd1 1491 if (is_utf8_pat || utf8_target) {
73104a1b
KW
1492 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1493 goto do_exactf_utf8;
1494 }
1495 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1496 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1497 goto do_exactf_non_utf8; /* isn't dealt with by these */
77a6d856 1498
73104a1b
KW
1499 case EXACTF:
1500 if (utf8_target) {
16d951b7 1501
73104a1b
KW
1502 /* regcomp.c already folded this if pattern is in UTF-8 */
1503 utf8_fold_flags = 0;
1504 goto do_exactf_utf8;
1505 }
1506 fold_array = PL_fold;
1507 folder = foldEQ;
1508 goto do_exactf_non_utf8;
1509
1510 case EXACTFL:
984e6dd1 1511 if (is_utf8_pat || utf8_target) {
73104a1b
KW
1512 utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
1513 goto do_exactf_utf8;
1514 }
1515 fold_array = PL_fold_locale;
1516 folder = foldEQ_locale;
1517 goto do_exactf_non_utf8;
3c760661 1518
73104a1b 1519 case EXACTFU_SS:
984e6dd1 1520 if (is_utf8_pat) {
73104a1b
KW
1521 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1522 }
1523 goto do_exactf_utf8;
16d951b7 1524
73104a1b
KW
1525 case EXACTFU_TRICKYFOLD:
1526 case EXACTFU:
984e6dd1
DM
1527 if (is_utf8_pat || utf8_target) {
1528 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
73104a1b
KW
1529 goto do_exactf_utf8;
1530 }
fac1af77 1531
73104a1b
KW
1532 /* Any 'ss' in the pattern should have been replaced by regcomp,
1533 * so we don't have to worry here about this single special case
1534 * in the Latin1 range */
1535 fold_array = PL_fold_latin1;
1536 folder = foldEQ_latin1;
1537
1538 /* FALL THROUGH */
1539
1540 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1541 are no glitches with fold-length differences
1542 between the target string and pattern */
1543
1544 /* The idea in the non-utf8 EXACTF* cases is to first find the
1545 * first character of the EXACTF* node and then, if necessary,
1546 * case-insensitively compare the full text of the node. c1 is the
1547 * first character. c2 is its fold. This logic will not work for
1548 * Unicode semantics and the german sharp ss, which hence should
1549 * not be compiled into a node that gets here. */
1550 pat_string = STRING(c);
1551 ln = STR_LEN(c); /* length to match in octets/bytes */
1552
1553 /* We know that we have to match at least 'ln' bytes (which is the
1554 * same as characters, since not utf8). If we have to match 3
1555 * characters, and there are only 2 availabe, we know without
1556 * trying that it will fail; so don't start a match past the
1557 * required minimum number from the far end */
1558 e = HOP3c(strend, -((I32)ln), s);
1559
02d5137b 1560 if (reginfo->intuit && e < s) {
73104a1b
KW
1561 e = s; /* Due to minlen logic of intuit() */
1562 }
fac1af77 1563
73104a1b
KW
1564 c1 = *pat_string;
1565 c2 = fold_array[c1];
1566 if (c1 == c2) { /* If char and fold are the same */
1567 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1568 }
1569 else {
1570 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1571 }
1572 break;
fac1af77 1573
73104a1b
KW
1574 do_exactf_utf8:
1575 {
1576 unsigned expansion;
1577
1578 /* If one of the operands is in utf8, we can't use the simpler folding
1579 * above, due to the fact that many different characters can have the
1580 * same fold, or portion of a fold, or different- length fold */
1581 pat_string = STRING(c);
1582 ln = STR_LEN(c); /* length to match in octets/bytes */
1583 pat_end = pat_string + ln;
984e6dd1 1584 lnc = is_utf8_pat /* length to match in characters */
73104a1b
KW
1585 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1586 : ln;
1587
1588 /* We have 'lnc' characters to match in the pattern, but because of
1589 * multi-character folding, each character in the target can match
1590 * up to 3 characters (Unicode guarantees it will never exceed
1591 * this) if it is utf8-encoded; and up to 2 if not (based on the
1592 * fact that the Latin 1 folds are already determined, and the
1593 * only multi-char fold in that range is the sharp-s folding to
1594 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1595 * string character. Adjust lnc accordingly, rounding up, so that
1596 * if we need to match at least 4+1/3 chars, that really is 5. */
1597 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1598 lnc = (lnc + expansion - 1) / expansion;
1599
1600 /* As in the non-UTF8 case, if we have to match 3 characters, and
1601 * only 2 are left, it's guaranteed to fail, so don't start a
1602 * match that would require us to go beyond the end of the string
1603 */
1604 e = HOP3c(strend, -((I32)lnc), s);
1605
02d5137b 1606 if (reginfo->intuit && e < s) {
73104a1b
KW
1607 e = s; /* Due to minlen logic of intuit() */
1608 }
0658cdde 1609
73104a1b
KW
1610 /* XXX Note that we could recalculate e to stop the loop earlier,
1611 * as the worst case expansion above will rarely be met, and as we
1612 * go along we would usually find that e moves further to the left.
1613 * This would happen only after we reached the point in the loop
1614 * where if there were no expansion we should fail. Unclear if
1615 * worth the expense */
1616
1617 while (s <= e) {
1618 char *my_strend= (char *)strend;
1619 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
984e6dd1 1620 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
02d5137b 1621 && (reginfo->intuit || regtry(reginfo, &s)) )
73104a1b
KW
1622 {
1623 goto got_it;
1624 }
1625 s += (utf8_target) ? UTF8SKIP(s) : 1;
1626 }
1627 break;
1628 }
1629 case BOUNDL:
272d35c9 1630 RXp_MATCH_TAINTED_on(prog);
0eb30aeb
KW
1631 FBC_BOUND(isWORDCHAR_LC,
1632 isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(tmp)),
1633 isWORDCHAR_LC_utf8((U8*)s));
73104a1b
KW
1634 break;
1635 case NBOUNDL:
272d35c9 1636 RXp_MATCH_TAINTED_on(prog);
0eb30aeb
KW
1637 FBC_NBOUND(isWORDCHAR_LC,
1638 isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(tmp)),
1639 isWORDCHAR_LC_utf8((U8*)s));
73104a1b
KW
1640 break;
1641 case BOUND:
1642 FBC_BOUND(isWORDCHAR,
0eb30aeb 1643 isWORDCHAR_uni(tmp),
03940dc2 1644 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
73104a1b
KW
1645 break;
1646 case BOUNDA:
1647 FBC_BOUND_NOLOAD(isWORDCHAR_A,
1648 isWORDCHAR_A(tmp),
1649 isWORDCHAR_A((U8*)s));
1650 break;
1651 case NBOUND:
1652 FBC_NBOUND(isWORDCHAR,
0eb30aeb 1653 isWORDCHAR_uni(tmp),
03940dc2 1654 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
73104a1b
KW
1655 break;
1656 case NBOUNDA:
1657 FBC_NBOUND_NOLOAD(isWORDCHAR_A,
1658 isWORDCHAR_A(tmp),
1659 isWORDCHAR_A((U8*)s));
1660 break;
1661 case BOUNDU:
1662 FBC_BOUND(isWORDCHAR_L1,
0eb30aeb 1663 isWORDCHAR_uni(tmp),
03940dc2 1664 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
73104a1b
KW
1665 break;
1666 case NBOUNDU:
1667 FBC_NBOUND(isWORDCHAR_L1,
0eb30aeb 1668 isWORDCHAR_uni(tmp),
03940dc2 1669 cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target)));
73104a1b 1670 break;
73104a1b
KW
1671 case LNBREAK:
1672 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1673 is_LNBREAK_latin1_safe(s, strend)
1674 );
1675 break;
3018b823
KW
1676
1677 /* The argument to all the POSIX node types is the class number to pass to
1678 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1679
1680 case NPOSIXL:
1681 to_complement = 1;
1682 /* FALLTHROUGH */
1683
1684 case POSIXL:
272d35c9 1685 RXp_MATCH_TAINTED_on(prog);
3018b823
KW
1686 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1687 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
73104a1b 1688 break;
3018b823
KW
1689
1690 case NPOSIXD:
1691 to_complement = 1;
1692 /* FALLTHROUGH */
1693
1694 case POSIXD:
1695 if (utf8_target) {
1696 goto posix_utf8;
1697 }
1698 goto posixa;
1699
1700 case NPOSIXA:
1701 if (utf8_target) {
1702 /* The complement of something that matches only ASCII matches all
1703 * UTF-8 variant code points, plus everything in ASCII that isn't
1704 * in the class */
1705 REXEC_FBC_UTF8_CLASS_SCAN(! UTF8_IS_INVARIANT(*s)
1706 || ! _generic_isCC_A(*s, FLAGS(c)));
1707 break;
1708 }
1709
1710 to_complement = 1;
1711 /* FALLTHROUGH */
1712
73104a1b 1713 case POSIXA:
3018b823 1714 posixa:
73104a1b 1715 /* Don't need to worry about utf8, as it can match only a single
3018b823
KW
1716 * byte invariant character. */
1717 REXEC_FBC_CLASS_SCAN(
1718 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
73104a1b 1719 break;
3018b823
KW
1720
1721 case NPOSIXU:
1722 to_complement = 1;
1723 /* FALLTHROUGH */
1724
1725 case POSIXU:
1726 if (! utf8_target) {
1727 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1728 FLAGS(c))));
1729 }
1730 else {
1731
1732 posix_utf8:
1733 classnum = (_char_class_number) FLAGS(c);
1734 if (classnum < _FIRST_NON_SWASH_CC) {
1735 while (s < strend) {
1736
1737 /* We avoid loading in the swash as long as possible, but
1738 * should we have to, we jump to a separate loop. This
1739 * extra 'if' statement is what keeps this code from being
1740 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1741 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1742 goto found_above_latin1;
1743 }
1744 if ((UTF8_IS_INVARIANT(*s)
1745 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
1746 classnum)))
1747 || (UTF8_IS_DOWNGRADEABLE_START(*s)
1748 && to_complement ^ cBOOL(
1749 _generic_isCC(TWO_BYTE_UTF8_TO_UNI(*s, *(s + 1)),
1750 classnum))))
1751 {
02d5137b 1752 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
3018b823
KW
1753 goto got_it;
1754 else {
1755 tmp = doevery;
1756 }
1757 }
1758 else {
1759 tmp = 1;
1760 }
1761 s += UTF8SKIP(s);
1762 }
1763 }
1764 else switch (classnum) { /* These classes are implemented as
1765 macros */
1766 case _CC_ENUM_SPACE: /* XXX would require separate code if we
1767 revert the change of \v matching this */
1768 /* FALL THROUGH */
1769
1770 case _CC_ENUM_PSXSPC:
1771 REXEC_FBC_UTF8_CLASS_SCAN(
1772 to_complement ^ cBOOL(isSPACE_utf8(s)));
1773 break;
1774
1775 case _CC_ENUM_BLANK:
1776 REXEC_FBC_UTF8_CLASS_SCAN(
1777 to_complement ^ cBOOL(isBLANK_utf8(s)));
1778 break;
1779
1780 case _CC_ENUM_XDIGIT:
1781 REXEC_FBC_UTF8_CLASS_SCAN(
1782 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
1783 break;
1784
1785 case _CC_ENUM_VERTSPACE:
1786 REXEC_FBC_UTF8_CLASS_SCAN(
1787 to_complement ^ cBOOL(isVERTWS_utf8(s)));
1788 break;
1789
1790 case _CC_ENUM_CNTRL:
1791 REXEC_FBC_UTF8_CLASS_SCAN(
1792 to_complement ^ cBOOL(isCNTRL_utf8(s)));
1793 break;
1794
1795 default:
1796 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
1797 assert(0); /* NOTREACHED */
1798 }
1799 }
1800 break;
1801
1802 found_above_latin1: /* Here we have to load a swash to get the result
1803 for the current code point */
1804 if (! PL_utf8_swash_ptrs[classnum]) {
1805 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
1806 PL_utf8_swash_ptrs[classnum] =
1807 _core_swash_init("utf8", swash_property_names[classnum],
1808 &PL_sv_undef, 1, 0, NULL, &flags);
1809 }
1810
1811 /* This is a copy of the loop above for swash classes, though using the
1812 * FBC macro instead of being expanded out. Since we've loaded the
1813 * swash, we don't have to check for that each time through the loop */
1814 REXEC_FBC_UTF8_CLASS_SCAN(
1815 to_complement ^ cBOOL(_generic_utf8(
1816 classnum,
1817 s,
1818 swash_fetch(PL_utf8_swash_ptrs[classnum],
1819 (U8 *) s, TRUE))));
73104a1b
KW
1820 break;
1821
1822 case AHOCORASICKC:
1823 case AHOCORASICK:
1824 {
1825 DECL_TRIE_TYPE(c);
1826 /* what trie are we using right now */
1827 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
1828 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
1829 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
1830
1831 const char *last_start = strend - trie->minlen;
6148ee25 1832#ifdef DEBUGGING
73104a1b 1833 const char *real_start = s;
6148ee25 1834#endif
73104a1b
KW
1835 STRLEN maxlen = trie->maxlen;
1836 SV *sv_points;
1837 U8 **points; /* map of where we were in the input string
1838 when reading a given char. For ASCII this
1839 is unnecessary overhead as the relationship
1840 is always 1:1, but for Unicode, especially
1841 case folded Unicode this is not true. */
1842 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1843 U8 *bitmap=NULL;
1844
1845
1846 GET_RE_DEBUG_FLAGS_DECL;
1847
1848 /* We can't just allocate points here. We need to wrap it in
1849 * an SV so it gets freed properly if there is a croak while
1850 * running the match */
1851 ENTER;
1852 SAVETMPS;
1853 sv_points=newSV(maxlen * sizeof(U8 *));
1854 SvCUR_set(sv_points,
1855 maxlen * sizeof(U8 *));
1856 SvPOK_on(sv_points);
1857 sv_2mortal(sv_points);
1858 points=(U8**)SvPV_nolen(sv_points );
1859 if ( trie_type != trie_utf8_fold
1860 && (trie->bitmap || OP(c)==AHOCORASICKC) )
1861 {
1862 if (trie->bitmap)
1863 bitmap=(U8*)trie->bitmap;
1864 else
1865 bitmap=(U8*)ANYOF_BITMAP(c);
1866 }
1867 /* this is the Aho-Corasick algorithm modified a touch
1868 to include special handling for long "unknown char" sequences.
1869 The basic idea being that we use AC as long as we are dealing
1870 with a possible matching char, when we encounter an unknown char
1871 (and we have not encountered an accepting state) we scan forward
1872 until we find a legal starting char.
1873 AC matching is basically that of trie matching, except that when
1874 we encounter a failing transition, we fall back to the current
1875 states "fail state", and try the current char again, a process
1876 we repeat until we reach the root state, state 1, or a legal
1877 transition. If we fail on the root state then we can either
1878 terminate if we have reached an accepting state previously, or
1879 restart the entire process from the beginning if we have not.
1880
1881 */
1882 while (s <= last_start) {
1883 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1884 U8 *uc = (U8*)s;
1885 U16 charid = 0;
1886 U32 base = 1;
1887 U32 state = 1;
1888 UV uvc = 0;
1889 STRLEN len = 0;
1890 STRLEN foldlen = 0;
1891 U8 *uscan = (U8*)NULL;
1892 U8 *leftmost = NULL;
1893#ifdef DEBUGGING
1894 U32 accepted_word= 0;
786e8c11 1895#endif
73104a1b
KW
1896 U32 pointpos = 0;
1897
1898 while ( state && uc <= (U8*)strend ) {
1899 int failed=0;
1900 U32 word = aho->states[ state ].wordnum;
1901
1902 if( state==1 ) {
1903 if ( bitmap ) {
1904 DEBUG_TRIE_EXECUTE_r(
1905 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1906 dump_exec_pos( (char *)uc, c, strend, real_start,
1907 (char *)uc, utf8_target );
1908 PerlIO_printf( Perl_debug_log,
1909 " Scanning for legal start char...\n");
1910 }
1911 );
1912 if (utf8_target) {
1913 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1914 uc += UTF8SKIP(uc);
1915 }
1916 } else {
1917 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
1918 uc++;
1919 }
786e8c11 1920 }
73104a1b 1921 s= (char *)uc;
07be1b83 1922 }
73104a1b
KW
1923 if (uc >(U8*)last_start) break;
1924 }
1925
1926 if ( word ) {
1927 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
1928 if (!leftmost || lpos < leftmost) {
1929 DEBUG_r(accepted_word=word);
1930 leftmost= lpos;
7016d6eb 1931 }
73104a1b 1932 if (base==0) break;
7016d6eb 1933
73104a1b
KW
1934 }
1935 points[pointpos++ % maxlen]= uc;
1936 if (foldlen || uc < (U8*)strend) {
1937 REXEC_TRIE_READ_CHAR(trie_type, trie,
1938 widecharmap, uc,
1939 uscan, len, uvc, charid, foldlen,
1940 foldbuf, uniflags);
1941 DEBUG_TRIE_EXECUTE_r({
1942 dump_exec_pos( (char *)uc, c, strend,
1943 real_start, s, utf8_target);
1944 PerlIO_printf(Perl_debug_log,
1945 " Charid:%3u CP:%4"UVxf" ",
1946 charid, uvc);
1947 });
1948 }
1949 else {
1950 len = 0;
1951 charid = 0;
1952 }
07be1b83 1953
73104a1b
KW
1954
1955 do {
6148ee25 1956#ifdef DEBUGGING
73104a1b 1957 word = aho->states[ state ].wordnum;
6148ee25 1958#endif
73104a1b
KW
1959 base = aho->states[ state ].trans.base;
1960
1961 DEBUG_TRIE_EXECUTE_r({
1962 if (failed)
1963 dump_exec_pos( (char *)uc, c, strend, real_start,
1964 s, utf8_target );
1965 PerlIO_printf( Perl_debug_log,
1966 "%sState: %4"UVxf", word=%"UVxf,
1967 failed ? " Fail transition to " : "",
1968 (UV)state, (UV)word);
1969 });
1970 if ( base ) {
1971 U32 tmp;
1972 I32 offset;
1973 if (charid &&
1974 ( ((offset = base + charid
1975 - 1 - trie->uniquecharcount)) >= 0)
1976 && ((U32)offset < trie->lasttrans)
1977 && trie->trans[offset].check == state
1978 && (tmp=trie->trans[offset].next))
1979 {
1980 DEBUG_TRIE_EXECUTE_r(
1981 PerlIO_printf( Perl_debug_log," - legal\n"));
1982 state = tmp;
1983 break;
07be1b83
YO
1984 }
1985 else {
786e8c11 1986 DEBUG_TRIE_EXECUTE_r(
73104a1b 1987 PerlIO_printf( Perl_debug_log," - fail\n"));
786e8c11 1988 failed = 1;
73104a1b 1989 state = aho->fail[state];
07be1b83 1990 }
07be1b83 1991 }
73104a1b
KW
1992 else {
1993 /* we must be accepting here */
1994 DEBUG_TRIE_EXECUTE_r(
1995 PerlIO_printf( Perl_debug_log," - accepting\n"));
1996 failed = 1;
1997 break;
786e8c11 1998 }
73104a1b
KW
1999 } while(state);
2000 uc += len;
2001 if (failed) {
2002 if (leftmost)
2003 break;
2004 if (!state) state = 1;
07be1b83 2005 }
73104a1b
KW
2006 }
2007 if ( aho->states[ state ].wordnum ) {
2008 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2009 if (!leftmost || lpos < leftmost) {
2010 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2011 leftmost = lpos;
07be1b83
YO
2012 }
2013 }
73104a1b
KW
2014 if (leftmost) {
2015 s = (char*)leftmost;
2016 DEBUG_TRIE_EXECUTE_r({
2017 PerlIO_printf(
2018 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2019 (UV)accepted_word, (IV)(s - real_start)
2020 );
2021 });
02d5137b 2022 if (reginfo->intuit || regtry(reginfo, &s)) {
73104a1b
KW
2023 FREETMPS;
2024 LEAVE;
2025 goto got_it;
2026 }
2027 s = HOPc(s,1);
2028 DEBUG_TRIE_EXECUTE_r({
2029 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2030 });
2031 } else {
2032 DEBUG_TRIE_EXECUTE_r(
2033 PerlIO_printf( Perl_debug_log,"No match.\n"));
2034 break;
2035 }
2036 }
2037 FREETMPS;
2038 LEAVE;
2039 }
2040 break;
2041 default:
2042 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2043 break;
2044 }
2045 return 0;
2046 got_it:
2047 return s;
6eb5f6b9
JH
2048}
2049
60165aa4
DM
2050/* set RX_SAVED_COPY, RX_SUBBEG etc.
2051 * flags have same meanings as with regexec_flags() */
2052
749f4950
DM
2053static void
2054S_reg_set_capture_string(pTHX_ REGEXP * const rx,
60165aa4
DM
2055 char *strbeg,
2056 char *strend,
2057 SV *sv,
2058 U32 flags,
2059 bool utf8_target)
2060{
2061 struct regexp *const prog = ReANY(rx);
2062
60165aa4
DM
2063 if (flags & REXEC_COPY_STR) {
2064#ifdef PERL_ANY_COW
2065 if (SvCANCOW(sv)) {
2066 if (DEBUG_C_TEST) {
2067 PerlIO_printf(Perl_debug_log,
2068 "Copy on write: regexp capture, type %d\n",
2069 (int) SvTYPE(sv));
2070 }
a76b0e90
DM
2071 /* skip creating new COW SV if a valid one already exists */
2072 if (! ( prog->saved_copy
2073 && SvIsCOW(sv)
2074 && SvPOKp(sv)
2075 && SvIsCOW(prog->saved_copy)
2076 && SvPOKp(prog->saved_copy)
2077 && SvPVX(sv) == SvPVX(prog->saved_copy)))
2078 {
2079 RX_MATCH_COPY_FREE(rx);
2080 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2081 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2082 assert (SvPOKp(prog->saved_copy));
2083 }
60165aa4
DM
2084 prog->sublen = strend - strbeg;
2085 prog->suboffset = 0;
2086 prog->subcoffset = 0;
2087 } else
2088#endif
2089 {
2090 I32 min = 0;
2091 I32 max = strend - strbeg;
2092 I32 sublen;
2093
2094 if ( (flags & REXEC_COPY_SKIP_POST)
e322109a 2095 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM
2096 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2097 ) { /* don't copy $' part of string */
2098 U32 n = 0;
2099 max = -1;
2100 /* calculate the right-most part of the string covered
2101 * by a capture. Due to look-ahead, this may be to
2102 * the right of $&, so we have to scan all captures */
2103 while (n <= prog->lastparen) {
2104 if (prog->offs[n].end > max)
2105 max = prog->offs[n].end;
2106 n++;
2107 }
2108 if (max == -1)
2109 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2110 ? prog->offs[0].start
2111 : 0;
2112 assert(max >= 0 && max <= strend - strbeg);
2113 }
2114
2115 if ( (flags & REXEC_COPY_SKIP_PRE)
e322109a 2116 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
60165aa4
DM
2117 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2118 ) { /* don't copy $` part of string */
2119 U32 n = 0;
2120 min = max;
2121 /* calculate the left-most part of the string covered
2122 * by a capture. Due to look-behind, this may be to
2123 * the left of $&, so we have to scan all captures */
2124 while (min && n <= prog->lastparen) {
2125 if ( prog->offs[n].start != -1
2126 && prog->offs[n].start < min)
2127 {
2128 min = prog->offs[n].start;
2129 }
2130 n++;
2131 }
2132 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2133 && min > prog->offs[0].end
2134 )
2135 min = prog->offs[0].end;
2136
2137 }
2138
2139 assert(min >= 0 && min <= max && min <= strend - strbeg);
2140 sublen = max - min;
2141
2142 if (RX_MATCH_COPIED(rx)) {
2143 if (sublen > prog->sublen)
2144 prog->subbeg =
2145 (char*)saferealloc(prog->subbeg, sublen+1);
2146 }
2147 else
2148 prog->subbeg = (char*)safemalloc(sublen+1);
2149 Copy(strbeg + min, prog->subbeg, sublen, char);
2150 prog->subbeg[sublen] = '\0';
2151 prog->suboffset = min;
2152 prog->sublen = sublen;
2153 RX_MATCH_COPIED_on(rx);
2154 }
2155 prog->subcoffset = prog->suboffset;
2156 if (prog->suboffset && utf8_target) {
2157 /* Convert byte offset to chars.
2158 * XXX ideally should only compute this if @-/@+
2159 * has been seen, a la PL_sawampersand ??? */
2160
2161 /* If there's a direct correspondence between the
2162 * string which we're matching and the original SV,
2163 * then we can use the utf8 len cache associated with
2164 * the SV. In particular, it means that under //g,
2165 * sv_pos_b2u() will use the previously cached
2166 * position to speed up working out the new length of
2167 * subcoffset, rather than counting from the start of
2168 * the string each time. This stops
2169 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2170 * from going quadratic */
2171 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2172 sv_pos_b2u(sv, &(prog->subcoffset));
2173 else
2174 prog->subcoffset = utf8_length((U8*)strbeg,
2175 (U8*)(strbeg+prog->suboffset));
2176 }
2177 }
2178 else {
2179 RX_MATCH_COPY_FREE(rx);
2180 prog->subbeg = strbeg;
2181 prog->suboffset = 0;
2182 prog->subcoffset = 0;
2183 prog->sublen = strend - strbeg;
2184 }
2185}
2186
2187
2188
fae667d5 2189
6eb5f6b9
JH
2190/*
2191 - regexec_flags - match a regexp against a string
2192 */
2193I32
5aaab254 2194Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
6eb5f6b9 2195 char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
8fd1a950
DM
2196/* stringarg: the point in the string at which to begin matching */
2197/* strend: pointer to null at end of string */
2198/* strbeg: real beginning of string */
2199/* minend: end of match must be >= minend bytes after stringarg. */
2200/* sv: SV being matched: only used for utf8 flag, pos() etc; string
2201 * itself is accessed via the pointers above */
2202/* data: May be used for some additional optimizations.
d058ec57 2203 Currently unused. */
8fd1a950
DM
2204/* nosave: For optimizations. */
2205
6eb5f6b9 2206{
97aff369 2207 dVAR;
8d919b0a 2208 struct regexp *const prog = ReANY(rx);
5aaab254 2209 char *s;
eb578fdb 2210 regnode *c;
03c83e26 2211 char *startpos;
6eb5f6b9
JH
2212 I32 minlen; /* must match at least this many chars */
2213 I32 dontbother = 0; /* how many characters not to try at end */
6eb5f6b9 2214 I32 end_shift = 0; /* Same for the end. */ /* CC */
f2ed9b32 2215 const bool utf8_target = cBOOL(DO_UTF8(sv));
2757e526 2216 I32 multiline;
f8fc2ecf 2217 RXi_GET_DECL(prog,progi);
02d5137b
DM
2218 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2219 regmatch_info *const reginfo = &reginfo_buf;
e9105d30 2220 regexp_paren_pair *swap = NULL;
006f26b2 2221 I32 oldsave;
a3621e74
YO
2222 GET_RE_DEBUG_FLAGS_DECL;
2223
7918f24d 2224 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
9d4ba2ae 2225 PERL_UNUSED_ARG(data);
6eb5f6b9
JH
2226
2227 /* Be paranoid... */
03c83e26 2228 if (prog == NULL || stringarg == NULL) {
6eb5f6b9
JH
2229 Perl_croak(aTHX_ "NULL regexp parameter");
2230 return 0;
2231 }
2232
6c3fea77 2233 DEBUG_EXECUTE_r(
03c83e26 2234 debug_start_match(rx, utf8_target, stringarg, strend,
6c3fea77
DM
2235 "Matching");
2236 );
8adc0f72 2237
b342a604
DM
2238 startpos = stringarg;
2239
03c83e26 2240 if (prog->extflags & RXf_GPOS_SEEN) {
d307c076
DM
2241 MAGIC *mg;
2242
03c83e26
DM
2243 /* in the presence of \G, we may need to start looking earlier in
2244 * the string than the suggested start point of stringarg:
2245 * if gofs->prog is set, then that's a known, fixed minimum
2246 * offset, such as
2247 * /..\G/: gofs = 2
2248 * /ab|c\G/: gofs = 1
2249 * or if the minimum offset isn't known, then we have to go back
2250 * to the start of the string, e.g. /w+\G/
2251 */
2252 if (prog->gofs) {
b342a604
DM
2253 if (startpos - prog->gofs < strbeg)
2254 startpos = strbeg;
2255 else
2256 startpos -= prog->gofs;
03c83e26 2257 }
b342a604
DM
2258 else if (prog->extflags & RXf_GPOS_FLOAT)
2259 startpos = strbeg;
d307c076
DM
2260
2261 /* set reginfo->ganch, the position where \G can match */
2262
5d96616f
DM
2263 reginfo->ganch =
2264 (flags & REXEC_IGNOREPOS)
2265 ? stringarg /* use start pos rather than pos() */
2266 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
2267 ? strbeg + mg->mg_len /* Defined pos() */
2268 : strbeg; /* pos() not defined; use start of string */
2269
2270 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2271 "GPOS ganch set to strbeg[%"IVdf"]\n", reginfo->ganch - strbeg));
03c83e26
DM
2272 }
2273
2274 minlen = prog->minlen;
b342a604 2275 if ((startpos + minlen) > strend || startpos < strbeg) {
03c83e26
DM
2276 DEBUG_r(PerlIO_printf(Perl_debug_log,
2277 "Regex match can't succeed, so not even tried\n"));
2278 return 0;
2279 }
2280
dfa77d06
DM
2281 s = startpos;
2282
e322109a 2283 if ((prog->extflags & RXf_USE_INTUIT)
7fadf4a7
DM
2284 && !(flags & REXEC_CHECKED))
2285 {
dfa77d06 2286 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
7fadf4a7 2287 flags, NULL);
dfa77d06 2288 if (!s)
7fadf4a7
DM
2289 return 0;
2290
e322109a 2291 if (prog->extflags & RXf_CHECK_ALL) {
7fadf4a7
DM
2292 /* we can match based purely on the result of INTUIT.
2293 * Set up captures etc just for $& and $-[0]
2294 * (an intuit-only match wont have $1,$2,..) */
2295 assert(!prog->nparens);
d5e7783a
DM
2296
2297 /* s/// doesn't like it if $& is earlier than where we asked it to
2298 * start searching (which can happen on something like /.\G/) */
2299 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2300 && (s < stringarg))
2301 {
2302 /* this should only be possible under \G */
2303 assert(prog->extflags & RXf_GPOS_SEEN);
2304 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2305 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2306 goto phooey;
2307 }
2308
7fadf4a7
DM
2309 /* match via INTUIT shouldn't have any captures.
2310 * Let @-, @+, $^N know */
2311 prog->lastparen = prog->lastcloseparen = 0;
2312 RX_MATCH_UTF8_set(rx, utf8_target);
2313 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 2314 S_reg_set_capture_string(aTHX_ rx,
7fadf4a7
DM
2315 strbeg, strend,
2316 sv, flags, utf8_target);
2317
dfa77d06 2318 prog->offs[0].start = s - strbeg;
7fadf4a7 2319 prog->offs[0].end = utf8_target
dfa77d06
DM
2320 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2321 : s - strbeg + prog->minlenret;
7fadf4a7
DM
2322 return 1;
2323 }
2324 }
2325
2757e526 2326
331b2dcc
DM
2327 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2328 * which will call destuctors to reset PL_regmatch_state, free higher
2329 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2330 * regmatch_info_aux_eval */
006f26b2 2331
006f26b2 2332 oldsave = PL_savestack_ix;
006f26b2 2333
6c3fea77 2334 multiline = prog->extflags & RXf_PMf_MULTILINE;
1de06328 2335
dfa77d06 2336 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
a3621e74 2337 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
a72c7584
JH
2338 "String too short [regexec_flags]...\n"));
2339 goto phooey;
1aa99e6b 2340 }
1de06328 2341
6eb5f6b9 2342 /* Check validity of program. */
f8fc2ecf 2343 if (UCHARAT(progi->program) != REG_MAGIC) {
6eb5f6b9
JH
2344 Perl_croak(aTHX_ "corrupted regexp program");
2345 }
2346
272d35c9 2347 RX_MATCH_TAINTED_off(rx);
6eb5f6b9 2348
6c3fea77
DM
2349 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2350 reginfo->intuit = 0;
2351 reginfo->is_utf8_target = cBOOL(utf8_target);
02d5137b
DM
2352 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2353 reginfo->warned = FALSE;
9d9163fb 2354 reginfo->strbeg = strbeg;
02d5137b 2355 reginfo->sv = sv;
1cb48e53 2356 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
220db18a 2357 reginfo->strend = strend;
6eb5f6b9 2358 /* see how far we have to get to not match where we matched before */
fe3974be 2359 reginfo->till = stringarg + minend;
6eb5f6b9 2360
82c23608
FC
2361 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv) && !IS_PADGV(sv)) {
2362 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2363 S_cleanup_regmatch_info_aux has executed (registered by
2364 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2365 magic belonging to this SV.
2366 Not newSVsv, either, as it does not COW.
2367 */
2368 reginfo->sv = newSV(0);
2369 sv_setsv(reginfo->sv, sv);
2370 SAVEFREESV(reginfo->sv);
2371 }
2372
331b2dcc
DM
2373 /* reserve next 2 or 3 slots in PL_regmatch_state:
2374 * slot N+0: may currently be in use: skip it
2375 * slot N+1: use for regmatch_info_aux struct
2376 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2377 * slot N+3: ready for use by regmatch()
2378 */
bf2039a9 2379
331b2dcc
DM
2380 {
2381 regmatch_state *old_regmatch_state;
2382 regmatch_slab *old_regmatch_slab;
2383 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2384
2385 /* on first ever match, allocate first slab */
2386 if (!PL_regmatch_slab) {
2387 Newx(PL_regmatch_slab, 1, regmatch_slab);
2388 PL_regmatch_slab->prev = NULL;
2389 PL_regmatch_slab->next = NULL;
2390 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2391 }
bf2039a9 2392
331b2dcc
DM
2393 old_regmatch_state = PL_regmatch_state;
2394 old_regmatch_slab = PL_regmatch_slab;
bf2039a9 2395
331b2dcc
DM
2396 for (i=0; i <= max; i++) {
2397 if (i == 1)
2398 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2399 else if (i ==2)
2400 reginfo->info_aux_eval =
2401 reginfo->info_aux->info_aux_eval =
2402 &(PL_regmatch_state->u.info_aux_eval);
bf2039a9 2403
331b2dcc
DM
2404 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2405 PL_regmatch_state = S_push_slab(aTHX);
2406 }
bf2039a9 2407
331b2dcc
DM
2408 /* note initial PL_regmatch_state position; at end of match we'll
2409 * pop back to there and free any higher slabs */
bf2039a9 2410
331b2dcc
DM
2411 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2412 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2ac8ff4b 2413 reginfo->info_aux->poscache = NULL;
bf2039a9 2414
331b2dcc 2415 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
bf2039a9 2416
331b2dcc
DM
2417 if ((prog->extflags & RXf_EVAL_SEEN))
2418 S_setup_eval_state(aTHX_ reginfo);
2419 else
2420 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
bf2039a9 2421 }
d3aa529c 2422
6eb5f6b9 2423 /* If there is a "must appear" string, look for it. */
6eb5f6b9 2424
288b8c02 2425 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
e9105d30
GG
2426 /* We have to be careful. If the previous successful match
2427 was from this regex we don't want a subsequent partially
2428 successful match to clobber the old results.
2429 So when we detect this possibility we add a swap buffer
d8da0584
KW
2430 to the re, and switch the buffer each match. If we fail,
2431 we switch it back; otherwise we leave it swapped.
e9105d30
GG
2432 */
2433 swap = prog->offs;
2434 /* do we need a save destructor here for eval dies? */
2435 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
495f47a5
DM
2436 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2437 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2438 PTR2UV(prog),
2439 PTR2UV(swap),
2440 PTR2UV(prog->offs)
2441 ));
c74340f9 2442 }
6eb5f6b9
JH
2443
2444 /* Simplest case: anchored match need be tried only once. */
2445 /* [unless only anchor is BOL and multiline is set] */
bbe252da 2446 if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
3542935d 2447 if (s == startpos && regtry(reginfo, &s))
6eb5f6b9 2448 goto got_it;
bbe252da
YO
2449 else if (multiline || (prog->intflags & PREGf_IMPLICIT)
2450 || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
6eb5f6b9
JH
2451 {
2452 char *end;
2453
2454 if (minlen)
2455 dontbother = minlen - 1;
1aa99e6b 2456 end = HOP3c(strend, -dontbother, strbeg) - 1;
6eb5f6b9 2457 /* for multiline we only have to try after newlines */
33b8afdf 2458 if (prog->check_substr || prog->check_utf8) {
92f3d482
YO
2459 /* because of the goto we can not easily reuse the macros for bifurcating the
2460 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2461 if (utf8_target) {
2462 if (s == startpos)
2463 goto after_try_utf8;
2464 while (1) {
02d5137b 2465 if (regtry(reginfo, &s)) {
92f3d482
YO
2466 goto got_it;
2467 }
2468 after_try_utf8:
2469 if (s > end) {
2470 goto phooey;
2471 }
2472 if (prog->extflags & RXf_USE_INTUIT) {
52a21eb3
DM
2473 s = re_intuit_start(rx, sv, strbeg,
2474 s + UTF8SKIP(s), strend, flags, NULL);
92f3d482
YO
2475 if (!s) {
2476 goto phooey;
2477 }
2478 }
2479 else {
2480 s += UTF8SKIP(s);
2481 }
2482 }
2483 } /* end search for check string in unicode */
2484 else {
2485 if (s == startpos) {
2486 goto after_try_latin;
2487 }
2488 while (1) {
02d5137b 2489 if (regtry(reginfo, &s)) {
92f3d482
YO
2490 goto got_it;
2491 }
2492 after_try_latin:
2493 if (s > end) {
2494 goto phooey;
2495 }
2496 if (prog->extflags & RXf_USE_INTUIT) {
52a21eb3
DM
2497 s = re_intuit_start(rx, sv, strbeg,
2498 s + 1, strend, flags, NULL);
92f3d482
YO
2499 if (!s) {
2500 goto phooey;
2501 }
2502 }
2503 else {
2504 s++;
2505 }
2506 }
2507 } /* end search for check string in latin*/
2508 } /* end search for check string */
2509 else { /* search for newline */
2510 if (s > startpos) {
2511 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
6eb5f6b9 2512 s--;
92f3d482 2513 }
21eede78
YO
2514 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2515 while (s <= end) { /* note it could be possible to match at the end of the string */
6eb5f6b9 2516 if (*s++ == '\n') { /* don't need PL_utf8skip here */
02d5137b 2517 if (regtry(reginfo, &s))
6eb5f6b9
JH
2518 goto got_it;
2519 }
92f3d482
YO
2520 }
2521 } /* end search for newline */
2522 } /* end anchored/multiline check string search */
6eb5f6b9 2523 goto phooey;
bbe252da 2524 } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
f9f4320a 2525 {
02d5137b 2526 /* the warning about reginfo->ganch being used without initialization
bbe252da 2527 is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
f9f4320a 2528 and we only enter this block when the same bit is set. */
02d5137b 2529 char *tmp_s = reginfo->ganch - prog->gofs;
c584a96e 2530
2f14ddcd 2531 if (s <= tmp_s && regtry(reginfo, &tmp_s))
6eb5f6b9
JH
2532 goto got_it;
2533 goto phooey;
2534 }
2535
2536 /* Messy cases: unanchored match. */
bbe252da 2537 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
6eb5f6b9 2538 /* we have /x+whatever/ */
984e6dd1 2539 /* it must be a one character string (XXXX Except is_utf8_pat?) */
33b8afdf 2540 char ch;
bf93d4cc
GS
2541#ifdef DEBUGGING
2542 int did_match = 0;
2543#endif
f2ed9b32 2544 if (utf8_target) {
7e0d5ad7
KW
2545 if (! prog->anchored_utf8) {
2546 to_utf8_substr(prog);
2547 }
2548 ch = SvPVX_const(prog->anchored_utf8)[0];
4cadc6a9 2549 REXEC_FBC_SCAN(
6eb5f6b9 2550 if (*s == ch) {
a3621e74 2551 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 2552 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH
2553 s += UTF8SKIP(s);
2554 while (s < strend && *s == ch)
2555 s += UTF8SKIP(s);
2556 }
4cadc6a9 2557 );
7e0d5ad7 2558
6eb5f6b9
JH
2559 }
2560 else {
7e0d5ad7
KW
2561 if (! prog->anchored_substr) {
2562 if (! to_byte_substr(prog)) {
6b54ddc5 2563 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2564 }
2565 }
2566 ch = SvPVX_const(prog->anchored_substr)[0];
4cadc6a9 2567 REXEC_FBC_SCAN(
6eb5f6b9 2568 if (*s == ch) {
a3621e74 2569 DEBUG_EXECUTE_r( did_match = 1 );
02d5137b 2570 if (regtry(reginfo, &s)) goto got_it;
6eb5f6b9
JH
2571 s++;
2572 while (s < strend && *s == ch)
2573 s++;
2574 }
4cadc6a9 2575 );
6eb5f6b9 2576 }
a3621e74 2577 DEBUG_EXECUTE_r(if (!did_match)
bf93d4cc 2578 PerlIO_printf(Perl_debug_log,
b7953727
JH
2579 "Did not find anchored character...\n")
2580 );
6eb5f6b9 2581 }
a0714e2c
SS
2582 else if (prog->anchored_substr != NULL
2583 || prog->anchored_utf8 != NULL
2584 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
33b8afdf
JH
2585 && prog->float_max_offset < strend - s)) {
2586 SV *must;
2587 I32 back_max;
2588 I32 back_min;
2589 char *last;
6eb5f6b9 2590 char *last1; /* Last position checked before */
bf93d4cc
GS
2591#ifdef DEBUGGING
2592 int did_match = 0;
2593#endif
33b8afdf 2594 if (prog->anchored_substr || prog->anchored_utf8) {
7e0d5ad7
KW
2595 if (utf8_target) {
2596 if (! prog->anchored_utf8) {
2597 to_utf8_substr(prog);
2598 }
2599 must = prog->anchored_utf8;
2600 }
2601 else {
2602 if (! prog->anchored_substr) {
2603 if (! to_byte_substr(prog)) {
6b54ddc5 2604 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2605 }
2606 }
2607 must = prog->anchored_substr;
2608 }
33b8afdf
JH
2609 back_max = back_min = prog->anchored_offset;
2610 } else {
7e0d5ad7
KW
2611 if (utf8_target) {
2612 if (! prog->float_utf8) {
2613 to_utf8_substr(prog);
2614 }
2615 must = prog->float_utf8;
2616 }
2617 else {
2618 if (! prog->float_substr) {
2619 if (! to_byte_substr(prog)) {
6b54ddc5 2620 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2621 }
2622 }
2623 must = prog->float_substr;
2624 }
33b8afdf
JH
2625 back_max = prog->float_max_offset;
2626 back_min = prog->float_min_offset;
2627 }
1de06328 2628
1de06328
YO
2629 if (back_min<0) {
2630 last = strend;
2631 } else {
2632 last = HOP3c(strend, /* Cannot start after this */
2633 -(I32)(CHR_SVLEN(must)
2634 - (SvTAIL(must) != 0) + back_min), strbeg);
2635 }
9d9163fb 2636 if (s > reginfo->strbeg)
6eb5f6b9
JH
2637 last1 = HOPc(s, -1);
2638 else
2639 last1 = s - 1; /* bogus */
2640
a0288114 2641 /* XXXX check_substr already used to find "s", can optimize if
6eb5f6b9 2642 check_substr==must. */
6eb5f6b9
JH
2643 dontbother = end_shift;
2644 strend = HOPc(strend, -dontbother);
2645 while ( (s <= last) &&
c33e64f0 2646 (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
9041c2e3 2647 (unsigned char*)strend, must,
c33e64f0 2648 multiline ? FBMrf_MULTILINE : 0)) ) {
a3621e74 2649 DEBUG_EXECUTE_r( did_match = 1 );
6eb5f6b9
JH
2650 if (HOPc(s, -back_max) > last1) {
2651 last1 = HOPc(s, -back_min);
2652 s = HOPc(s, -back_max);
2653 }
2654 else {
9d9163fb
DM
2655 char * const t = (last1 >= reginfo->strbeg)
2656 ? HOPc(last1, 1) : last1 + 1;
6eb5f6b9
JH
2657
2658 last1 = HOPc(s, -back_min);
52657f30 2659 s = t;
6eb5f6b9 2660 }
f2ed9b32 2661 if (utf8_target) {
6eb5f6b9 2662 while (s <= last1) {
02d5137b 2663 if (regtry(reginfo, &s))
6eb5f6b9 2664 goto got_it;
7016d6eb
DM
2665 if (s >= last1) {
2666 s++; /* to break out of outer loop */
2667 break;
2668 }
2669 s += UTF8SKIP(s);
6eb5f6b9
JH
2670 }
2671 }
2672 else {
2673 while (s <= last1) {
02d5137b 2674 if (regtry(reginfo, &s))
6eb5f6b9
JH
2675 goto got_it;
2676 s++;
2677 }
2678 }
2679 }
ab3bbdeb 2680 DEBUG_EXECUTE_r(if (!did_match) {
f2ed9b32 2681 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
ab3bbdeb
YO
2682 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2683 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
33b8afdf 2684 ((must == prog->anchored_substr || must == prog->anchored_utf8)
bf93d4cc 2685 ? "anchored" : "floating"),
ab3bbdeb
YO
2686 quoted, RE_SV_TAIL(must));
2687 });
6eb5f6b9
JH
2688 goto phooey;
2689 }
f8fc2ecf 2690 else if ( (c = progi->regstclass) ) {
f14c76ed 2691 if (minlen) {
f8fc2ecf 2692 const OPCODE op = OP(progi->regstclass);
66e933ab 2693 /* don't bother with what can't match */
786e8c11 2694 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
f14c76ed
RGS
2695 strend = HOPc(strend, -(minlen - 1));
2696 }
a3621e74 2697 DEBUG_EXECUTE_r({
be8e71aa 2698 SV * const prop = sv_newmortal();
32fc9b6a 2699 regprop(prog, prop, c);
0df25f3d 2700 {
f2ed9b32 2701 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb 2702 s,strend-s,60);
0df25f3d 2703 PerlIO_printf(Perl_debug_log,
1c8f8eb1 2704 "Matching stclass %.*s against %s (%d bytes)\n",
e4f74956 2705 (int)SvCUR(prop), SvPVX_const(prop),
ab3bbdeb 2706 quoted, (int)(strend - s));
0df25f3d 2707 }
ffc61ed2 2708 });
f9176b44 2709 if (find_byclass(prog, c, s, strend, reginfo))
6eb5f6b9 2710 goto got_it;
07be1b83 2711 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
d6a28714
JH
2712 }
2713 else {
2714 dontbother = 0;
a0714e2c 2715 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
33b8afdf 2716 /* Trim the end. */
6af40bd7 2717 char *last= NULL;
33b8afdf 2718 SV* float_real;
c33e64f0
FC
2719 STRLEN len;
2720 const char *little;
33b8afdf 2721
7e0d5ad7
KW
2722 if (utf8_target) {
2723 if (! prog->float_utf8) {
2724 to_utf8_substr(prog);
2725 }
2726 float_real = prog->float_utf8;
2727 }
2728 else {
2729 if (! prog->float_substr) {
2730 if (! to_byte_substr(prog)) {
6b54ddc5 2731 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
7e0d5ad7
KW
2732 }
2733 }
2734 float_real = prog->float_substr;
2735 }
d6a28714 2736
c33e64f0
FC
2737 little = SvPV_const(float_real, len);
2738 if (SvTAIL(float_real)) {
7f18ad16
KW
2739 /* This means that float_real contains an artificial \n on
2740 * the end due to the presence of something like this:
2741 * /foo$/ where we can match both "foo" and "foo\n" at the
2742 * end of the string. So we have to compare the end of the
2743 * string first against the float_real without the \n and
2744 * then against the full float_real with the string. We
2745 * have to watch out for cases where the string might be
2746 * smaller than the float_real or the float_real without
2747 * the \n. */
1a13b075
YO
2748 char *checkpos= strend - len;
2749 DEBUG_OPTIMISE_r(
2750 PerlIO_printf(Perl_debug_log,
2751 "%sChecking for float_real.%s\n",
2752 PL_colors[4], PL_colors[5]));
2753 if (checkpos + 1 < strbeg) {
7f18ad16
KW
2754 /* can't match, even if we remove the trailing \n
2755 * string is too short to match */
1a13b075
YO
2756 DEBUG_EXECUTE_r(
2757 PerlIO_printf(Perl_debug_log,
2758 "%sString shorter than required trailing substring, cannot match.%s\n",
2759 PL_colors[4], PL_colors[5]));
2760 goto phooey;
2761 } else if (memEQ(checkpos + 1, little, len - 1)) {
7f18ad16
KW
2762 /* can match, the end of the string matches without the
2763 * "\n" */
1a13b075
YO
2764 last = checkpos + 1;
2765 } else if (checkpos < strbeg) {
7f18ad16
KW
2766 /* cant match, string is too short when the "\n" is
2767 * included */
1a13b075
YO
2768 DEBUG_EXECUTE_r(
2769 PerlIO_printf(Perl_debug_log,
2770 "%sString does not contain required trailing substring, cannot match.%s\n",
2771 PL_colors[4], PL_colors[5]));
2772 goto phooey;
2773 } else if (!multiline) {
7f18ad16
KW
2774 /* non multiline match, so compare with the "\n" at the
2775 * end of the string */
1a13b075
YO
2776 if (memEQ(checkpos, little, len)) {
2777 last= checkpos;
2778 } else {
2779 DEBUG_EXECUTE_r(
2780 PerlIO_printf(Perl_debug_log,
2781 "%sString does not contain required trailing substring, cannot match.%s\n",
2782 PL_colors[4], PL_colors[5]));
2783 goto phooey;
2784 }
2785 } else {
7f18ad16
KW
2786 /* multiline match, so we have to search for a place
2787 * where the full string is located */
d6a28714 2788 goto find_last;
1a13b075 2789 }
c33e64f0 2790 } else {
d6a28714 2791 find_last:
9041c2e3 2792 if (len)
d6a28714 2793 last = rninstr(s, strend, little, little + len);
b8c5462f 2794 else
a0288114 2795 last = strend; /* matching "$" */
b8c5462f 2796 }
6af40bd7 2797 if (!last) {
7f18ad16
KW
2798 /* at one point this block contained a comment which was
2799 * probably incorrect, which said that this was a "should not
2800 * happen" case. Even if it was true when it was written I am
2801 * pretty sure it is not anymore, so I have removed the comment
2802 * and replaced it with this one. Yves */
6bda09f9
YO
2803 DEBUG_EXECUTE_r(
2804 PerlIO_printf(Perl_debug_log,
6af40bd7
YO
2805 "String does not contain required substring, cannot match.\n"
2806 ));
2807 goto phooey;
bf93d4cc 2808 }
d6a28714
JH
2809 dontbother = strend - last + prog->float_min_offset;
2810 }
2811 if (minlen && (dontbother < minlen))
2812 dontbother = minlen - 1;
2813 strend -= dontbother; /* this one's always in bytes! */
2814 /* We don't know much -- general case. */
f2ed9b32 2815 if (utf8_target) {
d6a28714 2816 for (;;) {
02d5137b 2817 if (regtry(reginfo, &s))
d6a28714
JH
2818 goto got_it;
2819 if (s >= strend)
2820 break;
b8c5462f 2821 s += UTF8SKIP(s);
d6a28714
JH
2822 };
2823 }
2824 else {
2825 do {
02d5137b 2826 if (regtry(reginfo, &s))
d6a28714
JH
2827 goto got_it;
2828 } while (s++ < strend);
2829 }
2830 }
2831
2832 /* Failure. */
2833 goto phooey;
2834
2835got_it:
d5e7783a
DM
2836 /* s/// doesn't like it if $& is earlier than where we asked it to
2837 * start searching (which can happen on something like /.\G/) */
2838 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2839 && (prog->offs[0].start < stringarg - strbeg))
2840 {
2841 /* this should only be possible under \G */
2842 assert(prog->extflags & RXf_GPOS_SEEN);
2843 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2844 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2845 goto phooey;
2846 }
2847
495f47a5
DM
2848 DEBUG_BUFFERS_r(
2849 if (swap)
2850 PerlIO_printf(Perl_debug_log,
2851 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
2852 PTR2UV(prog),
2853 PTR2UV(swap)
2854 );
2855 );
e9105d30 2856 Safefree(swap);
d6a28714 2857
bf2039a9
DM
2858 /* clean up; this will trigger destructors that will free all slabs
2859 * above the current one, and cleanup the regmatch_info_aux
2860 * and regmatch_info_aux_eval sructs */
8adc0f72 2861
006f26b2
DM
2862 LEAVE_SCOPE(oldsave);
2863
5daac39c
NC
2864 if (RXp_PAREN_NAMES(prog))
2865 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
d6a28714 2866
0254aed9
DM
2867 RX_MATCH_UTF8_set(rx, utf8_target);
2868
d6a28714 2869 /* make sure $`, $&, $', and $digit will work later */
60165aa4 2870 if ( !(flags & REXEC_NOT_FIRST) )
749f4950 2871 S_reg_set_capture_string(aTHX_ rx,
60165aa4
DM
2872 strbeg, reginfo->strend,
2873 sv, flags, utf8_target);
9041c2e3 2874
d6a28714
JH
2875 return 1;
2876
2877phooey:
a3621e74 2878 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
e4584336 2879 PL_colors[4], PL_colors[5]));
8adc0f72 2880
bf2039a9
DM
2881 /* clean up; this will trigger destructors that will free all slabs
2882 * above the current one, and cleanup the regmatch_info_aux
2883 * and regmatch_info_aux_eval sructs */
8adc0f72 2884
006f26b2
DM
2885 LEAVE_SCOPE(oldsave);
2886
e9105d30 2887 if (swap) {
c74340f9 2888 /* we failed :-( roll it back */
495f47a5
DM
2889 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2890 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
2891 PTR2UV(prog),
2892 PTR2UV(prog->offs),
2893 PTR2UV(swap)
2894 ));
e9105d30
GG
2895 Safefree(prog->offs);
2896 prog->offs = swap;
2897 }
d6a28714
JH
2898 return 0;
2899}
2900
6bda09f9 2901
b3d298be 2902/* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
ec43f78b
DM
2903 * Do inc before dec, in case old and new rex are the same */
2904#define SET_reg_curpm(Re2) \
bf2039a9 2905 if (reginfo->info_aux_eval) { \
ec43f78b
DM
2906 (void)ReREFCNT_inc(Re2); \
2907 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
2908 PM_SETRE((PL_reg_curpm), (Re2)); \
2909 }
2910
2911
d6a28714
JH
2912/*
2913 - regtry - try match at specific point
2914 */
2915STATIC I32 /* 0 failure, 1 success */
f73aaa43 2916S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
d6a28714 2917{
97aff369 2918 dVAR;
d6a28714 2919 CHECKPOINT lastcp;
288b8c02 2920 REGEXP *const rx = reginfo->prog;
8d919b0a 2921 regexp *const prog = ReANY(rx);
f73aaa43 2922 I32 result;
f8fc2ecf 2923 RXi_GET_DECL(prog,progi);
a3621e74 2924 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2925
2926 PERL_ARGS_ASSERT_REGTRY;
2927
24b23f37 2928 reginfo->cutpoint=NULL;
d6a28714 2929
9d9163fb 2930 prog->offs[0].start = *startposp - reginfo->strbeg;
d6a28714 2931 prog->lastparen = 0;
03994de8 2932 prog->lastcloseparen = 0;
d6a28714
JH
2933
2934 /* XXXX What this code is doing here?!!! There should be no need
b93070ed 2935 to do this again and again, prog->lastparen should take care of
3dd2943c 2936 this! --ilya*/
dafc8851
JH
2937
2938 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
2939 * Actually, the code in regcppop() (which Ilya may be meaning by
b93070ed 2940 * prog->lastparen), is not needed at all by the test suite
225593e1
DM
2941 * (op/regexp, op/pat, op/split), but that code is needed otherwise
2942 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
2943 * Meanwhile, this code *is* needed for the
daf18116
JH
2944 * above-mentioned test suite tests to succeed. The common theme
2945 * on those tests seems to be returning null fields from matches.
225593e1 2946 * --jhi updated by dapm */
dafc8851 2947#if 1
d6a28714 2948 if (prog->nparens) {
b93070ed 2949 regexp_paren_pair *pp = prog->offs;
eb578fdb 2950 I32 i;
b93070ed 2951 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
f0ab9afb
NC
2952 ++pp;
2953 pp->start = -1;
2954 pp->end = -1;
d6a28714
JH
2955 }
2956 }
dafc8851 2957#endif
02db2b7b 2958 REGCP_SET(lastcp);
f73aaa43
DM
2959 result = regmatch(reginfo, *startposp, progi->program + 1);
2960 if (result != -1) {
2961 prog->offs[0].end = result;
d6a28714
JH
2962 return 1;
2963 }
24b23f37 2964 if (reginfo->cutpoint)
f73aaa43 2965 *startposp= reginfo->cutpoint;
02db2b7b 2966 REGCP_UNWIND(lastcp);
d6a28714
JH
2967 return 0;
2968}
2969
02db2b7b 2970
8ba1375e
MJD
2971#define sayYES goto yes
2972#define sayNO goto no
262b90c4 2973#define sayNO_SILENT goto no_silent
8ba1375e 2974
f9f4320a
YO
2975/* we dont use STMT_START/END here because it leads to
2976 "unreachable code" warnings, which are bogus, but distracting. */
2977#define CACHEsayNO \
c476f425 2978 if (ST.cache_mask) \
2ac8ff4b 2979 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
f9f4320a 2980 sayNO
3298f257 2981
a3621e74 2982/* this is used to determine how far from the left messages like
265c4333
YO
2983 'failed...' are printed. It should be set such that messages
2984 are inline with the regop output that created them.
a3621e74 2985*/
265c4333 2986#define REPORT_CODE_OFF 32
a3621e74
YO
2987
2988
40a82448
DM
2989#define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
2990#define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
79a2a0e8
KW
2991#define CHRTEST_NOT_A_CP_1 -999
2992#define CHRTEST_NOT_A_CP_2 -998
9e137952 2993
5d9a96ca
DM
2994/* grab a new slab and return the first slot in it */
2995
2996STATIC regmatch_state *
2997S_push_slab(pTHX)
2998{
a35a87e7 2999#if PERL_VERSION < 9 && !defined(PERL_CORE)
54df2634
NC
3000 dMY_CXT;
3001#endif
5d9a96ca
DM
3002 regmatch_slab *s = PL_regmatch_slab->next;
3003 if (!s) {
3004 Newx(s, 1, regmatch_slab);
3005 s->prev = PL_regmatch_slab;
3006 s->next = NULL;
3007 PL_regmatch_slab->next = s;
3008 }
3009 PL_regmatch_slab = s;
86545054 3010 return SLAB_FIRST(s);
5d9a96ca 3011}
5b47454d 3012
95b24440 3013
40a82448
DM
3014/* push a new state then goto it */
3015
4d5016e5
DM
3016#define PUSH_STATE_GOTO(state, node, input) \
3017 pushinput = input; \
40a82448
DM
3018 scan = node; \
3019 st->resume_state = state; \
3020 goto push_state;
3021
3022/* push a new state with success backtracking, then goto it */
3023
4d5016e5
DM
3024#define PUSH_YES_STATE_GOTO(state, node, input) \
3025 pushinput = input; \
40a82448
DM
3026 scan = node; \
3027 st->resume_state = state; \
3028 goto push_yes_state;
3029
aa283a38 3030
aa283a38 3031
4d5016e5 3032
d6a28714 3033/*
95b24440 3034
bf1f174e
DM
3035regmatch() - main matching routine
3036
3037This is basically one big switch statement in a loop. We execute an op,
3038set 'next' to point the next op, and continue. If we come to a point which
3039we may need to backtrack to on failure such as (A|B|C), we push a
3040backtrack state onto the backtrack stack. On failure, we pop the top
3041state, and re-enter the loop at the state indicated. If there are no more
3042states to pop, we return failure.
3043
3044Sometimes we also need to backtrack on success; for example /A+/, where
3045after successfully matching one A, we need to go back and try to
3046match another one; similarly for lookahead assertions: if the assertion
3047completes successfully, we backtrack to the state just before the assertion
3048and then carry on. In these cases, the pushed state is marked as
3049'backtrack on success too'. This marking is in fact done by a chain of
3050pointers, each pointing to the previous 'yes' state. On success, we pop to
3051the nearest yes state, discarding any intermediate failure-only states.
3052Sometimes a yes state is pushed just to force some cleanup code to be
3053called at the end of a successful match or submatch; e.g. (??{$re}) uses
3054it to free the inner regex.
3055
3056Note that failure backtracking rewinds the cursor position, while
3057success backtracking leaves it alone.
3058
3059A pattern is complete when the END op is executed, while a subpattern
3060such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3061ops trigger the "pop to last yes state if any, otherwise return true"
3062behaviour.
3063
3064A common convention in this function is to use A and B to refer to the two
3065subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3066the subpattern to be matched possibly multiple times, while B is the entire
3067rest of the pattern. Variable and state names reflect this convention.
3068
3069The states in the main switch are the union of ops and failure/success of
3070substates associated with with that op. For example, IFMATCH is the op
3071that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3072'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3073successfully matched A and IFMATCH_A_fail is a state saying that we have
3074just failed to match A. Resume states always come in pairs. The backtrack
3075state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3076at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3077on success or failure.
3078
3079The struct that holds a backtracking state is actually a big union, with
3080one variant for each major type of op. The variable st points to the
3081top-most backtrack struct. To make the code clearer, within each
3082block of code we #define ST to alias the relevant union.
3083
3084Here's a concrete example of a (vastly oversimplified) IFMATCH
3085implementation:
3086
3087 switch (state) {
3088 ....
3089
3090#define ST st->u.ifmatch
3091
3092 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3093 ST.foo = ...; // some state we wish to save
95b24440 3094 ...
bf1f174e
DM
3095 // push a yes backtrack state with a resume value of
3096 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3097 // first node of A:
4d5016e5 3098 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
bf1f174e
DM
3099 // NOTREACHED
3100
3101 case IFMATCH_A: // we have successfully executed A; now continue with B
3102 next = B;
3103 bar = ST.foo; // do something with the preserved value
3104 break;
3105
3106 case IFMATCH_A_fail: // A failed, so the assertion failed
3107 ...; // do some housekeeping, then ...
3108 sayNO; // propagate the failure
3109
3110#undef ST
95b24440 3111
bf1f174e
DM
3112 ...
3113 }
95b24440 3114
bf1f174e
DM
3115For any old-timers reading this who are familiar with the old recursive
3116approach, the code above is equivalent to:
95b24440 3117
bf1f174e
DM
3118 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3119 {
3120 int foo = ...
95b24440 3121 ...
bf1f174e
DM
3122 if (regmatch(A)) {
3123 next = B;
3124 bar = foo;
3125 break;
95b24440 3126 }
bf1f174e
DM
3127 ...; // do some housekeeping, then ...
3128 sayNO; // propagate the failure
95b24440 3129 }
bf1f174e
DM
3130
3131The topmost backtrack state, pointed to by st, is usually free. If you
3132want to claim it, populate any ST.foo fields in it with values you wish to
3133save, then do one of
3134
4d5016e5
DM
3135 PUSH_STATE_GOTO(resume_state, node, newinput);
3136 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
bf1f174e
DM
3137
3138which sets that backtrack state's resume value to 'resume_state', pushes a
3139new free entry to the top of the backtrack stack, then goes to 'node'.
3140On backtracking, the free slot is popped, and the saved state becomes the
3141new free state. An ST.foo field in this new top state can be temporarily
3142accessed to retrieve values, but once the main loop is re-entered, it
3143becomes available for reuse.
3144
3145Note that the depth of the backtrack stack constantly increases during the
3146left-to-right execution of the pattern, rather than going up and down with
3147the pattern nesting. For example the stack is at its maximum at Z at the
3148end of the pattern, rather than at X in the following:
3149
3150 /(((X)+)+)+....(Y)+....Z/
3151
3152The only exceptions to this are lookahead/behind assertions and the cut,
3153(?>A), which pop all the backtrack states associated with A before
3154continuing.
3155
486ec47a 3156Backtrack state structs are allocated in slabs of about 4K in size.
bf1f174e
DM
3157PL_regmatch_state and st always point to the currently active state,
3158and PL_regmatch_slab points to the slab currently containing
3159PL_regmatch_state. The first time regmatch() is called, the first slab is
3160allocated, and is never freed until interpreter destruction. When the slab
3161is full, a new one is allocated and chained to the end. At exit from
3162regmatch(), slabs allocated since entry are freed.
3163
3164*/
95b24440 3165
40a82448 3166
5bc10b2c 3167#define DEBUG_STATE_pp(pp) \
265c4333 3168 DEBUG_STATE_r({ \
f2ed9b32 3169 DUMP_EXEC_POS(locinput, scan, utf8_target); \
5bc10b2c 3170 PerlIO_printf(Perl_debug_log, \
5d458dd8 3171 " %*s"pp" %s%s%s%s%s\n", \
5bc10b2c 3172 depth*2, "", \
13d6edb4 3173 PL_reg_name[st->resume_state], \
5d458dd8
YO
3174 ((st==yes_state||st==mark_state) ? "[" : ""), \
3175 ((st==yes_state) ? "Y" : ""), \
3176 ((st==mark_state) ? "M" : ""), \
3177 ((st==yes_state||st==mark_state) ? "]" : "") \
3178 ); \
265c4333 3179 });
5bc10b2c 3180
40a82448 3181
3dab1dad 3182#define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
95b24440 3183
3df15adc 3184#ifdef DEBUGGING
5bc10b2c 3185
ab3bbdeb 3186STATIC void
f2ed9b32 3187S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
ab3bbdeb
YO
3188 const char *start, const char *end, const char *blurb)
3189{
efd26800 3190 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
7918f24d
NC
3191
3192 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3193
ab3bbdeb
YO
3194 if (!PL_colorset)
3195 reginitcolors();
3196 {
3197 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
d2c6dc5e 3198 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
ab3bbdeb 3199
f2ed9b32 3200 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
ab3bbdeb
YO
3201 start, end - start, 60);
3202
3203 PerlIO_printf(Perl_debug_log,
3204 "%s%s REx%s %s against %s\n",
3205 PL_colors[4], blurb, PL_colors[5], s0, s1);
3206
f2ed9b32 3207 if (utf8_target||utf8_pat)
1de06328
YO
3208 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3209 utf8_pat ? "pattern" : "",
f2ed9b32
KW
3210 utf8_pat && utf8_target ? " and " : "",
3211 utf8_target ? "string" : ""
ab3bbdeb
YO
3212 );
3213 }
3214}
3df15adc
YO
3215
3216STATIC void
786e8c11
YO
3217S_dump_exec_pos(pTHX_ const char *locinput,
3218 const regnode *scan,
3219 const char *loc_regeol,
3220 const char *loc_bostr,
3221 const char *loc_reg_starttry,
f2ed9b32 3222 const bool utf8_target)
07be1b83 3223{
786e8c11 3224 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
07be1b83 3225 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
786e8c11 3226 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
07be1b83
YO
3227 /* The part of the string before starttry has one color
3228 (pref0_len chars), between starttry and current
3229 position another one (pref_len - pref0_len chars),
3230 after the current position the third one.
3231 We assume that pref0_len <= pref_len, otherwise we
3232 decrease pref0_len. */
786e8c11
YO
3233 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3234 ? (5 + taill) - l : locinput - loc_bostr;
07be1b83
YO
3235 int pref0_len;
3236
7918f24d
NC
3237 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3238
f2ed9b32 3239 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
07be1b83 3240 pref_len++;
786e8c11
YO
3241 pref0_len = pref_len - (locinput - loc_reg_starttry);
3242 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3243 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3244 ? (5 + taill) - pref_len : loc_regeol - locinput);
f2ed9b32 3245 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
07be1b83
YO
3246 l--;
3247 if (pref0_len < 0)
3248 pref0_len = 0;
3249 if (pref0_len > pref_len)
3250 pref0_len = pref_len;
3251 {
f2ed9b32 3252 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
0df25f3d 3253
ab3bbdeb 3254 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
1de06328 3255 (locinput - pref_len),pref0_len, 60, 4, 5);
0df25f3d 3256
ab3bbdeb 3257 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3df15adc 3258 (locinput - pref_len + pref0_len),
1de06328 3259 pref_len - pref0_len, 60, 2, 3);
0df25f3d 3260
ab3bbdeb 3261 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
1de06328 3262 locinput, loc_regeol - locinput, 10, 0, 1);
0df25f3d 3263
1de06328 3264 const STRLEN tlen=len0+len1+len2;
3df15adc 3265 PerlIO_printf(Perl_debug_log,
ab3bbdeb 3266 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
786e8c11 3267 (IV)(locinput - loc_bostr),
07be1b83 3268 len0, s0,
07be1b83 3269 len1, s1,
07be1b83 3270 (docolor ? "" : "> <"),
07be1b83 3271 len2, s2,
f9f4320a 3272 (int)(tlen > 19 ? 0 : 19 - tlen),
07be1b83
YO
3273 "");
3274 }
3275}
3df15adc 3276
07be1b83
YO
3277#endif
3278
0a4db386
YO
3279/* reg_check_named_buff_matched()
3280 * Checks to see if a named buffer has matched. The data array of
3281 * buffer numbers corresponding to the buffer is expected to reside
3282 * in the regexp->data->data array in the slot stored in the ARG() of
3283 * node involved. Note that this routine doesn't actually care about the
3284 * name, that information is not preserved from compilation to execution.
3285 * Returns the index of the leftmost defined buffer with the given name
3286 * or 0 if non of the buffers matched.
3287 */
3288STATIC I32
7918f24d
NC
3289S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
3290{
0a4db386 3291 I32 n;
f8fc2ecf 3292 RXi_GET_DECL(rex,rexi);
ad64d0ec 3293 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
0a4db386 3294 I32 *nums=(I32*)SvPVX(sv_dat);
7918f24d
NC
3295
3296 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3297
0a4db386 3298 for ( n=0; n<SvIVX(sv_dat); n++ ) {
b93070ed
DM
3299 if ((I32)rex->lastparen >= nums[n] &&
3300 rex->offs[nums[n]].end != -1)
0a4db386
YO
3301 {
3302 return nums[n];
3303 }
3304 }
3305 return 0;
3306}
3307
2f554ef7 3308
c74f6de9 3309static bool
984e6dd1 3310S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
aed7b151 3311 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
c74f6de9 3312{
79a2a0e8
KW
3313 /* This function determines if there are one or two characters that match
3314 * the first character of the passed-in EXACTish node <text_node>, and if
3315 * so, returns them in the passed-in pointers.
c74f6de9 3316 *
79a2a0e8
KW
3317 * If it determines that no possible character in the target string can
3318 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3319 * the first character in <text_node> requires UTF-8 to represent, and the
3320 * target string isn't in UTF-8.)
c74f6de9 3321 *
79a2a0e8
KW
3322 * If there are more than two characters that could match the beginning of
3323 * <text_node>, or if more context is required to determine a match or not,
3324 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3325 *
3326 * The motiviation behind this function is to allow the caller to set up
3327 * tight loops for matching. If <text_node> is of type EXACT, there is
3328 * only one possible character that can match its first character, and so
3329 * the situation is quite simple. But things get much more complicated if
3330 * folding is involved. It may be that the first character of an EXACTFish
3331 * node doesn't participate in any possible fold, e.g., punctuation, so it
3332 * can be matched only by itself. The vast majority of characters that are
3333 * in folds match just two things, their lower and upper-case equivalents.
3334 * But not all are like that; some have multiple possible matches, or match
3335 * sequences of more than one character. This function sorts all that out.
3336 *
3337 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3338 * loop of trying to match A*, we know we can't exit where the thing
3339 * following it isn't a B. And something can't be a B unless it is the
3340 * beginning of B. By putting a quick test for that beginning in a tight
3341 * loop, we can rule out things that can't possibly be B without having to
3342 * break out of the loop, thus avoiding work. Similarly, if A is a single
3343 * character, we can make a tight loop matching A*, using the outputs of
3344 * this function.
3345 *
3346 * If the target string to match isn't in UTF-8, and there aren't
3347 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3348 * the one or two possible octets (which are characters in this situation)
3349 * that can match. In all cases, if there is only one character that can
3350 * match, *<c1p> and *<c2p> will be identical.
3351 *
3352 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3353 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3354 * can match the beginning of <text_node>. They should be declared with at
3355 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3356 * undefined what these contain.) If one or both of the buffers are
3357 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3358 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3359 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3360 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3361 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
c74f6de9 3362
ba44c216 3363 const bool utf8_target = reginfo->is_utf8_target;
79a2a0e8 3364
ddb0d839
KW
3365 UV c1 = CHRTEST_NOT_A_CP_1;
3366 UV c2 = CHRTEST_NOT_A_CP_2;
79a2a0e8 3367 bool use_chrtest_void = FALSE;
aed7b151 3368 const bool is_utf8_pat = reginfo->is_utf8_pat;
79a2a0e8
KW
3369
3370 /* Used when we have both utf8 input and utf8 output, to avoid converting
3371 * to/from code points */
3372 bool utf8_has_been_setup = FALSE;
3373
c74f6de9
KW
3374 dVAR;
3375
b4291290 3376 U8 *pat = (U8*)STRING(text_node);
c74f6de9 3377
79a2a0e8
KW
3378 if (OP(text_node) == EXACT) {
3379
3380 /* In an exact node, only one thing can be matched, that first
3381 * character. If both the pat and the target are UTF-8, we can just
3382 * copy the input to the output, avoiding finding the code point of
3383 * that character */
984e6dd1 3384 if (!is_utf8_pat) {
79a2a0e8
KW
3385 c2 = c1 = *pat;
3386 }
3387 else if (utf8_target) {
3388 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3389 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3390 utf8_has_been_setup = TRUE;
3391 }
3392 else {
3393 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
c74f6de9 3394 }
79a2a0e8
KW
3395 }
3396 else /* an EXACTFish node */
984e6dd1 3397 if ((is_utf8_pat
79a2a0e8
KW
3398 && is_MULTI_CHAR_FOLD_utf8_safe(pat,
3399 pat + STR_LEN(text_node)))
984e6dd1 3400 || (!is_utf8_pat
79a2a0e8
KW
3401 && is_MULTI_CHAR_FOLD_latin1_safe(pat,
3402 pat + STR_LEN(text_node))))
3403 {
3404 /* Multi-character folds require more context to sort out. Also
3405 * PL_utf8_foldclosures used below doesn't handle them, so have to be
3406 * handled outside this routine */
3407 use_chrtest_void = TRUE;
3408 }
3409 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
984e6dd1 3410 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
79a2a0e8
KW
3411 if (c1 > 256) {
3412 /* Load the folds hash, if not already done */
3413 SV** listp;
3414 if (! PL_utf8_foldclosures) {
3415 if (! PL_utf8_tofold) {
3416 U8 dummy[UTF8_MAXBYTES+1];
3417
3418 /* Force loading this by folding an above-Latin1 char */
3419 to_utf8_fold((U8*) HYPHEN_UTF8, dummy, NULL);
3420 assert(PL_utf8_tofold); /* Verify that worked */
3421 }
3422 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
3423 }
3424
3425 /* The fold closures data structure is a hash with the keys being
3426 * the UTF-8 of every character that is folded to, like 'k', and
3427 * the values each an array of all code points that fold to its
3428 * key. e.g. [ 'k', 'K', KELVIN_SIGN ]. Multi-character folds are
3429 * not included */
3430 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3431 (char *) pat,
3432 UTF8SKIP(pat),
3433 FALSE))))
3434 {
3435 /* Not found in the hash, therefore there are no folds
3436 * containing it, so there is only a single character that
3437 * could match */
3438 c2 = c1;
3439 }
3440 else { /* Does participate in folds */
3441 AV* list = (AV*) *listp;
3442 if (av_len(list) != 1) {
3443
3444 /* If there aren't exactly two folds to this, it is outside
3445 * the scope of this function */
3446 use_chrtest_void = TRUE;
3447 }
3448 else { /* There are two. Get them */
3449 SV** c_p = av_fetch(list, 0, FALSE);
3450 if (c_p == NULL) {
3451 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3452 }
3453 c1 = SvUV(*c_p);
3454
3455 c_p = av_fetch(list, 1, FALSE);
3456 if (c_p == NULL) {
3457 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3458 }
3459 c2 = SvUV(*c_p);
3460
3461 /* Folds that cross the 255/256 boundary are forbidden if
3462 * EXACTFL, or EXACTFA and one is ASCIII. Since the
3463 * pattern character is above 256, and its only other match
3464 * is below 256, the only legal match will be to itself.
3465 * We have thrown away the original, so have to compute
3466 * which is the one above 255 */
3467 if ((c1 < 256) != (c2 < 256)) {
3468 if (OP(text_node) == EXACTFL
3469 || (OP(text_node) == EXACTFA
3470 && (isASCII(c1) || isASCII(c2))))
3471 {
3472 if (c1 < 256) {
3473 c1 = c2;
3474 }
3475 else {
3476 c2 = c1;
3477 }
3478 }
3479 }
3480 }
3481 }
3482 }
3483 else /* Here, c1 is < 255 */
3484 if (utf8_target
3485 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3486 && OP(text_node) != EXACTFL
3487 && (OP(text_node) != EXACTFA || ! isASCII(c1)))
c74f6de9
KW
3488 {
3489 /* Here, there could be something above Latin1 in the target which
79a2a0e8
KW
3490 * folds to this character in the pattern. All such cases except
3491 * LATIN SMALL LETTER Y WITH DIAERESIS have more than two characters
3492 * involved in their folds, so are outside the scope of this
3493 * function */
3494 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3495 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3496 }
3497 else {
3498 use_chrtest_void = TRUE;
3499 }
c74f6de9
KW
3500 }
3501 else { /* Here nothing above Latin1 can fold to the pattern character */
3502 switch (OP(text_node)) {
3503
3504 case EXACTFL: /* /l rules */
79a2a0e8 3505 c2 = PL_fold_locale[c1];
c74f6de9
KW
3506 break;
3507
3508 case EXACTF:
3509 if (! utf8_target) { /* /d rules */
79a2a0e8 3510 c2 = PL_fold[c1];
c74f6de9
KW
3511 break;
3512 }
3513 /* FALLTHROUGH */
3514 /* /u rules for all these. This happens to work for
79a2a0e8 3515 * EXACTFA as nothing in Latin1 folds to ASCII */
c74f6de9
KW
3516 case EXACTFA:
3517 case EXACTFU_TRICKYFOLD:
79a2a0e8 3518 case EXACTFU_SS:
c74f6de9 3519 case EXACTFU:
79a2a0e8 3520 c2 = PL_fold_latin1[c1];
c74f6de9
KW
3521 break;
3522
878531d3
KW
3523 default:
3524 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3525 assert(0); /* NOTREACHED */
c74f6de9
KW
3526 }
3527 }
3528 }
79a2a0e8
KW
3529
3530 /* Here have figured things out. Set up the returns */
3531 if (use_chrtest_void) {
3532 *c2p = *c1p = CHRTEST_VOID;
3533 }
3534 else if (utf8_target) {
3535 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3536 uvchr_to_utf8(c1_utf8, c1);
3537 uvchr_to_utf8(c2_utf8, c2);
c74f6de9 3538 }
c74f6de9 3539
79a2a0e8
KW
3540 /* Invariants are stored in both the utf8 and byte outputs; Use
3541 * negative numbers otherwise for the byte ones. Make sure that the
3542 * byte ones are the same iff the utf8 ones are the same */
3543 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3544 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3545 ? *c2_utf8
3546 : (c1 == c2)