Commit | Line | Data |
---|---|---|
a0d0e21e LW |
1 | /* regexec.c |
2 | */ | |
3 | ||
4 | /* | |
4ac71550 TC |
5 | * One Ring to rule them all, One Ring to find them |
6 | & | |
7 | * [p.v of _The Lord of the Rings_, opening poem] | |
8 | * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"] | |
9 | * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"] | |
a0d0e21e LW |
10 | */ |
11 | ||
61296642 DM |
12 | /* This file contains functions for executing a regular expression. See |
13 | * also regcomp.c which funnily enough, contains functions for compiling | |
166f8a29 | 14 | * a regular expression. |
e4a054ea DM |
15 | * |
16 | * This file is also copied at build time to ext/re/re_exec.c, where | |
17 | * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT. | |
18 | * This causes the main functions to be compiled under new names and with | |
19 | * debugging support added, which makes "use re 'debug'" work. | |
166f8a29 DM |
20 | */ |
21 | ||
a687059c LW |
22 | /* NOTE: this is derived from Henry Spencer's regexp code, and should not |
23 | * confused with the original package (see point 3 below). Thanks, Henry! | |
24 | */ | |
25 | ||
26 | /* Additional note: this code is very heavily munged from Henry's version | |
27 | * in places. In some spots I've traded clarity for efficiency, so don't | |
28 | * blame Henry for some of the lack of readability. | |
29 | */ | |
30 | ||
e50aee73 AD |
31 | /* The names of the functions have been changed from regcomp and |
32 | * regexec to pregcomp and pregexec in order to avoid conflicts | |
33 | * with the POSIX routines of the same names. | |
34 | */ | |
35 | ||
b9d5759e | 36 | #ifdef PERL_EXT_RE_BUILD |
54df2634 | 37 | #include "re_top.h" |
9041c2e3 | 38 | #endif |
56953603 | 39 | |
a687059c | 40 | /* |
e50aee73 | 41 | * pregcomp and pregexec -- regsub and regerror are not used in perl |
a687059c LW |
42 | * |
43 | * Copyright (c) 1986 by University of Toronto. | |
44 | * Written by Henry Spencer. Not derived from licensed software. | |
45 | * | |
46 | * Permission is granted to anyone to use this software for any | |
47 | * purpose on any computer system, and to redistribute it freely, | |
48 | * subject to the following restrictions: | |
49 | * | |
50 | * 1. The author is not responsible for the consequences of use of | |
51 | * this software, no matter how awful, even if they arise | |
52 | * from defects in it. | |
53 | * | |
54 | * 2. The origin of this software must not be misrepresented, either | |
55 | * by explicit claim or by omission. | |
56 | * | |
57 | * 3. Altered versions must be plainly marked as such, and must not | |
58 | * be misrepresented as being the original software. | |
59 | * | |
60 | **** Alterations to Henry's code are... | |
61 | **** | |
4bb101f2 | 62 | **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, |
1129b882 NC |
63 | **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 |
64 | **** by Larry Wall and others | |
a687059c | 65 | **** |
9ef589d8 LW |
66 | **** You may distribute under the terms of either the GNU General Public |
67 | **** License or the Artistic License, as specified in the README file. | |
a687059c LW |
68 | * |
69 | * Beware that some of this code is subtly aware of the way operator | |
70 | * precedence is structured in regular expressions. Serious changes in | |
71 | * regular-expression syntax might require a total rethink. | |
72 | */ | |
73 | #include "EXTERN.h" | |
864dbfa3 | 74 | #define PERL_IN_REGEXEC_C |
a687059c | 75 | #include "perl.h" |
0f5d15d6 | 76 | |
54df2634 NC |
77 | #ifdef PERL_IN_XSUB_RE |
78 | # include "re_comp.h" | |
79 | #else | |
80 | # include "regcomp.h" | |
81 | #endif | |
a687059c | 82 | |
81e983c1 | 83 | #include "inline_invlist.c" |
1b0f46bf | 84 | #include "unicode_constants.h" |
81e983c1 | 85 | |
e1cf74e3 CB |
86 | #ifdef DEBUGGING |
87 | /* At least one required character in the target string is expressible only in | |
88 | * UTF-8. */ | |
89 | static const char* const non_utf8_target_but_utf8_required | |
90 | = "Can't match, because target string needs to be in UTF-8\n"; | |
91 | #endif | |
92 | ||
93 | #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \ | |
94 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\ | |
95 | goto target; \ | |
96 | } STMT_END | |
97 | ||
c74f6de9 KW |
98 | #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i) |
99 | ||
a687059c LW |
100 | #ifndef STATIC |
101 | #define STATIC static | |
102 | #endif | |
103 | ||
2d66f61e | 104 | /* Valid only for non-utf8 strings: avoids the reginclass |
7e2509c1 KW |
105 | * call if there are no complications: i.e., if everything matchable is |
106 | * straight forward in the bitmap */ | |
3db24e1e | 107 | #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,c+1,0) \ |
af364d03 | 108 | : ANYOF_BITMAP_TEST(p,*(c))) |
7d3e948e | 109 | |
c277df42 IZ |
110 | /* |
111 | * Forwards. | |
112 | */ | |
113 | ||
f2ed9b32 | 114 | #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv)) |
ba44c216 | 115 | #define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b) |
a0ed51b3 | 116 | |
3dab1dad | 117 | #define HOPc(pos,off) \ |
ba44c216 | 118 | (char *)(reginfo->is_utf8_target \ |
220db18a | 119 | ? reghop3((U8*)pos, off, \ |
9d9163fb | 120 | (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \ |
3dab1dad | 121 | : (U8*)(pos + off)) |
557f47af | 122 | |
3dab1dad | 123 | #define HOPBACKc(pos, off) \ |
ba44c216 | 124 | (char*)(reginfo->is_utf8_target \ |
9d9163fb DM |
125 | ? reghopmaybe3((U8*)pos, -off, (U8*)(reginfo->strbeg)) \ |
126 | : (pos - off >= reginfo->strbeg) \ | |
8e11feef | 127 | ? (U8*)pos - off \ |
3dab1dad | 128 | : NULL) |
efb30f32 | 129 | |
ba44c216 | 130 | #define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off)) |
1aa99e6b | 131 | #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim)) |
1aa99e6b | 132 | |
557f47af DM |
133 | /* lim must be +ve. Returns NULL on overshoot */ |
134 | #define HOPMAYBE3(pos,off,lim) \ | |
135 | (reginfo->is_utf8_target \ | |
136 | ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \ | |
137 | : ((U8*)pos + off <= lim) \ | |
138 | ? (U8*)pos + off \ | |
139 | : NULL) | |
140 | ||
8e9f2289 DM |
141 | /* like HOP3, but limits the result to <= lim even for the non-utf8 case. |
142 | * off must be >=0; args should be vars rather than expressions */ | |
143 | #define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \ | |
144 | ? reghop3((U8*)(pos), off, (U8*)(lim)) \ | |
145 | : (U8*)((pos + off) > lim ? lim : (pos + off))) | |
146 | ||
2974eaec DM |
147 | #define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \ |
148 | ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \ | |
149 | : (U8*)(pos + off)) | |
150 | #define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim)) | |
7016d6eb DM |
151 | |
152 | #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */ | |
153 | #define NEXTCHR_IS_EOS (nextchr < 0) | |
154 | ||
155 | #define SET_nextchr \ | |
220db18a | 156 | nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS) |
7016d6eb DM |
157 | |
158 | #define SET_locinput(p) \ | |
159 | locinput = (p); \ | |
160 | SET_nextchr | |
161 | ||
162 | ||
2a16ac92 | 163 | #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) STMT_START { \ |
c7304fe2 KW |
164 | if (!swash_ptr) { \ |
165 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \ | |
c7304fe2 | 166 | swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \ |
2a16ac92 | 167 | 1, 0, invlist, &flags); \ |
c7304fe2 KW |
168 | assert(swash_ptr); \ |
169 | } \ | |
170 | } STMT_END | |
171 | ||
172 | /* If in debug mode, we test that a known character properly matches */ | |
173 | #ifdef DEBUGGING | |
174 | # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \ | |
175 | property_name, \ | |
2a16ac92 | 176 | invlist, \ |
c7304fe2 | 177 | utf8_char_in_property) \ |
2a16ac92 | 178 | LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist); \ |
c7304fe2 KW |
179 | assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE)); |
180 | #else | |
181 | # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \ | |
182 | property_name, \ | |
2a16ac92 | 183 | invlist, \ |
c7304fe2 | 184 | utf8_char_in_property) \ |
2a16ac92 | 185 | LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) |
c7304fe2 | 186 | #endif |
d1eb3177 | 187 | |
c7304fe2 KW |
188 | #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \ |
189 | PL_utf8_swash_ptrs[_CC_WORDCHAR], \ | |
2a16ac92 KW |
190 | "", \ |
191 | PL_XPosix_ptrs[_CC_WORDCHAR], \ | |
df38da56 | 192 | LATIN_CAPITAL_LETTER_SHARP_S_UTF8); |
c7304fe2 KW |
193 | |
194 | #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \ | |
195 | STMT_START { \ | |
196 | LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \ | |
197 | "_X_regular_begin", \ | |
2a16ac92 | 198 | NULL, \ |
df38da56 | 199 | LATIN_CAPITAL_LETTER_SHARP_S_UTF8); \ |
c7304fe2 KW |
200 | LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \ |
201 | "_X_extend", \ | |
2a16ac92 | 202 | NULL, \ |
c7304fe2 KW |
203 | COMBINING_GRAVE_ACCENT_UTF8); \ |
204 | } STMT_END | |
d1eb3177 | 205 | |
c7304fe2 | 206 | #define PLACEHOLDER /* Something for the preprocessor to grab onto */ |
3dab1dad YO |
207 | /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */ |
208 | ||
5f80c4cf | 209 | /* for use after a quantifier and before an EXACT-like node -- japhy */ |
c35dcbe2 YO |
210 | /* it would be nice to rework regcomp.sym to generate this stuff. sigh |
211 | * | |
212 | * NOTE that *nothing* that affects backtracking should be in here, specifically | |
213 | * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a | |
214 | * node that is in between two EXACT like nodes when ascertaining what the required | |
215 | * "follow" character is. This should probably be moved to regex compile time | |
216 | * although it may be done at run time beause of the REF possibility - more | |
217 | * investigation required. -- demerphq | |
218 | */ | |
baa60164 KW |
219 | #define JUMPABLE(rn) ( \ |
220 | OP(rn) == OPEN || \ | |
3e901dc0 | 221 | (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \ |
baa60164 KW |
222 | OP(rn) == EVAL || \ |
223 | OP(rn) == SUSPEND || OP(rn) == IFMATCH || \ | |
224 | OP(rn) == PLUS || OP(rn) == MINMOD || \ | |
225 | OP(rn) == KEEPS || \ | |
226 | (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \ | |
e2d8ce26 | 227 | ) |
ee9b8eae | 228 | #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT) |
e2d8ce26 | 229 | |
ee9b8eae YO |
230 | #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF ) |
231 | ||
232 | #if 0 | |
233 | /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so | |
234 | we don't need this definition. */ | |
235 | #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF ) | |
098b07d5 | 236 | #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFA || OP(rn)==EXACTFA_NO_TRIE || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF ) |
ee9b8eae YO |
237 | #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL ) |
238 | ||
239 | #else | |
240 | /* ... so we use this as its faster. */ | |
241 | #define IS_TEXT(rn) ( OP(rn)==EXACT ) | |
098b07d5 | 242 | #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn) == EXACTFA || OP(rn) == EXACTFA_NO_TRIE) |
ee9b8eae YO |
243 | #define IS_TEXTF(rn) ( OP(rn)==EXACTF ) |
244 | #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL ) | |
245 | ||
246 | #endif | |
e2d8ce26 | 247 | |
a84d97b6 HS |
248 | /* |
249 | Search for mandatory following text node; for lookahead, the text must | |
250 | follow but for lookbehind (rn->flags != 0) we skip to the next step. | |
251 | */ | |
baa60164 | 252 | #define FIND_NEXT_IMPT(rn) STMT_START { \ |
3dab1dad YO |
253 | while (JUMPABLE(rn)) { \ |
254 | const OPCODE type = OP(rn); \ | |
255 | if (type == SUSPEND || PL_regkind[type] == CURLY) \ | |
e2d8ce26 | 256 | rn = NEXTOPER(NEXTOPER(rn)); \ |
3dab1dad | 257 | else if (type == PLUS) \ |
cca55fe3 | 258 | rn = NEXTOPER(rn); \ |
3dab1dad | 259 | else if (type == IFMATCH) \ |
a84d97b6 | 260 | rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \ |
e2d8ce26 | 261 | else rn += NEXT_OFF(rn); \ |
3dab1dad | 262 | } \ |
5f80c4cf | 263 | } STMT_END |
74750237 | 264 | |
22913b96 KW |
265 | /* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode. |
266 | * These are for the pre-composed Hangul syllables, which are all in a | |
267 | * contiguous block and arranged there in such a way so as to facilitate | |
268 | * alorithmic determination of their characteristics. As such, they don't need | |
269 | * a swash, but can be determined by simple arithmetic. Almost all are | |
270 | * GCB=LVT, but every 28th one is a GCB=LV */ | |
271 | #define SBASE 0xAC00 /* Start of block */ | |
272 | #define SCount 11172 /* Length of block */ | |
273 | #define TCount 28 | |
c476f425 | 274 | |
006f26b2 DM |
275 | #define SLAB_FIRST(s) (&(s)->states[0]) |
276 | #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1]) | |
277 | ||
a75351a1 | 278 | static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo); |
bf2039a9 | 279 | static void S_cleanup_regmatch_info_aux(pTHX_ void *arg); |
bf2039a9 | 280 | static regmatch_state * S_push_slab(pTHX); |
51371543 | 281 | |
87c0511b | 282 | #define REGCP_PAREN_ELEMS 3 |
f067efbf | 283 | #define REGCP_OTHER_ELEMS 3 |
e0fa7e2b | 284 | #define REGCP_FRAME_ELEMS 1 |
620d5b66 NC |
285 | /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and |
286 | * are needed for the regexp context stack bookkeeping. */ | |
287 | ||
76e3520e | 288 | STATIC CHECKPOINT |
92da3157 | 289 | S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen) |
a0d0e21e | 290 | { |
a3b680e6 | 291 | const int retval = PL_savestack_ix; |
92da3157 DM |
292 | const int paren_elems_to_push = |
293 | (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS; | |
e0fa7e2b NC |
294 | const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS; |
295 | const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT; | |
87c0511b | 296 | I32 p; |
40a82448 | 297 | GET_RE_DEBUG_FLAGS_DECL; |
a0d0e21e | 298 | |
b93070ed DM |
299 | PERL_ARGS_ASSERT_REGCPPUSH; |
300 | ||
e49a9654 | 301 | if (paren_elems_to_push < 0) |
e8a85d26 JH |
302 | Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u", |
303 | (int)paren_elems_to_push, (int)maxopenparen, | |
304 | (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS); | |
e49a9654 | 305 | |
e0fa7e2b NC |
306 | if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems) |
307 | Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf | |
5df417d0 | 308 | " out of range (%lu-%ld)", |
92da3157 DM |
309 | total_elems, |
310 | (unsigned long)maxopenparen, | |
311 | (long)parenfloor); | |
e0fa7e2b | 312 | |
620d5b66 | 313 | SSGROW(total_elems + REGCP_FRAME_ELEMS); |
7f69552c | 314 | |
495f47a5 | 315 | DEBUG_BUFFERS_r( |
92da3157 | 316 | if ((int)maxopenparen > (int)parenfloor) |
495f47a5 DM |
317 | PerlIO_printf(Perl_debug_log, |
318 | "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n", | |
319 | PTR2UV(rex), | |
320 | PTR2UV(rex->offs) | |
321 | ); | |
322 | ); | |
92da3157 | 323 | for (p = parenfloor+1; p <= (I32)maxopenparen; p++) { |
b1ce53c5 | 324 | /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */ |
99a90e59 FC |
325 | SSPUSHIV(rex->offs[p].end); |
326 | SSPUSHIV(rex->offs[p].start); | |
1ca2007e | 327 | SSPUSHINT(rex->offs[p].start_tmp); |
e7707071 | 328 | DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, |
495f47a5 DM |
329 | " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n", |
330 | (UV)p, | |
331 | (IV)rex->offs[p].start, | |
332 | (IV)rex->offs[p].start_tmp, | |
333 | (IV)rex->offs[p].end | |
40a82448 | 334 | )); |
a0d0e21e | 335 | } |
b1ce53c5 | 336 | /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */ |
92da3157 | 337 | SSPUSHINT(maxopenparen); |
b93070ed DM |
338 | SSPUSHINT(rex->lastparen); |
339 | SSPUSHINT(rex->lastcloseparen); | |
e0fa7e2b | 340 | SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */ |
41123dfd | 341 | |
a0d0e21e LW |
342 | return retval; |
343 | } | |
344 | ||
c277df42 | 345 | /* These are needed since we do not localize EVAL nodes: */ |
ab3bbdeb YO |
346 | #define REGCP_SET(cp) \ |
347 | DEBUG_STATE_r( \ | |
ab3bbdeb | 348 | PerlIO_printf(Perl_debug_log, \ |
e4f74956 | 349 | " Setting an EVAL scope, savestack=%"IVdf"\n", \ |
ab3bbdeb YO |
350 | (IV)PL_savestack_ix)); \ |
351 | cp = PL_savestack_ix | |
c3464db5 | 352 | |
ab3bbdeb | 353 | #define REGCP_UNWIND(cp) \ |
e4f74956 | 354 | DEBUG_STATE_r( \ |
ab3bbdeb | 355 | if (cp != PL_savestack_ix) \ |
e4f74956 YO |
356 | PerlIO_printf(Perl_debug_log, \ |
357 | " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \ | |
ab3bbdeb YO |
358 | (IV)(cp), (IV)PL_savestack_ix)); \ |
359 | regcpblow(cp) | |
c277df42 | 360 | |
a8d1f4b4 DM |
361 | #define UNWIND_PAREN(lp, lcp) \ |
362 | for (n = rex->lastparen; n > lp; n--) \ | |
363 | rex->offs[n].end = -1; \ | |
364 | rex->lastparen = n; \ | |
365 | rex->lastcloseparen = lcp; | |
366 | ||
367 | ||
f067efbf | 368 | STATIC void |
92da3157 | 369 | S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p) |
a0d0e21e | 370 | { |
e0fa7e2b | 371 | UV i; |
87c0511b | 372 | U32 paren; |
a3621e74 YO |
373 | GET_RE_DEBUG_FLAGS_DECL; |
374 | ||
7918f24d NC |
375 | PERL_ARGS_ASSERT_REGCPPOP; |
376 | ||
b1ce53c5 | 377 | /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */ |
c6bf6a65 | 378 | i = SSPOPUV; |
e0fa7e2b NC |
379 | assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */ |
380 | i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */ | |
b93070ed DM |
381 | rex->lastcloseparen = SSPOPINT; |
382 | rex->lastparen = SSPOPINT; | |
92da3157 | 383 | *maxopenparen_p = SSPOPINT; |
b1ce53c5 | 384 | |
620d5b66 | 385 | i -= REGCP_OTHER_ELEMS; |
b1ce53c5 | 386 | /* Now restore the parentheses context. */ |
495f47a5 DM |
387 | DEBUG_BUFFERS_r( |
388 | if (i || rex->lastparen + 1 <= rex->nparens) | |
389 | PerlIO_printf(Perl_debug_log, | |
390 | "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n", | |
391 | PTR2UV(rex), | |
392 | PTR2UV(rex->offs) | |
393 | ); | |
394 | ); | |
92da3157 | 395 | paren = *maxopenparen_p; |
620d5b66 | 396 | for ( ; i > 0; i -= REGCP_PAREN_ELEMS) { |
ea3daa5d | 397 | SSize_t tmps; |
1ca2007e | 398 | rex->offs[paren].start_tmp = SSPOPINT; |
99a90e59 FC |
399 | rex->offs[paren].start = SSPOPIV; |
400 | tmps = SSPOPIV; | |
b93070ed DM |
401 | if (paren <= rex->lastparen) |
402 | rex->offs[paren].end = tmps; | |
495f47a5 DM |
403 | DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log, |
404 | " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n", | |
405 | (UV)paren, | |
406 | (IV)rex->offs[paren].start, | |
407 | (IV)rex->offs[paren].start_tmp, | |
408 | (IV)rex->offs[paren].end, | |
409 | (paren > rex->lastparen ? "(skipped)" : "")); | |
c277df42 | 410 | ); |
87c0511b | 411 | paren--; |
a0d0e21e | 412 | } |
daf18116 | 413 | #if 1 |
dafc8851 JH |
414 | /* It would seem that the similar code in regtry() |
415 | * already takes care of this, and in fact it is in | |
416 | * a better location to since this code can #if 0-ed out | |
417 | * but the code in regtry() is needed or otherwise tests | |
418 | * requiring null fields (pat.t#187 and split.t#{13,14} | |
daf18116 JH |
419 | * (as of patchlevel 7877) will fail. Then again, |
420 | * this code seems to be necessary or otherwise | |
225593e1 DM |
421 | * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/ |
422 | * --jhi updated by dapm */ | |
b93070ed | 423 | for (i = rex->lastparen + 1; i <= rex->nparens; i++) { |
92da3157 | 424 | if (i > *maxopenparen_p) |
b93070ed DM |
425 | rex->offs[i].start = -1; |
426 | rex->offs[i].end = -1; | |
495f47a5 DM |
427 | DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log, |
428 | " \\%"UVuf": %s ..-1 undeffing\n", | |
429 | (UV)i, | |
92da3157 | 430 | (i > *maxopenparen_p) ? "-1" : " " |
495f47a5 | 431 | )); |
a0d0e21e | 432 | } |
dafc8851 | 433 | #endif |
a0d0e21e LW |
434 | } |
435 | ||
74088413 DM |
436 | /* restore the parens and associated vars at savestack position ix, |
437 | * but without popping the stack */ | |
438 | ||
439 | STATIC void | |
92da3157 | 440 | S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p) |
74088413 DM |
441 | { |
442 | I32 tmpix = PL_savestack_ix; | |
443 | PL_savestack_ix = ix; | |
92da3157 | 444 | regcppop(rex, maxopenparen_p); |
74088413 DM |
445 | PL_savestack_ix = tmpix; |
446 | } | |
447 | ||
02db2b7b | 448 | #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */ |
a0d0e21e | 449 | |
31c7f561 KW |
450 | STATIC bool |
451 | S_isFOO_lc(pTHX_ const U8 classnum, const U8 character) | |
452 | { | |
453 | /* Returns a boolean as to whether or not 'character' is a member of the | |
454 | * Posix character class given by 'classnum' that should be equivalent to a | |
455 | * value in the typedef '_char_class_number'. | |
456 | * | |
457 | * Ideally this could be replaced by a just an array of function pointers | |
458 | * to the C library functions that implement the macros this calls. | |
459 | * However, to compile, the precise function signatures are required, and | |
460 | * these may vary from platform to to platform. To avoid having to figure | |
461 | * out what those all are on each platform, I (khw) am using this method, | |
7aee35ff KW |
462 | * which adds an extra layer of function call overhead (unless the C |
463 | * optimizer strips it away). But we don't particularly care about | |
464 | * performance with locales anyway. */ | |
31c7f561 KW |
465 | |
466 | switch ((_char_class_number) classnum) { | |
15861f94 | 467 | case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character); |
31c7f561 | 468 | case _CC_ENUM_ALPHA: return isALPHA_LC(character); |
e8d596e0 KW |
469 | case _CC_ENUM_ASCII: return isASCII_LC(character); |
470 | case _CC_ENUM_BLANK: return isBLANK_LC(character); | |
b0d691b2 KW |
471 | case _CC_ENUM_CASED: return isLOWER_LC(character) |
472 | || isUPPER_LC(character); | |
e8d596e0 | 473 | case _CC_ENUM_CNTRL: return isCNTRL_LC(character); |
31c7f561 KW |
474 | case _CC_ENUM_DIGIT: return isDIGIT_LC(character); |
475 | case _CC_ENUM_GRAPH: return isGRAPH_LC(character); | |
476 | case _CC_ENUM_LOWER: return isLOWER_LC(character); | |
477 | case _CC_ENUM_PRINT: return isPRINT_LC(character); | |
e8d596e0 | 478 | case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character); |
31c7f561 | 479 | case _CC_ENUM_PUNCT: return isPUNCT_LC(character); |
e8d596e0 | 480 | case _CC_ENUM_SPACE: return isSPACE_LC(character); |
31c7f561 KW |
481 | case _CC_ENUM_UPPER: return isUPPER_LC(character); |
482 | case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character); | |
31c7f561 | 483 | case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character); |
31c7f561 KW |
484 | default: /* VERTSPACE should never occur in locales */ |
485 | Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum); | |
486 | } | |
487 | ||
e5964223 | 488 | NOT_REACHED; /* NOTREACHED */ |
31c7f561 KW |
489 | return FALSE; |
490 | } | |
491 | ||
3018b823 KW |
492 | STATIC bool |
493 | S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character) | |
494 | { | |
495 | /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded | |
496 | * 'character' is a member of the Posix character class given by 'classnum' | |
497 | * that should be equivalent to a value in the typedef | |
498 | * '_char_class_number'. | |
499 | * | |
500 | * This just calls isFOO_lc on the code point for the character if it is in | |
501 | * the range 0-255. Outside that range, all characters avoid Unicode | |
502 | * rules, ignoring any locale. So use the Unicode function if this class | |
503 | * requires a swash, and use the Unicode macro otherwise. */ | |
504 | ||
505 | PERL_ARGS_ASSERT_ISFOO_UTF8_LC; | |
506 | ||
507 | if (UTF8_IS_INVARIANT(*character)) { | |
508 | return isFOO_lc(classnum, *character); | |
509 | } | |
510 | else if (UTF8_IS_DOWNGRADEABLE_START(*character)) { | |
511 | return isFOO_lc(classnum, | |
94bb8c36 | 512 | TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1))); |
3018b823 KW |
513 | } |
514 | ||
515 | if (classnum < _FIRST_NON_SWASH_CC) { | |
516 | ||
517 | /* Initialize the swash unless done already */ | |
518 | if (! PL_utf8_swash_ptrs[classnum]) { | |
519 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; | |
2a16ac92 KW |
520 | PL_utf8_swash_ptrs[classnum] = |
521 | _core_swash_init("utf8", | |
522 | "", | |
523 | &PL_sv_undef, 1, 0, | |
524 | PL_XPosix_ptrs[classnum], &flags); | |
3018b823 KW |
525 | } |
526 | ||
92a2046b KW |
527 | return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *) |
528 | character, | |
529 | TRUE /* is UTF */ )); | |
3018b823 KW |
530 | } |
531 | ||
532 | switch ((_char_class_number) classnum) { | |
533 | case _CC_ENUM_SPACE: | |
534 | case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character); | |
535 | ||
536 | case _CC_ENUM_BLANK: return is_HORIZWS_high(character); | |
537 | case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character); | |
538 | case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character); | |
e1ee3960 | 539 | default: break; |
3018b823 KW |
540 | } |
541 | ||
e1ee3960 | 542 | return FALSE; /* Things like CNTRL are always below 256 */ |
3018b823 KW |
543 | } |
544 | ||
a687059c | 545 | /* |
e50aee73 | 546 | * pregexec and friends |
a687059c LW |
547 | */ |
548 | ||
76234dfb | 549 | #ifndef PERL_IN_XSUB_RE |
a687059c | 550 | /* |
c277df42 | 551 | - pregexec - match a regexp against a string |
a687059c | 552 | */ |
c277df42 | 553 | I32 |
5aaab254 | 554 | Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend, |
ea3daa5d | 555 | char *strbeg, SSize_t minend, SV *screamer, U32 nosave) |
8fd1a950 DM |
556 | /* stringarg: the point in the string at which to begin matching */ |
557 | /* strend: pointer to null at end of string */ | |
558 | /* strbeg: real beginning of string */ | |
559 | /* minend: end of match must be >= minend bytes after stringarg. */ | |
560 | /* screamer: SV being matched: only used for utf8 flag, pos() etc; string | |
561 | * itself is accessed via the pointers above */ | |
562 | /* nosave: For optimizations. */ | |
c277df42 | 563 | { |
7918f24d NC |
564 | PERL_ARGS_ASSERT_PREGEXEC; |
565 | ||
c277df42 | 566 | return |
9041c2e3 | 567 | regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL, |
c277df42 IZ |
568 | nosave ? 0 : REXEC_COPY_STR); |
569 | } | |
76234dfb | 570 | #endif |
22e551b9 | 571 | |
cad2e5aa | 572 | |
6eb5f6b9 | 573 | |
1a4edc3c DM |
574 | /* re_intuit_start(): |
575 | * | |
576 | * Based on some optimiser hints, try to find the earliest position in the | |
577 | * string where the regex could match. | |
578 | * | |
579 | * rx: the regex to match against | |
580 | * sv: the SV being matched: only used for utf8 flag; the string | |
581 | * itself is accessed via the pointers below. Note that on | |
582 | * something like an overloaded SV, SvPOK(sv) may be false | |
583 | * and the string pointers may point to something unrelated to | |
584 | * the SV itself. | |
585 | * strbeg: real beginning of string | |
586 | * strpos: the point in the string at which to begin matching | |
587 | * strend: pointer to the byte following the last char of the string | |
588 | * flags currently unused; set to 0 | |
589 | * data: currently unused; set to NULL | |
590 | * | |
591 | * The basic idea of re_intuit_start() is to use some known information | |
592 | * about the pattern, namely: | |
593 | * | |
594 | * a) the longest known anchored substring (i.e. one that's at a | |
595 | * constant offset from the beginning of the pattern; but not | |
596 | * necessarily at a fixed offset from the beginning of the | |
597 | * string); | |
598 | * b) the longest floating substring (i.e. one that's not at a constant | |
599 | * offset from the beginning of the pattern); | |
600 | * c) Whether the pattern is anchored to the string; either | |
601 | * an absolute anchor: /^../, or anchored to \n: /^.../m, | |
602 | * or anchored to pos(): /\G/; | |
603 | * d) A start class: a real or synthetic character class which | |
604 | * represents which characters are legal at the start of the pattern; | |
605 | * | |
606 | * to either quickly reject the match, or to find the earliest position | |
607 | * within the string at which the pattern might match, thus avoiding | |
608 | * running the full NFA engine at those earlier locations, only to | |
609 | * eventually fail and retry further along. | |
610 | * | |
611 | * Returns NULL if the pattern can't match, or returns the address within | |
612 | * the string which is the earliest place the match could occur. | |
613 | * | |
614 | * The longest of the anchored and floating substrings is called 'check' | |
615 | * and is checked first. The other is called 'other' and is checked | |
616 | * second. The 'other' substring may not be present. For example, | |
617 | * | |
618 | * /(abc|xyz)ABC\d{0,3}DEFG/ | |
619 | * | |
620 | * will have | |
621 | * | |
622 | * check substr (float) = "DEFG", offset 6..9 chars | |
623 | * other substr (anchored) = "ABC", offset 3..3 chars | |
624 | * stclass = [ax] | |
625 | * | |
626 | * Be aware that during the course of this function, sometimes 'anchored' | |
627 | * refers to a substring being anchored relative to the start of the | |
628 | * pattern, and sometimes to the pattern itself being anchored relative to | |
629 | * the string. For example: | |
630 | * | |
631 | * /\dabc/: "abc" is anchored to the pattern; | |
632 | * /^\dabc/: "abc" is anchored to the pattern and the string; | |
633 | * /\d+abc/: "abc" is anchored to neither the pattern nor the string; | |
634 | * /^\d+abc/: "abc" is anchored to neither the pattern nor the string, | |
635 | * but the pattern is anchored to the string. | |
52a21eb3 DM |
636 | */ |
637 | ||
cad2e5aa | 638 | char * |
52a21eb3 DM |
639 | Perl_re_intuit_start(pTHX_ |
640 | REGEXP * const rx, | |
641 | SV *sv, | |
642 | const char * const strbeg, | |
643 | char *strpos, | |
644 | char *strend, | |
645 | const U32 flags, | |
646 | re_scream_pos_data *data) | |
cad2e5aa | 647 | { |
8d919b0a | 648 | struct regexp *const prog = ReANY(rx); |
6b071d16 | 649 | SSize_t start_shift = prog->check_offset_min; |
cad2e5aa | 650 | /* Should be nonnegative! */ |
ea3daa5d | 651 | SSize_t end_shift = 0; |
0fc004dd DM |
652 | /* current lowest pos in string where the regex can start matching */ |
653 | char *rx_origin = strpos; | |
eb578fdb | 654 | SV *check; |
f2ed9b32 | 655 | const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */ |
6480a6c4 | 656 | U8 other_ix = 1 - prog->substrs->check_ix; |
6ad5ffb3 | 657 | bool ml_anch = 0; |
8f4bf5fc | 658 | char *other_last = strpos;/* latest pos 'other' substr already checked to */ |
bd61b366 | 659 | char *check_at = NULL; /* check substr found at this pos */ |
bbe252da | 660 | const I32 multiline = prog->extflags & RXf_PMf_MULTILINE; |
f8fc2ecf | 661 | RXi_GET_DECL(prog,progi); |
02d5137b DM |
662 | regmatch_info reginfo_buf; /* create some info to pass to find_byclass */ |
663 | regmatch_info *const reginfo = ®info_buf; | |
a3621e74 YO |
664 | GET_RE_DEBUG_FLAGS_DECL; |
665 | ||
7918f24d | 666 | PERL_ARGS_ASSERT_RE_INTUIT_START; |
c33e64f0 FC |
667 | PERL_UNUSED_ARG(flags); |
668 | PERL_UNUSED_ARG(data); | |
7918f24d | 669 | |
1dc475d0 DM |
670 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
671 | "Intuit: trying to determine minimum start position...\n")); | |
672 | ||
fb9bbddb DM |
673 | /* for now, assume that all substr offsets are positive. If at some point |
674 | * in the future someone wants to do clever things with look-behind and | |
675 | * -ve offsets, they'll need to fix up any code in this function | |
676 | * which uses these offsets. See the thread beginning | |
677 | * <20140113145929.GF27210@iabyn.com> | |
678 | */ | |
679 | assert(prog->substrs->data[0].min_offset >= 0); | |
680 | assert(prog->substrs->data[0].max_offset >= 0); | |
681 | assert(prog->substrs->data[1].min_offset >= 0); | |
682 | assert(prog->substrs->data[1].max_offset >= 0); | |
683 | assert(prog->substrs->data[2].min_offset >= 0); | |
684 | assert(prog->substrs->data[2].max_offset >= 0); | |
685 | ||
f7022b5a | 686 | /* for now, assume that if both present, that the floating substring |
83f2232d | 687 | * doesn't start before the anchored substring. |
f7022b5a DM |
688 | * If you break this assumption (e.g. doing better optimisations |
689 | * with lookahead/behind), then you'll need to audit the code in this | |
690 | * function carefully first | |
691 | */ | |
692 | assert( | |
693 | ! ( (prog->anchored_utf8 || prog->anchored_substr) | |
694 | && (prog->float_utf8 || prog->float_substr)) | |
695 | || (prog->float_min_offset >= prog->anchored_offset)); | |
696 | ||
1a4edc3c DM |
697 | /* byte rather than char calculation for efficiency. It fails |
698 | * to quickly reject some cases that can't match, but will reject | |
699 | * them later after doing full char arithmetic */ | |
c344f387 | 700 | if (prog->minlen > strend - strpos) { |
a3621e74 | 701 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
1dc475d0 | 702 | " String too short...\n")); |
cad2e5aa | 703 | goto fail; |
2c2d71f5 | 704 | } |
d8da0584 | 705 | |
ab4e48c1 | 706 | RX_MATCH_UTF8_set(rx,utf8_target); |
6c3fea77 | 707 | reginfo->is_utf8_target = cBOOL(utf8_target); |
bf2039a9 | 708 | reginfo->info_aux = NULL; |
9d9163fb | 709 | reginfo->strbeg = strbeg; |
220db18a | 710 | reginfo->strend = strend; |
aed7b151 | 711 | reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx)); |
02d5137b | 712 | reginfo->intuit = 1; |
1cb48e53 DM |
713 | /* not actually used within intuit, but zero for safety anyway */ |
714 | reginfo->poscache_maxiter = 0; | |
02d5137b | 715 | |
f2ed9b32 | 716 | if (utf8_target) { |
33b8afdf JH |
717 | if (!prog->check_utf8 && prog->check_substr) |
718 | to_utf8_substr(prog); | |
719 | check = prog->check_utf8; | |
720 | } else { | |
7e0d5ad7 KW |
721 | if (!prog->check_substr && prog->check_utf8) { |
722 | if (! to_byte_substr(prog)) { | |
6b54ddc5 | 723 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail); |
7e0d5ad7 KW |
724 | } |
725 | } | |
33b8afdf JH |
726 | check = prog->check_substr; |
727 | } | |
274cd312 | 728 | |
1dc475d0 DM |
729 | /* dump the various substring data */ |
730 | DEBUG_OPTIMISE_MORE_r({ | |
731 | int i; | |
732 | for (i=0; i<=2; i++) { | |
733 | SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr | |
734 | : prog->substrs->data[i].substr); | |
735 | if (!sv) | |
736 | continue; | |
737 | ||
738 | PerlIO_printf(Perl_debug_log, | |
739 | " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf | |
740 | " useful=%"IVdf" utf8=%d [%s]\n", | |
741 | i, | |
742 | (IV)prog->substrs->data[i].min_offset, | |
743 | (IV)prog->substrs->data[i].max_offset, | |
744 | (IV)prog->substrs->data[i].end_shift, | |
745 | BmUSEFUL(sv), | |
746 | utf8_target ? 1 : 0, | |
747 | SvPEEK(sv)); | |
748 | } | |
749 | }); | |
750 | ||
8e1490ee | 751 | if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */ |
9fc7410e DM |
752 | |
753 | /* ml_anch: check after \n? | |
754 | * | |
755 | * A note about IMPLICIT: on an un-anchored pattern beginning | |
756 | * with /.*.../, these flags will have been added by the | |
757 | * compiler: | |
758 | * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL | |
759 | * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL | |
760 | */ | |
7d2d37f5 DM |
761 | ml_anch = (prog->intflags & PREGf_ANCH_MBOL) |
762 | && !(prog->intflags & PREGf_IMPLICIT); | |
cad2e5aa | 763 | |
343c8a29 | 764 | if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) { |
c889ccc8 DM |
765 | /* we are only allowed to match at BOS or \G */ |
766 | ||
57fcbfa7 | 767 | /* trivially reject if there's a BOS anchor and we're not at BOS. |
7bb3b9eb DM |
768 | * |
769 | * Note that we don't try to do a similar quick reject for | |
770 | * \G, since generally the caller will have calculated strpos | |
771 | * based on pos() and gofs, so the string is already correctly | |
772 | * anchored by definition; and handling the exceptions would | |
773 | * be too fiddly (e.g. REXEC_IGNOREPOS). | |
57fcbfa7 | 774 | */ |
7bb3b9eb | 775 | if ( strpos != strbeg |
d3d47aac | 776 | && (prog->intflags & PREGf_ANCH_SBOL)) |
c889ccc8 DM |
777 | { |
778 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, | |
1dc475d0 | 779 | " Not at start...\n")); |
c889ccc8 DM |
780 | goto fail; |
781 | } | |
782 | ||
a5d12a4b DM |
783 | /* in the presence of an anchor, the anchored (relative to the |
784 | * start of the regex) substr must also be anchored relative | |
66b7ec5c DM |
785 | * to strpos. So quickly reject if substr isn't found there. |
786 | * This works for \G too, because the caller will already have | |
787 | * subtracted gofs from pos, and gofs is the offset from the | |
788 | * \G to the start of the regex. For example, in /.abc\Gdef/, | |
789 | * where substr="abcdef", pos()=3, gofs=4, offset_min=1: | |
790 | * caller will have set strpos=pos()-4; we look for the substr | |
791 | * at position pos()-4+1, which lines up with the "a" */ | |
a5d12a4b | 792 | |
c889ccc8 | 793 | if (prog->check_offset_min == prog->check_offset_max |
d0d44648 | 794 | && !(prog->intflags & PREGf_CANY_SEEN)) |
c889ccc8 DM |
795 | { |
796 | /* Substring at constant offset from beg-of-str... */ | |
d307bf57 | 797 | SSize_t slen = SvCUR(check); |
343c8a29 | 798 | char *s = HOP3c(strpos, prog->check_offset_min, strend); |
1de06328 | 799 | |
1dc475d0 DM |
800 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
801 | " Looking for check substr at fixed offset %"IVdf"...\n", | |
802 | (IV)prog->check_offset_min)); | |
803 | ||
7742aa66 DM |
804 | if (SvTAIL(check)) { |
805 | /* In this case, the regex is anchored at the end too. | |
806 | * Unless it's a multiline match, the lengths must match | |
807 | * exactly, give or take a \n. NB: slen >= 1 since | |
808 | * the last char of check is \n */ | |
809 | if (!multiline | |
810 | && ( strend - s > slen | |
811 | || strend - s < slen - 1 | |
812 | || (strend - s == slen && strend[-1] != '\n'))) | |
c889ccc8 DM |
813 | { |
814 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, | |
1dc475d0 | 815 | " String too long...\n")); |
c889ccc8 DM |
816 | goto fail_finish; |
817 | } | |
818 | /* Now should match s[0..slen-2] */ | |
819 | slen--; | |
c889ccc8 | 820 | } |
d307bf57 DM |
821 | if (slen && (*SvPVX_const(check) != *s |
822 | || (slen > 1 && memNE(SvPVX_const(check), s, slen)))) | |
823 | { | |
824 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, | |
1dc475d0 | 825 | " String not equal...\n")); |
d307bf57 DM |
826 | goto fail_finish; |
827 | } | |
c889ccc8 DM |
828 | |
829 | check_at = s; | |
830 | goto success_at_start; | |
cad2e5aa | 831 | } |
cad2e5aa | 832 | } |
cad2e5aa | 833 | } |
0fc004dd | 834 | |
c0e0ec46 | 835 | end_shift = prog->check_end_shift; |
cad2e5aa | 836 | |
19188028 | 837 | #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */ |
0033605d | 838 | if (end_shift < 0) |
1de06328 | 839 | Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ", |
220fc49f | 840 | (IV)end_shift, RX_PRECOMP(prog)); |
2c2d71f5 JH |
841 | #endif |
842 | ||
2c2d71f5 | 843 | restart: |
1de06328 | 844 | |
66b7ec5c DM |
845 | /* This is the (re)entry point of the main loop in this function. |
846 | * The goal of this loop is to: | |
847 | * 1) find the "check" substring in the region rx_origin..strend | |
848 | * (adjusted by start_shift / end_shift). If not found, reject | |
849 | * immediately. | |
850 | * 2) If it exists, look for the "other" substr too if defined; for | |
851 | * example, if the check substr maps to the anchored substr, then | |
852 | * check the floating substr, and vice-versa. If not found, go | |
853 | * back to (1) with rx_origin suitably incremented. | |
854 | * 3) If we find an rx_origin position that doesn't contradict | |
855 | * either of the substrings, then check the possible additional | |
856 | * constraints on rx_origin of /^.../m or a known start class. | |
857 | * If these fail, then depending on which constraints fail, jump | |
858 | * back to here, or to various other re-entry points further along | |
859 | * that skip some of the first steps. | |
860 | * 4) If we pass all those tests, update the BmUSEFUL() count on the | |
861 | * substring. If the start position was determined to be at the | |
862 | * beginning of the string - so, not rejected, but not optimised, | |
863 | * since we have to run regmatch from position 0 - decrement the | |
864 | * BmUSEFUL() count. Otherwise increment it. | |
865 | */ | |
866 | ||
1a4edc3c DM |
867 | |
868 | /* first, look for the 'check' substring */ | |
869 | ||
1de06328 | 870 | { |
c33e64f0 FC |
871 | U8* start_point; |
872 | U8* end_point; | |
c889ccc8 | 873 | |
c889ccc8 | 874 | DEBUG_OPTIMISE_MORE_r({ |
1dc475d0 | 875 | PerlIO_printf(Perl_debug_log, |
ae5d4331 | 876 | " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf |
1dc475d0 | 877 | " Start shift: %"IVdf" End shift %"IVdf |
4d281893 | 878 | " Real end Shift: %"IVdf"\n", |
197641b6 | 879 | (IV)(rx_origin - strpos), |
c889ccc8 | 880 | (IV)prog->check_offset_min, |
12fbc530 DM |
881 | (IV)start_shift, |
882 | (IV)end_shift, | |
c889ccc8 DM |
883 | (IV)prog->check_end_shift); |
884 | }); | |
1de06328 | 885 | |
0d331aaf | 886 | if (prog->intflags & PREGf_CANY_SEEN) { |
0fc004dd | 887 | start_point= (U8*)(rx_origin + start_shift); |
12fbc530 | 888 | end_point= (U8*)(strend - end_shift); |
557f47af DM |
889 | if (start_point > end_point) |
890 | goto fail_finish; | |
1de06328 | 891 | } else { |
557f47af DM |
892 | end_point = HOP3(strend, -end_shift, strbeg); |
893 | start_point = HOPMAYBE3(rx_origin, start_shift, end_point); | |
894 | if (!start_point) | |
895 | goto fail_finish; | |
1de06328 | 896 | } |
c889ccc8 | 897 | |
557f47af | 898 | |
e0362b86 | 899 | /* If the regex is absolutely anchored to either the start of the |
d3d47aac | 900 | * string (SBOL) or to pos() (ANCH_GPOS), then |
e0362b86 DM |
901 | * check_offset_max represents an upper bound on the string where |
902 | * the substr could start. For the ANCH_GPOS case, we assume that | |
903 | * the caller of intuit will have already set strpos to | |
904 | * pos()-gofs, so in this case strpos + offset_max will still be | |
905 | * an upper bound on the substr. | |
906 | */ | |
c19c836a DM |
907 | if (!ml_anch |
908 | && prog->intflags & PREGf_ANCH | |
e0362b86 | 909 | && prog->check_offset_max != SSize_t_MAX) |
c19c836a | 910 | { |
1a08ba3a | 911 | SSize_t len = SvCUR(check) - !!SvTAIL(check); |
e0362b86 DM |
912 | const char * const anchor = |
913 | (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg); | |
914 | ||
915 | /* do a bytes rather than chars comparison. It's conservative; | |
916 | * so it skips doing the HOP if the result can't possibly end | |
917 | * up earlier than the old value of end_point. | |
918 | */ | |
919 | if ((char*)end_point - anchor > prog->check_offset_max) { | |
920 | end_point = HOP3lim((U8*)anchor, | |
921 | prog->check_offset_max, | |
922 | end_point -len) | |
923 | + len; | |
924 | } | |
d6ef1678 DM |
925 | } |
926 | ||
6bda09f9 | 927 | DEBUG_OPTIMISE_MORE_r({ |
1dc475d0 | 928 | PerlIO_printf(Perl_debug_log, " fbm_instr len=%d str=<%.*s>\n", |
1de06328 | 929 | (int)(end_point - start_point), |
fc8cd66c | 930 | (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point), |
1de06328 YO |
931 | start_point); |
932 | }); | |
933 | ||
ae5d4331 | 934 | check_at = fbm_instr( start_point, end_point, |
7fba1cd6 | 935 | check, multiline ? FBMrf_MULTILINE : 0); |
c889ccc8 | 936 | |
8fd34720 DM |
937 | /* Update the count-of-usability, remove useless subpatterns, |
938 | unshift s. */ | |
939 | ||
940 | DEBUG_EXECUTE_r({ | |
941 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), | |
942 | SvPVX_const(check), RE_SV_DUMPLEN(check), 30); | |
943 | PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s", | |
944 | (check_at ? "Found" : "Did not find"), | |
945 | (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) | |
946 | ? "anchored" : "floating"), | |
947 | quoted, | |
948 | RE_SV_TAIL(check), | |
949 | (check_at ? " at offset " : "...\n") ); | |
950 | }); | |
2c2d71f5 | 951 | |
8fd34720 DM |
952 | if (!check_at) |
953 | goto fail_finish; | |
954 | /* Finish the diagnostic message */ | |
955 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(check_at - strpos)) ); | |
2c2d71f5 | 956 | |
8fd34720 DM |
957 | /* set rx_origin to the minimum position where the regex could start |
958 | * matching, given the constraint of the just-matched check substring. | |
959 | * But don't set it lower than previously. | |
960 | */ | |
fdc003fd | 961 | |
8fd34720 DM |
962 | if (check_at - rx_origin > prog->check_offset_max) |
963 | rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin); | |
964 | } | |
fdc003fd DM |
965 | |
966 | ||
1a4edc3c | 967 | /* now look for the 'other' substring if defined */ |
2c2d71f5 | 968 | |
6480a6c4 DM |
969 | if (utf8_target ? prog->substrs->data[other_ix].utf8_substr |
970 | : prog->substrs->data[other_ix].substr) | |
1de06328 | 971 | { |
30944b6d | 972 | /* Take into account the "other" substring. */ |
6c3343a6 DM |
973 | char *last, *last1; |
974 | char *s; | |
975 | SV* must; | |
976 | struct reg_substr_datum *other; | |
977 | ||
978 | do_other_substr: | |
979 | other = &prog->substrs->data[other_ix]; | |
980 | ||
981 | /* if "other" is anchored: | |
982 | * we've previously found a floating substr starting at check_at. | |
983 | * This means that the regex origin must lie somewhere | |
984 | * between min (rx_origin): HOP3(check_at, -check_offset_max) | |
985 | * and max: HOP3(check_at, -check_offset_min) | |
986 | * (except that min will be >= strpos) | |
987 | * So the fixed substr must lie somewhere between | |
988 | * HOP3(min, anchored_offset) | |
989 | * HOP3(max, anchored_offset) + SvCUR(substr) | |
990 | */ | |
991 | ||
992 | /* if "other" is floating | |
993 | * Calculate last1, the absolute latest point where the | |
994 | * floating substr could start in the string, ignoring any | |
995 | * constraints from the earlier fixed match. It is calculated | |
996 | * as follows: | |
997 | * | |
998 | * strend - prog->minlen (in chars) is the absolute latest | |
999 | * position within the string where the origin of the regex | |
1000 | * could appear. The latest start point for the floating | |
1001 | * substr is float_min_offset(*) on from the start of the | |
1002 | * regex. last1 simply combines thee two offsets. | |
1003 | * | |
1004 | * (*) You might think the latest start point should be | |
1005 | * float_max_offset from the regex origin, and technically | |
1006 | * you'd be correct. However, consider | |
1007 | * /a\d{2,4}bcd\w/ | |
1008 | * Here, float min, max are 3,5 and minlen is 7. | |
1009 | * This can match either | |
1010 | * /a\d\dbcd\w/ | |
1011 | * /a\d\d\dbcd\w/ | |
1012 | * /a\d\d\d\dbcd\w/ | |
1013 | * In the first case, the regex matches minlen chars; in the | |
1014 | * second, minlen+1, in the third, minlen+2. | |
1015 | * In the first case, the floating offset is 3 (which equals | |
1016 | * float_min), in the second, 4, and in the third, 5 (which | |
1017 | * equals float_max). In all cases, the floating string bcd | |
1018 | * can never start more than 4 chars from the end of the | |
1019 | * string, which equals minlen - float_min. As the substring | |
1020 | * starts to match more than float_min from the start of the | |
1021 | * regex, it makes the regex match more than minlen chars, | |
1022 | * and the two cancel each other out. So we can always use | |
1023 | * float_min - minlen, rather than float_max - minlen for the | |
1024 | * latest position in the string. | |
1025 | * | |
1026 | * Note that -minlen + float_min_offset is equivalent (AFAIKT) | |
1027 | * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift | |
1028 | */ | |
1029 | ||
e7a14a9c | 1030 | assert(prog->minlen >= other->min_offset); |
6c3343a6 DM |
1031 | last1 = HOP3c(strend, |
1032 | other->min_offset - prog->minlen, strbeg); | |
1033 | ||
4d006249 | 1034 | if (other_ix) {/* i.e. if (other-is-float) */ |
6c3343a6 DM |
1035 | /* last is the latest point where the floating substr could |
1036 | * start, *given* any constraints from the earlier fixed | |
1037 | * match. This constraint is that the floating string starts | |
1038 | * <= float_max_offset chars from the regex origin (rx_origin). | |
1039 | * If this value is less than last1, use it instead. | |
eb3831ce | 1040 | */ |
6c3343a6 DM |
1041 | assert(rx_origin <= last1); |
1042 | last = | |
1043 | /* this condition handles the offset==infinity case, and | |
1044 | * is a short-cut otherwise. Although it's comparing a | |
1045 | * byte offset to a char length, it does so in a safe way, | |
1046 | * since 1 char always occupies 1 or more bytes, | |
1047 | * so if a string range is (last1 - rx_origin) bytes, | |
1048 | * it will be less than or equal to (last1 - rx_origin) | |
1049 | * chars; meaning it errs towards doing the accurate HOP3 | |
1050 | * rather than just using last1 as a short-cut */ | |
1051 | (last1 - rx_origin) < other->max_offset | |
1052 | ? last1 | |
1053 | : (char*)HOP3lim(rx_origin, other->max_offset, last1); | |
1054 | } | |
1055 | else { | |
1056 | assert(strpos + start_shift <= check_at); | |
1057 | last = HOP4c(check_at, other->min_offset - start_shift, | |
1058 | strbeg, strend); | |
1059 | } | |
ead917d0 | 1060 | |
6c3343a6 DM |
1061 | s = HOP3c(rx_origin, other->min_offset, strend); |
1062 | if (s < other_last) /* These positions already checked */ | |
1063 | s = other_last; | |
1064 | ||
1065 | must = utf8_target ? other->utf8_substr : other->substr; | |
1066 | assert(SvPOK(must)); | |
1067 | s = fbm_instr( | |
1068 | (unsigned char*)s, | |
1069 | (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0), | |
1070 | must, | |
1071 | multiline ? FBMrf_MULTILINE : 0 | |
1072 | ); | |
1073 | DEBUG_EXECUTE_r({ | |
1074 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), | |
1075 | SvPVX_const(must), RE_SV_DUMPLEN(must), 30); | |
1076 | PerlIO_printf(Perl_debug_log, " %s %s substr %s%s", | |
1077 | s ? "Found" : "Contradicts", | |
1078 | other_ix ? "floating" : "anchored", | |
1079 | quoted, RE_SV_TAIL(must)); | |
1080 | }); | |
ead917d0 | 1081 | |
ead917d0 | 1082 | |
6c3343a6 DM |
1083 | if (!s) { |
1084 | /* last1 is latest possible substr location. If we didn't | |
1085 | * find it before there, we never will */ | |
1086 | if (last >= last1) { | |
1087 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, | |
1088 | ", giving up...\n")); | |
1089 | goto fail_finish; | |
ead917d0 DM |
1090 | } |
1091 | ||
6c3343a6 DM |
1092 | /* try to find the check substr again at a later |
1093 | * position. Maybe next time we'll find the "other" substr | |
1094 | * in range too */ | |
1095 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, | |
1096 | ", trying %s at offset %ld...\n", | |
1097 | (other_ix ? "floating" : "anchored"), | |
197641b6 | 1098 | (long)(HOP3c(check_at, 1, strend) - strpos))); |
6c3343a6 DM |
1099 | |
1100 | other_last = HOP3c(last, 1, strend) /* highest failure */; | |
1101 | rx_origin = | |
4d006249 | 1102 | other_ix /* i.e. if other-is-float */ |
6c3343a6 DM |
1103 | ? HOP3c(rx_origin, 1, strend) |
1104 | : HOP4c(last, 1 - other->min_offset, strbeg, strend); | |
1105 | goto restart; | |
1106 | } | |
1107 | else { | |
1108 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n", | |
197641b6 | 1109 | (long)(s - strpos))); |
ead917d0 | 1110 | |
4d006249 | 1111 | if (other_ix) { /* if (other-is-float) */ |
6c3343a6 DM |
1112 | /* other_last is set to s, not s+1, since its possible for |
1113 | * a floating substr to fail first time, then succeed | |
1114 | * second time at the same floating position; e.g.: | |
1115 | * "-AB--AABZ" =~ /\wAB\d*Z/ | |
1116 | * The first time round, anchored and float match at | |
1117 | * "-(AB)--AAB(Z)" then fail on the initial \w character | |
1118 | * class. Second time round, they match at "-AB--A(AB)(Z)". | |
1119 | */ | |
1120 | other_last = s; | |
ead917d0 DM |
1121 | } |
1122 | else { | |
6c3343a6 DM |
1123 | rx_origin = HOP3c(s, -other->min_offset, strbeg); |
1124 | other_last = HOP3c(s, 1, strend); | |
ead917d0 | 1125 | } |
6c3343a6 | 1126 | } |
cad2e5aa | 1127 | } |
acba93e8 DM |
1128 | else { |
1129 | DEBUG_OPTIMISE_MORE_r( | |
1130 | PerlIO_printf(Perl_debug_log, | |
1131 | " Check-only match: offset min:%"IVdf" max:%"IVdf | |
1c1c599d | 1132 | " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf |
acba93e8 DM |
1133 | " strend-strpos:%"IVdf"\n", |
1134 | (IV)prog->check_offset_min, | |
1135 | (IV)prog->check_offset_max, | |
1c1c599d | 1136 | (IV)(check_at-strpos), |
acba93e8 | 1137 | (IV)(rx_origin-strpos), |
1c1c599d | 1138 | (IV)(rx_origin-check_at), |
acba93e8 DM |
1139 | (IV)(strend-strpos) |
1140 | ) | |
1141 | ); | |
1142 | } | |
2c2d71f5 | 1143 | |
acba93e8 | 1144 | postprocess_substr_matches: |
0991020e | 1145 | |
1a4edc3c | 1146 | /* handle the extra constraint of /^.../m if present */ |
e3c6feb0 | 1147 | |
7d2d37f5 | 1148 | if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') { |
4620cb61 DM |
1149 | char *s; |
1150 | ||
a62659bd DM |
1151 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
1152 | " looking for /^/m anchor")); | |
d0880ea7 DM |
1153 | |
1154 | /* we have failed the constraint of a \n before rx_origin. | |
2e759faa DM |
1155 | * Find the next \n, if any, even if it's beyond the current |
1156 | * anchored and/or floating substrings. Whether we should be | |
1157 | * scanning ahead for the next \n or the next substr is debatable. | |
1158 | * On the one hand you'd expect rare substrings to appear less | |
1159 | * often than \n's. On the other hand, searching for \n means | |
1160 | * we're effectively flipping been check_substr and "\n" on each | |
1161 | * iteration as the current "rarest" string candidate, which | |
1162 | * means for example that we'll quickly reject the whole string if | |
1163 | * hasn't got a \n, rather than trying every substr position | |
1164 | * first | |
1165 | */ | |
d0880ea7 | 1166 | |
4620cb61 DM |
1167 | s = HOP3c(strend, - prog->minlen, strpos); |
1168 | if (s <= rx_origin || | |
1169 | ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin))) | |
1170 | { | |
d0880ea7 DM |
1171 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
1172 | " Did not find /%s^%s/m...\n", | |
1173 | PL_colors[0], PL_colors[1])); | |
a62659bd DM |
1174 | goto fail_finish; |
1175 | } | |
d0880ea7 | 1176 | |
4ada1233 DM |
1177 | /* earliest possible origin is 1 char after the \n. |
1178 | * (since *rx_origin == '\n', it's safe to ++ here rather than | |
1179 | * HOP(rx_origin, 1)) */ | |
1180 | rx_origin++; | |
d0880ea7 | 1181 | |
f4f115de | 1182 | if (prog->substrs->check_ix == 0 /* check is anchored */ |
4ada1233 | 1183 | || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos)) |
f4f115de | 1184 | { |
d0880ea7 DM |
1185 | /* Position contradicts check-string; either because |
1186 | * check was anchored (and thus has no wiggle room), | |
4ada1233 | 1187 | * or check was float and rx_origin is above the float range */ |
d0880ea7 DM |
1188 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
1189 | " Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n", | |
4ada1233 | 1190 | PL_colors[0], PL_colors[1], (long)(rx_origin - strpos))); |
d0880ea7 DM |
1191 | goto restart; |
1192 | } | |
1193 | ||
1194 | /* if we get here, the check substr must have been float, | |
2e759faa | 1195 | * is in range, and we may or may not have had an anchored |
d0880ea7 DM |
1196 | * "other" substr which still contradicts */ |
1197 | assert(prog->substrs->check_ix); /* check is float */ | |
1198 | ||
1199 | if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) { | |
1200 | /* whoops, the anchored "other" substr exists, so we still | |
1201 | * contradict. On the other hand, the float "check" substr | |
1202 | * didn't contradict, so just retry the anchored "other" | |
1203 | * substr */ | |
1204 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, | |
1205 | " Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n", | |
1206 | PL_colors[0], PL_colors[1], | |
1207 | (long)(rx_origin - strpos), | |
1208 | (long)(rx_origin - strpos + prog->anchored_offset))); | |
1209 | goto do_other_substr; | |
1210 | } | |
1211 | ||
1212 | /* success: we don't contradict the found floating substring | |
1213 | * (and there's no anchored substr). */ | |
d0880ea7 DM |
1214 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
1215 | " Found /%s^%s/m at offset %ld...\n", | |
1216 | PL_colors[0], PL_colors[1], (long)(rx_origin - strpos))); | |
e3c6feb0 DM |
1217 | } |
1218 | else { | |
2e759faa | 1219 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
fe4f3442 | 1220 | " (multiline anchor test skipped)\n")); |
e3c6feb0 DM |
1221 | } |
1222 | ||
ffad1e6a | 1223 | success_at_start: |
e3c6feb0 | 1224 | |
cad2e5aa | 1225 | |
dd170ff5 DM |
1226 | /* if we have a starting character class, then test that extra constraint. |
1227 | * (trie stclasses are too expensive to use here, we are better off to | |
1228 | * leave it to regmatch itself) */ | |
1229 | ||
f8fc2ecf | 1230 | if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) { |
f8fc2ecf | 1231 | const U8* const str = (U8*)STRING(progi->regstclass); |
0991020e | 1232 | |
2c75e362 | 1233 | /* XXX this value could be pre-computed */ |
f8fc2ecf | 1234 | const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT |
2c75e362 DM |
1235 | ? (reginfo->is_utf8_pat |
1236 | ? utf8_distance(str + STR_LEN(progi->regstclass), str) | |
1237 | : STR_LEN(progi->regstclass)) | |
66e933ab | 1238 | : 1); |
1de06328 | 1239 | char * endpos; |
fa3bb21d | 1240 | char *s; |
000dfd2d DM |
1241 | /* latest pos that a matching float substr constrains rx start to */ |
1242 | char *rx_max_float = NULL; | |
1243 | ||
c75a3985 DM |
1244 | /* if the current rx_origin is anchored, either by satisfying an |
1245 | * anchored substring constraint, or a /^.../m constraint, then we | |
1246 | * can reject the current origin if the start class isn't found | |
1247 | * at the current position. If we have a float-only match, then | |
1248 | * rx_origin is constrained to a range; so look for the start class | |
1249 | * in that range. if neither, then look for the start class in the | |
1250 | * whole rest of the string */ | |
1251 | ||
dd170ff5 DM |
1252 | /* XXX DAPM it's not clear what the minlen test is for, and why |
1253 | * it's not used in the floating case. Nothing in the test suite | |
1254 | * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>. | |
1255 | * Here are some old comments, which may or may not be correct: | |
1256 | * | |
1257 | * minlen == 0 is possible if regstclass is \b or \B, | |
1258 | * and the fixed substr is ''$. | |
1259 | * Since minlen is already taken into account, rx_origin+1 is | |
1260 | * before strend; accidentally, minlen >= 1 guaranties no false | |
1261 | * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 : | |
1262 | * 0) below assumes that regstclass does not come from lookahead... | |
1263 | * If regstclass takes bytelength more than 1: If charlength==1, OK. | |
1264 | * This leaves EXACTF-ish only, which are dealt with in | |
1265 | * find_byclass(). | |
1266 | */ | |
1267 | ||
7d2d37f5 | 1268 | if (prog->anchored_substr || prog->anchored_utf8 || ml_anch) |
fa3bb21d | 1269 | endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend); |
000dfd2d DM |
1270 | else if (prog->float_substr || prog->float_utf8) { |
1271 | rx_max_float = HOP3c(check_at, -start_shift, strbeg); | |
1272 | endpos= HOP3c(rx_max_float, cl_l, strend); | |
1273 | } | |
1de06328 YO |
1274 | else |
1275 | endpos= strend; | |
1276 | ||
1dc475d0 DM |
1277 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
1278 | " looking for class: start_shift: %"IVdf" check_at: %"IVdf | |
c43b5520 | 1279 | " rx_origin: %"IVdf" endpos: %"IVdf"\n", |
1dc475d0 | 1280 | (IV)start_shift, (IV)(check_at - strbeg), |
c43b5520 | 1281 | (IV)(rx_origin - strbeg), (IV)(endpos - strbeg))); |
d8080198 | 1282 | |
c43b5520 | 1283 | s = find_byclass(prog, progi->regstclass, rx_origin, endpos, |
f9176b44 | 1284 | reginfo); |
be778b1a | 1285 | if (!s) { |
6eb5f6b9 | 1286 | if (endpos == strend) { |
a3621e74 | 1287 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
1dc475d0 | 1288 | " Could not match STCLASS...\n") ); |
6eb5f6b9 JH |
1289 | goto fail; |
1290 | } | |
a3621e74 | 1291 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
1dc475d0 | 1292 | " This position contradicts STCLASS...\n") ); |
e0eb31e7 DM |
1293 | if ((prog->intflags & PREGf_ANCH) && !ml_anch |
1294 | && !(prog->intflags & PREGf_IMPLICIT)) | |
653099ff | 1295 | goto fail; |
9fed8d02 | 1296 | |
6eb5f6b9 | 1297 | /* Contradict one of substrings */ |
97136c8a DM |
1298 | if (prog->anchored_substr || prog->anchored_utf8) { |
1299 | if (prog->substrs->check_ix == 1) { /* check is float */ | |
1300 | /* Have both, check_string is floating */ | |
1301 | assert(rx_origin + start_shift <= check_at); | |
1302 | if (rx_origin + start_shift != check_at) { | |
1303 | /* not at latest position float substr could match: | |
c75a3985 DM |
1304 | * Recheck anchored substring, but not floating. |
1305 | * The condition above is in bytes rather than | |
1306 | * chars for efficiency. It's conservative, in | |
1307 | * that it errs on the side of doing 'goto | |
1308 | * do_other_substr', where a more accurate | |
1309 | * char-based calculation will be done */ | |
97136c8a DM |
1310 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
1311 | " Looking for anchored substr starting at offset %ld...\n", | |
1312 | (long)(other_last - strpos)) ); | |
1313 | goto do_other_substr; | |
3369914b | 1314 | } |
3369914b DM |
1315 | } |
1316 | } | |
97136c8a | 1317 | else { |
9fed8d02 DM |
1318 | /* float-only */ |
1319 | ||
7d2d37f5 | 1320 | if (ml_anch) { |
c75a3985 DM |
1321 | /* In the presence of ml_anch, we might be able to |
1322 | * find another \n without breaking the current float | |
1323 | * constraint. */ | |
1324 | ||
1325 | /* strictly speaking this should be HOP3c(..., 1, ...), | |
1326 | * but since we goto a block of code that's going to | |
1327 | * search for the next \n if any, its safe here */ | |
9fed8d02 | 1328 | rx_origin++; |
9fed8d02 DM |
1329 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
1330 | " Looking for /%s^%s/m starting at offset %ld...\n", | |
1331 | PL_colors[0], PL_colors[1], | |
1332 | (long)(rx_origin - strpos)) ); | |
9fed8d02 | 1333 | goto postprocess_substr_matches; |
ab60c45a | 1334 | } |
c75a3985 DM |
1335 | |
1336 | /* strictly speaking this can never be true; but might | |
1337 | * be if we ever allow intuit without substrings */ | |
1338 | if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) | |
9fed8d02 | 1339 | goto fail; |
c75a3985 | 1340 | |
000dfd2d | 1341 | rx_origin = rx_max_float; |
9fed8d02 DM |
1342 | } |
1343 | ||
c75a3985 DM |
1344 | /* at this point, any matching substrings have been |
1345 | * contradicted. Start again... */ | |
1346 | ||
9fed8d02 | 1347 | rx_origin = HOP3c(rx_origin, 1, strend); |
557f47af DM |
1348 | |
1349 | /* uses bytes rather than char calculations for efficiency. | |
1350 | * It's conservative: it errs on the side of doing 'goto restart', | |
1351 | * where there is code that does a proper char-based test */ | |
9fed8d02 | 1352 | if (rx_origin + start_shift + end_shift > strend) { |
40268e5b | 1353 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
9fed8d02 DM |
1354 | " Could not match STCLASS...\n") ); |
1355 | goto fail; | |
1356 | } | |
1357 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, | |
1358 | " Looking for %s substr starting at offset %ld...\n", | |
1359 | (prog->substrs->check_ix ? "floating" : "anchored"), | |
1360 | (long)(rx_origin + start_shift - strpos)) ); | |
1361 | goto restart; | |
6eb5f6b9 | 1362 | } |
9fed8d02 | 1363 | |
c75a3985 DM |
1364 | /* Success !!! */ |
1365 | ||
5f9c6575 | 1366 | if (rx_origin != s) { |
a3621e74 | 1367 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
1dc475d0 | 1368 | " By STCLASS: moving %ld --> %ld\n", |
5f9c6575 | 1369 | (long)(rx_origin - strpos), (long)(s - strpos)) |
b7953727 JH |
1370 | ); |
1371 | } | |
1372 | else { | |
a3621e74 | 1373 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
1dc475d0 | 1374 | " Does not contradict STCLASS...\n"); |
b7953727 JH |
1375 | ); |
1376 | } | |
6eb5f6b9 | 1377 | } |
ffad1e6a DM |
1378 | |
1379 | /* Decide whether using the substrings helped */ | |
1380 | ||
1381 | if (rx_origin != strpos) { | |
1382 | /* Fixed substring is found far enough so that the match | |
1383 | cannot start at strpos. */ | |
1384 | ||
1385 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n")); | |
1386 | ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */ | |
1387 | } | |
1388 | else { | |
70563e16 DM |
1389 | /* The found rx_origin position does not prohibit matching at |
1390 | * strpos, so calling intuit didn't gain us anything. Decrement | |
1391 | * the BmUSEFUL() count on the check substring, and if we reach | |
1392 | * zero, free it. */ | |
1393 | if (!(prog->intflags & PREGf_NAUGHTY) | |
ffad1e6a DM |
1394 | && (utf8_target ? ( |
1395 | prog->check_utf8 /* Could be deleted already */ | |
1396 | && --BmUSEFUL(prog->check_utf8) < 0 | |
1397 | && (prog->check_utf8 == prog->float_utf8) | |
1398 | ) : ( | |
1399 | prog->check_substr /* Could be deleted already */ | |
1400 | && --BmUSEFUL(prog->check_substr) < 0 | |
1401 | && (prog->check_substr == prog->float_substr) | |
1402 | ))) | |
1403 | { | |
1404 | /* If flags & SOMETHING - do not do it many times on the same match */ | |
1405 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n")); | |
1406 | /* XXX Does the destruction order has to change with utf8_target? */ | |
1407 | SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr); | |
1408 | SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8); | |
1409 | prog->check_substr = prog->check_utf8 = NULL; /* disable */ | |
1410 | prog->float_substr = prog->float_utf8 = NULL; /* clear */ | |
1411 | check = NULL; /* abort */ | |
ffad1e6a DM |
1412 | /* XXXX This is a remnant of the old implementation. It |
1413 | looks wasteful, since now INTUIT can use many | |
1414 | other heuristics. */ | |
1415 | prog->extflags &= ~RXf_USE_INTUIT; | |
ffad1e6a DM |
1416 | } |
1417 | } | |
1418 | ||
1419 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, | |
1420 | "Intuit: %sSuccessfully guessed:%s match at offset %ld\n", | |
1421 | PL_colors[4], PL_colors[5], (long)(rx_origin - strpos)) ); | |
1422 | ||
c765d6e0 | 1423 | return rx_origin; |
2c2d71f5 JH |
1424 | |
1425 | fail_finish: /* Substring not found */ | |
33b8afdf | 1426 | if (prog->check_substr || prog->check_utf8) /* could be removed already */ |
f2ed9b32 | 1427 | BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */ |
cad2e5aa | 1428 | fail: |
a3621e74 | 1429 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n", |
e4584336 | 1430 | PL_colors[4], PL_colors[5])); |
bd61b366 | 1431 | return NULL; |
cad2e5aa | 1432 | } |
9661b544 | 1433 | |
70563e16 | 1434 | |
a0a388a1 | 1435 | #define DECL_TRIE_TYPE(scan) \ |
098b07d5 KW |
1436 | const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \ |
1437 | trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold } \ | |
fab2782b YO |
1438 | trie_type = ((scan->flags == EXACT) \ |
1439 | ? (utf8_target ? trie_utf8 : trie_plain) \ | |
098b07d5 KW |
1440 | : (scan->flags == EXACTFA) \ |
1441 | ? (utf8_target ? trie_utf8_exactfa_fold : trie_latin_utf8_exactfa_fold) \ | |
1442 | : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold)) | |
fab2782b | 1443 | |
fd3249ee | 1444 | #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \ |
baa60164 | 1445 | STMT_START { \ |
fab2782b | 1446 | STRLEN skiplen; \ |
baa60164 | 1447 | U8 flags = FOLD_FLAGS_FULL; \ |
fab2782b | 1448 | switch (trie_type) { \ |
31f05a37 | 1449 | case trie_utf8_exactfa_fold: \ |
baa60164 | 1450 | flags |= FOLD_FLAGS_NOMIX_ASCII; \ |
924ba076 | 1451 | /* FALLTHROUGH */ \ |
fab2782b YO |
1452 | case trie_utf8_fold: \ |
1453 | if ( foldlen>0 ) { \ | |
c80e42f3 | 1454 | uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \ |
fab2782b YO |
1455 | foldlen -= len; \ |
1456 | uscan += len; \ | |
1457 | len=0; \ | |
1458 | } else { \ | |
445bf929 | 1459 | uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \ |
fab2782b YO |
1460 | len = UTF8SKIP(uc); \ |
1461 | skiplen = UNISKIP( uvc ); \ | |
1462 | foldlen -= skiplen; \ | |
1463 | uscan = foldbuf + skiplen; \ | |
1464 | } \ | |
1465 | break; \ | |
baa60164 KW |
1466 | case trie_latin_utf8_exactfa_fold: \ |
1467 | flags |= FOLD_FLAGS_NOMIX_ASCII; \ | |
924ba076 | 1468 | /* FALLTHROUGH */ \ |
fab2782b YO |
1469 | case trie_latin_utf8_fold: \ |
1470 | if ( foldlen>0 ) { \ | |
c80e42f3 | 1471 | uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \ |
fab2782b YO |
1472 | foldlen -= len; \ |
1473 | uscan += len; \ | |
1474 | len=0; \ | |
1475 | } else { \ | |
1476 | len = 1; \ | |
31f05a37 | 1477 | uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \ |
fab2782b YO |
1478 | skiplen = UNISKIP( uvc ); \ |
1479 | foldlen -= skiplen; \ | |
1480 | uscan = foldbuf + skiplen; \ | |
1481 | } \ | |
1482 | break; \ | |
1483 | case trie_utf8: \ | |
c80e42f3 | 1484 | uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \ |
fab2782b YO |
1485 | break; \ |
1486 | case trie_plain: \ | |
1487 | uvc = (UV)*uc; \ | |
1488 | len = 1; \ | |
1489 | } \ | |
1490 | if (uvc < 256) { \ | |
1491 | charid = trie->charmap[ uvc ]; \ | |
1492 | } \ | |
1493 | else { \ | |
1494 | charid = 0; \ | |
1495 | if (widecharmap) { \ | |
1496 | SV** const svpp = hv_fetch(widecharmap, \ | |
1497 | (char*)&uvc, sizeof(UV), 0); \ | |
1498 | if (svpp) \ | |
1499 | charid = (U16)SvIV(*svpp); \ | |
1500 | } \ | |
1501 | } \ | |
4cadc6a9 YO |
1502 | } STMT_END |
1503 | ||
ae7c5b9b KW |
1504 | #define DUMP_EXEC_POS(li,s,doutf8) \ |
1505 | dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \ | |
1506 | startpos, doutf8) | |
1507 | ||
c84a03c5 | 1508 | #define REXEC_FBC_EXACTISH_SCAN(COND) \ |
4cadc6a9 YO |
1509 | STMT_START { \ |
1510 | while (s <= e) { \ | |
c84a03c5 | 1511 | if ( (COND) \ |
fac1af77 | 1512 | && (ln == 1 || folder(s, pat_string, ln)) \ |
02d5137b | 1513 | && (reginfo->intuit || regtry(reginfo, &s)) )\ |
4cadc6a9 YO |
1514 | goto got_it; \ |
1515 | s++; \ | |
1516 | } \ | |
1517 | } STMT_END | |
1518 | ||
c84a03c5 | 1519 | #define REXEC_FBC_UTF8_SCAN(CODE) \ |
4cadc6a9 | 1520 | STMT_START { \ |
9a902117 | 1521 | while (s < strend) { \ |
c84a03c5 | 1522 | CODE \ |
9a902117 | 1523 | s += UTF8SKIP(s); \ |
4cadc6a9 YO |
1524 | } \ |
1525 | } STMT_END | |
1526 | ||
c84a03c5 | 1527 | #define REXEC_FBC_SCAN(CODE) \ |
4cadc6a9 YO |
1528 | STMT_START { \ |
1529 | while (s < strend) { \ | |
c84a03c5 | 1530 | CODE \ |
4cadc6a9 YO |
1531 | s++; \ |
1532 | } \ | |
1533 | } STMT_END | |
1534 | ||
05bd126c KW |
1535 | #define REXEC_FBC_UTF8_CLASS_SCAN(COND) \ |
1536 | REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \ | |
1537 | if (COND) { \ | |
1538 | if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \ | |
1539 | goto got_it; \ | |
1540 | else \ | |
1541 | tmp = doevery; \ | |
1542 | } \ | |
1543 | else \ | |
1544 | tmp = 1; \ | |
4cadc6a9 YO |
1545 | ) |
1546 | ||
05bd126c KW |
1547 | #define REXEC_FBC_CLASS_SCAN(COND) \ |
1548 | REXEC_FBC_SCAN( /* Loops while (s < strend) */ \ | |
1549 | if (COND) { \ | |
1550 | if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \ | |
1551 | goto got_it; \ | |
1552 | else \ | |
1553 | tmp = doevery; \ | |
1554 | } \ | |
1555 | else \ | |
1556 | tmp = 1; \ | |
4cadc6a9 YO |
1557 | ) |
1558 | ||
c84a03c5 | 1559 | #define REXEC_FBC_CSCAN(CONDUTF8,COND) \ |
baa60164 | 1560 | if (utf8_target) { \ |
c84a03c5 | 1561 | REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \ |
e1d1eefb YO |
1562 | } \ |
1563 | else { \ | |
c84a03c5 | 1564 | REXEC_FBC_CLASS_SCAN(COND); \ |
d981ef24 | 1565 | } |
05bd126c | 1566 | |
05bd126c KW |
1567 | /* The three macros below are slightly different versions of the same logic. |
1568 | * | |
1569 | * The first is for /a and /aa when the target string is UTF-8. This can only | |
1570 | * match ascii, but it must advance based on UTF-8. The other two handle the | |
1571 | * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking | |
1572 | * for the boundary (or non-boundary) between a word and non-word character. | |
1573 | * The utf8 and non-utf8 cases have the same logic, but the details must be | |
1574 | * different. Find the "wordness" of the character just prior to this one, and | |
1575 | * compare it with the wordness of this one. If they differ, we have a | |
1576 | * boundary. At the beginning of the string, pretend that the previous | |
1577 | * character was a new-line. | |
1578 | * | |
1579 | * All these macros uncleanly have side-effects with each other and outside | |
1580 | * variables. So far it's been too much trouble to clean-up | |
1581 | * | |
1582 | * TEST_NON_UTF8 is the macro or function to call to test if its byte input is | |
1583 | * a word character or not. | |
1584 | * IF_SUCCESS is code to do if it finds that we are at a boundary between | |
1585 | * word/non-word | |
1586 | * IF_FAIL is code to do if we aren't at a boundary between word/non-word | |
1587 | * | |
1588 | * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we | |
1589 | * are looking for a boundary or for a non-boundary. If we are looking for a | |
1590 | * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and | |
1591 | * see if this tentative match actually works, and if so, to quit the loop | |
1592 | * here. And vice-versa if we are looking for a non-boundary. | |
1593 | * | |
1594 | * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and | |
1595 | * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of | |
1596 | * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be | |
1597 | * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal | |
1598 | * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that | |
1599 | * complement. But in that branch we complement tmp, meaning that at the | |
1600 | * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s), | |
1601 | * which means at the top of the loop in the next iteration, it is | |
1602 | * TEST_NON_UTF8(s-1) */ | |
b2f4e957 | 1603 | #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \ |
05bd126c KW |
1604 | tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \ |
1605 | tmp = TEST_NON_UTF8(tmp); \ | |
1606 | REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \ | |
1607 | if (tmp == ! TEST_NON_UTF8((U8) *s)) { \ | |
1608 | tmp = !tmp; \ | |
1609 | IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \ | |
1610 | } \ | |
1611 | else { \ | |
1612 | IF_FAIL; \ | |
1613 | } \ | |
1614 | ); \ | |
1615 | ||
1616 | /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and | |
1617 | * TEST_UTF8 is a macro that for the same input code points returns identically | |
1618 | * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */ | |
236d82fd | 1619 | #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \ |
05bd126c KW |
1620 | if (s == reginfo->strbeg) { \ |
1621 | tmp = '\n'; \ | |
1622 | } \ | |
1623 | else { /* Back-up to the start of the previous character */ \ | |
1624 | U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \ | |
1625 | tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \ | |
3db24e1e | 1626 | 0, UTF8_ALLOW_DEFAULT); \ |
05bd126c KW |
1627 | } \ |
1628 | tmp = TEST_UV(tmp); \ | |
1629 | LOAD_UTF8_CHARCLASS_ALNUM(); \ | |
1630 | REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \ | |
1631 | if (tmp == ! (TEST_UTF8((U8 *) s))) { \ | |
1632 | tmp = !tmp; \ | |
1633 | IF_SUCCESS; \ | |
1634 | } \ | |
1635 | else { \ | |
1636 | IF_FAIL; \ | |
1637 | } \ | |
1638 | ); | |
cfaf538b | 1639 | |
05bd126c KW |
1640 | /* Like the above two macros. UTF8_CODE is the complete code for handling |
1641 | * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc | |
1642 | * macros below */ | |
baa60164 | 1643 | #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \ |
63ac0dad | 1644 | if (utf8_target) { \ |
05bd126c | 1645 | UTF8_CODE \ |
63ac0dad KW |
1646 | } \ |
1647 | else { /* Not utf8 */ \ | |
9d9163fb | 1648 | tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \ |
63ac0dad | 1649 | tmp = TEST_NON_UTF8(tmp); \ |
05bd126c | 1650 | REXEC_FBC_SCAN( /* advances s while s < strend */ \ |
63ac0dad | 1651 | if (tmp == ! TEST_NON_UTF8((U8) *s)) { \ |
63ac0dad | 1652 | IF_SUCCESS; \ |
760cfa8e | 1653 | tmp = !tmp; \ |
63ac0dad KW |
1654 | } \ |
1655 | else { \ | |
1656 | IF_FAIL; \ | |
1657 | } \ | |
1658 | ); \ | |
1659 | } \ | |
c8519dc7 KW |
1660 | /* Here, things have been set up by the previous code so that tmp is the \ |
1661 | * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \ | |
1662 | * utf8ness of the target). We also have to check if this matches against \ | |
1663 | * the EOS, which we treat as a \n (which is the same value in both UTF-8 \ | |
1664 | * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \ | |
1665 | * string */ \ | |
1666 | if (tmp == ! TEST_NON_UTF8('\n')) { \ | |
1667 | IF_SUCCESS; \ | |
1668 | } \ | |
1669 | else { \ | |
1670 | IF_FAIL; \ | |
1671 | } | |
63ac0dad | 1672 | |
ae7c5b9b KW |
1673 | /* This is the macro to use when we want to see if something that looks like it |
1674 | * could match, actually does, and if so exits the loop */ | |
1675 | #define REXEC_FBC_TRYIT \ | |
1676 | if ((reginfo->intuit || regtry(reginfo, &s))) \ | |
1677 | goto got_it | |
1678 | ||
1679 | /* The only difference between the BOUND and NBOUND cases is that | |
1680 | * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in | |
1681 | * NBOUND. This is accomplished by passing it as either the if or else clause, | |
1682 | * with the other one being empty (PLACEHOLDER is defined as empty). | |
1683 | * | |
1684 | * The TEST_FOO parameters are for operating on different forms of input, but | |
1685 | * all should be ones that return identically for the same underlying code | |
1686 | * points */ | |
1687 | #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \ | |
1688 | FBC_BOUND_COMMON( \ | |
1689 | FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \ | |
1690 | TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER) | |
1691 | ||
1692 | #define FBC_BOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \ | |
1693 | FBC_BOUND_COMMON( \ | |
1694 | FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \ | |
1695 | TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER) | |
1696 | ||
1697 | #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \ | |
1698 | FBC_BOUND_COMMON( \ | |
1699 | FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \ | |
1700 | TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT) | |
1701 | ||
1702 | #define FBC_NBOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \ | |
1703 | FBC_BOUND_COMMON( \ | |
1704 | FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \ | |
1705 | TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT) | |
1706 | ||
05bd126c | 1707 | |
786e8c11 | 1708 | /* We know what class REx starts with. Try to find this position... */ |
02d5137b | 1709 | /* if reginfo->intuit, its a dryrun */ |
786e8c11 YO |
1710 | /* annoyingly all the vars in this routine have different names from their counterparts |
1711 | in regmatch. /grrr */ | |
3c3eec57 | 1712 | STATIC char * |
07be1b83 | 1713 | S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, |
f9176b44 | 1714 | const char *strend, regmatch_info *reginfo) |
a687059c | 1715 | { |
73104a1b KW |
1716 | dVAR; |
1717 | const I32 doevery = (prog->intflags & PREGf_SKIP) == 0; | |
1718 | char *pat_string; /* The pattern's exactish string */ | |
1719 | char *pat_end; /* ptr to end char of pat_string */ | |
1720 | re_fold_t folder; /* Function for computing non-utf8 folds */ | |
1721 | const U8 *fold_array; /* array for folding ords < 256 */ | |
1722 | STRLEN ln; | |
1723 | STRLEN lnc; | |
73104a1b KW |
1724 | U8 c1; |
1725 | U8 c2; | |
1726 | char *e; | |
1727 | I32 tmp = 1; /* Scratch variable? */ | |
ba44c216 | 1728 | const bool utf8_target = reginfo->is_utf8_target; |
73104a1b | 1729 | UV utf8_fold_flags = 0; |
f9176b44 | 1730 | const bool is_utf8_pat = reginfo->is_utf8_pat; |
3018b823 KW |
1731 | bool to_complement = FALSE; /* Invert the result? Taking the xor of this |
1732 | with a result inverts that result, as 0^1 = | |
1733 | 1 and 1^1 = 0 */ | |
1734 | _char_class_number classnum; | |
1735 | ||
73104a1b | 1736 | RXi_GET_DECL(prog,progi); |
2f7f8cb1 | 1737 | |
73104a1b | 1738 | PERL_ARGS_ASSERT_FIND_BYCLASS; |
2f7f8cb1 | 1739 | |
73104a1b KW |
1740 | /* We know what class it must start with. */ |
1741 | switch (OP(c)) { | |
1742 | case ANYOF: | |
1743 | if (utf8_target) { | |
1744 | REXEC_FBC_UTF8_CLASS_SCAN( | |
3db24e1e | 1745 | reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target)); |
73104a1b KW |
1746 | } |
1747 | else { | |
1748 | REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s)); | |
1749 | } | |
1750 | break; | |
1751 | case CANY: | |
1752 | REXEC_FBC_SCAN( | |
02d5137b | 1753 | if (tmp && (reginfo->intuit || regtry(reginfo, &s))) |
73104a1b KW |
1754 | goto got_it; |
1755 | else | |
1756 | tmp = doevery; | |
1757 | ); | |
1758 | break; | |
1759 | ||
098b07d5 KW |
1760 | case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */ |
1761 | assert(! is_utf8_pat); | |
924ba076 | 1762 | /* FALLTHROUGH */ |
73104a1b | 1763 | case EXACTFA: |
984e6dd1 | 1764 | if (is_utf8_pat || utf8_target) { |
73104a1b KW |
1765 | utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
1766 | goto do_exactf_utf8; | |
1767 | } | |
1768 | fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */ | |
1769 | folder = foldEQ_latin1; /* /a, except the sharp s one which */ | |
1770 | goto do_exactf_non_utf8; /* isn't dealt with by these */ | |
77a6d856 | 1771 | |
2fdb7295 KW |
1772 | case EXACTF: /* This node only generated for non-utf8 patterns */ |
1773 | assert(! is_utf8_pat); | |
73104a1b | 1774 | if (utf8_target) { |
73104a1b KW |
1775 | utf8_fold_flags = 0; |
1776 | goto do_exactf_utf8; | |
1777 | } | |
1778 | fold_array = PL_fold; | |
1779 | folder = foldEQ; | |
1780 | goto do_exactf_non_utf8; | |
1781 | ||
1782 | case EXACTFL: | |
31f05a37 | 1783 | if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) { |
cea315b6 | 1784 | utf8_fold_flags = FOLDEQ_LOCALE; |
73104a1b KW |
1785 | goto do_exactf_utf8; |
1786 | } | |
1787 | fold_array = PL_fold_locale; | |
1788 | folder = foldEQ_locale; | |
1789 | goto do_exactf_non_utf8; | |
3c760661 | 1790 | |
73104a1b | 1791 | case EXACTFU_SS: |
984e6dd1 | 1792 | if (is_utf8_pat) { |
73104a1b KW |
1793 | utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED; |
1794 | } | |
1795 | goto do_exactf_utf8; | |
16d951b7 | 1796 | |
73104a1b | 1797 | case EXACTFU: |
984e6dd1 DM |
1798 | if (is_utf8_pat || utf8_target) { |
1799 | utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0; | |
73104a1b KW |
1800 | goto do_exactf_utf8; |
1801 | } | |
fac1af77 | 1802 | |
73104a1b KW |
1803 | /* Any 'ss' in the pattern should have been replaced by regcomp, |
1804 | * so we don't have to worry here about this single special case | |
1805 | * in the Latin1 range */ | |
1806 | fold_array = PL_fold_latin1; | |
1807 | folder = foldEQ_latin1; | |
1808 | ||
924ba076 | 1809 | /* FALLTHROUGH */ |
73104a1b KW |
1810 | |
1811 | do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there | |
1812 | are no glitches with fold-length differences | |
1813 | between the target string and pattern */ | |
1814 | ||
1815 | /* The idea in the non-utf8 EXACTF* cases is to first find the | |
1816 | * first character of the EXACTF* node and then, if necessary, | |
1817 | * case-insensitively compare the full text of the node. c1 is the | |
1818 | * first character. c2 is its fold. This logic will not work for | |
1819 | * Unicode semantics and the german sharp ss, which hence should | |
1820 | * not be compiled into a node that gets here. */ | |
1821 | pat_string = STRING(c); | |
1822 | ln = STR_LEN(c); /* length to match in octets/bytes */ | |
1823 | ||
1824 | /* We know that we have to match at least 'ln' bytes (which is the | |
1825 | * same as characters, since not utf8). If we have to match 3 | |
1826 | * characters, and there are only 2 availabe, we know without | |
1827 | * trying that it will fail; so don't start a match past the | |
1828 | * required minimum number from the far end */ | |
ea3daa5d | 1829 | e = HOP3c(strend, -((SSize_t)ln), s); |
73104a1b | 1830 | |
02d5137b | 1831 | if (reginfo->intuit && e < s) { |
73104a1b KW |
1832 | e = s; /* Due to minlen logic of intuit() */ |
1833 | } | |
fac1af77 | 1834 | |
73104a1b KW |
1835 | c1 = *pat_string; |
1836 | c2 = fold_array[c1]; | |
1837 | if (c1 == c2) { /* If char and fold are the same */ | |
1838 | REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1); | |
1839 | } | |
1840 | else { | |
1841 | REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2); | |
1842 | } | |
1843 | break; | |
fac1af77 | 1844 | |
73104a1b KW |
1845 | do_exactf_utf8: |
1846 | { | |
1847 | unsigned expansion; | |
1848 | ||
1849 | /* If one of the operands is in utf8, we can't use the simpler folding | |
1850 | * above, due to the fact that many different characters can have the | |
1851 | * same fold, or portion of a fold, or different- length fold */ | |
1852 | pat_string = STRING(c); | |
1853 | ln = STR_LEN(c); /* length to match in octets/bytes */ | |
1854 | pat_end = pat_string + ln; | |
984e6dd1 | 1855 | lnc = is_utf8_pat /* length to match in characters */ |
73104a1b KW |
1856 | ? utf8_length((U8 *) pat_string, (U8 *) pat_end) |
1857 | : ln; | |
1858 | ||
1859 | /* We have 'lnc' characters to match in the pattern, but because of | |
1860 | * multi-character folding, each character in the target can match | |
1861 | * up to 3 characters (Unicode guarantees it will never exceed | |
1862 | * this) if it is utf8-encoded; and up to 2 if not (based on the | |
1863 | * fact that the Latin 1 folds are already determined, and the | |
1864 | * only multi-char fold in that range is the sharp-s folding to | |
1865 | * 'ss'. Thus, a pattern character can match as little as 1/3 of a | |
1866 | * string character. Adjust lnc accordingly, rounding up, so that | |
1867 | * if we need to match at least 4+1/3 chars, that really is 5. */ | |
1868 | expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2; | |
1869 | lnc = (lnc + expansion - 1) / expansion; | |
1870 | ||
1871 | /* As in the non-UTF8 case, if we have to match 3 characters, and | |
1872 | * only 2 are left, it's guaranteed to fail, so don't start a | |
1873 | * match that would require us to go beyond the end of the string | |
1874 | */ | |
ea3daa5d | 1875 | e = HOP3c(strend, -((SSize_t)lnc), s); |
73104a1b | 1876 | |
02d5137b | 1877 | if (reginfo->intuit && e < s) { |
73104a1b KW |
1878 | e = s; /* Due to minlen logic of intuit() */ |
1879 | } | |
0658cdde | 1880 | |
73104a1b KW |
1881 | /* XXX Note that we could recalculate e to stop the loop earlier, |
1882 | * as the worst case expansion above will rarely be met, and as we | |
1883 | * go along we would usually find that e moves further to the left. | |
1884 | * This would happen only after we reached the point in the loop | |
1885 | * where if there were no expansion we should fail. Unclear if | |
1886 | * worth the expense */ | |
1887 | ||
1888 | while (s <= e) { | |
1889 | char *my_strend= (char *)strend; | |
1890 | if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target, | |
984e6dd1 | 1891 | pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags) |
02d5137b | 1892 | && (reginfo->intuit || regtry(reginfo, &s)) ) |
73104a1b KW |
1893 | { |
1894 | goto got_it; | |
1895 | } | |
1896 | s += (utf8_target) ? UTF8SKIP(s) : 1; | |
1897 | } | |
1898 | break; | |
1899 | } | |
236d82fd | 1900 | |
73104a1b | 1901 | case BOUNDL: |
236d82fd | 1902 | FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8); |
73104a1b KW |
1903 | break; |
1904 | case NBOUNDL: | |
236d82fd | 1905 | FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8); |
73104a1b KW |
1906 | break; |
1907 | case BOUND: | |
236d82fd | 1908 | FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8); |
73104a1b KW |
1909 | break; |
1910 | case BOUNDA: | |
236d82fd | 1911 | FBC_BOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A); |
73104a1b KW |
1912 | break; |
1913 | case NBOUND: | |
236d82fd | 1914 | FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8); |
73104a1b KW |
1915 | break; |
1916 | case NBOUNDA: | |
236d82fd | 1917 | FBC_NBOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A); |
73104a1b KW |
1918 | break; |
1919 | case BOUNDU: | |
236d82fd | 1920 | FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8); |
73104a1b KW |
1921 | break; |
1922 | case NBOUNDU: | |
236d82fd | 1923 | FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8); |
73104a1b | 1924 | break; |
73104a1b KW |
1925 | case LNBREAK: |
1926 | REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend), | |
1927 | is_LNBREAK_latin1_safe(s, strend) | |
1928 | ); | |
1929 | break; | |
3018b823 KW |
1930 | |
1931 | /* The argument to all the POSIX node types is the class number to pass to | |
1932 | * _generic_isCC() to build a mask for searching in PL_charclass[] */ | |
1933 | ||
1934 | case NPOSIXL: | |
1935 | to_complement = 1; | |
1936 | /* FALLTHROUGH */ | |
1937 | ||
1938 | case POSIXL: | |
3018b823 KW |
1939 | REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)), |
1940 | to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s))); | |
73104a1b | 1941 | break; |
3018b823 KW |
1942 | |
1943 | case NPOSIXD: | |
1944 | to_complement = 1; | |
1945 | /* FALLTHROUGH */ | |
1946 | ||
1947 | case POSIXD: | |
1948 | if (utf8_target) { | |
1949 | goto posix_utf8; | |
1950 | } | |
1951 | goto posixa; | |
1952 | ||
1953 | case NPOSIXA: | |
1954 | if (utf8_target) { | |
1955 | /* The complement of something that matches only ASCII matches all | |
837226c8 KW |
1956 | * non-ASCII, plus everything in ASCII that isn't in the class. */ |
1957 | REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s) | |
3018b823 KW |
1958 | || ! _generic_isCC_A(*s, FLAGS(c))); |
1959 | break; | |
1960 | } | |
1961 | ||
1962 | to_complement = 1; | |
1963 | /* FALLTHROUGH */ | |
1964 | ||
73104a1b | 1965 | case POSIXA: |
3018b823 | 1966 | posixa: |
73104a1b | 1967 | /* Don't need to worry about utf8, as it can match only a single |
3018b823 KW |
1968 | * byte invariant character. */ |
1969 | REXEC_FBC_CLASS_SCAN( | |
1970 | to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c)))); | |
73104a1b | 1971 | break; |
3018b823 KW |
1972 | |
1973 | case NPOSIXU: | |
1974 | to_complement = 1; | |
1975 | /* FALLTHROUGH */ | |
1976 | ||
1977 | case POSIXU: | |
1978 | if (! utf8_target) { | |
1979 | REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s, | |
1980 | FLAGS(c)))); | |
1981 | } | |
1982 | else { | |
1983 | ||
1984 | posix_utf8: | |
1985 | classnum = (_char_class_number) FLAGS(c); | |
1986 | if (classnum < _FIRST_NON_SWASH_CC) { | |
1987 | while (s < strend) { | |
1988 | ||
1989 | /* We avoid loading in the swash as long as possible, but | |
1990 | * should we have to, we jump to a separate loop. This | |
1991 | * extra 'if' statement is what keeps this code from being | |
1992 | * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */ | |
1993 | if (UTF8_IS_ABOVE_LATIN1(*s)) { | |
1994 | goto found_above_latin1; | |
1995 | } | |
1996 | if ((UTF8_IS_INVARIANT(*s) | |
1997 | && to_complement ^ cBOOL(_generic_isCC((U8) *s, | |
1998 | classnum))) | |
1999 | || (UTF8_IS_DOWNGRADEABLE_START(*s) | |
2000 | && to_complement ^ cBOOL( | |
94bb8c36 KW |
2001 | _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s, |
2002 | *(s + 1)), | |
3018b823 KW |
2003 | classnum)))) |
2004 | { | |
02d5137b | 2005 | if (tmp && (reginfo->intuit || regtry(reginfo, &s))) |
3018b823 KW |
2006 | goto got_it; |
2007 | else { | |
2008 | tmp = doevery; | |
2009 | } | |
2010 | } | |
2011 | else { | |
2012 | tmp = 1; | |
2013 | } | |
2014 | s += UTF8SKIP(s); | |
2015 | } | |
2016 | } | |
2017 | else switch (classnum) { /* These classes are implemented as | |
2018 | macros */ | |
2019 | case _CC_ENUM_SPACE: /* XXX would require separate code if we | |
2020 | revert the change of \v matching this */ | |
924ba076 | 2021 | /* FALLTHROUGH */ |
3018b823 KW |
2022 | |
2023 | case _CC_ENUM_PSXSPC: | |
2024 | REXEC_FBC_UTF8_CLASS_SCAN( | |
2025 | to_complement ^ cBOOL(isSPACE_utf8(s))); | |
2026 | break; | |
2027 | ||
2028 | case _CC_ENUM_BLANK: | |
2029 | REXEC_FBC_UTF8_CLASS_SCAN( | |
2030 | to_complement ^ cBOOL(isBLANK_utf8(s))); | |
2031 | break; | |
2032 | ||
2033 | case _CC_ENUM_XDIGIT: | |
2034 | REXEC_FBC_UTF8_CLASS_SCAN( | |
2035 | to_complement ^ cBOOL(isXDIGIT_utf8(s))); | |
2036 | break; | |
2037 | ||
2038 | case _CC_ENUM_VERTSPACE: | |
2039 | REXEC_FBC_UTF8_CLASS_SCAN( | |
2040 | to_complement ^ cBOOL(isVERTWS_utf8(s))); | |
2041 | break; | |
2042 | ||
2043 | case _CC_ENUM_CNTRL: | |
2044 | REXEC_FBC_UTF8_CLASS_SCAN( | |
2045 | to_complement ^ cBOOL(isCNTRL_utf8(s))); | |
2046 | break; | |
2047 | ||
2048 | default: | |
2049 | Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum); | |
e5964223 | 2050 | NOT_REACHED; /* NOTREACHED */ |
3018b823 KW |
2051 | } |
2052 | } | |
2053 | break; | |
2054 | ||
2055 | found_above_latin1: /* Here we have to load a swash to get the result | |
2056 | for the current code point */ | |
2057 | if (! PL_utf8_swash_ptrs[classnum]) { | |
2058 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; | |
2059 | PL_utf8_swash_ptrs[classnum] = | |
2a16ac92 KW |
2060 | _core_swash_init("utf8", |
2061 | "", | |
2062 | &PL_sv_undef, 1, 0, | |
2063 | PL_XPosix_ptrs[classnum], &flags); | |
3018b823 KW |
2064 | } |
2065 | ||
2066 | /* This is a copy of the loop above for swash classes, though using the | |
2067 | * FBC macro instead of being expanded out. Since we've loaded the | |
2068 | * swash, we don't have to check for that each time through the loop */ | |
2069 | REXEC_FBC_UTF8_CLASS_SCAN( | |
2070 | to_complement ^ cBOOL(_generic_utf8( | |
2071 | classnum, | |
2072 | s, | |
2073 | swash_fetch(PL_utf8_swash_ptrs[classnum], | |
2074 | (U8 *) s, TRUE)))); | |
73104a1b KW |
2075 | break; |
2076 | ||
2077 | case AHOCORASICKC: | |
2078 | case AHOCORASICK: | |
2079 | { | |
2080 | DECL_TRIE_TYPE(c); | |
2081 | /* what trie are we using right now */ | |
2082 | reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ]; | |
2083 | reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ]; | |
2084 | HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]); | |
2085 | ||
2086 | const char *last_start = strend - trie->minlen; | |
6148ee25 | 2087 | #ifdef DEBUGGING |
73104a1b | 2088 | const char *real_start = s; |
6148ee25 | 2089 | #endif |
73104a1b KW |
2090 | STRLEN maxlen = trie->maxlen; |
2091 | SV *sv_points; | |
2092 | U8 **points; /* map of where we were in the input string | |
2093 | when reading a given char. For ASCII this | |
2094 | is unnecessary overhead as the relationship | |
2095 | is always 1:1, but for Unicode, especially | |
2096 | case folded Unicode this is not true. */ | |
2097 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; | |
2098 | U8 *bitmap=NULL; | |
2099 | ||
2100 | ||
2101 | GET_RE_DEBUG_FLAGS_DECL; | |
2102 | ||
2103 | /* We can't just allocate points here. We need to wrap it in | |
2104 | * an SV so it gets freed properly if there is a croak while | |
2105 | * running the match */ | |
2106 | ENTER; | |
2107 | SAVETMPS; | |
2108 | sv_points=newSV(maxlen * sizeof(U8 *)); | |
2109 | SvCUR_set(sv_points, | |
2110 | maxlen * sizeof(U8 *)); | |
2111 | SvPOK_on(sv_points); | |
2112 | sv_2mortal(sv_points); | |
2113 | points=(U8**)SvPV_nolen(sv_points ); | |
2114 | if ( trie_type != trie_utf8_fold | |
2115 | && (trie->bitmap || OP(c)==AHOCORASICKC) ) | |
2116 | { | |
2117 | if (trie->bitmap) | |
2118 | bitmap=(U8*)trie->bitmap; | |
2119 | else | |
2120 | bitmap=(U8*)ANYOF_BITMAP(c); | |
2121 | } | |
2122 | /* this is the Aho-Corasick algorithm modified a touch | |
2123 | to include special handling for long "unknown char" sequences. | |
2124 | The basic idea being that we use AC as long as we are dealing | |
2125 | with a possible matching char, when we encounter an unknown char | |
2126 | (and we have not encountered an accepting state) we scan forward | |
2127 | until we find a legal starting char. | |
2128 | AC matching is basically that of trie matching, except that when | |
2129 | we encounter a failing transition, we fall back to the current | |
2130 | states "fail state", and try the current char again, a process | |
2131 | we repeat until we reach the root state, state 1, or a legal | |
2132 | transition. If we fail on the root state then we can either | |
2133 | terminate if we have reached an accepting state previously, or | |
2134 | restart the entire process from the beginning if we have not. | |
2135 | ||
2136 | */ | |
2137 | while (s <= last_start) { | |
2138 | const U32 uniflags = UTF8_ALLOW_DEFAULT; | |
2139 | U8 *uc = (U8*)s; | |
2140 | U16 charid = 0; | |
2141 | U32 base = 1; | |
2142 | U32 state = 1; | |
2143 | UV uvc = 0; | |
2144 | STRLEN len = 0; | |
2145 | STRLEN foldlen = 0; | |
2146 | U8 *uscan = (U8*)NULL; | |
2147 | U8 *leftmost = NULL; | |
2148 | #ifdef DEBUGGING | |
2149 | U32 accepted_word= 0; | |
786e8c11 | 2150 | #endif |
73104a1b KW |
2151 | U32 pointpos = 0; |
2152 | ||
2153 | while ( state && uc <= (U8*)strend ) { | |
2154 | int failed=0; | |
2155 | U32 word = aho->states[ state ].wordnum; | |
2156 | ||
2157 | if( state==1 ) { | |
2158 | if ( bitmap ) { | |
2159 | DEBUG_TRIE_EXECUTE_r( | |
2160 | if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { | |
2161 | dump_exec_pos( (char *)uc, c, strend, real_start, | |
2162 | (char *)uc, utf8_target ); | |
2163 | PerlIO_printf( Perl_debug_log, | |
2164 | " Scanning for legal start char...\n"); | |
2165 | } | |
2166 | ); | |
2167 | if (utf8_target) { | |
2168 | while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { | |
2169 | uc += UTF8SKIP(uc); | |
2170 | } | |
2171 | } else { | |
2172 | while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { | |
2173 | uc++; | |
2174 | } | |
786e8c11 | 2175 | } |
73104a1b | 2176 | s= (char *)uc; |
07be1b83 | 2177 | } |
73104a1b KW |
2178 | if (uc >(U8*)last_start) break; |
2179 | } | |
2180 | ||
2181 | if ( word ) { | |
2182 | U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ]; | |
2183 | if (!leftmost || lpos < leftmost) { | |
2184 | DEBUG_r(accepted_word=word); | |
2185 | leftmost= lpos; | |
7016d6eb | 2186 | } |
73104a1b | 2187 | if (base==0) break; |
7016d6eb | 2188 | |
73104a1b KW |
2189 | } |
2190 | points[pointpos++ % maxlen]= uc; | |
2191 | if (foldlen || uc < (U8*)strend) { | |
2192 | REXEC_TRIE_READ_CHAR(trie_type, trie, | |
2193 | widecharmap, uc, | |
2194 | uscan, len, uvc, charid, foldlen, | |
2195 | foldbuf, uniflags); | |
2196 | DEBUG_TRIE_EXECUTE_r({ | |
2197 | dump_exec_pos( (char *)uc, c, strend, | |
2198 | real_start, s, utf8_target); | |
2199 | PerlIO_printf(Perl_debug_log, | |
2200 | " Charid:%3u CP:%4"UVxf" ", | |
2201 | charid, uvc); | |
2202 | }); | |
2203 | } | |
2204 | else { | |
2205 | len = 0; | |
2206 | charid = 0; | |
2207 | } | |
07be1b83 | 2208 | |
73104a1b KW |
2209 | |
2210 | do { | |
6148ee25 | 2211 | #ifdef DEBUGGING |
73104a1b | 2212 | word = aho->states[ state ].wordnum; |
6148ee25 | 2213 | #endif |
73104a1b KW |
2214 | base = aho->states[ state ].trans.base; |
2215 | ||
2216 | DEBUG_TRIE_EXECUTE_r({ | |
2217 | if (failed) | |
2218 | dump_exec_pos( (char *)uc, c, strend, real_start, | |
2219 | s, utf8_target ); | |
2220 | PerlIO_printf( Perl_debug_log, | |
2221 | "%sState: %4"UVxf", word=%"UVxf, | |
2222 | failed ? " Fail transition to " : "", | |
2223 | (UV)state, (UV)word); | |
2224 | }); | |
2225 | if ( base ) { | |
2226 | U32 tmp; | |
2227 | I32 offset; | |
2228 | if (charid && | |
2229 | ( ((offset = base + charid | |
2230 | - 1 - trie->uniquecharcount)) >= 0) | |
2231 | && ((U32)offset < trie->lasttrans) | |
2232 | && trie->trans[offset].check == state | |
2233 | && (tmp=trie->trans[offset].next)) | |
2234 | { | |
2235 | DEBUG_TRIE_EXECUTE_r( | |
2236 | PerlIO_printf( Perl_debug_log," - legal\n")); | |
2237 | state = tmp; | |
2238 | break; | |
07be1b83 YO |
2239 | } |
2240 | else { | |
786e8c11 | 2241 | DEBUG_TRIE_EXECUTE_r( |
73104a1b | 2242 | PerlIO_printf( Perl_debug_log," - fail\n")); |
786e8c11 | 2243 | failed = 1; |
73104a1b | 2244 | state = aho->fail[state]; |
07be1b83 | 2245 | } |
07be1b83 | 2246 | } |
73104a1b KW |
2247 | else { |
2248 | /* we must be accepting here */ | |
2249 | DEBUG_TRIE_EXECUTE_r( | |
2250 | PerlIO_printf( Perl_debug_log," - accepting\n")); | |
2251 | failed = 1; | |
2252 | break; | |
786e8c11 | 2253 | } |
73104a1b KW |
2254 | } while(state); |
2255 | uc += len; | |
2256 | if (failed) { | |
2257 | if (leftmost) | |
2258 | break; | |
2259 | if (!state) state = 1; | |
07be1b83 | 2260 | } |
73104a1b KW |
2261 | } |
2262 | if ( aho->states[ state ].wordnum ) { | |
2263 | U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ]; | |
2264 | if (!leftmost || lpos < leftmost) { | |
2265 | DEBUG_r(accepted_word=aho->states[ state ].wordnum); | |
2266 | leftmost = lpos; | |
07be1b83 YO |
2267 | } |
2268 | } | |
73104a1b KW |
2269 | if (leftmost) { |
2270 | s = (char*)leftmost; | |
2271 | DEBUG_TRIE_EXECUTE_r({ | |
2272 | PerlIO_printf( | |
2273 | Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n", | |
2274 | (UV)accepted_word, (IV)(s - real_start) | |
2275 | ); | |
2276 | }); | |
02d5137b | 2277 | if (reginfo->intuit || regtry(reginfo, &s)) { |
73104a1b KW |
2278 | FREETMPS; |
2279 | LEAVE; | |
2280 | goto got_it; | |
2281 | } | |
2282 | s = HOPc(s,1); | |
2283 | DEBUG_TRIE_EXECUTE_r({ | |
2284 | PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n"); | |
2285 | }); | |
2286 | } else { | |
2287 | DEBUG_TRIE_EXECUTE_r( | |
2288 | PerlIO_printf( Perl_debug_log,"No match.\n")); | |
2289 | break; | |
2290 | } | |
2291 | } | |
2292 | FREETMPS; | |
2293 | LEAVE; | |
2294 | } | |
2295 | break; | |
2296 | default: | |
2297 | Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c)); | |
73104a1b KW |
2298 | } |
2299 | return 0; | |
2300 | got_it: | |
2301 | return s; | |
6eb5f6b9 JH |
2302 | } |
2303 | ||
60165aa4 DM |
2304 | /* set RX_SAVED_COPY, RX_SUBBEG etc. |
2305 | * flags have same meanings as with regexec_flags() */ | |
2306 | ||
749f4950 DM |
2307 | static void |
2308 | S_reg_set_capture_string(pTHX_ REGEXP * const rx, | |
60165aa4 DM |
2309 | char *strbeg, |
2310 | char *strend, | |
2311 | SV *sv, | |
2312 | U32 flags, | |
2313 | bool utf8_target) | |
2314 | { | |
2315 | struct regexp *const prog = ReANY(rx); | |
2316 | ||
60165aa4 DM |
2317 | if (flags & REXEC_COPY_STR) { |
2318 | #ifdef PERL_ANY_COW | |
2319 | if (SvCANCOW(sv)) { | |
2320 | if (DEBUG_C_TEST) { | |
2321 | PerlIO_printf(Perl_debug_log, | |
2322 | "Copy on write: regexp capture, type %d\n", | |
2323 | (int) SvTYPE(sv)); | |
2324 | } | |
5411a0e5 DM |
2325 | /* Create a new COW SV to share the match string and store |
2326 | * in saved_copy, unless the current COW SV in saved_copy | |
2327 | * is valid and suitable for our purpose */ | |
2328 | if (( prog->saved_copy | |
2329 | && SvIsCOW(prog->saved_copy) | |
2330 | && SvPOKp(prog->saved_copy) | |
2331 | && SvIsCOW(sv) | |
2332 | && SvPOKp(sv) | |
2333 | && SvPVX(sv) == SvPVX(prog->saved_copy))) | |
a76b0e90 | 2334 | { |
5411a0e5 DM |
2335 | /* just reuse saved_copy SV */ |
2336 | if (RXp_MATCH_COPIED(prog)) { | |
2337 | Safefree(prog->subbeg); | |
2338 | RXp_MATCH_COPIED_off(prog); | |
2339 | } | |
2340 | } | |
2341 | else { | |
2342 | /* create new COW SV to share string */ | |
a76b0e90 DM |
2343 | RX_MATCH_COPY_FREE(rx); |
2344 | prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv); | |
a76b0e90 | 2345 | } |
5411a0e5 DM |
2346 | prog->subbeg = (char *)SvPVX_const(prog->saved_copy); |
2347 | assert (SvPOKp(prog->saved_copy)); | |
60165aa4 DM |
2348 | prog->sublen = strend - strbeg; |
2349 | prog->suboffset = 0; | |
2350 | prog->subcoffset = 0; | |
2351 | } else | |
2352 | #endif | |
2353 | { | |
99a90e59 FC |
2354 | SSize_t min = 0; |
2355 | SSize_t max = strend - strbeg; | |
ea3daa5d | 2356 | SSize_t sublen; |
60165aa4 DM |
2357 | |
2358 | if ( (flags & REXEC_COPY_SKIP_POST) | |
e322109a | 2359 | && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */ |
60165aa4 DM |
2360 | && !(PL_sawampersand & SAWAMPERSAND_RIGHT) |
2361 | ) { /* don't copy $' part of string */ | |
2362 | U32 n = 0; | |
2363 | max = -1; | |
2364 | /* calculate the right-most part of the string covered | |
2365 | * by a capture. Due to look-ahead, this may be to | |
2366 | * the right of $&, so we have to scan all captures */ | |
2367 | while (n <= prog->lastparen) { | |
2368 | if (prog->offs[n].end > max) | |
2369 | max = prog->offs[n].end; | |
2370 | n++; | |
2371 | } | |
2372 | if (max == -1) | |
2373 | max = (PL_sawampersand & SAWAMPERSAND_LEFT) | |
2374 | ? prog->offs[0].start | |
2375 | : 0; | |
2376 | assert(max >= 0 && max <= strend - strbeg); | |
2377 | } | |
2378 | ||
2379 | if ( (flags & REXEC_COPY_SKIP_PRE) | |
e322109a | 2380 | && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */ |
60165aa4 DM |
2381 | && !(PL_sawampersand & SAWAMPERSAND_LEFT) |
2382 | ) { /* don't copy $` part of string */ | |
2383 | U32 n = 0; | |
2384 | min = max; | |
2385 | /* calculate the left-most part of the string covered | |
2386 | * by a capture. Due to look-behind, this may be to | |
2387 | * the left of $&, so we have to scan all captures */ | |
2388 | while (min && n <= prog->lastparen) { | |
2389 | if ( prog->offs[n].start != -1 | |
2390 | && prog->offs[n].start < min) | |
2391 | { | |
2392 | min = prog->offs[n].start; | |
2393 | } | |
2394 | n++; | |
2395 | } | |
2396 | if ((PL_sawampersand & SAWAMPERSAND_RIGHT) | |
2397 | && min > prog->offs[0].end | |
2398 | ) | |
2399 | min = prog->offs[0].end; | |
2400 | ||
2401 | } | |
2402 | ||
2403 | assert(min >= 0 && min <= max && min <= strend - strbeg); | |
2404 | sublen = max - min; | |
2405 | ||
2406 | if (RX_MATCH_COPIED(rx)) { | |
2407 | if (sublen > prog->sublen) | |
2408 | prog->subbeg = | |
2409 | (char*)saferealloc(prog->subbeg, sublen+1); | |
2410 | } | |
2411 | else | |
2412 | prog->subbeg = (char*)safemalloc(sublen+1); | |
2413 | Copy(strbeg + min, prog->subbeg, sublen, char); | |
2414 | prog->subbeg[sublen] = '\0'; | |
2415 | prog->suboffset = min; | |
2416 | prog->sublen = sublen; | |
2417 | RX_MATCH_COPIED_on(rx); | |
2418 | } | |
2419 | prog->subcoffset = prog->suboffset; | |
2420 | if (prog->suboffset && utf8_target) { | |
2421 | /* Convert byte offset to chars. | |
2422 | * XXX ideally should only compute this if @-/@+ | |
2423 | * has been seen, a la PL_sawampersand ??? */ | |
2424 | ||
2425 | /* If there's a direct correspondence between the | |
2426 | * string which we're matching and the original SV, | |
2427 | * then we can use the utf8 len cache associated with | |
2428 | * the SV. In particular, it means that under //g, | |
2429 | * sv_pos_b2u() will use the previously cached | |
2430 | * position to speed up working out the new length of | |
2431 | * subcoffset, rather than counting from the start of | |
2432 | * the string each time. This stops | |
2433 | * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g; | |
2434 | * from going quadratic */ | |
2435 | if (SvPOKp(sv) && SvPVX(sv) == strbeg) | |
ea3daa5d FC |
2436 | prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset, |
2437 | SV_GMAGIC|SV_CONST_RETURN); | |
60165aa4 DM |
2438 | else |
2439 | prog->subcoffset = utf8_length((U8*)strbeg, | |
2440 | (U8*)(strbeg+prog->suboffset)); | |
2441 | } | |
2442 | } | |
2443 | else { | |
2444 | RX_MATCH_COPY_FREE(rx); | |
2445 | prog->subbeg = strbeg; | |
2446 | prog->suboffset = 0; | |
2447 | prog->subcoffset = 0; | |
2448 | prog->sublen = strend - strbeg; | |
2449 | } | |
2450 | } | |
2451 | ||
2452 | ||
2453 | ||
fae667d5 | 2454 | |
6eb5f6b9 JH |
2455 | /* |
2456 | - regexec_flags - match a regexp against a string | |
2457 | */ | |
2458 | I32 | |
5aaab254 | 2459 | Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend, |
ea3daa5d | 2460 | char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags) |
8fd1a950 DM |
2461 | /* stringarg: the point in the string at which to begin matching */ |
2462 | /* strend: pointer to null at end of string */ | |
2463 | /* strbeg: real beginning of string */ | |
2464 | /* minend: end of match must be >= minend bytes after stringarg. */ | |
2465 | /* sv: SV being matched: only used for utf8 flag, pos() etc; string | |
2466 | * itself is accessed via the pointers above */ | |
2467 | /* data: May be used for some additional optimizations. | |
d058ec57 | 2468 | Currently unused. */ |
a340edde | 2469 | /* flags: For optimizations. See REXEC_* in regexp.h */ |
8fd1a950 | 2470 | |
6eb5f6b9 | 2471 | { |
8d919b0a | 2472 | struct regexp *const prog = ReANY(rx); |
5aaab254 | 2473 | char *s; |
eb578fdb | 2474 | regnode *c; |
03c83e26 | 2475 | char *startpos; |
ea3daa5d FC |
2476 | SSize_t minlen; /* must match at least this many chars */ |
2477 | SSize_t dontbother = 0; /* how many characters not to try at end */ | |
f2ed9b32 | 2478 | const bool utf8_target = cBOOL(DO_UTF8(sv)); |
2757e526 | 2479 | I32 multiline; |
f8fc2ecf | 2480 | RXi_GET_DECL(prog,progi); |
02d5137b DM |
2481 | regmatch_info reginfo_buf; /* create some info to pass to regtry etc */ |
2482 | regmatch_info *const reginfo = ®info_buf; | |
e9105d30 | 2483 | regexp_paren_pair *swap = NULL; |
006f26b2 | 2484 | I32 oldsave; |
a3621e74 YO |
2485 | GET_RE_DEBUG_FLAGS_DECL; |
2486 | ||
7918f24d | 2487 | PERL_ARGS_ASSERT_REGEXEC_FLAGS; |
9d4ba2ae | 2488 | PERL_UNUSED_ARG(data); |
6eb5f6b9 JH |
2489 | |
2490 | /* Be paranoid... */ | |
03c83e26 | 2491 | if (prog == NULL || stringarg == NULL) { |
6eb5f6b9 | 2492 | Perl_croak(aTHX_ "NULL regexp parameter"); |
6eb5f6b9 JH |
2493 | } |
2494 | ||
6c3fea77 | 2495 | DEBUG_EXECUTE_r( |
03c83e26 | 2496 | debug_start_match(rx, utf8_target, stringarg, strend, |
6c3fea77 DM |
2497 | "Matching"); |
2498 | ); | |
8adc0f72 | 2499 | |
b342a604 DM |
2500 | startpos = stringarg; |
2501 | ||
58430ea8 | 2502 | if (prog->intflags & PREGf_GPOS_SEEN) { |
d307c076 DM |
2503 | MAGIC *mg; |
2504 | ||
fef7148b DM |
2505 | /* set reginfo->ganch, the position where \G can match */ |
2506 | ||
2507 | reginfo->ganch = | |
2508 | (flags & REXEC_IGNOREPOS) | |
2509 | ? stringarg /* use start pos rather than pos() */ | |
2510 | : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0) | |
25fdce4a FC |
2511 | /* Defined pos(): */ |
2512 | ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg) | |
fef7148b DM |
2513 | : strbeg; /* pos() not defined; use start of string */ |
2514 | ||
2515 | DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log, | |
7b0eb0b8 | 2516 | "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg))); |
fef7148b | 2517 | |
03c83e26 DM |
2518 | /* in the presence of \G, we may need to start looking earlier in |
2519 | * the string than the suggested start point of stringarg: | |
0b2c2a84 | 2520 | * if prog->gofs is set, then that's a known, fixed minimum |
03c83e26 DM |
2521 | * offset, such as |
2522 | * /..\G/: gofs = 2 | |
2523 | * /ab|c\G/: gofs = 1 | |
2524 | * or if the minimum offset isn't known, then we have to go back | |
2525 | * to the start of the string, e.g. /w+\G/ | |
2526 | */ | |
2bfbe302 | 2527 | |
8e1490ee | 2528 | if (prog->intflags & PREGf_ANCH_GPOS) { |
2bfbe302 DM |
2529 | startpos = reginfo->ganch - prog->gofs; |
2530 | if (startpos < | |
2531 | ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg)) | |
2532 | { | |
2533 | DEBUG_r(PerlIO_printf(Perl_debug_log, | |
2534 | "fail: ganch-gofs before earliest possible start\n")); | |
2535 | return 0; | |
2536 | } | |
2537 | } | |
2538 | else if (prog->gofs) { | |
b342a604 DM |
2539 | if (startpos - prog->gofs < strbeg) |
2540 | startpos = strbeg; | |
2541 | else | |
2542 | startpos -= prog->gofs; | |
03c83e26 | 2543 | } |
58430ea8 | 2544 | else if (prog->intflags & PREGf_GPOS_FLOAT) |
b342a604 | 2545 | startpos = strbeg; |
03c83e26 DM |
2546 | } |
2547 | ||
2548 | minlen = prog->minlen; | |
b342a604 | 2549 | if ((startpos + minlen) > strend || startpos < strbeg) { |
03c83e26 DM |
2550 | DEBUG_r(PerlIO_printf(Perl_debug_log, |
2551 | "Regex match can't succeed, so not even tried\n")); | |
2552 | return 0; | |
2553 | } | |
2554 | ||
63a3746a DM |
2555 | /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave), |
2556 | * which will call destuctors to reset PL_regmatch_state, free higher | |
2557 | * PL_regmatch_slabs, and clean up regmatch_info_aux and | |
2558 | * regmatch_info_aux_eval */ | |
2559 | ||
2560 | oldsave = PL_savestack_ix; | |
2561 | ||
dfa77d06 DM |
2562 | s = startpos; |
2563 | ||
e322109a | 2564 | if ((prog->extflags & RXf_USE_INTUIT) |
7fadf4a7 DM |
2565 | && !(flags & REXEC_CHECKED)) |
2566 | { | |
dfa77d06 | 2567 | s = re_intuit_start(rx, sv, strbeg, startpos, strend, |
7fadf4a7 | 2568 | flags, NULL); |
dfa77d06 | 2569 | if (!s) |
7fadf4a7 DM |
2570 | return 0; |
2571 | ||
e322109a | 2572 | if (prog->extflags & RXf_CHECK_ALL) { |
7fadf4a7 DM |
2573 | /* we can match based purely on the result of INTUIT. |
2574 | * Set up captures etc just for $& and $-[0] | |
2575 | * (an intuit-only match wont have $1,$2,..) */ | |
2576 | assert(!prog->nparens); | |
d5e7783a DM |
2577 | |
2578 | /* s/// doesn't like it if $& is earlier than where we asked it to | |
2579 | * start searching (which can happen on something like /.\G/) */ | |
2580 | if ( (flags & REXEC_FAIL_ON_UNDERFLOW) | |
2581 | && (s < stringarg)) | |
2582 | { | |
2583 | /* this should only be possible under \G */ | |
58430ea8 | 2584 | assert(prog->intflags & PREGf_GPOS_SEEN); |
d5e7783a DM |
2585 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
2586 | "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n")); | |
2587 | goto phooey; | |
2588 | } | |
2589 | ||
7fadf4a7 DM |
2590 | /* match via INTUIT shouldn't have any captures. |
2591 | * Let @-, @+, $^N know */ | |
2592 | prog->lastparen = prog->lastcloseparen = 0; | |
2593 | RX_MATCH_UTF8_set(rx, utf8_target); | |
3ff69bd6 DM |
2594 | prog->offs[0].start = s - strbeg; |
2595 | prog->offs[0].end = utf8_target | |
2596 | ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg | |
2597 | : s - strbeg + prog->minlenret; | |
7fadf4a7 | 2598 | if ( !(flags & REXEC_NOT_FIRST) ) |
749f4950 | 2599 | S_reg_set_capture_string(aTHX_ rx, |
7fadf4a7 DM |
2600 | strbeg, strend, |
2601 | sv, flags, utf8_target); | |
2602 | ||
7fadf4a7 DM |
2603 | return 1; |
2604 | } | |
2605 | } | |
2606 | ||
6c3fea77 | 2607 | multiline = prog->extflags & RXf_PMf_MULTILINE; |
1de06328 | 2608 | |
dfa77d06 | 2609 | if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) { |
a3621e74 | 2610 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
a72c7584 JH |
2611 | "String too short [regexec_flags]...\n")); |
2612 | goto phooey; | |
1aa99e6b | 2613 | } |
1de06328 | 2614 | |
6eb5f6b9 | 2615 | /* Check validity of program. */ |
f8fc2ecf | 2616 | if (UCHARAT(progi->program) != REG_MAGIC) { |
6eb5f6b9 JH |
2617 | Perl_croak(aTHX_ "corrupted regexp program"); |
2618 | } | |
2619 | ||
1738e041 | 2620 | RX_MATCH_TAINTED_off(rx); |
ab4e48c1 | 2621 | RX_MATCH_UTF8_set(rx, utf8_target); |
1738e041 | 2622 | |
6c3fea77 DM |
2623 | reginfo->prog = rx; /* Yes, sorry that this is confusing. */ |
2624 | reginfo->intuit = 0; | |
2625 | reginfo->is_utf8_target = cBOOL(utf8_target); | |
02d5137b DM |
2626 | reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx)); |
2627 | reginfo->warned = FALSE; | |
9d9163fb | 2628 | reginfo->strbeg = strbeg; |
02d5137b | 2629 | reginfo->sv = sv; |
1cb48e53 | 2630 | reginfo->poscache_maxiter = 0; /* not yet started a countdown */ |
220db18a | 2631 | reginfo->strend = strend; |
6eb5f6b9 | 2632 | /* see how far we have to get to not match where we matched before */ |
fe3974be | 2633 | reginfo->till = stringarg + minend; |
6eb5f6b9 | 2634 | |
60779a30 | 2635 | if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) { |
82c23608 FC |
2636 | /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after |
2637 | S_cleanup_regmatch_info_aux has executed (registered by | |
2638 | SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies | |
2639 | magic belonging to this SV. | |
2640 | Not newSVsv, either, as it does not COW. | |
2641 | */ | |
2642 | reginfo->sv = newSV(0); | |
4cba5ac0 | 2643 | SvSetSV_nosteal(reginfo->sv, sv); |
82c23608 FC |
2644 | SAVEFREESV(reginfo->sv); |
2645 | } | |
2646 | ||
331b2dcc DM |
2647 | /* reserve next 2 or 3 slots in PL_regmatch_state: |
2648 | * slot N+0: may currently be in use: skip it | |
2649 | * slot N+1: use for regmatch_info_aux struct | |
2650 | * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s | |
2651 | * slot N+3: ready for use by regmatch() | |
2652 | */ | |
bf2039a9 | 2653 | |
331b2dcc DM |
2654 | { |
2655 | regmatch_state *old_regmatch_state; | |
2656 | regmatch_slab *old_regmatch_slab; | |
2657 | int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1; | |
2658 | ||
2659 | /* on first ever match, allocate first slab */ | |
2660 | if (!PL_regmatch_slab) { | |
2661 | Newx(PL_regmatch_slab, 1, regmatch_slab); | |
2662 | PL_regmatch_slab->prev = NULL; | |
2663 | PL_regmatch_slab->next = NULL; | |
2664 | PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab); | |
2665 | } | |
bf2039a9 | 2666 | |
331b2dcc DM |
2667 | old_regmatch_state = PL_regmatch_state; |
2668 | old_regmatch_slab = PL_regmatch_slab; | |
bf2039a9 | 2669 | |
331b2dcc DM |
2670 | for (i=0; i <= max; i++) { |
2671 | if (i == 1) | |
2672 | reginfo->info_aux = &(PL_regmatch_state->u.info_aux); | |
2673 | else if (i ==2) | |
2674 | reginfo->info_aux_eval = | |
2675 | reginfo->info_aux->info_aux_eval = | |
2676 | &(PL_regmatch_state->u.info_aux_eval); | |
bf2039a9 | 2677 | |
331b2dcc DM |
2678 | if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab)) |
2679 | PL_regmatch_state = S_push_slab(aTHX); | |
2680 | } | |
bf2039a9 | 2681 | |
331b2dcc DM |
2682 | /* note initial PL_regmatch_state position; at end of match we'll |
2683 | * pop back to there and free any higher slabs */ | |
bf2039a9 | 2684 | |
331b2dcc DM |
2685 | reginfo->info_aux->old_regmatch_state = old_regmatch_state; |
2686 | reginfo->info_aux->old_regmatch_slab = old_regmatch_slab; | |
2ac8ff4b | 2687 | reginfo->info_aux->poscache = NULL; |
bf2039a9 | 2688 | |
331b2dcc | 2689 | SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux); |
bf2039a9 | 2690 | |
331b2dcc DM |
2691 | if ((prog->extflags & RXf_EVAL_SEEN)) |
2692 | S_setup_eval_state(aTHX_ reginfo); | |
2693 | else | |
2694 | reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL; | |
bf2039a9 | 2695 | } |
d3aa529c | 2696 | |
6eb5f6b9 | 2697 | /* If there is a "must appear" string, look for it. */ |
6eb5f6b9 | 2698 | |
288b8c02 | 2699 | if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) { |
e9105d30 GG |
2700 | /* We have to be careful. If the previous successful match |
2701 | was from this regex we don't want a subsequent partially | |
2702 | successful match to clobber the old results. | |
2703 | So when we detect this possibility we add a swap buffer | |
d8da0584 KW |
2704 | to the re, and switch the buffer each match. If we fail, |
2705 | we switch it back; otherwise we leave it swapped. | |
e9105d30 GG |
2706 | */ |
2707 | swap = prog->offs; | |
2708 | /* do we need a save destructor here for eval dies? */ | |
2709 | Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair); | |
495f47a5 DM |
2710 | DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, |
2711 | "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n", | |
2712 | PTR2UV(prog), | |
2713 | PTR2UV(swap), | |
2714 | PTR2UV(prog->offs) | |
2715 | )); | |
c74340f9 | 2716 | } |
6eb5f6b9 JH |
2717 | |
2718 | /* Simplest case: anchored match need be tried only once. */ | |
d3d47aac | 2719 | /* [unless only anchor is MBOL - implying multiline is set] */ |
8e1490ee | 2720 | if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) { |
3542935d | 2721 | if (s == startpos && regtry(reginfo, &s)) |
6eb5f6b9 | 2722 | goto got_it; |
8e1490ee | 2723 | else if (multiline || (prog->intflags & (PREGf_IMPLICIT | PREGf_ANCH_MBOL))) /* XXXX SBOL? */ |
6eb5f6b9 JH |
2724 | { |
2725 | char *end; | |
2726 | ||
2727 | if (minlen) | |
2728 | dontbother = minlen - 1; | |
1aa99e6b | 2729 | end = HOP3c(strend, -dontbother, strbeg) - 1; |
6eb5f6b9 | 2730 | /* for multiline we only have to try after newlines */ |
33b8afdf | 2731 | if (prog->check_substr || prog->check_utf8) { |
92f3d482 YO |
2732 | /* because of the goto we can not easily reuse the macros for bifurcating the |
2733 | unicode/non-unicode match modes here like we do elsewhere - demerphq */ | |
2734 | if (utf8_target) { | |
2735 | if (s == startpos) | |
2736 | goto after_try_utf8; | |
2737 | while (1) { | |
02d5137b | 2738 | if (regtry(reginfo, &s)) { |
92f3d482 YO |
2739 | goto got_it; |
2740 | } | |
2741 | after_try_utf8: | |
2742 | if (s > end) { | |
2743 | goto phooey; | |
2744 | } | |
2745 | if (prog->extflags & RXf_USE_INTUIT) { | |
52a21eb3 DM |
2746 | s = re_intuit_start(rx, sv, strbeg, |
2747 | s + UTF8SKIP(s), strend, flags, NULL); | |
92f3d482 YO |
2748 | if (!s) { |
2749 | goto phooey; | |
2750 | } | |
2751 | } | |
2752 | else { | |
2753 | s += UTF8SKIP(s); | |
2754 | } | |
2755 | } | |
2756 | } /* end search for check string in unicode */ | |
2757 | else { | |
2758 | if (s == startpos) { | |
2759 | goto after_try_latin; | |
2760 | } | |
2761 | while (1) { | |
02d5137b | 2762 | if (regtry(reginfo, &s)) { |
92f3d482 YO |
2763 | goto got_it; |
2764 | } | |
2765 | after_try_latin: | |
2766 | if (s > end) { | |
2767 | goto phooey; | |
2768 | } | |
2769 | if (prog->extflags & RXf_USE_INTUIT) { | |
52a21eb3 DM |
2770 | s = re_intuit_start(rx, sv, strbeg, |
2771 | s + 1, strend, flags, NULL); | |
92f3d482 YO |
2772 | if (!s) { |
2773 | goto phooey; | |
2774 | } | |
2775 | } | |
2776 | else { | |
2777 | s++; | |
2778 | } | |
2779 | } | |
2780 | } /* end search for check string in latin*/ | |
2781 | } /* end search for check string */ | |
2782 | else { /* search for newline */ | |
2783 | if (s > startpos) { | |
2784 | /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/ | |
6eb5f6b9 | 2785 | s--; |
92f3d482 | 2786 | } |
21eede78 YO |
2787 | /* We can use a more efficient search as newlines are the same in unicode as they are in latin */ |
2788 | while (s <= end) { /* note it could be possible to match at the end of the string */ | |
6eb5f6b9 | 2789 | if (*s++ == '\n') { /* don't need PL_utf8skip here */ |
02d5137b | 2790 | if (regtry(reginfo, &s)) |
6eb5f6b9 JH |
2791 | goto got_it; |
2792 | } | |
92f3d482 YO |
2793 | } |
2794 | } /* end search for newline */ | |
2795 | } /* end anchored/multiline check string search */ | |
6eb5f6b9 | 2796 | goto phooey; |
a8430a8b | 2797 | } else if (prog->intflags & PREGf_ANCH_GPOS) |
f9f4320a | 2798 | { |
a8430a8b YO |
2799 | /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */ |
2800 | assert(prog->intflags & PREGf_GPOS_SEEN); | |
2bfbe302 DM |
2801 | /* For anchored \G, the only position it can match from is |
2802 | * (ganch-gofs); we already set startpos to this above; if intuit | |
2803 | * moved us on from there, we can't possibly succeed */ | |
2804 | assert(startpos == reginfo->ganch - prog->gofs); | |
2805 | if (s == startpos && regtry(reginfo, &s)) | |
6eb5f6b9 JH |
2806 | goto got_it; |
2807 | goto phooey; | |
2808 | } | |
2809 | ||
2810 | /* Messy cases: unanchored match. */ | |
bbe252da | 2811 | if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) { |
6eb5f6b9 | 2812 | /* we have /x+whatever/ */ |
984e6dd1 | 2813 | /* it must be a one character string (XXXX Except is_utf8_pat?) */ |
33b8afdf | 2814 | char ch; |
bf93d4cc GS |
2815 | #ifdef DEBUGGING |
2816 | int did_match = 0; | |
2817 | #endif | |
f2ed9b32 | 2818 | if (utf8_target) { |
7e0d5ad7 KW |
2819 | if (! prog->anchored_utf8) { |
2820 | to_utf8_substr(prog); | |
2821 | } | |
2822 | ch = SvPVX_const(prog->anchored_utf8)[0]; | |
4cadc6a9 | 2823 | REXEC_FBC_SCAN( |
6eb5f6b9 | 2824 | if (*s == ch) { |
a3621e74 | 2825 | DEBUG_EXECUTE_r( did_match = 1 ); |
02d5137b | 2826 | if (regtry(reginfo, &s)) goto got_it; |
6eb5f6b9 JH |
2827 | s += UTF8SKIP(s); |
2828 | while (s < strend && *s == ch) | |
2829 | s += UTF8SKIP(s); | |
2830 | } | |
4cadc6a9 | 2831 | ); |
7e0d5ad7 | 2832 | |
6eb5f6b9 JH |
2833 | } |
2834 | else { | |
7e0d5ad7 KW |
2835 | if (! prog->anchored_substr) { |
2836 | if (! to_byte_substr(prog)) { | |
6b54ddc5 | 2837 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); |
7e0d5ad7 KW |
2838 | } |
2839 | } | |
2840 | ch = SvPVX_const(prog->anchored_substr)[0]; | |
4cadc6a9 | 2841 | REXEC_FBC_SCAN( |
6eb5f6b9 | 2842 | if (*s == ch) { |
a3621e74 | 2843 | DEBUG_EXECUTE_r( did_match = 1 ); |
02d5137b | 2844 | if (regtry(reginfo, &s)) goto got_it; |
6eb5f6b9 JH |
2845 | s++; |
2846 | while (s < strend && *s == ch) | |
2847 | s++; | |
2848 | } | |
4cadc6a9 | 2849 | ); |
6eb5f6b9 | 2850 | } |
a3621e74 | 2851 | DEBUG_EXECUTE_r(if (!did_match) |
bf93d4cc | 2852 | PerlIO_printf(Perl_debug_log, |
b7953727 JH |
2853 | "Did not find anchored character...\n") |
2854 | ); | |
6eb5f6b9 | 2855 | } |
a0714e2c SS |
2856 | else if (prog->anchored_substr != NULL |
2857 | || prog->anchored_utf8 != NULL | |
2858 | || ((prog->float_substr != NULL || prog->float_utf8 != NULL) | |
33b8afdf JH |
2859 | && prog->float_max_offset < strend - s)) { |
2860 | SV *must; | |
ea3daa5d FC |
2861 | SSize_t back_max; |
2862 | SSize_t back_min; | |
33b8afdf | 2863 | char *last; |
6eb5f6b9 | 2864 | char *last1; /* Last position checked before */ |
bf93d4cc GS |
2865 | #ifdef DEBUGGING |
2866 | int did_match = 0; | |
2867 | #endif | |
33b8afdf | 2868 | if (prog->anchored_substr || prog->anchored_utf8) { |
7e0d5ad7 KW |
2869 | if (utf8_target) { |
2870 | if (! prog->anchored_utf8) { | |
2871 | to_utf8_substr(prog); | |
2872 | } | |
2873 | must = prog->anchored_utf8; | |
2874 | } | |
2875 | else { | |
2876 | if (! prog->anchored_substr) { | |
2877 | if (! to_byte_substr(prog)) { | |
6b54ddc5 | 2878 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); |
7e0d5ad7 KW |
2879 | } |
2880 | } | |
2881 | must = prog->anchored_substr; | |
2882 | } | |
33b8afdf JH |
2883 | back_max = back_min = prog->anchored_offset; |
2884 | } else { | |
7e0d5ad7 KW |
2885 | if (utf8_target) { |
2886 | if (! prog->float_utf8) { | |
2887 | to_utf8_substr(prog); | |
2888 | } | |
2889 | must = prog->float_utf8; | |
2890 | } | |
2891 | else { | |
2892 | if (! prog->float_substr) { | |
2893 | if (! to_byte_substr(prog)) { | |
6b54ddc5 | 2894 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); |
7e0d5ad7 KW |
2895 | } |
2896 | } | |
2897 | must = prog->float_substr; | |
2898 | } | |
33b8afdf JH |
2899 | back_max = prog->float_max_offset; |
2900 | back_min = prog->float_min_offset; | |
2901 | } | |
1de06328 | 2902 | |
1de06328 YO |
2903 | if (back_min<0) { |
2904 | last = strend; | |
2905 | } else { | |
2906 | last = HOP3c(strend, /* Cannot start after this */ | |
ea3daa5d | 2907 | -(SSize_t)(CHR_SVLEN(must) |
1de06328 YO |
2908 | - (SvTAIL(must) != 0) + back_min), strbeg); |
2909 | } | |
9d9163fb | 2910 | if (s > reginfo->strbeg) |
6eb5f6b9 JH |
2911 | last1 = HOPc(s, -1); |
2912 | else | |
2913 | last1 = s - 1; /* bogus */ | |
2914 | ||
a0288114 | 2915 | /* XXXX check_substr already used to find "s", can optimize if |
6eb5f6b9 | 2916 | check_substr==must. */ |
bf05793b | 2917 | dontbother = 0; |
6eb5f6b9 JH |
2918 | strend = HOPc(strend, -dontbother); |
2919 | while ( (s <= last) && | |
e50d57d4 | 2920 | (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend), |
9041c2e3 | 2921 | (unsigned char*)strend, must, |
c33e64f0 | 2922 | multiline ? FBMrf_MULTILINE : 0)) ) { |
a3621e74 | 2923 | DEBUG_EXECUTE_r( did_match = 1 ); |
6eb5f6b9 JH |
2924 | if (HOPc(s, -back_max) > last1) { |
2925 | last1 = HOPc(s, -back_min); | |
2926 | s = HOPc(s, -back_max); | |
2927 | } | |
2928 | else { | |
9d9163fb DM |
2929 | char * const t = (last1 >= reginfo->strbeg) |
2930 | ? HOPc(last1, 1) : last1 + 1; | |
6eb5f6b9 JH |
2931 | |
2932 | last1 = HOPc(s, -back_min); | |
52657f30 | 2933 | s = t; |
6eb5f6b9 | 2934 | } |
f2ed9b32 | 2935 | if (utf8_target) { |
6eb5f6b9 | 2936 | while (s <= last1) { |
02d5137b | 2937 | if (regtry(reginfo, &s)) |
6eb5f6b9 | 2938 | goto got_it; |
7016d6eb DM |
2939 | if (s >= last1) { |
2940 | s++; /* to break out of outer loop */ | |
2941 | break; | |
2942 | } | |
2943 | s += UTF8SKIP(s); | |
6eb5f6b9 JH |
2944 | } |
2945 | } | |
2946 | else { | |
2947 | while (s <= last1) { | |
02d5137b | 2948 | if (regtry(reginfo, &s)) |
6eb5f6b9 JH |
2949 | goto got_it; |
2950 | s++; | |
2951 | } | |
2952 | } | |
2953 | } | |
ab3bbdeb | 2954 | DEBUG_EXECUTE_r(if (!did_match) { |
f2ed9b32 | 2955 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), |
ab3bbdeb YO |
2956 | SvPVX_const(must), RE_SV_DUMPLEN(must), 30); |
2957 | PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n", | |
33b8afdf | 2958 | ((must == prog->anchored_substr || must == prog->anchored_utf8) |
bf93d4cc | 2959 | ? "anchored" : "floating"), |
ab3bbdeb YO |
2960 | quoted, RE_SV_TAIL(must)); |
2961 | }); | |
6eb5f6b9 JH |
2962 | goto phooey; |
2963 | } | |
f8fc2ecf | 2964 | else if ( (c = progi->regstclass) ) { |
f14c76ed | 2965 | if (minlen) { |
f8fc2ecf | 2966 | const OPCODE op = OP(progi->regstclass); |
66e933ab | 2967 | /* don't bother with what can't match */ |
786e8c11 | 2968 | if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE) |
f14c76ed RGS |
2969 | strend = HOPc(strend, -(minlen - 1)); |
2970 | } | |
a3621e74 | 2971 | DEBUG_EXECUTE_r({ |
be8e71aa | 2972 | SV * const prop = sv_newmortal(); |
8b9781c9 | 2973 | regprop(prog, prop, c, reginfo, NULL); |
0df25f3d | 2974 | { |
f2ed9b32 | 2975 | RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1), |
ab3bbdeb | 2976 | s,strend-s,60); |
0df25f3d | 2977 | PerlIO_printf(Perl_debug_log, |
1c8f8eb1 | 2978 | "Matching stclass %.*s against %s (%d bytes)\n", |
e4f74956 | 2979 | (int)SvCUR(prop), SvPVX_const(prop), |
ab3bbdeb | 2980 | quoted, (int)(strend - s)); |
0df25f3d | 2981 | } |
ffc61ed2 | 2982 | }); |
f9176b44 | 2983 | if (find_byclass(prog, c, s, strend, reginfo)) |
6eb5f6b9 | 2984 | goto got_it; |
07be1b83 | 2985 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n")); |
d6a28714 JH |
2986 | } |
2987 | else { | |
2988 | dontbother = 0; | |
a0714e2c | 2989 | if (prog->float_substr != NULL || prog->float_utf8 != NULL) { |
33b8afdf | 2990 | /* Trim the end. */ |
6af40bd7 | 2991 | char *last= NULL; |
33b8afdf | 2992 | SV* float_real; |
c33e64f0 FC |
2993 | STRLEN len; |
2994 | const char *little; | |
33b8afdf | 2995 | |
7e0d5ad7 KW |
2996 | if (utf8_target) { |
2997 | if (! prog->float_utf8) { | |
2998 | to_utf8_substr(prog); | |
2999 | } | |
3000 | float_real = prog->float_utf8; | |
3001 | } | |
3002 | else { | |
3003 | if (! prog->float_substr) { | |
3004 | if (! to_byte_substr(prog)) { | |
6b54ddc5 | 3005 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); |
7e0d5ad7 KW |
3006 | } |
3007 | } | |
3008 | float_real = prog->float_substr; | |
3009 | } | |
d6a28714 | 3010 | |
c33e64f0 FC |
3011 | little = SvPV_const(float_real, len); |
3012 | if (SvTAIL(float_real)) { | |
7f18ad16 KW |
3013 | /* This means that float_real contains an artificial \n on |
3014 | * the end due to the presence of something like this: | |
3015 | * /foo$/ where we can match both "foo" and "foo\n" at the | |
3016 | * end of the string. So we have to compare the end of the | |
3017 | * string first against the float_real without the \n and | |
3018 | * then against the full float_real with the string. We | |
3019 | * have to watch out for cases where the string might be | |
3020 | * smaller than the float_real or the float_real without | |
3021 | * the \n. */ | |
1a13b075 YO |
3022 | char *checkpos= strend - len; |
3023 | DEBUG_OPTIMISE_r( | |
3024 | PerlIO_printf(Perl_debug_log, | |
3025 | "%sChecking for float_real.%s\n", | |
3026 | PL_colors[4], PL_colors[5])); | |
3027 | if (checkpos + 1 < strbeg) { | |
7f18ad16 KW |
3028 | /* can't match, even if we remove the trailing \n |
3029 | * string is too short to match */ | |
1a13b075 YO |
3030 | DEBUG_EXECUTE_r( |
3031 | PerlIO_printf(Perl_debug_log, | |
3032 | "%sString shorter than required trailing substring, cannot match.%s\n", | |
3033 | PL_colors[4], PL_colors[5])); | |
3034 | goto phooey; | |
3035 | } else if (memEQ(checkpos + 1, little, len - 1)) { | |
7f18ad16 KW |
3036 | /* can match, the end of the string matches without the |
3037 | * "\n" */ | |
1a13b075 YO |
3038 | last = checkpos + 1; |
3039 | } else if (checkpos < strbeg) { | |
7f18ad16 KW |
3040 | /* cant match, string is too short when the "\n" is |
3041 | * included */ | |
1a13b075 YO |
3042 | DEBUG_EXECUTE_r( |
3043 | PerlIO_printf(Perl_debug_log, | |
3044 | "%sString does not contain required trailing substring, cannot match.%s\n", | |
3045 | PL_colors[4], PL_colors[5])); | |
3046 | goto phooey; | |
3047 | } else if (!multiline) { | |
7f18ad16 KW |
3048 | /* non multiline match, so compare with the "\n" at the |
3049 | * end of the string */ | |
1a13b075 YO |
3050 | if (memEQ(checkpos, little, len)) { |
3051 | last= checkpos; | |
3052 | } else { | |
3053 | DEBUG_EXECUTE_r( | |
3054 | PerlIO_printf(Perl_debug_log, | |
3055 | "%sString does not contain required trailing substring, cannot match.%s\n", | |
3056 | PL_colors[4], PL_colors[5])); | |
3057 | goto phooey; | |
3058 | } | |
3059 | } else { | |
7f18ad16 KW |
3060 | /* multiline match, so we have to search for a place |
3061 | * where the full string is located */ | |
d6a28714 | 3062 | goto find_last; |
1a13b075 | 3063 | } |
c33e64f0 | 3064 | } else { |
d6a28714 | 3065 | find_last: |
9041c2e3 | 3066 | if (len) |
d6a28714 | 3067 | last = rninstr(s, strend, little, little + len); |
b8c5462f | 3068 | else |
a0288114 | 3069 | last = strend; /* matching "$" */ |
b8c5462f | 3070 | } |
6af40bd7 | 3071 | if (!last) { |
7f18ad16 KW |
3072 | /* at one point this block contained a comment which was |
3073 | * probably incorrect, which said that this was a "should not | |
3074 | * happen" case. Even if it was true when it was written I am | |
3075 | * pretty sure it is not anymore, so I have removed the comment | |
3076 | * and replaced it with this one. Yves */ | |
6bda09f9 YO |
3077 | DEBUG_EXECUTE_r( |
3078 | PerlIO_printf(Perl_debug_log, | |
b729e729 YO |
3079 | "%sString does not contain required substring, cannot match.%s\n", |
3080 | PL_colors[4], PL_colors[5] | |
6af40bd7 YO |
3081 | )); |
3082 | goto phooey; | |
bf93d4cc | 3083 | } |
d6a28714 JH |
3084 | dontbother = strend - last + prog->float_min_offset; |
3085 | } | |
3086 | if (minlen && (dontbother < minlen)) | |
3087 | dontbother = minlen - 1; | |
3088 | strend -= dontbother; /* this one's always in bytes! */ | |
3089 | /* We don't know much -- general case. */ | |
f2ed9b32 | 3090 | if (utf8_target) { |
d6a28714 | 3091 | for (;;) { |
02d5137b | 3092 | if (regtry(reginfo, &s)) |
d6a28714 JH |
3093 | goto got_it; |
3094 | if (s >= strend) | |
3095 | break; | |
b8c5462f | 3096 | s += UTF8SKIP(s); |
d6a28714 JH |
3097 | }; |
3098 | } | |
3099 | else { | |
3100 | do { | |
02d5137b | 3101 | if (regtry(reginfo, &s)) |
d6a28714 JH |
3102 | goto got_it; |
3103 | } while (s++ < strend); | |
3104 | } | |
3105 | } | |
3106 | ||
3107 | /* Failure. */ | |
3108 | goto phooey; | |
3109 | ||
3110 | got_it: | |
d5e7783a DM |
3111 | /* s/// doesn't like it if $& is earlier than where we asked it to |
3112 | * start searching (which can happen on something like /.\G/) */ | |
3113 | if ( (flags & REXEC_FAIL_ON_UNDERFLOW) | |
3114 | && (prog->offs[0].start < stringarg - strbeg)) | |
3115 | { | |
3116 | /* this should only be possible under \G */ | |
58430ea8 | 3117 | assert(prog->intflags & PREGf_GPOS_SEEN); |
d5e7783a DM |
3118 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
3119 | "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n")); | |
3120 | goto phooey; | |
3121 | } | |
3122 | ||
495f47a5 DM |
3123 | DEBUG_BUFFERS_r( |
3124 | if (swap) | |
3125 | PerlIO_printf(Perl_debug_log, | |
3126 | "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n", | |
3127 | PTR2UV(prog), | |
3128 | PTR2UV(swap) | |
3129 | ); | |
3130 | ); | |
e9105d30 | 3131 | Safefree(swap); |
d6a28714 | 3132 | |
bf2039a9 DM |
3133 | /* clean up; this will trigger destructors that will free all slabs |
3134 | * above the current one, and cleanup the regmatch_info_aux | |
3135 | * and regmatch_info_aux_eval sructs */ | |
8adc0f72 | 3136 | |
006f26b2 DM |
3137 | LEAVE_SCOPE(oldsave); |
3138 | ||
5daac39c NC |
3139 | if (RXp_PAREN_NAMES(prog)) |
3140 | (void)hv_iterinit(RXp_PAREN_NAMES(prog)); | |
d6a28714 JH |
3141 | |
3142 | /* make sure $`, $&, $', and $digit will work later */ | |
60165aa4 | 3143 | if ( !(flags & REXEC_NOT_FIRST) ) |
749f4950 | 3144 | S_reg_set_capture_string(aTHX_ rx, |
60165aa4 DM |
3145 | strbeg, reginfo->strend, |
3146 | sv, flags, utf8_target); | |
9041c2e3 | 3147 | |
d6a28714 JH |
3148 | return 1; |
3149 | ||
3150 | phooey: | |
a3621e74 | 3151 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n", |
e4584336 | 3152 | PL_colors[4], PL_colors[5])); |
8adc0f72 | 3153 | |
bf2039a9 DM |
3154 | /* clean up; this will trigger destructors that will free all slabs |
3155 | * above the current one, and cleanup the regmatch_info_aux | |
3156 | * and regmatch_info_aux_eval sructs */ | |
8adc0f72 | 3157 | |
006f26b2 DM |
3158 | LEAVE_SCOPE(oldsave); |
3159 | ||
e9105d30 | 3160 | if (swap) { |
c74340f9 | 3161 | /* we failed :-( roll it back */ |
495f47a5 DM |
3162 | DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, |
3163 | "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n", | |
3164 | PTR2UV(prog), | |
3165 | PTR2UV(prog->offs), | |
3166 | PTR2UV(swap) | |
3167 | )); | |
e9105d30 GG |
3168 | Safefree(prog->offs); |
3169 | prog->offs = swap; | |
3170 | } | |
d6a28714 JH |
3171 | return 0; |
3172 | } | |
3173 | ||
6bda09f9 | 3174 | |
b3d298be | 3175 | /* Set which rex is pointed to by PL_reg_curpm, handling ref counting. |
ec43f78b | 3176 | * Do inc before dec, in case old and new rex are the same */ |
baa60164 | 3177 | #define SET_reg_curpm(Re2) \ |
bf2039a9 | 3178 | if (reginfo->info_aux_eval) { \ |
ec43f78b DM |
3179 | (void)ReREFCNT_inc(Re2); \ |
3180 | ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \ | |
3181 | PM_SETRE((PL_reg_curpm), (Re2)); \ | |
3182 | } | |
3183 | ||
3184 | ||
d6a28714 JH |
3185 | /* |
3186 | - regtry - try match at specific point | |
3187 | */ | |
3188 | STATIC I32 /* 0 failure, 1 success */ | |
f73aaa43 | 3189 | S_regtry(pTHX_ regmatch_info *reginfo, char **startposp) |
d6a28714 | 3190 | { |
d6a28714 | 3191 | CHECKPOINT lastcp; |
288b8c02 | 3192 | REGEXP *const rx = reginfo->prog; |
8d919b0a | 3193 | regexp *const prog = ReANY(rx); |
99a90e59 | 3194 | SSize_t result; |
f8fc2ecf | 3195 | RXi_GET_DECL(prog,progi); |
a3621e74 | 3196 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
3197 | |
3198 | PERL_ARGS_ASSERT_REGTRY; | |
3199 | ||
24b23f37 | 3200 | reginfo->cutpoint=NULL; |
d6a28714 | 3201 | |
9d9163fb | 3202 | prog->offs[0].start = *startposp - reginfo->strbeg; |
d6a28714 | 3203 | prog->lastparen = 0; |
03994de8 | 3204 | prog->lastcloseparen = 0; |
d6a28714 JH |
3205 | |
3206 | /* XXXX What this code is doing here?!!! There should be no need | |
b93070ed | 3207 | to do this again and again, prog->lastparen should take care of |
3dd2943c | 3208 | this! --ilya*/ |
dafc8851 JH |
3209 | |
3210 | /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code. | |
3211 | * Actually, the code in regcppop() (which Ilya may be meaning by | |
b93070ed | 3212 | * prog->lastparen), is not needed at all by the test suite |
225593e1 DM |
3213 | * (op/regexp, op/pat, op/split), but that code is needed otherwise |
3214 | * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/ | |
3215 | * Meanwhile, this code *is* needed for the | |
daf18116 JH |
3216 | * above-mentioned test suite tests to succeed. The common theme |
3217 | * on those tests seems to be returning null fields from matches. | |
225593e1 | 3218 | * --jhi updated by dapm */ |
dafc8851 | 3219 | #if 1 |
d6a28714 | 3220 | if (prog->nparens) { |
b93070ed | 3221 | regexp_paren_pair *pp = prog->offs; |
eb578fdb | 3222 | I32 i; |
b93070ed | 3223 | for (i = prog->nparens; i > (I32)prog->lastparen; i--) { |
f0ab9afb NC |
3224 | ++pp; |
3225 | pp->start = -1; | |
3226 | pp->end = -1; | |
d6a28714 JH |
3227 | } |
3228 | } | |
dafc8851 | 3229 | #endif |
02db2b7b | 3230 | REGCP_SET(lastcp); |
f73aaa43 DM |
3231 | result = regmatch(reginfo, *startposp, progi->program + 1); |
3232 | if (result != -1) { | |
3233 | prog->offs[0].end = result; | |
d6a28714 JH |
3234 | return 1; |
3235 | } | |
24b23f37 | 3236 | if (reginfo->cutpoint) |
f73aaa43 | 3237 | *startposp= reginfo->cutpoint; |
02db2b7b | 3238 | REGCP_UNWIND(lastcp); |
d6a28714 JH |
3239 | return 0; |
3240 | } | |
3241 | ||
02db2b7b | 3242 | |
8ba1375e MJD |
3243 | #define sayYES goto yes |
3244 | #define sayNO goto no | |
262b90c4 | 3245 | #define sayNO_SILENT goto no_silent |
8ba1375e | 3246 | |
f9f4320a YO |
3247 | /* we dont use STMT_START/END here because it leads to |
3248 | "unreachable code" warnings, which are bogus, but distracting. */ | |
3249 | #define CACHEsayNO \ | |
c476f425 | 3250 | if (ST.cache_mask) \ |
2ac8ff4b | 3251 | reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \ |
f9f4320a | 3252 | sayNO |
3298f257 | 3253 | |
a3621e74 | 3254 | /* this is used to determine how far from the left messages like |
265c4333 YO |
3255 | 'failed...' are printed. It should be set such that messages |
3256 | are inline with the regop output that created them. | |
a3621e74 | 3257 | */ |
265c4333 | 3258 | #define REPORT_CODE_OFF 32 |
a3621e74 YO |
3259 | |
3260 | ||
40a82448 DM |
3261 | #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */ |
3262 | #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */ | |
79a2a0e8 KW |
3263 | #define CHRTEST_NOT_A_CP_1 -999 |
3264 | #define CHRTEST_NOT_A_CP_2 -998 | |
9e137952 | 3265 | |
5d9a96ca DM |
3266 | /* grab a new slab and return the first slot in it */ |
3267 | ||
3268 | STATIC regmatch_state * | |
3269 | S_push_slab(pTHX) | |
3270 | { | |
a35a87e7 | 3271 | #if PERL_VERSION < 9 && !defined(PERL_CORE) |
54df2634 NC |
3272 | dMY_CXT; |
3273 | #endif | |
5d9a96ca DM |
3274 | regmatch_slab *s = PL_regmatch_slab->next; |
3275 | if (!s) { | |
3276 | Newx(s, 1, regmatch_slab); | |
3277 | s->prev = PL_regmatch_slab; | |
3278 | s->next = NULL; | |
3279 | PL_regmatch_slab->next = s; | |
3280 | } | |
3281 | PL_regmatch_slab = s; | |
86545054 | 3282 | return SLAB_FIRST(s); |
5d9a96ca | 3283 | } |
5b47454d | 3284 | |
95b24440 | 3285 | |
40a82448 DM |
3286 | /* push a new state then goto it */ |
3287 | ||
4d5016e5 DM |
3288 | #define PUSH_STATE_GOTO(state, node, input) \ |
3289 | pushinput = input; \ | |
40a82448 DM |
3290 | scan = node; \ |
3291 | st->resume_state = state; \ | |
3292 | goto push_state; | |
3293 | ||
3294 | /* push a new state with success backtracking, then goto it */ | |
3295 | ||
4d5016e5 DM |
3296 | #define PUSH_YES_STATE_GOTO(state, node, input) \ |
3297 | pushinput = input; \ | |
40a82448 DM |
3298 | scan = node; \ |
3299 | st->resume_state = state; \ | |
3300 | goto push_yes_state; | |
3301 | ||
aa283a38 | 3302 | |
aa283a38 | 3303 | |
4d5016e5 | 3304 | |
d6a28714 | 3305 | /* |
95b24440 | 3306 | |
bf1f174e DM |
3307 | regmatch() - main matching routine |
3308 | ||
3309 | This is basically one big switch statement in a loop. We execute an op, | |
3310 | set 'next' to point the next op, and continue. If we come to a point which | |
3311 | we may need to backtrack to on failure such as (A|B|C), we push a | |
3312 | backtrack state onto the backtrack stack. On failure, we pop the top | |
3313 | state, and re-enter the loop at the state indicated. If there are no more | |
3314 | states to pop, we return failure. | |
3315 | ||
3316 | Sometimes we also need to backtrack on success; for example /A+/, where | |
3317 | after successfully matching one A, we need to go back and try to | |
3318 | match another one; similarly for lookahead assertions: if the assertion | |
3319 | completes successfully, we backtrack to the state just before the assertion | |
3320 | and then carry on. In these cases, the pushed state is marked as | |
3321 | 'backtrack on success too'. This marking is in fact done by a chain of | |
3322 | pointers, each pointing to the previous 'yes' state. On success, we pop to | |
3323 | the nearest yes state, discarding any intermediate failure-only states. | |
3324 | Sometimes a yes state is pushed just to force some cleanup code to be | |
3325 | called at the end of a successful match or submatch; e.g. (??{$re}) uses | |
3326 | it to free the inner regex. | |
3327 | ||
3328 | Note that failure backtracking rewinds the cursor position, while | |
3329 | success backtracking leaves it alone. | |
3330 | ||
3331 | A pattern is complete when the END op is executed, while a subpattern | |
3332 | such as (?=foo) is complete when the SUCCESS op is executed. Both of these | |
3333 | ops trigger the "pop to last yes state if any, otherwise return true" | |
3334 | behaviour. | |
3335 | ||
3336 | A common convention in this function is to use A and B to refer to the two | |
3337 | subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is | |
3338 | the subpattern to be matched possibly multiple times, while B is the entire | |
3339 | rest of the pattern. Variable and state names reflect this convention. | |
3340 | ||
3341 | The states in the main switch are the union of ops and failure/success of | |
3342 | substates associated with with that op. For example, IFMATCH is the op | |
3343 | that does lookahead assertions /(?=A)B/ and so the IFMATCH state means | |
3344 | 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just | |
3345 | successfully matched A and IFMATCH_A_fail is a state saying that we have | |
3346 | just failed to match A. Resume states always come in pairs. The backtrack | |
3347 | state we push is marked as 'IFMATCH_A', but when that is popped, we resume | |
3348 | at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking | |
3349 | on success or failure. | |
3350 | ||
3351 | The struct that holds a backtracking state is actually a big union, with | |
3352 | one variant for each major type of op. The variable st points to the | |
3353 | top-most backtrack struct. To make the code clearer, within each | |
3354 | block of code we #define ST to alias the relevant union. | |
3355 | ||
3356 | Here's a concrete example of a (vastly oversimplified) IFMATCH | |
3357 | implementation: | |
3358 | ||
3359 | switch (state) { | |
3360 | .... | |
3361 | ||
3362 | #define ST st->u.ifmatch | |
3363 | ||
3364 | case IFMATCH: // we are executing the IFMATCH op, (?=A)B | |
3365 | ST.foo = ...; // some state we wish to save | |
95b24440 | 3366 | ... |
bf1f174e DM |
3367 | // push a yes backtrack state with a resume value of |
3368 | // IFMATCH_A/IFMATCH_A_fail, then continue execution at the | |
3369 | // first node of A: | |
4d5016e5 | 3370 | PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput); |
bf1f174e DM |
3371 | // NOTREACHED |
3372 | ||
3373 | case IFMATCH_A: // we have successfully executed A; now continue with B | |
3374 | next = B; | |
3375 | bar = ST.foo; // do something with the preserved value | |
3376 | break; | |
3377 | ||
3378 | case IFMATCH_A_fail: // A failed, so the assertion failed | |
3379 | ...; // do some housekeeping, then ... | |
3380 | sayNO; // propagate the failure | |
3381 | ||
3382 | #undef ST | |
95b24440 | 3383 | |
bf1f174e DM |
3384 | ... |
3385 | } | |
95b24440 | 3386 | |
bf1f174e DM |
3387 | For any old-timers reading this who are familiar with the old recursive |
3388 | approach, the code above is equivalent to: | |
95b24440 | 3389 | |
bf1f174e DM |
3390 | case IFMATCH: // we are executing the IFMATCH op, (?=A)B |
3391 | { | |
3392 | int foo = ... | |
95b24440 | 3393 | ... |
bf1f174e DM |
3394 | if (regmatch(A)) { |
3395 | next = B; | |
3396 | bar = foo; | |
3397 | break; | |
95b24440 | 3398 | } |
bf1f174e DM |
3399 | ...; // do some housekeeping, then ... |
3400 | sayNO; // propagate the failure | |
95b24440 | 3401 | } |
bf1f174e DM |
3402 | |
3403 | The topmost backtrack state, pointed to by st, is usually free. If you | |
3404 | want to claim it, populate any ST.foo fields in it with values you wish to | |
3405 | save, then do one of | |
3406 | ||
4d5016e5 DM |
3407 | PUSH_STATE_GOTO(resume_state, node, newinput); |
3408 | PUSH_YES_STATE_GOTO(resume_state, node, newinput); | |
bf1f174e DM |
3409 | |
3410 | which sets that backtrack state's resume value to 'resume_state', pushes a | |
3411 | new free entry to the top of the backtrack stack, then goes to 'node'. | |
3412 | On backtracking, the free slot is popped, and the saved state becomes the | |
3413 | new free state. An ST.foo field in this new top state can be temporarily | |
3414 | accessed to retrieve values, but once the main loop is re-entered, it | |
3415 | becomes available for reuse. | |
3416 | ||
3417 | Note that the depth of the backtrack stack constantly increases during the | |
3418 | left-to-right execution of the pattern, rather than going up and down with | |
3419 | the pattern nesting. For example the stack is at its maximum at Z at the | |
3420 | end of the pattern, rather than at X in the following: | |
3421 | ||
3422 | /(((X)+)+)+....(Y)+....Z/ | |
3423 | ||
3424 | The only exceptions to this are lookahead/behind assertions and the cut, | |
3425 | (?>A), which pop all the backtrack states associated with A before | |
3426 | continuing. | |
3427 | ||
486ec47a | 3428 | Backtrack state structs are allocated in slabs of about 4K in size. |
bf1f174e DM |
3429 | PL_regmatch_state and st always point to the currently active state, |
3430 | and PL_regmatch_slab points to the slab currently containing | |
3431 | PL_regmatch_state. The first time regmatch() is called, the first slab is | |
3432 | allocated, and is never freed until interpreter destruction. When the slab | |
3433 | is full, a new one is allocated and chained to the end. At exit from | |
3434 | regmatch(), slabs allocated since entry are freed. | |
3435 | ||
3436 | */ | |
95b24440 | 3437 | |
40a82448 | 3438 | |
5bc10b2c | 3439 | #define DEBUG_STATE_pp(pp) \ |
265c4333 | 3440 | DEBUG_STATE_r({ \ |
baa60164 | 3441 | DUMP_EXEC_POS(locinput, scan, utf8_target); \ |
5bc10b2c | 3442 | PerlIO_printf(Perl_debug_log, \ |
5d458dd8 | 3443 | " %*s"pp" %s%s%s%s%s\n", \ |
5bc10b2c | 3444 | depth*2, "", \ |
baa60164 | 3445 | PL_reg_name[st->resume_state], \ |
5d458dd8 YO |
3446 | ((st==yes_state||st==mark_state) ? "[" : ""), \ |
3447 | ((st==yes_state) ? "Y" : ""), \ | |
3448 | ((st==mark_state) ? "M" : ""), \ | |
3449 | ((st==yes_state||st==mark_state) ? "]" : "") \ | |
3450 | ); \ | |
265c4333 | 3451 | }); |
5bc10b2c | 3452 | |
40a82448 | 3453 | |
3dab1dad | 3454 | #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1) |
95b24440 | 3455 | |
3df15adc | 3456 | #ifdef DEBUGGING |
5bc10b2c | 3457 | |
ab3bbdeb | 3458 | STATIC void |
f2ed9b32 | 3459 | S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target, |
ab3bbdeb YO |
3460 | const char *start, const char *end, const char *blurb) |
3461 | { | |
efd26800 | 3462 | const bool utf8_pat = RX_UTF8(prog) ? 1 : 0; |
7918f24d NC |
3463 | |
3464 | PERL_ARGS_ASSERT_DEBUG_START_MATCH; | |
3465 | ||
ab3bbdeb YO |
3466 | if (!PL_colorset) |
3467 | reginitcolors(); | |
3468 | { | |
3469 | RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0), | |
d2c6dc5e | 3470 | RX_PRECOMP_const(prog), RX_PRELEN(prog), 60); |
ab3bbdeb | 3471 | |
f2ed9b32 | 3472 | RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1), |
ab3bbdeb YO |
3473 | start, end - start, 60); |
3474 | ||
3475 | PerlIO_printf(Perl_debug_log, | |
3476 | "%s%s REx%s %s against %s\n", | |
3477 | PL_colors[4], blurb, PL_colors[5], s0, s1); | |
3478 | ||
f2ed9b32 | 3479 | if (utf8_target||utf8_pat) |
1de06328 YO |
3480 | PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n", |
3481 | utf8_pat ? "pattern" : "", | |
f2ed9b32 KW |
3482 | utf8_pat && utf8_target ? " and " : "", |
3483 | utf8_target ? "string" : "" | |
ab3bbdeb YO |
3484 | ); |
3485 | } | |
3486 | } | |
3df15adc YO |
3487 | |
3488 | STATIC void | |
786e8c11 YO |
3489 | S_dump_exec_pos(pTHX_ const char *locinput, |
3490 | const regnode *scan, | |
3491 | const char *loc_regeol, | |
3492 | const char *loc_bostr, | |
3493 | const char *loc_reg_starttry, | |
f2ed9b32 | 3494 | const bool utf8_target) |
07be1b83 | 3495 | { |
786e8c11 | 3496 | const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4]; |
07be1b83 | 3497 | const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */ |
786e8c11 | 3498 | int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput); |
07be1b83 YO |
3499 | /* The part of the string before starttry has one color |
3500 | (pref0_len chars), between starttry and current | |
3501 | position another one (pref_len - pref0_len chars), | |
3502 | after the current position the third one. | |
3503 | We assume that pref0_len <= pref_len, otherwise we | |
3504 | decrease pref0_len. */ | |
786e8c11 YO |
3505 | int pref_len = (locinput - loc_bostr) > (5 + taill) - l |
3506 | ? (5 + taill) - l : locinput - loc_bostr; | |
07be1b83 YO |
3507 | int pref0_len; |
3508 | ||
7918f24d NC |
3509 | PERL_ARGS_ASSERT_DUMP_EXEC_POS; |
3510 | ||
f2ed9b32 | 3511 | while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len))) |
07be1b83 | 3512 | pref_len++; |
786e8c11 YO |
3513 | pref0_len = pref_len - (locinput - loc_reg_starttry); |
3514 | if (l + pref_len < (5 + taill) && l < loc_regeol - locinput) | |
3515 | l = ( loc_regeol - locinput > (5 + taill) - pref_len | |
3516 | ? (5 + taill) - pref_len : loc_regeol - locinput); | |
f2ed9b32 | 3517 | while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l))) |
07be1b83 YO |
3518 | l--; |
3519 | if (pref0_len < 0) | |
3520 | pref0_len = 0; | |
3521 | if (pref0_len > pref_len) | |
3522 | pref0_len = pref_len; | |
3523 | { | |
f2ed9b32 | 3524 | const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0; |
0df25f3d | 3525 | |
ab3bbdeb | 3526 | RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0), |
1de06328 | 3527 | (locinput - pref_len),pref0_len, 60, 4, 5); |
0df25f3d | 3528 | |
ab3bbdeb | 3529 | RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1), |
3df15adc | 3530 | (locinput - pref_len + pref0_len), |
1de06328 | 3531 | pref_len - pref0_len, 60, 2, 3); |
0df25f3d | 3532 | |
ab3bbdeb | 3533 | RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2), |
1de06328 | 3534 | locinput, loc_regeol - locinput, 10, 0, 1); |
0df25f3d | 3535 | |
1de06328 | 3536 | const STRLEN tlen=len0+len1+len2; |
3df15adc | 3537 | PerlIO_printf(Perl_debug_log, |
ab3bbdeb | 3538 | "%4"IVdf" <%.*s%.*s%s%.*s>%*s|", |
786e8c11 | 3539 | (IV)(locinput - loc_bostr), |
07be1b83 | 3540 | len0, s0, |
07be1b83 | 3541 | len1, s1, |
07be1b83 | 3542 | (docolor ? "" : "> <"), |
07be1b83 | 3543 | len2, s2, |
f9f4320a | 3544 | (int)(tlen > 19 ? 0 : 19 - tlen), |
07be1b83 YO |
3545 | ""); |
3546 | } | |
3547 | } | |
3df15adc | 3548 | |
07be1b83 YO |
3549 | #endif |
3550 | ||
0a4db386 YO |
3551 | /* reg_check_named_buff_matched() |
3552 | * Checks to see if a named buffer has matched. The data array of | |
3553 | * buffer numbers corresponding to the buffer is expected to reside | |
3554 | * in the regexp->data->data array in the slot stored in the ARG() of | |
3555 | * node involved. Note that this routine doesn't actually care about the | |
3556 | * name, that information is not preserved from compilation to execution. | |
3557 | * Returns the index of the leftmost defined buffer with the given name | |
3558 | * or 0 if non of the buffers matched. | |
3559 | */ | |
3560 | STATIC I32 | |
dc3bf405 | 3561 | S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan) |
7918f24d | 3562 | { |
0a4db386 | 3563 | I32 n; |
f8fc2ecf | 3564 | RXi_GET_DECL(rex,rexi); |
ad64d0ec | 3565 | SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
0a4db386 | 3566 | I32 *nums=(I32*)SvPVX(sv_dat); |
7918f24d NC |
3567 | |
3568 | PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED; | |
3569 | ||
0a4db386 | 3570 | for ( n=0; n<SvIVX(sv_dat); n++ ) { |
b93070ed DM |
3571 | if ((I32)rex->lastparen >= nums[n] && |
3572 | rex->offs[nums[n]].end != -1) | |
0a4db386 YO |
3573 | { |
3574 | return nums[n]; | |
3575 | } | |
3576 | } | |
3577 | return 0; | |
3578 | } | |
3579 | ||
2f554ef7 | 3580 | |
c74f6de9 | 3581 | static bool |
984e6dd1 | 3582 | S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p, |
aed7b151 | 3583 | U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo) |
c74f6de9 | 3584 | { |
79a2a0e8 KW |
3585 | /* This function determines if there are one or two characters that match |
3586 | * the first character of the passed-in EXACTish node <text_node>, and if | |
3587 | * so, returns them in the passed-in pointers. | |
c74f6de9 | 3588 | * |
79a2a0e8 KW |
3589 | * If it determines that no possible character in the target string can |
3590 | * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if | |
3591 | * the first character in <text_node> requires UTF-8 to represent, and the | |
3592 | * target string isn't in UTF-8.) | |
c74f6de9 | 3593 | * |
79a2a0e8 KW |
3594 | * If there are more than two characters that could match the beginning of |
3595 | * <text_node>, or if more context is required to determine a match or not, | |
3596 | * it sets both *<c1p> and *<c2p> to CHRTEST_VOID. | |
3597 | * | |
3598 | * The motiviation behind this function is to allow the caller to set up | |
3599 | * tight loops for matching. If <text_node> is of type EXACT, there is | |
3600 | * only one possible character that can match its first character, and so | |
3601 | * the situation is quite simple. But things get much more complicated if | |
3602 | * folding is involved. It may be that the first character of an EXACTFish | |
3603 | * node doesn't participate in any possible fold, e.g., punctuation, so it | |
3604 | * can be matched only by itself. The vast majority of characters that are | |
3605 | * in folds match just two things, their lower and upper-case equivalents. | |
3606 | * But not all are like that; some have multiple possible matches, or match | |
3607 | * sequences of more than one character. This function sorts all that out. | |
3608 | * | |
3609 | * Consider the patterns A*B or A*?B where A and B are arbitrary. In a | |
3610 | * loop of trying to match A*, we know we can't exit where the thing | |
3611 | * following it isn't a B. And something can't be a B unless it is the | |
3612 | * beginning of B. By putting a quick test for that beginning in a tight | |
3613 | * loop, we can rule out things that can't possibly be B without having to | |
3614 | * break out of the loop, thus avoiding work. Similarly, if A is a single | |
3615 | * character, we can make a tight loop matching A*, using the outputs of | |
3616 | * this function. | |
3617 | * | |
3618 | * If the target string to match isn't in UTF-8, and there aren't | |
3619 | * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to | |
3620 | * the one or two possible octets (which are characters in this situation) | |
3621 | * that can match. In all cases, if there is only one character that can | |
3622 | * match, *<c1p> and *<c2p> will be identical. | |
3623 | * | |
3624 | * If the target string is in UTF-8, the buffers pointed to by <c1_utf8> | |
3625 | * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that | |
3626 | * can match the beginning of <text_node>. They should be declared with at | |
3627 | * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is | |
3628 | * undefined what these contain.) If one or both of the buffers are | |
3629 | * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the | |
3630 | * corresponding invariant. If variant, the corresponding *<c1p> and/or | |
3631 | * *<c2p> will be set to a negative number(s) that shouldn't match any code | |
3632 | * point (unless inappropriately coerced to unsigned). *<c1p> will equal | |
3633 | * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */ | |
c74f6de9 | 3634 | |
ba44c216 | 3635 | const bool utf8_target = reginfo->is_utf8_target; |
79a2a0e8 | 3636 | |
9cd990bf JH |
3637 | UV c1 = (UV)CHRTEST_NOT_A_CP_1; |
3638 | UV c2 = (UV)CHRTEST_NOT_A_CP_2; | |
79a2a0e8 | 3639 | bool use_chrtest_void = FALSE; |
aed7b151 | 3640 | const bool is_utf8_pat = reginfo->is_utf8_pat; |
79a2a0e8 KW |
3641 | |
3642 | /* Used when we have both utf8 input and utf8 output, to avoid converting | |
3643 | * to/from code points */ | |
3644 | bool utf8_has_been_setup = FALSE; | |
3645 | ||
c74f6de9 KW |
3646 | dVAR; |
3647 | ||
b4291290 | 3648 | U8 *pat = (U8*)STRING(text_node); |
a6715020 | 3649 | U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' }; |
c74f6de9 | 3650 | |
79a2a0e8 KW |
3651 | if (OP(text_node) == EXACT) { |
3652 | ||
3653 | /* In an exact node, only one thing can be matched, that first | |
3654 | * character. If both the pat and the target are UTF-8, we can just | |
3655 | * copy the input to the output, avoiding finding the code point of | |
3656 | * that character */ | |
984e6dd1 | 3657 | if (!is_utf8_pat) { |
79a2a0e8 KW |
3658 | c2 = c1 = *pat; |
3659 | } | |
3660 | else if (utf8_target) { | |
3661 | Copy(pat, c1_utf8, UTF8SKIP(pat), U8); | |
3662 | Copy(pat, c2_utf8, UTF8SKIP(pat), U8); | |
3663 | utf8_has_been_setup = TRUE; | |
3664 | } | |
3665 | else { | |
3666 | c2 = c1 = valid_utf8_to_uvchr(pat, NULL); | |
c74f6de9 | 3667 | } |
79a2a0e8 | 3668 | } |
31f05a37 KW |
3669 | else { /* an EXACTFish node */ |
3670 | U8 *pat_end = pat + STR_LEN(text_node); | |
3671 | ||
3672 | /* An EXACTFL node has at least some characters unfolded, because what | |
3673 | * they match is not known until now. So, now is the time to fold | |
3674 | * the first few of them, as many as are needed to determine 'c1' and | |
3675 | * 'c2' later in the routine. If the pattern isn't UTF-8, we only need | |
3676 | * to fold if in a UTF-8 locale, and then only the Sharp S; everything | |
3677 | * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we | |
3678 | * need to fold as many characters as a single character can fold to, | |
3679 | * so that later we can check if the first ones are such a multi-char | |
3680 | * fold. But, in such a pattern only locale-problematic characters | |
3681 | * aren't folded, so we can skip this completely if the first character | |
3682 | * in the node isn't one of the tricky ones */ | |
3683 | if (OP(text_node) == EXACTFL) { | |
3684 | ||
3685 | if (! is_utf8_pat) { | |
3686 | if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S) | |
3687 | { | |
3688 | folded[0] = folded[1] = 's'; | |
3689 | pat = folded; | |
3690 | pat_end = folded + 2; | |
3691 | } | |
3692 | } | |
3693 | else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) { | |
3694 | U8 *s = pat; | |
3695 | U8 *d = folded; | |
3696 | int i; | |
3697 | ||
3698 | for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) { | |
3699 | if (isASCII(*s)) { | |
3700 | *(d++) = (U8) toFOLD_LC(*s); | |
3701 | s++; | |
3702 | } | |
3703 | else { | |
3704 | STRLEN len; | |
3705 | _to_utf8_fold_flags(s, | |
3706 | d, | |
3707 | &len, | |
3708 | FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE); | |
3709 | d += len; | |
3710 | s += UTF8SKIP(s); | |
3711 | } | |
3712 | } | |
3713 | ||
3714 | pat = folded; | |
3715 | pat_end = d; | |
3716 | } | |
3717 | } | |
3718 | ||
251b239f KW |
3719 | if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end)) |
3720 | || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end))) | |
baa60164 KW |
3721 | { |
3722 | /* Multi-character folds require more context to sort out. Also | |
3723 | * PL_utf8_foldclosures used below doesn't handle them, so have to | |
3724 | * be handled outside this routine */ | |
3725 | use_chrtest_void = TRUE; | |
3726 | } | |
3727 | else { /* an EXACTFish node which doesn't begin with a multi-char fold */ | |
3728 | c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat; | |
e8ea8356 | 3729 | if (c1 > 255) { |
baa60164 KW |
3730 | /* Load the folds hash, if not already done */ |
3731 | SV** listp; | |
3732 | if (! PL_utf8_foldclosures) { | |
31667e6b | 3733 | _load_PL_utf8_foldclosures(); |
79a2a0e8 | 3734 | } |
79a2a0e8 | 3735 | |
baa60164 KW |
3736 | /* The fold closures data structure is a hash with the keys |
3737 | * being the UTF-8 of every character that is folded to, like | |
3738 | * 'k', and the values each an array of all code points that | |
3739 | * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ]. | |
3740 | * Multi-character folds are not included */ | |
3741 | if ((! (listp = hv_fetch(PL_utf8_foldclosures, | |
3742 | (char *) pat, | |
3743 | UTF8SKIP(pat), | |
3744 | FALSE)))) | |
3745 | { | |
3746 | /* Not found in the hash, therefore there are no folds | |
3747 | * containing it, so there is only a single character that | |
3748 | * could match */ | |
3749 | c2 = c1; | |
79a2a0e8 | 3750 | } |
baa60164 KW |
3751 | else { /* Does participate in folds */ |
3752 | AV* list = (AV*) *listp; | |
b9f2b683 | 3753 | if (av_tindex(list) != 1) { |
79a2a0e8 | 3754 | |
baa60164 KW |
3755 | /* If there aren't exactly two folds to this, it is |
3756 | * outside the scope of this function */ | |
3757 | use_chrtest_void = TRUE; | |
79a2a0e8 | 3758 | } |
baa60164 KW |
3759 | else { /* There are two. Get them */ |
3760 | SV** c_p = av_fetch(list, 0, FALSE); | |
3761 | if (c_p == NULL) { | |
3762 | Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); | |
3763 | } | |
3764 | c1 = SvUV(*c_p); | |
3765 | ||
3766 | c_p = av_fetch(list, 1, FALSE); | |
3767 | if (c_p == NULL) { | |
3768 | Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); | |
3769 | } | |
3770 | c2 = SvUV(*c_p); | |
3771 | ||
3772 | /* Folds that cross the 255/256 boundary are forbidden | |
3773 | * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and | |
3774 | * one is ASCIII. Since the pattern character is above | |
e8ea8356 | 3775 | * 255, and its only other match is below 256, the only |
baa60164 KW |
3776 | * legal match will be to itself. We have thrown away |
3777 | * the original, so have to compute which is the one | |
e8ea8356 | 3778 | * above 255. */ |
baa60164 KW |
3779 | if ((c1 < 256) != (c2 < 256)) { |
3780 | if ((OP(text_node) == EXACTFL | |
3781 | && ! IN_UTF8_CTYPE_LOCALE) | |
3782 | || ((OP(text_node) == EXACTFA | |
3783 | || OP(text_node) == EXACTFA_NO_TRIE) | |
3784 | && (isASCII(c1) || isASCII(c2)))) | |
3785 | { | |
3786 | if (c1 < 256) { | |
3787 | c1 = c2; | |
3788 | } | |
3789 | else { | |
3790 | c2 = c1; | |
3791 | } | |
79a2a0e8 KW |
3792 | } |
3793 | } | |
3794 | } | |
3795 | } | |
3796 | } | |
e8ea8356 | 3797 | else /* Here, c1 is <= 255 */ |
baa60164 KW |
3798 | if (utf8_target |
3799 | && HAS_NONLATIN1_FOLD_CLOSURE(c1) | |
3800 | && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE)) | |
3801 | && ((OP(text_node) != EXACTFA | |
3802 | && OP(text_node) != EXACTFA_NO_TRIE) | |
3803 | || ! isASCII(c1))) | |
3804 | { | |
3805 | /* Here, there could be something above Latin1 in the target | |
3806 | * which folds to this character in the pattern. All such | |
3807 | * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more | |
3808 | * than two characters involved in their folds, so are outside | |
3809 | * the scope of this function */ | |
3810 | if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) { | |
3811 | c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS; | |
3812 | } | |
3813 | else { | |
3814 | use_chrtest_void = TRUE; | |
3815 | } | |
79a2a0e8 | 3816 | } |
baa60164 KW |
3817 | else { /* Here nothing above Latin1 can fold to the pattern |
3818 | character */ | |
3819 | switch (OP(text_node)) { | |
c74f6de9 | 3820 | |
baa60164 KW |
3821 | case EXACTFL: /* /l rules */ |
3822 | c2 = PL_fold_locale[c1]; | |
3823 | break; | |
c74f6de9 | 3824 | |
baa60164 KW |
3825 | case EXACTF: /* This node only generated for non-utf8 |
3826 | patterns */ | |
3827 | assert(! is_utf8_pat); | |
3828 | if (! utf8_target) { /* /d rules */ | |
3829 | c2 = PL_fold[c1]; | |
3830 | break; | |
3831 | } | |
3832 | /* FALLTHROUGH */ | |
3833 | /* /u rules for all these. This happens to work for | |
3834 | * EXACTFA as nothing in Latin1 folds to ASCII */ | |
3835 | case EXACTFA_NO_TRIE: /* This node only generated for | |
3836 | non-utf8 patterns */ | |
3837 | assert(! is_utf8_pat); | |
924ba076 | 3838 | /* FALLTHROUGH */ |
baa60164 KW |
3839 | case EXACTFA: |
3840 | case EXACTFU_SS: | |
3841 | case EXACTFU: | |
3842 | c2 = PL_fold_latin1[c1]; | |
c74f6de9 | 3843 | break; |
c74f6de9 | 3844 | |
baa60164 KW |
3845 | default: |
3846 | Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node)); | |
e5964223 | 3847 | NOT_REACHED; /* NOTREACHED */ |
baa60164 | 3848 | } |
c74f6de9 KW |
3849 | } |
3850 | } | |
3851 | } | |
79a2a0e8 KW |
3852 | |
3853 | /* Here have figured things out. Set up the returns */ | |
3854 | if (use_chrtest_void) { | |
3855 | *c2p = *c1p = CHRTEST_VOID; | |
3856 | } | |
3857 | else if (utf8_target) { | |
3858 | if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */ | |
3859 | uvchr_to_utf8(c1_utf8, c1); | |
3860 | uvchr_to_utf8(c2_utf8, c2); | |
c74f6de9 | 3861 | } |
c74f6de9 | 3862 | |
79a2a0e8 KW |
3863 | /* Invariants are stored in both the utf8 and byte outputs; Use |
3864 | * negative numbers otherwise for the byte ones. Make sure that the | |
3865 | * byte ones are the same iff the utf8 ones are the same */ | |
3866 | *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1; | |
3867 | *c2p = (UTF8_IS_INVARIANT(*c2_utf8)) | |
3868 | ? *c2_utf8 | |
3869 | : (c1 == c2) | |
3870 | ? CHRTEST_NOT_A_CP_1 | |
3871 | : CHRTEST_NOT_A_CP_2; | |
3872 | } | |
3873 | else if (c1 > 255) { | |
3874 | if (c2 > 255) { /* both possibilities are above what a non-utf8 string | |
3875 | can represent */ | |
3876 | return FALSE; | |
3877 | } | |
c74f6de9 | 3878 | |
79a2a0e8 KW |
3879 | *c1p = *c2p = c2; /* c2 is the only representable value */ |
3880 | } | |
3881 | else { /* c1 is representable; see about c2 */ | |
3882 | *c1p = c1; | |
3883 | *c2p = (c2 < 256) ? c2 : c1; | |
c74f6de9 | 3884 | } |
2f554ef7 | 3885 | |
c74f6de9 KW |
3886 | return TRUE; |
3887 | } | |
2f554ef7 | 3888 | |
f73aaa43 | 3889 | /* returns -1 on failure, $+[0] on success */ |
99a90e59 | 3890 | STATIC SSize_t |
f73aaa43 | 3891 | S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog) |
d6a28714 | 3892 | { |
a35a87e7 | 3893 | #if PERL_VERSION < 9 && !defined(PERL_CORE) |
54df2634 NC |
3894 | dMY_CXT; |
3895 | #endif | |
27da23d5 | 3896 | dVAR; |
ba44c216 | 3897 | const bool utf8_target = reginfo->is_utf8_target; |
4ad0818d | 3898 | const U32 uniflags = UTF8_ALLOW_DEFAULT; |
288b8c02 | 3899 | REGEXP *rex_sv = reginfo->prog; |
8d919b0a | 3900 | regexp *rex = ReANY(rex_sv); |
f8fc2ecf | 3901 | RXi_GET_DECL(rex,rexi); |
5d9a96ca | 3902 | /* the current state. This is a cached copy of PL_regmatch_state */ |
eb578fdb | 3903 | regmatch_state *st; |
5d9a96ca | 3904 | /* cache heavy used fields of st in registers */ |
eb578fdb KW |
3905 | regnode *scan; |
3906 | regnode *next; | |
3907 | U32 n = 0; /* general value; init to avoid compiler warning */ | |
ea3daa5d | 3908 | SSize_t ln = 0; /* len or last; init to avoid compiler warning */ |
d60de1d1 | 3909 | char *locinput = startpos; |
4d5016e5 | 3910 | char *pushinput; /* where to continue after a PUSH */ |
eb578fdb | 3911 | I32 nextchr; /* is always set to UCHARAT(locinput) */ |
24d3c4a9 | 3912 | |
b69b0499 | 3913 | bool result = 0; /* return value of S_regmatch */ |
24d3c4a9 | 3914 | int depth = 0; /* depth of backtrack stack */ |
4b196cd4 YO |
3915 | U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */ |
3916 | const U32 max_nochange_depth = | |
3917 | (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ? | |
3918 | 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH; | |
77cb431f DM |
3919 | regmatch_state *yes_state = NULL; /* state to pop to on success of |
3920 | subpattern */ | |
e2e6a0f1 YO |
3921 | /* mark_state piggy backs on the yes_state logic so that when we unwind |
3922 | the stack on success we can update the mark_state as we go */ | |
3923 | regmatch_state *mark_state = NULL; /* last mark state we have seen */ | |
faec1544 | 3924 | regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */ |
b8591aee | 3925 | struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */ |
40a82448 | 3926 | U32 state_num; |
5d458dd8 YO |
3927 | bool no_final = 0; /* prevent failure from backtracking? */ |
3928 | bool do_cutgroup = 0; /* no_final only until next branch/trie entry */ | |
d60de1d1 | 3929 | char *startpoint = locinput; |
5d458dd8 YO |
3930 | SV *popmark = NULL; /* are we looking for a mark? */ |
3931 | SV *sv_commit = NULL; /* last mark name seen in failure */ | |
3932 | SV *sv_yes_mark = NULL; /* last mark name we have seen | |
486ec47a | 3933 | during a successful match */ |
5d458dd8 YO |
3934 | U32 lastopen = 0; /* last open we saw */ |
3935 | bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0; | |
4a2b275c | 3936 | SV* const oreplsv = GvSVn(PL_replgv); |
24d3c4a9 DM |
3937 | /* these three flags are set by various ops to signal information to |
3938 | * the very next op. They have a useful lifetime of exactly one loop | |
3939 | * iteration, and are not preserved or restored by state pushes/pops | |
3940 | */ | |
3941 | bool sw = 0; /* the condition value in (?(cond)a|b) */ | |
3942 | bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */ | |
3943 | int logical = 0; /* the following EVAL is: | |
3944 | 0: (?{...}) | |
3945 | 1: (?(?{...})X|Y) | |
3946 | 2: (??{...}) | |
3947 | or the following IFMATCH/UNLESSM is: | |
3948 | false: plain (?=foo) | |
3949 | true: used as a condition: (?(?=foo)) | |
3950 | */ | |
81ed78b2 DM |
3951 | PAD* last_pad = NULL; |
3952 | dMULTICALL; | |
3953 | I32 gimme = G_SCALAR; | |
3954 | CV *caller_cv = NULL; /* who called us */ | |
3955 | CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */ | |
74088413 | 3956 | CHECKPOINT runops_cp; /* savestack position before executing EVAL */ |
92da3157 | 3957 | U32 maxopenparen = 0; /* max '(' index seen so far */ |
3018b823 KW |
3958 | int to_complement; /* Invert the result? */ |
3959 | _char_class_number classnum; | |
984e6dd1 | 3960 | bool is_utf8_pat = reginfo->is_utf8_pat; |
81ed78b2 | 3961 | |
95b24440 | 3962 | #ifdef DEBUGGING |
e68ec53f | 3963 | GET_RE_DEBUG_FLAGS_DECL; |
d6a28714 JH |
3964 | #endif |
3965 | ||
7e68f152 FC |
3966 | /* protect against undef(*^R) */ |
3967 | SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv)); | |
3968 | ||
81ed78b2 DM |
3969 | /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */ |
3970 | multicall_oldcatch = 0; | |
3971 | multicall_cv = NULL; | |
3972 | cx = NULL; | |
4f8dbb2d JL |
3973 | PERL_UNUSED_VAR(multicall_cop); |
3974 | PERL_UNUSED_VAR(newsp); | |
81ed78b2 DM |
3975 | |
3976 | ||
7918f24d NC |
3977 | PERL_ARGS_ASSERT_REGMATCH; |
3978 | ||
3b57cd43 | 3979 | DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({ |
24b23f37 | 3980 | PerlIO_printf(Perl_debug_log,"regmatch start\n"); |
3b57cd43 | 3981 | })); |
5d9a96ca | 3982 | |
331b2dcc | 3983 | st = PL_regmatch_state; |
5d9a96ca | 3984 | |
d6a28714 | 3985 | /* Note that nextchr is a byte even in UTF */ |
7016d6eb | 3986 | SET_nextchr; |
d6a28714 JH |
3987 | scan = prog; |
3988 | while (scan != NULL) { | |
8ba1375e | 3989 | |
a3621e74 | 3990 | DEBUG_EXECUTE_r( { |
6136c704 | 3991 | SV * const prop = sv_newmortal(); |
1de06328 | 3992 | regnode *rnext=regnext(scan); |
f2ed9b32 | 3993 | DUMP_EXEC_POS( locinput, scan, utf8_target ); |
8b9781c9 | 3994 | regprop(rex, prop, scan, reginfo, NULL); |
07be1b83 YO |
3995 | |
3996 | PerlIO_printf(Perl_debug_log, | |
3997 | "%3"IVdf":%*s%s(%"IVdf")\n", | |
f8fc2ecf | 3998 | (IV)(scan - rexi->program), depth*2, "", |
07be1b83 | 3999 | SvPVX_const(prop), |
1de06328 | 4000 | (PL_regkind[OP(scan)] == END || !rnext) ? |
f8fc2ecf | 4001 | 0 : (IV)(rnext - rexi->program)); |
2a782b5b | 4002 | }); |
d6a28714 JH |
4003 | |
4004 | next = scan + NEXT_OFF(scan); | |
4005 | if (next == scan) | |
4006 | next = NULL; | |
40a82448 | 4007 | state_num = OP(scan); |
d6a28714 | 4008 | |
40a82448 | 4009 | reenter_switch: |
3018b823 | 4010 | to_complement = 0; |
34a81e2b | 4011 | |
7016d6eb | 4012 | SET_nextchr; |
e6ca698c | 4013 | assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS)); |
bf798dc4 | 4014 | |
40a82448 | 4015 | switch (state_num) { |
d3d47aac | 4016 | case SBOL: /* /^../ and /\A../ */ |
9d9163fb | 4017 | if (locinput == reginfo->strbeg) |
b8c5462f | 4018 | break; |
d6a28714 | 4019 | sayNO; |
3c0563b9 DM |
4020 | |
4021 | case MBOL: /* /^../m */ | |
9d9163fb | 4022 | if (locinput == reginfo->strbeg || |
7016d6eb | 4023 | (!NEXTCHR_IS_EOS && locinput[-1] == '\n')) |
d6a28714 | 4024 | { |
b8c5462f JH |
4025 | break; |
4026 | } | |
d6a28714 | 4027 | sayNO; |
3c0563b9 | 4028 | |
3c0563b9 | 4029 | case GPOS: /* \G */ |
3b0527fe | 4030 | if (locinput == reginfo->ganch) |
d6a28714 JH |
4031 | break; |
4032 | sayNO; | |
ee9b8eae | 4033 | |
3c0563b9 | 4034 | case KEEPS: /* \K */ |
ee9b8eae | 4035 | /* update the startpoint */ |
b93070ed | 4036 | st->u.keeper.val = rex->offs[0].start; |
9d9163fb | 4037 | rex->offs[0].start = locinput - reginfo->strbeg; |
4d5016e5 | 4038 | PUSH_STATE_GOTO(KEEPS_next, next, locinput); |
a74ff37d | 4039 | /* NOTREACHED */ |
e5964223 | 4040 | NOT_REACHED; |
a74ff37d | 4041 | |
ee9b8eae YO |
4042 | case KEEPS_next_fail: |
4043 | /* rollback the start point change */ | |
b93070ed | 4044 | rex->offs[0].start = st->u.keeper.val; |
ee9b8eae | 4045 | sayNO_SILENT; |
a74ff37d | 4046 | /* NOTREACHED */ |
e5964223 | 4047 | NOT_REACHED; |
3c0563b9 | 4048 | |
3c0563b9 | 4049 | case MEOL: /* /..$/m */ |
7016d6eb | 4050 | if (!NEXTCHR_IS_EOS && nextchr != '\n') |
b8c5462f | 4051 | sayNO; |
b8c5462f | 4052 | break; |
3c0563b9 | 4053 | |
d3d47aac | 4054 | case SEOL: /* /..$/ */ |
7016d6eb | 4055 | if (!NEXTCHR_IS_EOS && nextchr != '\n') |
b8c5462f | 4056 | sayNO; |
220db18a | 4057 | if (reginfo->strend - locinput > 1) |
b8c5462f | 4058 | sayNO; |
b8c5462f | 4059 | break; |
3c0563b9 DM |
4060 | |
4061 | case EOS: /* \z */ | |
7016d6eb | 4062 | if (!NEXTCHR_IS_EOS) |
b8c5462f | 4063 | sayNO; |
d6a28714 | 4064 | break; |
3c0563b9 DM |
4065 | |
4066 | case SANY: /* /./s */ | |
7016d6eb | 4067 | if (NEXTCHR_IS_EOS) |
4633a7c4 | 4068 | sayNO; |
28b98f76 | 4069 | goto increment_locinput; |
3c0563b9 DM |
4070 | |
4071 | case CANY: /* \C */ | |
7016d6eb | 4072 | if (NEXTCHR_IS_EOS) |
f33976b4 | 4073 | sayNO; |
3640db6b | 4074 | locinput++; |
a0d0e21e | 4075 | break; |
3c0563b9 DM |
4076 | |
4077 | case REG_ANY: /* /./ */ | |
7016d6eb | 4078 | if ((NEXTCHR_IS_EOS) || nextchr == '\n') |
1aa99e6b | 4079 | sayNO; |
28b98f76 DM |
4080 | goto increment_locinput; |
4081 | ||
166ba7cd DM |
4082 | |
4083 | #undef ST | |
4084 | #define ST st->u.trie | |
3c0563b9 | 4085 | case TRIEC: /* (ab|cd) with known charclass */ |
786e8c11 YO |
4086 | /* In this case the charclass data is available inline so |
4087 | we can fail fast without a lot of extra overhead. | |
4088 | */ | |
7016d6eb | 4089 | if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) { |
fab2782b YO |
4090 | DEBUG_EXECUTE_r( |
4091 | PerlIO_printf(Perl_debug_log, | |
4092 | "%*s %sfailed to match trie start class...%s\n", | |
4093 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) | |
4094 | ); | |
4095 | sayNO_SILENT; | |
a74ff37d | 4096 | /* NOTREACHED */ |
e5964223 | 4097 | NOT_REACHED; |
786e8c11 | 4098 | } |
924ba076 | 4099 | /* FALLTHROUGH */ |
3c0563b9 | 4100 | case TRIE: /* (ab|cd) */ |
2e64971a DM |
4101 | /* the basic plan of execution of the trie is: |
4102 | * At the beginning, run though all the states, and | |
4103 | * find the longest-matching word. Also remember the position | |
4104 | * of the shortest matching word. For example, this pattern: | |
4105 | * 1 2 3 4 5 | |
4106 | * ab|a|x|abcd|abc | |
4107 | * when matched against the string "abcde", will generate | |
4108 | * accept states for all words except 3, with the longest | |
895cc420 | 4109 | * matching word being 4, and the shortest being 2 (with |
2e64971a DM |
4110 | * the position being after char 1 of the string). |
4111 | * | |
4112 | * Then for each matching word, in word order (i.e. 1,2,4,5), | |
4113 | * we run the remainder of the pattern; on each try setting | |
4114 | * the current position to the character following the word, | |
4115 | * returning to try the next word on failure. | |
4116 | * | |
4117 | * We avoid having to build a list of words at runtime by | |
4118 | * using a compile-time structure, wordinfo[].prev, which | |
4119 | * gives, for each word, the previous accepting word (if any). | |
4120 | * In the case above it would contain the mappings 1->2, 2->0, | |
4121 | * 3->0, 4->5, 5->1. We can use this table to generate, from | |
4122 | * the longest word (4 above), a list of all words, by | |
4123 | * following the list of prev pointers; this gives us the | |
4124 | * unordered list 4,5,1,2. Then given the current word we have | |
4125 | * just tried, we can go through the list and find the | |
4126 | * next-biggest word to try (so if we just failed on word 2, | |
4127 | * the next in the list is 4). | |
4128 | * | |
4129 | * Since at runtime we don't record the matching position in | |
4130 | * the string for each word, we have to work that out for | |
4131 | * each word we're about to process. The wordinfo table holds | |
4132 | * the character length of each word; given that we recorded | |
4133 | * at the start: the position of the shortest word and its | |
4134 | * length in chars, we just need to move the pointer the | |
4135 | * difference between the two char lengths. Depending on | |
4136 | * Unicode status and folding, that's cheap or expensive. | |
4137 | * | |
4138 | * This algorithm is optimised for the case where are only a | |
4139 | * small number of accept states, i.e. 0,1, or maybe 2. | |
4140 | * With lots of accepts states, and having to try all of them, | |
4141 | * it becomes quadratic on number of accept states to find all | |
4142 | * the next words. | |
4143 | */ | |
4144 | ||
3dab1dad | 4145 | { |
07be1b83 | 4146 | /* what type of TRIE am I? (utf8 makes this contextual) */ |
a0a388a1 | 4147 | DECL_TRIE_TYPE(scan); |
3dab1dad YO |
4148 | |
4149 | /* what trie are we using right now */ | |
be8e71aa | 4150 | reg_trie_data * const trie |
f8fc2ecf | 4151 | = (reg_trie_data*)rexi->data->data[ ARG( scan ) ]; |
85fbaab2 | 4152 | HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]); |
3dab1dad | 4153 | U32 state = trie->startstate; |
166ba7cd | 4154 | |
7016d6eb DM |
4155 | if ( trie->bitmap |
4156 | && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr))) | |
4157 | { | |
3dab1dad YO |
4158 | if (trie->states[ state ].wordnum) { |
4159 | DEBUG_EXECUTE_r( | |
4160 | PerlIO_printf(Perl_debug_log, | |
4161 | "%*s %smatched empty string...%s\n", | |
5bc10b2c | 4162 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) |
3dab1dad | 4163 | ); |
20dbff7c YO |
4164 | if (!trie->jump) |
4165 | break; | |
3dab1dad YO |
4166 | } else { |
4167 | DEBUG_EXECUTE_r( | |
4168 | PerlIO_printf(Perl_debug_log, | |
786e8c11 | 4169 | "%*s %sfailed to match trie start class...%s\n", |
5bc10b2c | 4170 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) |
3dab1dad YO |
4171 | ); |
4172 | sayNO_SILENT; | |
4173 | } | |
4174 | } | |
166ba7cd | 4175 | |
786e8c11 YO |
4176 | { |
4177 | U8 *uc = ( U8* )locinput; | |
4178 | ||
4179 | STRLEN len = 0; | |
4180 | STRLEN foldlen = 0; | |
4181 | U8 *uscan = (U8*)NULL; | |
786e8c11 | 4182 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
2e64971a DM |
4183 | U32 charcount = 0; /* how many input chars we have matched */ |
4184 | U32 accepted = 0; /* have we seen any accepting states? */ | |
786e8c11 | 4185 | |
786e8c11 | 4186 | ST.jump = trie->jump; |
786e8c11 | 4187 | ST.me = scan; |
2e64971a DM |
4188 | ST.firstpos = NULL; |
4189 | ST.longfold = FALSE; /* char longer if folded => it's harder */ | |
4190 | ST.nextword = 0; | |
4191 | ||
4192 | /* fully traverse the TRIE; note the position of the | |
4193 | shortest accept state and the wordnum of the longest | |
4194 | accept state */ | |
07be1b83 | 4195 | |
220db18a | 4196 | while ( state && uc <= (U8*)(reginfo->strend) ) { |
786e8c11 | 4197 | U32 base = trie->states[ state ].trans.base; |
f9f4320a | 4198 | UV uvc = 0; |
acb909b4 | 4199 | U16 charid = 0; |
2e64971a DM |
4200 | U16 wordnum; |
4201 | wordnum = trie->states[ state ].wordnum; | |
4202 | ||
4203 | if (wordnum) { /* it's an accept state */ | |
4204 | if (!accepted) { | |
4205 | accepted = 1; | |
4206 | /* record first match position */ | |
4207 | if (ST.longfold) { | |
4208 | ST.firstpos = (U8*)locinput; | |
4209 | ST.firstchars = 0; | |
5b47454d | 4210 | } |
2e64971a DM |
4211 | else { |
4212 | ST.firstpos = uc; | |
4213 | ST.firstchars = charcount; | |
4214 | } | |
4215 | } | |
4216 | if (!ST.nextword || wordnum < ST.nextword) | |
4217 | ST.nextword = wordnum; | |
4218 | ST.topword = wordnum; | |
786e8c11 | 4219 | } |
a3621e74 | 4220 | |
07be1b83 | 4221 | DEBUG_TRIE_EXECUTE_r({ |
f2ed9b32 | 4222 | DUMP_EXEC_POS( (char *)uc, scan, utf8_target ); |
a3621e74 | 4223 | PerlIO_printf( Perl_debug_log, |
2e64971a | 4224 | "%*s %sState: %4"UVxf" Accepted: %c ", |
5bc10b2c | 4225 | 2+depth * 2, "", PL_colors[4], |
2e64971a | 4226 | (UV)state, (accepted ? 'Y' : 'N')); |
07be1b83 | 4227 | }); |
a3621e74 | 4228 | |
2e64971a | 4229 | /* read a char and goto next state */ |
220db18a | 4230 | if ( base && (foldlen || uc < (U8*)(reginfo->strend))) { |
6dd2be57 | 4231 | I32 offset; |
55eed653 NC |
4232 | REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, |
4233 | uscan, len, uvc, charid, foldlen, | |
4234 | foldbuf, uniflags); | |
2e64971a DM |
4235 | charcount++; |
4236 | if (foldlen>0) | |
4237 | ST.longfold = TRUE; | |
5b47454d | 4238 | if (charid && |
6dd2be57 DM |
4239 | ( ((offset = |
4240 | base + charid - 1 - trie->uniquecharcount)) >= 0) | |
4241 | ||
4242 | && ((U32)offset < trie->lasttrans) | |
4243 | && trie->trans[offset].check == state) | |
5b47454d | 4244 | { |
6dd2be57 | 4245 | state = trie->trans[offset].next; |
5b47454d DM |
4246 | } |
4247 | else { | |
4248 | state = 0; | |
4249 | } | |
4250 | uc += len; | |
4251 | ||
4252 | } | |
4253 | else { | |
a3621e74 YO |
4254 | state = 0; |
4255 | } | |
4256 | DEBUG_TRIE_EXECUTE_r( | |
e4584336 | 4257 | PerlIO_printf( Perl_debug_log, |
786e8c11 | 4258 | "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n", |
e4584336 | 4259 | charid, uvc, (UV)state, PL_colors[5] ); |
a3621e74 YO |
4260 | ); |
4261 | } | |
2e64971a | 4262 | if (!accepted) |
a3621e74 | 4263 | sayNO; |
a3621e74 | 4264 | |
2e64971a DM |
4265 | /* calculate total number of accept states */ |
4266 | { | |
4267 | U16 w = ST.topword; | |
4268 | accepted = 0; | |
4269 | while (w) { | |
4270 | w = trie->wordinfo[w].prev; | |
4271 | accepted++; | |
4272 | } | |
4273 | ST.accepted = accepted; | |
4274 | } | |
4275 | ||
166ba7cd DM |
4276 | DEBUG_EXECUTE_r( |
4277 | PerlIO_printf( Perl_debug_log, | |
4278 | "%*s %sgot %"IVdf" possible matches%s\n", | |
5bc10b2c | 4279 | REPORT_CODE_OFF + depth * 2, "", |
166ba7cd DM |
4280 | PL_colors[4], (IV)ST.accepted, PL_colors[5] ); |
4281 | ); | |
2e64971a | 4282 | goto trie_first_try; /* jump into the fail handler */ |
786e8c11 | 4283 | }} |
a74ff37d | 4284 | /* NOTREACHED */ |
e5964223 | 4285 | NOT_REACHED; |
2e64971a DM |
4286 | |
4287 | case TRIE_next_fail: /* we failed - try next alternative */ | |
a059a757 DM |
4288 | { |
4289 | U8 *uc; | |
fae667d5 YO |
4290 | if ( ST.jump) { |
4291 | REGCP_UNWIND(ST.cp); | |
a8d1f4b4 | 4292 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
fae667d5 | 4293 | } |
2e64971a DM |
4294 | if (!--ST.accepted) { |
4295 | DEBUG_EXECUTE_r({ | |
4296 | PerlIO_printf( Perl_debug_log, | |
4297 | "%*s %sTRIE failed...%s\n", | |
4298 | REPORT_CODE_OFF+depth*2, "", | |
4299 | PL_colors[4], | |
4300 | PL_colors[5] ); | |
4301 | }); | |
4302 | sayNO_SILENT; | |
4303 | } | |
4304 | { | |
4305 | /* Find next-highest word to process. Note that this code | |
4306 | * is O(N^2) per trie run (O(N) per branch), so keep tight */ | |
eb578fdb KW |
4307 | U16 min = 0; |
4308 | U16 word; | |
4309 | U16 const nextword = ST.nextword; | |
4310 | reg_trie_wordinfo * const wordinfo | |
2e64971a DM |
4311 | = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo; |
4312 | for (word=ST.topword; word; word=wordinfo[word].prev) { | |
4313 | if (word > nextword && (!min || word < min)) | |
4314 | min = word; | |
4315 | } | |
4316 | ST.nextword = min; | |
4317 | } | |
4318 | ||
fae667d5 | 4319 | trie_first_try: |
5d458dd8 YO |
4320 | if (do_cutgroup) { |
4321 | do_cutgroup = 0; | |
4322 | no_final = 0; | |
4323 | } | |
fae667d5 YO |
4324 | |
4325 | if ( ST.jump) { | |
b93070ed | 4326 | ST.lastparen = rex->lastparen; |
f6033a9d | 4327 | ST.lastcloseparen = rex->lastcloseparen; |
fae667d5 | 4328 | REGCP_SET(ST.cp); |
2e64971a | 4329 | } |
a3621e74 | 4330 | |
2e64971a | 4331 | /* find start char of end of current word */ |
166ba7cd | 4332 | { |
2e64971a | 4333 | U32 chars; /* how many chars to skip */ |
2e64971a DM |
4334 | reg_trie_data * const trie |
4335 | = (reg_trie_data*)rexi->data->data[ARG(ST.me)]; | |
4336 | ||
4337 | assert((trie->wordinfo[ST.nextword].len - trie->prefixlen) | |
4338 | >= ST.firstchars); | |
4339 | chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen) | |
4340 | - ST.firstchars; | |
a059a757 | 4341 | uc = ST.firstpos; |
2e64971a DM |
4342 | |
4343 | if (ST.longfold) { | |
4344 | /* the hard option - fold each char in turn and find | |
4345 | * its folded length (which may be different */ | |
4346 | U8 foldbuf[UTF8_MAXBYTES_CASE + 1]; | |
4347 | STRLEN foldlen; | |
4348 | STRLEN len; | |
d9a396a3 | 4349 | UV uvc; |
2e64971a DM |
4350 | U8 *uscan; |
4351 | ||
4352 | while (chars) { | |
f2ed9b32 | 4353 | if (utf8_target) { |
c80e42f3 | 4354 | uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len, |
2e64971a DM |
4355 | uniflags); |
4356 | uc += len; | |
4357 | } | |
4358 | else { | |
4359 | uvc = *uc; | |
4360 | uc++; | |
4361 | } | |
4362 | uvc = to_uni_fold(uvc, foldbuf, &foldlen); | |
4363 | uscan = foldbuf; | |
4364 | while (foldlen) { | |
4365 | if (!--chars) | |
4366 | break; | |
c80e42f3 | 4367 | uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len, |
2e64971a DM |
4368 | uniflags); |
4369 | uscan += len; | |
4370 | foldlen -= len; | |
4371 | } | |
4372 | } | |
a3621e74 | 4373 | } |
2e64971a | 4374 | else { |
f2ed9b32 | 4375 | if (utf8_target) |
2e64971a DM |
4376 | while (chars--) |
4377 | uc += UTF8SKIP(uc); | |
4378 | else | |
4379 | uc += chars; | |
4380 | } | |
2e64971a | 4381 | } |
166ba7cd | 4382 | |
6603fe3e DM |
4383 | scan = ST.me + ((ST.jump && ST.jump[ST.nextword]) |
4384 | ? ST.jump[ST.nextword] | |
4385 | : NEXT_OFF(ST.me)); | |
166ba7cd | 4386 | |
2e64971a DM |
4387 | DEBUG_EXECUTE_r({ |
4388 | PerlIO_printf( Perl_debug_log, | |
4389 | "%*s %sTRIE matched word #%d, continuing%s\n", | |
4390 | REPORT_CODE_OFF+depth*2, "", | |
4391 | PL_colors[4], | |
4392 | ST.nextword, | |
4393 | PL_colors[5] | |
4394 | ); | |
4395 | }); | |
4396 | ||
4397 | if (ST.accepted > 1 || has_cutgroup) { | |
a059a757 | 4398 | PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc); |
a74ff37d | 4399 | /* NOTREACHED */ |
e5964223 | 4400 | NOT_REACHED; |
166ba7cd | 4401 | } |
2e64971a DM |
4402 | /* only one choice left - just continue */ |
4403 | DEBUG_EXECUTE_r({ | |
4404 | AV *const trie_words | |
4405 | = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]); | |
4406 | SV ** const tmp = av_fetch( trie_words, | |
4407 | ST.nextword-1, 0 ); | |
4408 | SV *sv= tmp ? sv_newmortal() : NULL; | |
4409 | ||
4410 | PerlIO_printf( Perl_debug_log, | |
4411 | "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n", | |
4412 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], | |
4413 | ST.nextword, | |
4414 | tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0, | |
4415 | PL_colors[0], PL_colors[1], | |
c89df6cf | 4416 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII |
2e64971a DM |
4417 | ) |
4418 | : "not compiled under -Dr", | |
4419 | PL_colors[5] ); | |
4420 | }); | |
4421 | ||
a059a757 | 4422 | locinput = (char*)uc; |
2e64971a | 4423 | continue; /* execute rest of RE */ |
a74ff37d | 4424 | /* NOTREACHED */ |
a059a757 | 4425 | } |
166ba7cd DM |
4426 | #undef ST |
4427 | ||
3c0563b9 | 4428 | case EXACT: { /* /abc/ */ |
95b24440 | 4429 | char *s = STRING(scan); |
24d3c4a9 | 4430 | ln = STR_LEN(scan); |
984e6dd1 | 4431 | if (utf8_target != is_utf8_pat) { |
bc517b45 | 4432 | /* The target and the pattern have differing utf8ness. */ |
1aa99e6b | 4433 | char *l = locinput; |
24d3c4a9 | 4434 | const char * const e = s + ln; |
a72c7584 | 4435 | |
f2ed9b32 | 4436 | if (utf8_target) { |
e6a3850e KW |
4437 | /* The target is utf8, the pattern is not utf8. |
4438 | * Above-Latin1 code points can't match the pattern; | |
4439 | * invariants match exactly, and the other Latin1 ones need | |
4440 | * to be downgraded to a single byte in order to do the | |
4441 | * comparison. (If we could be confident that the target | |
4442 | * is not malformed, this could be refactored to have fewer | |
4443 | * tests by just assuming that if the first bytes match, it | |
4444 | * is an invariant, but there are tests in the test suite | |
4445 | * dealing with (??{...}) which violate this) */ | |
1aa99e6b | 4446 | while (s < e) { |
220db18a DM |
4447 | if (l >= reginfo->strend |
4448 | || UTF8_IS_ABOVE_LATIN1(* (U8*) l)) | |
4449 | { | |
e6a3850e KW |
4450 | sayNO; |
4451 | } | |
4452 | if (UTF8_IS_INVARIANT(*(U8*)l)) { | |
4453 | if (*l != *s) { | |
4454 | sayNO; | |
4455 | } | |
4456 | l++; | |
4457 | } | |
4458 | else { | |
94bb8c36 KW |
4459 | if (TWO_BYTE_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s) |
4460 | { | |
e6a3850e KW |
4461 | sayNO; |
4462 | } | |
4463 | l += 2; | |
4464 | } | |
4465 | s++; | |
1aa99e6b | 4466 | } |
5ff6fc6d JH |
4467 | } |
4468 | else { | |
4469 | /* The target is not utf8, the pattern is utf8. */ | |
1aa99e6b | 4470 | while (s < e) { |
220db18a DM |
4471 | if (l >= reginfo->strend |
4472 | || UTF8_IS_ABOVE_LATIN1(* (U8*) s)) | |
e6a3850e KW |
4473 | { |
4474 | sayNO; | |
4475 | } | |
4476 | if (UTF8_IS_INVARIANT(*(U8*)s)) { | |
4477 | if (*s != *l) { | |
4478 | sayNO; | |
4479 | } | |
4480 | s++; | |
4481 | } | |
4482 | else { | |
94bb8c36 KW |
4483 | if (TWO_BYTE_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l) |
4484 | { | |
e6a3850e KW |
4485 | sayNO; |
4486 | } | |
4487 | s += 2; | |
4488 | } | |
4489 | l++; | |
1aa99e6b | 4490 | } |
5ff6fc6d | 4491 | } |
1aa99e6b | 4492 | locinput = l; |
1aa99e6b | 4493 | } |
5ac65bff KW |
4494 | else { |
4495 | /* The target and the pattern have the same utf8ness. */ | |
4496 | /* Inline the first character, for speed. */ | |
220db18a | 4497 | if (reginfo->strend - locinput < ln |
5ac65bff KW |
4498 | || UCHARAT(s) != nextchr |
4499 | || (ln > 1 && memNE(s, locinput, ln))) | |
4500 | { | |
4501 | sayNO; | |
4502 | } | |
4503 | locinput += ln; | |
4504 | } | |
d6a28714 | 4505 | break; |
95b24440 | 4506 | } |
7016d6eb | 4507 | |
3c0563b9 | 4508 | case EXACTFL: { /* /abc/il */ |
a932d541 | 4509 | re_fold_t folder; |
9a5a5549 KW |
4510 | const U8 * fold_array; |
4511 | const char * s; | |
d513472c | 4512 | U32 fold_utf8_flags; |
9a5a5549 | 4513 | |
f67f9e53 KW |
4514 | folder = foldEQ_locale; |
4515 | fold_array = PL_fold_locale; | |
cea315b6 | 4516 | fold_utf8_flags = FOLDEQ_LOCALE; |
9a5a5549 KW |
4517 | goto do_exactf; |
4518 | ||
3c0563b9 | 4519 | case EXACTFU_SS: /* /\x{df}/iu */ |
3c0563b9 | 4520 | case EXACTFU: /* /abc/iu */ |
9a5a5549 KW |
4521 | folder = foldEQ_latin1; |
4522 | fold_array = PL_fold_latin1; | |
984e6dd1 | 4523 | fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0; |
9a5a5549 KW |
4524 | goto do_exactf; |
4525 | ||
098b07d5 KW |
4526 | case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 |
4527 | patterns */ | |
4528 | assert(! is_utf8_pat); | |
924ba076 | 4529 | /* FALLTHROUGH */ |
3c0563b9 | 4530 | case EXACTFA: /* /abc/iaa */ |
2f7f8cb1 KW |
4531 | folder = foldEQ_latin1; |
4532 | fold_array = PL_fold_latin1; | |
57014d77 | 4533 | fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
2f7f8cb1 KW |
4534 | goto do_exactf; |
4535 | ||
2fdb7295 KW |
4536 | case EXACTF: /* /abc/i This node only generated for |
4537 | non-utf8 patterns */ | |
4538 | assert(! is_utf8_pat); | |
9a5a5549 KW |
4539 | folder = foldEQ; |
4540 | fold_array = PL_fold; | |
62bf7766 | 4541 | fold_utf8_flags = 0; |
9a5a5549 KW |
4542 | |
4543 | do_exactf: | |
4544 | s = STRING(scan); | |
24d3c4a9 | 4545 | ln = STR_LEN(scan); |
d6a28714 | 4546 | |
31f05a37 KW |
4547 | if (utf8_target |
4548 | || is_utf8_pat | |
4549 | || state_num == EXACTFU_SS | |
4550 | || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE)) | |
4551 | { | |
3c760661 KW |
4552 | /* Either target or the pattern are utf8, or has the issue where |
4553 | * the fold lengths may differ. */ | |
be8e71aa | 4554 | const char * const l = locinput; |
220db18a | 4555 | char *e = reginfo->strend; |
bc517b45 | 4556 | |
984e6dd1 | 4557 | if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat, |
fa5b1667 | 4558 | l, &e, 0, utf8_target, fold_utf8_flags)) |
c3e1d013 KW |
4559 | { |
4560 | sayNO; | |
5486206c | 4561 | } |
d07ddd77 | 4562 | locinput = e; |
d07ddd77 | 4563 | break; |
a0ed51b3 | 4564 | } |
d6a28714 | 4565 | |
0a138b74 | 4566 | /* Neither the target nor the pattern are utf8 */ |
1443c94c DM |
4567 | if (UCHARAT(s) != nextchr |
4568 | && !NEXTCHR_IS_EOS | |
4569 | && UCHARAT(s) != fold_array[nextchr]) | |
9a5a5549 | 4570 | { |
a0ed51b3 | 4571 | sayNO; |
9a5a5549 | 4572 | } |
220db18a | 4573 | if (reginfo->strend - locinput < ln) |
b8c5462f | 4574 | sayNO; |
9a5a5549 | 4575 | if (ln > 1 && ! folder(s, locinput, ln)) |
4633a7c4 | 4576 | sayNO; |
24d3c4a9 | 4577 | locinput += ln; |
a0d0e21e | 4578 | break; |
9a5a5549 | 4579 | } |
63ac0dad KW |
4580 | |
4581 | /* XXX Could improve efficiency by separating these all out using a | |
4582 | * macro or in-line function. At that point regcomp.c would no longer | |
4583 | * have to set the FLAGS fields of these */ | |
3c0563b9 DM |
4584 | case BOUNDL: /* /\b/l */ |
4585 | case NBOUNDL: /* /\B/l */ | |
3c0563b9 DM |
4586 | case BOUND: /* /\b/ */ |
4587 | case BOUNDU: /* /\b/u */ | |
4588 | case BOUNDA: /* /\b/a */ | |
4589 | case NBOUND: /* /\B/ */ | |
4590 | case NBOUNDU: /* /\B/u */ | |
4591 | case NBOUNDA: /* /\B/a */ | |
b2680017 | 4592 | /* was last char in word? */ |
f2e96b5d KW |
4593 | if (utf8_target |
4594 | && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET | |
4595 | && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET) | |
4596 | { | |
9d9163fb | 4597 | if (locinput == reginfo->strbeg) |
b2680017 YO |
4598 | ln = '\n'; |
4599 | else { | |
9d9163fb DM |
4600 | const U8 * const r = |
4601 | reghop3((U8*)locinput, -1, (U8*)(reginfo->strbeg)); | |
b2680017 | 4602 | |
3db24e1e KW |
4603 | ln = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, |
4604 | 0, uniflags); | |
b2680017 | 4605 | } |
63ac0dad | 4606 | if (FLAGS(scan) != REGEX_LOCALE_CHARSET) { |
0eb30aeb | 4607 | ln = isWORDCHAR_uni(ln); |
7016d6eb DM |
4608 | if (NEXTCHR_IS_EOS) |
4609 | n = 0; | |
4610 | else { | |
4611 | LOAD_UTF8_CHARCLASS_ALNUM(); | |
03940dc2 | 4612 | n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput, |
7016d6eb DM |
4613 | utf8_target); |
4614 | } | |
b2680017 YO |
4615 | } |
4616 | else { | |
f4cd282c | 4617 | ln = isWORDCHAR_LC_uvchr(ln); |
0eb30aeb | 4618 | n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput); |
b2680017 YO |
4619 | } |
4620 | } | |
4621 | else { | |
cfaf538b KW |
4622 | |
4623 | /* Here the string isn't utf8, or is utf8 and only ascii | |
4624 | * characters are to match \w. In the latter case looking at | |
4625 | * the byte just prior to the current one may be just the final | |
4626 | * byte of a multi-byte character. This is ok. There are two | |
4627 | * cases: | |
4628 | * 1) it is a single byte character, and then the test is doing | |
4629 | * just what it's supposed to. | |
4630 | * 2) it is a multi-byte character, in which case the final | |
4631 | * byte is never mistakable for ASCII, and so the test | |
4632 | * will say it is not a word character, which is the | |
4633 | * correct answer. */ | |
9d9163fb | 4634 | ln = (locinput != reginfo->strbeg) ? |
b2680017 | 4635 | UCHARAT(locinput - 1) : '\n'; |
63ac0dad KW |
4636 | switch (FLAGS(scan)) { |
4637 | case REGEX_UNICODE_CHARSET: | |
4638 | ln = isWORDCHAR_L1(ln); | |
7016d6eb | 4639 | n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr); |
63ac0dad KW |
4640 | break; |
4641 | case REGEX_LOCALE_CHARSET: | |
0eb30aeb KW |
4642 | ln = isWORDCHAR_LC(ln); |
4643 | n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr); | |
63ac0dad KW |
4644 | break; |
4645 | case REGEX_DEPENDS_CHARSET: | |
0eb30aeb KW |
4646 | ln = isWORDCHAR(ln); |
4647 | n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr); | |
63ac0dad | 4648 | break; |
cfaf538b | 4649 | case REGEX_ASCII_RESTRICTED_CHARSET: |
c973bd4f | 4650 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
cfaf538b | 4651 | ln = isWORDCHAR_A(ln); |
7016d6eb | 4652 | n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr); |
cfaf538b | 4653 | break; |
63ac0dad KW |
4654 | default: |
4655 | Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan)); | |
b2680017 YO |
4656 | } |
4657 | } | |
63ac0dad KW |
4658 | /* Note requires that all BOUNDs be lower than all NBOUNDs in |
4659 | * regcomp.sym */ | |
4660 | if (((!ln) == (!n)) == (OP(scan) < NBOUND)) | |
b2680017 YO |
4661 | sayNO; |
4662 | break; | |
3c0563b9 | 4663 | |
3c0563b9 | 4664 | case ANYOF: /* /[abc]/ */ |
7016d6eb DM |
4665 | if (NEXTCHR_IS_EOS) |
4666 | sayNO; | |
e0193e47 | 4667 | if (utf8_target) { |
3db24e1e KW |
4668 | if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend, |
4669 | utf8_target)) | |
09b08e9b | 4670 | sayNO; |
635cd5d4 | 4671 | locinput += UTF8SKIP(locinput); |
ffc61ed2 JH |
4672 | } |
4673 | else { | |
20ed0b26 | 4674 | if (!REGINCLASS(rex, scan, (U8*)locinput)) |
09b08e9b | 4675 | sayNO; |
3640db6b | 4676 | locinput++; |
e0f9d4a8 | 4677 | } |
b8c5462f | 4678 | break; |
3c0563b9 | 4679 | |
3018b823 KW |
4680 | /* The argument (FLAGS) to all the POSIX node types is the class number |
4681 | * */ | |
ee9a90b8 | 4682 | |
3018b823 KW |
4683 | case NPOSIXL: /* \W or [:^punct:] etc. under /l */ |
4684 | to_complement = 1; | |
4685 | /* FALLTHROUGH */ | |
4686 | ||
4687 | case POSIXL: /* \w or [:punct:] etc. under /l */ | |
4688 | if (NEXTCHR_IS_EOS) | |
bedac28b | 4689 | sayNO; |
bedac28b | 4690 | |
3018b823 KW |
4691 | /* Use isFOO_lc() for characters within Latin1. (Note that |
4692 | * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else | |
4693 | * wouldn't be invariant) */ | |
4694 | if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) { | |
eb4e9c04 | 4695 | if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) { |
bedac28b KW |
4696 | sayNO; |
4697 | } | |
4698 | } | |
3018b823 KW |
4699 | else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) { |
4700 | if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), | |
94bb8c36 | 4701 | (U8) TWO_BYTE_UTF8_TO_NATIVE(nextchr, |
3018b823 KW |
4702 | *(locinput + 1)))))) |
4703 | { | |
bedac28b | 4704 | sayNO; |
3018b823 | 4705 | } |
bedac28b | 4706 | } |
3018b823 KW |
4707 | else { /* Here, must be an above Latin-1 code point */ |
4708 | goto utf8_posix_not_eos; | |
bedac28b | 4709 | } |
3018b823 KW |
4710 | |
4711 | /* Here, must be utf8 */ | |
4712 | locinput += UTF8SKIP(locinput); | |
bedac28b KW |
4713 | break; |
4714 | ||
3018b823 KW |
4715 | case NPOSIXD: /* \W or [:^punct:] etc. under /d */ |
4716 | to_complement = 1; | |
4717 | /* FALLTHROUGH */ | |
4718 | ||
4719 | case POSIXD: /* \w or [:punct:] etc. under /d */ | |
bedac28b | 4720 | if (utf8_target) { |
3018b823 | 4721 | goto utf8_posix; |
bedac28b | 4722 | } |
3018b823 KW |
4723 | goto posixa; |
4724 | ||
4725 | case NPOSIXA: /* \W or [:^punct:] etc. under /a */ | |
bedac28b | 4726 | |
3018b823 | 4727 | if (NEXTCHR_IS_EOS) { |
bedac28b KW |
4728 | sayNO; |
4729 | } | |
bedac28b | 4730 | |
3018b823 KW |
4731 | /* All UTF-8 variants match */ |
4732 | if (! UTF8_IS_INVARIANT(nextchr)) { | |
4733 | goto increment_locinput; | |
bedac28b | 4734 | } |
ee9a90b8 | 4735 | |
3018b823 KW |
4736 | to_complement = 1; |
4737 | /* FALLTHROUGH */ | |
4738 | ||
4739 | case POSIXA: /* \w or [:punct:] etc. under /a */ | |
4740 | ||
4741 | posixa: | |
4742 | /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in | |
4743 | * UTF-8, and also from NPOSIXA even in UTF-8 when the current | |
4744 | * character is a single byte */ | |
20d0b1e9 | 4745 | |
3018b823 KW |
4746 | if (NEXTCHR_IS_EOS |
4747 | || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr, | |
4748 | FLAGS(scan))))) | |
4749 | { | |
0658cdde KW |
4750 | sayNO; |
4751 | } | |
3018b823 KW |
4752 | |
4753 | /* Here we are either not in utf8, or we matched a utf8-invariant, | |
4754 | * so the next char is the next byte */ | |
3640db6b | 4755 | locinput++; |
0658cdde | 4756 | break; |
3c0563b9 | 4757 | |
3018b823 KW |
4758 | case NPOSIXU: /* \W or [:^punct:] etc. under /u */ |
4759 | to_complement = 1; | |
4760 | /* FALLTHROUGH */ | |
4761 | ||
4762 | case POSIXU: /* \w or [:punct:] etc. under /u */ | |
4763 | utf8_posix: | |
4764 | if (NEXTCHR_IS_EOS) { | |
0658cdde KW |
4765 | sayNO; |
4766 | } | |
3018b823 KW |
4767 | utf8_posix_not_eos: |
4768 | ||
4769 | /* Use _generic_isCC() for characters within Latin1. (Note that | |
4770 | * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else | |
4771 | * wouldn't be invariant) */ | |
4772 | if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) { | |
4773 | if (! (to_complement ^ cBOOL(_generic_isCC(nextchr, | |
4774 | FLAGS(scan))))) | |
4775 | { | |
4776 | sayNO; | |
4777 | } | |
4778 | locinput++; | |
4779 | } | |
4780 | else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) { | |
4781 | if (! (to_complement | |
94bb8c36 | 4782 | ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(nextchr, |
3018b823 | 4783 | *(locinput + 1)), |
94bb8c36 | 4784 | FLAGS(scan))))) |
3018b823 KW |
4785 | { |
4786 | sayNO; | |
4787 | } | |
4788 | locinput += 2; | |
4789 | } | |
4790 | else { /* Handle above Latin-1 code points */ | |
4791 | classnum = (_char_class_number) FLAGS(scan); | |
4792 | if (classnum < _FIRST_NON_SWASH_CC) { | |
4793 | ||
4794 | /* Here, uses a swash to find such code points. Load if if | |
4795 | * not done already */ | |
4796 | if (! PL_utf8_swash_ptrs[classnum]) { | |
4797 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; | |
4798 | PL_utf8_swash_ptrs[classnum] | |
4799 | = _core_swash_init("utf8", | |
2a16ac92 KW |
4800 | "", |
4801 | &PL_sv_undef, 1, 0, | |
4802 | PL_XPosix_ptrs[classnum], &flags); | |
3018b823 KW |
4803 | } |
4804 | if (! (to_complement | |
4805 | ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], | |
4806 | (U8 *) locinput, TRUE)))) | |
4807 | { | |
4808 | sayNO; | |
4809 | } | |
4810 | } | |
4811 | else { /* Here, uses macros to find above Latin-1 code points */ | |
4812 | switch (classnum) { | |
4813 | case _CC_ENUM_SPACE: /* XXX would require separate | |
4814 | code if we revert the change | |
4815 | of \v matching this */ | |
4816 | case _CC_ENUM_PSXSPC: | |
4817 | if (! (to_complement | |
4818 | ^ cBOOL(is_XPERLSPACE_high(locinput)))) | |
4819 | { | |
4820 | sayNO; | |
4821 | } | |
4822 | break; | |
4823 | case _CC_ENUM_BLANK: | |
4824 | if (! (to_complement | |
4825 | ^ cBOOL(is_HORIZWS_high(locinput)))) | |
4826 | { | |
4827 | sayNO; | |
4828 | } | |
4829 | break; | |
4830 | case _CC_ENUM_XDIGIT: | |
4831 | if (! (to_complement | |
4832 | ^ cBOOL(is_XDIGIT_high(locinput)))) | |
4833 | { | |
4834 | sayNO; | |
4835 | } | |
4836 | break; | |
4837 | case _CC_ENUM_VERTSPACE: | |
4838 | if (! (to_complement | |
4839 | ^ cBOOL(is_VERTWS_high(locinput)))) | |
4840 | { | |
4841 | sayNO; | |
4842 | } | |
4843 | break; | |
4844 | default: /* The rest, e.g. [:cntrl:], can't match | |
4845 | above Latin1 */ | |
4846 | if (! to_complement) { | |
4847 | sayNO; | |
4848 | } | |
4849 | break; | |
4850 | } | |
4851 | } | |
4852 | locinput += UTF8SKIP(locinput); | |
4853 | } | |
4854 | break; | |
0658cdde | 4855 | |
37e2e78e KW |
4856 | case CLUMP: /* Match \X: logical Unicode character. This is defined as |
4857 | a Unicode extended Grapheme Cluster */ | |
4858 | /* From http://www.unicode.org/reports/tr29 (5.2 version). An | |
4859 | extended Grapheme Cluster is: | |
4860 | ||
7aee35ff KW |
4861 | CR LF |
4862 | | Prepend* Begin Extend* | |
4863 | | . | |
37e2e78e | 4864 | |
7aee35ff KW |
4865 | Begin is: ( Special_Begin | ! Control ) |
4866 | Special_Begin is: ( Regional-Indicator+ | Hangul-syllable ) | |
4867 | Extend is: ( Grapheme_Extend | Spacing_Mark ) | |
4868 | Control is: [ GCB_Control | CR | LF ] | |
4869 | Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) )) | |
37e2e78e | 4870 | |
27d4fc33 KW |
4871 | If we create a 'Regular_Begin' = Begin - Special_Begin, then |
4872 | we can rewrite | |
4873 | ||
4874 | Begin is ( Regular_Begin + Special Begin ) | |
4875 | ||
4876 | It turns out that 98.4% of all Unicode code points match | |
4877 | Regular_Begin. Doing it this way eliminates a table match in | |
c101f46d | 4878 | the previous implementation for almost all Unicode code points. |
27d4fc33 | 4879 | |
37e2e78e KW |
4880 | There is a subtlety with Prepend* which showed up in testing. |
4881 | Note that the Begin, and only the Begin is required in: | |
4882 | | Prepend* Begin Extend* | |
cc3b396d KW |
4883 | Also, Begin contains '! Control'. A Prepend must be a |
4884 | '! Control', which means it must also be a Begin. What it | |
4885 | comes down to is that if we match Prepend* and then find no | |
4886 | suitable Begin afterwards, that if we backtrack the last | |
4887 | Prepend, that one will be a suitable Begin. | |
37e2e78e KW |
4888 | */ |
4889 | ||
7016d6eb | 4890 | if (NEXTCHR_IS_EOS) |
a0ed51b3 | 4891 | sayNO; |
f2ed9b32 | 4892 | if (! utf8_target) { |
37e2e78e KW |
4893 | |
4894 | /* Match either CR LF or '.', as all the other possibilities | |
4895 | * require utf8 */ | |
4896 | locinput++; /* Match the . or CR */ | |
cc3b396d KW |
4897 | if (nextchr == '\r' /* And if it was CR, and the next is LF, |
4898 | match the LF */ | |
220db18a | 4899 | && locinput < reginfo->strend |
e699a1d5 KW |
4900 | && UCHARAT(locinput) == '\n') |
4901 | { | |
4902 | locinput++; | |
4903 | } | |
37e2e78e KW |
4904 | } |
4905 | else { | |
4906 | ||
4907 | /* Utf8: See if is ( CR LF ); already know that locinput < | |
220db18a DM |
4908 | * reginfo->strend, so locinput+1 is in bounds */ |
4909 | if ( nextchr == '\r' && locinput+1 < reginfo->strend | |
f67f9e53 | 4910 | && UCHARAT(locinput + 1) == '\n') |
7016d6eb | 4911 | { |
37e2e78e KW |
4912 | locinput += 2; |
4913 | } | |
4914 | else { | |
45fdf108 KW |
4915 | STRLEN len; |
4916 | ||
37e2e78e KW |
4917 | /* In case have to backtrack to beginning, then match '.' */ |
4918 | char *starting = locinput; | |
4919 | ||
4920 | /* In case have to backtrack the last prepend */ | |
e699a1d5 | 4921 | char *previous_prepend = NULL; |
37e2e78e KW |
4922 | |
4923 | LOAD_UTF8_CHARCLASS_GCB(); | |
4924 | ||
45fdf108 | 4925 | /* Match (prepend)* */ |
220db18a | 4926 | while (locinput < reginfo->strend |
45fdf108 KW |
4927 | && (len = is_GCB_Prepend_utf8(locinput))) |
4928 | { | |
4929 | previous_prepend = locinput; | |
4930 | locinput += len; | |
a1853d78 | 4931 | } |
37e2e78e KW |
4932 | |
4933 | /* As noted above, if we matched a prepend character, but | |
4934 | * the next thing won't match, back off the last prepend we | |
4935 | * matched, as it is guaranteed to match the begin */ | |
4936 | if (previous_prepend | |
220db18a | 4937 | && (locinput >= reginfo->strend |
c101f46d KW |
4938 | || (! swash_fetch(PL_utf8_X_regular_begin, |
4939 | (U8*)locinput, utf8_target) | |
bff53399 | 4940 | && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput))) |
c101f46d | 4941 | ) |
37e2e78e KW |
4942 | { |
4943 | locinput = previous_prepend; | |
4944 | } | |
4945 | ||
220db18a | 4946 | /* Note that here we know reginfo->strend > locinput, as we |
37e2e78e KW |
4947 | * tested that upon input to this switch case, and if we |
4948 | * moved locinput forward, we tested the result just above | |
4949 | * and it either passed, or we backed off so that it will | |
4950 | * now pass */ | |
11dfcd49 KW |
4951 | if (swash_fetch(PL_utf8_X_regular_begin, |
4952 | (U8*)locinput, utf8_target)) { | |
27d4fc33 KW |
4953 | locinput += UTF8SKIP(locinput); |
4954 | } | |
bff53399 | 4955 | else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) { |
37e2e78e KW |
4956 | |
4957 | /* Here did not match the required 'Begin' in the | |
4958 | * second term. So just match the very first | |
4959 | * character, the '.' of the final term of the regex */ | |
4960 | locinput = starting + UTF8SKIP(starting); | |
27d4fc33 | 4961 | goto exit_utf8; |
37e2e78e KW |
4962 | } else { |
4963 | ||
11dfcd49 KW |
4964 | /* Here is a special begin. It can be composed of |
4965 | * several individual characters. One possibility is | |
4966 | * RI+ */ | |
45fdf108 KW |
4967 | if ((len = is_GCB_RI_utf8(locinput))) { |
4968 | locinput += len; | |
220db18a | 4969 | while (locinput < reginfo->strend |
45fdf108 | 4970 | && (len = is_GCB_RI_utf8(locinput))) |
11dfcd49 | 4971 | { |
45fdf108 | 4972 | locinput += len; |
11dfcd49 | 4973 | } |
45fdf108 KW |
4974 | } else if ((len = is_GCB_T_utf8(locinput))) { |
4975 | /* Another possibility is T+ */ | |
4976 | locinput += len; | |
220db18a | 4977 | while (locinput < reginfo->strend |
45fdf108 | 4978 | && (len = is_GCB_T_utf8(locinput))) |
11dfcd49 | 4979 | { |
45fdf108 | 4980 | locinput += len; |
11dfcd49 KW |
4981 | } |
4982 | } else { | |
4983 | ||
4984 | /* Here, neither RI+ nor T+; must be some other | |
4985 | * Hangul. That means it is one of the others: L, | |
4986 | * LV, LVT or V, and matches: | |
4987 | * L* (L | LVT T* | V * V* T* | LV V* T*) */ | |
4988 | ||
4989 | /* Match L* */ | |
220db18a | 4990 | while (locinput < reginfo->strend |
45fdf108 | 4991 | && (len = is_GCB_L_utf8(locinput))) |
11dfcd49 | 4992 | { |
45fdf108 | 4993 | locinput += len; |
11dfcd49 | 4994 | } |
37e2e78e | 4995 | |
11dfcd49 KW |
4996 | /* Here, have exhausted L*. If the next character |
4997 | * is not an LV, LVT nor V, it means we had to have | |
4998 | * at least one L, so matches L+ in the original | |
4999 | * equation, we have a complete hangul syllable. | |
5000 | * Are done. */ | |
5001 | ||
220db18a | 5002 | if (locinput < reginfo->strend |
45fdf108 | 5003 | && is_GCB_LV_LVT_V_utf8(locinput)) |
11dfcd49 | 5004 | { |
11dfcd49 | 5005 | /* Otherwise keep going. Must be LV, LVT or V. |
7d43c479 KW |
5006 | * See if LVT, by first ruling out V, then LV */ |
5007 | if (! is_GCB_V_utf8(locinput) | |
5008 | /* All but every TCount one is LV */ | |
5009 | && (valid_utf8_to_uvchr((U8 *) locinput, | |
5010 | NULL) | |
5011 | - SBASE) | |
5012 | % TCount != 0) | |
5013 | { | |
11dfcd49 KW |
5014 | locinput += UTF8SKIP(locinput); |
5015 | } else { | |
5016 | ||
5017 | /* Must be V or LV. Take it, then match | |
5018 | * V* */ | |
5019 | locinput += UTF8SKIP(locinput); | |
220db18a | 5020 | while (locinput < reginfo->strend |
45fdf108 | 5021 | && (len = is_GCB_V_utf8(locinput))) |
11dfcd49 | 5022 | { |
45fdf108 | 5023 | locinput += len; |
11dfcd49 KW |
5024 | } |
5025 | } | |
37e2e78e | 5026 | |
11dfcd49 | 5027 | /* And any of LV, LVT, or V can be followed |
45fdf108 | 5028 | * by T* */ |
220db18a | 5029 | while (locinput < reginfo->strend |
45fdf108 | 5030 | && (len = is_GCB_T_utf8(locinput))) |
11dfcd49 | 5031 | { |
45fdf108 | 5032 | locinput += len; |
11dfcd49 KW |
5033 | } |
5034 | } | |
cd94d768 | 5035 | } |
11dfcd49 | 5036 | } |
37e2e78e | 5037 | |
11dfcd49 | 5038 | /* Match any extender */ |
220db18a | 5039 | while (locinput < reginfo->strend |
11dfcd49 KW |
5040 | && swash_fetch(PL_utf8_X_extend, |
5041 | (U8*)locinput, utf8_target)) | |
5042 | { | |
5043 | locinput += UTF8SKIP(locinput); | |
5044 | } | |
37e2e78e | 5045 | } |
27d4fc33 | 5046 | exit_utf8: |
220db18a | 5047 | if (locinput > reginfo->strend) sayNO; |
37e2e78e | 5048 | } |
a0ed51b3 | 5049 | break; |
81714fb9 | 5050 | |
3c0563b9 | 5051 | case NREFFL: /* /\g{name}/il */ |
d7ef4b73 KW |
5052 | { /* The capture buffer cases. The ones beginning with N for the |
5053 | named buffers just convert to the equivalent numbered and | |
5054 | pretend they were called as the corresponding numbered buffer | |
5055 | op. */ | |
26ecd678 TC |
5056 | /* don't initialize these in the declaration, it makes C++ |
5057 | unhappy */ | |
9d9163fb | 5058 | const char *s; |
ff1157ca | 5059 | char type; |
8368298a TC |
5060 | re_fold_t folder; |
5061 | const U8 *fold_array; | |
26ecd678 | 5062 | UV utf8_fold_flags; |
8368298a | 5063 | |
d7ef4b73 KW |
5064 | folder = foldEQ_locale; |
5065 | fold_array = PL_fold_locale; | |
5066 | type = REFFL; | |
cea315b6 | 5067 | utf8_fold_flags = FOLDEQ_LOCALE; |
d7ef4b73 KW |
5068 | goto do_nref; |
5069 | ||
3c0563b9 | 5070 | case NREFFA: /* /\g{name}/iaa */ |
2f7f8cb1 KW |
5071 | folder = foldEQ_latin1; |
5072 | fold_array = PL_fold_latin1; | |
5073 | type = REFFA; | |
5074 | utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; | |
5075 | goto do_nref; | |
5076 | ||
3c0563b9 | 5077 | case NREFFU: /* /\g{name}/iu */ |
d7ef4b73 KW |
5078 | folder = foldEQ_latin1; |
5079 | fold_array = PL_fold_latin1; | |
5080 | type = REFFU; | |
d513472c | 5081 | utf8_fold_flags = 0; |
d7ef4b73 KW |
5082 | goto do_nref; |
5083 | ||
3c0563b9 | 5084 | case NREFF: /* /\g{name}/i */ |
d7ef4b73 KW |
5085 | folder = foldEQ; |
5086 | fold_array = PL_fold; | |
5087 | type = REFF; | |
d513472c | 5088 | utf8_fold_flags = 0; |
d7ef4b73 KW |
5089 | goto do_nref; |
5090 | ||
3c0563b9 | 5091 | case NREF: /* /\g{name}/ */ |
d7ef4b73 | 5092 | type = REF; |
83d7b90b KW |
5093 | folder = NULL; |
5094 | fold_array = NULL; | |
d513472c | 5095 | utf8_fold_flags = 0; |
d7ef4b73 KW |
5096 | do_nref: |
5097 | ||
5098 | /* For the named back references, find the corresponding buffer | |
5099 | * number */ | |
0a4db386 YO |
5100 | n = reg_check_named_buff_matched(rex,scan); |
5101 | ||
d7ef4b73 | 5102 | if ( ! n ) { |
81714fb9 | 5103 | sayNO; |
d7ef4b73 KW |
5104 | } |
5105 | goto do_nref_ref_common; | |
5106 | ||
3c0563b9 | 5107 | case REFFL: /* /\1/il */ |
d7ef4b73 KW |
5108 | folder = foldEQ_locale; |
5109 | fold_array = PL_fold_locale; | |
cea315b6 | 5110 | utf8_fold_flags = FOLDEQ_LOCALE; |
d7ef4b73 KW |
5111 | goto do_ref; |
5112 | ||
3c0563b9 | 5113 | case REFFA: /* /\1/iaa */ |
2f7f8cb1 KW |
5114 | folder = foldEQ_latin1; |
5115 | fold_array = PL_fold_latin1; | |
5116 | utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; | |
5117 | goto do_ref; | |
5118 | ||
3c0563b9 | 5119 | case REFFU: /* /\1/iu */ |
d7ef4b73 KW |
5120 | folder = foldEQ_latin1; |
5121 | fold_array = PL_fold_latin1; | |
d513472c | 5122 | utf8_fold_flags = 0; |
d7ef4b73 KW |
5123 | goto do_ref; |
5124 | ||
3c0563b9 | 5125 | case REFF: /* /\1/i */ |
d7ef4b73 KW |
5126 | folder = foldEQ; |
5127 | fold_array = PL_fold; | |
d513472c | 5128 | utf8_fold_flags = 0; |
83d7b90b | 5129 | goto do_ref; |
d7ef4b73 | 5130 | |
3c0563b9 | 5131 | case REF: /* /\1/ */ |
83d7b90b KW |
5132 | folder = NULL; |
5133 | fold_array = NULL; | |
d513472c | 5134 | utf8_fold_flags = 0; |
83d7b90b | 5135 | |
d7ef4b73 | 5136 | do_ref: |
81714fb9 | 5137 | type = OP(scan); |
d7ef4b73 KW |
5138 | n = ARG(scan); /* which paren pair */ |
5139 | ||
5140 | do_nref_ref_common: | |
b93070ed | 5141 | ln = rex->offs[n].start; |
1cb48e53 | 5142 | reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */ |
b93070ed | 5143 | if (rex->lastparen < n || ln == -1) |
af3f8c16 | 5144 | sayNO; /* Do not match unless seen CLOSEn. */ |
b93070ed | 5145 | if (ln == rex->offs[n].end) |
a0d0e21e | 5146 | break; |
a0ed51b3 | 5147 | |
9d9163fb | 5148 | s = reginfo->strbeg + ln; |
d7ef4b73 | 5149 | if (type != REF /* REF can do byte comparison */ |
31f05a37 KW |
5150 | && (utf8_target || type == REFFU || type == REFFL)) |
5151 | { | |
220db18a | 5152 | char * limit = reginfo->strend; |
d7ef4b73 KW |
5153 | |
5154 | /* This call case insensitively compares the entire buffer | |
5155 | * at s, with the current input starting at locinput, but | |
220db18a DM |
5156 | * not going off the end given by reginfo->strend, and |
5157 | * returns in <limit> upon success, how much of the | |
5158 | * current input was matched */ | |
b93070ed | 5159 | if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target, |
d513472c | 5160 | locinput, &limit, 0, utf8_target, utf8_fold_flags)) |
d7ef4b73 KW |
5161 | { |
5162 | sayNO; | |
a0ed51b3 | 5163 | } |
d7ef4b73 | 5164 | locinput = limit; |
a0ed51b3 LW |
5165 | break; |
5166 | } | |
5167 | ||
d7ef4b73 | 5168 | /* Not utf8: Inline the first character, for speed. */ |
7016d6eb DM |
5169 | if (!NEXTCHR_IS_EOS && |
5170 | UCHARAT(s) != nextchr && | |
81714fb9 | 5171 | (type == REF || |
d7ef4b73 | 5172 | UCHARAT(s) != fold_array[nextchr])) |
4633a7c4 | 5173 | sayNO; |
b93070ed | 5174 | ln = rex->offs[n].end - ln; |
220db18a | 5175 | if (locinput + ln > reginfo->strend) |
4633a7c4 | 5176 | sayNO; |
81714fb9 | 5177 | if (ln > 1 && (type == REF |
24d3c4a9 | 5178 | ? memNE(s, locinput, ln) |
d7ef4b73 | 5179 | : ! folder(s, locinput, ln))) |
4633a7c4 | 5180 | sayNO; |
24d3c4a9 | 5181 | locinput += ln; |
a0d0e21e | 5182 | break; |
81714fb9 | 5183 | } |
3c0563b9 DM |
5184 | |
5185 | case NOTHING: /* null op; e.g. the 'nothing' following | |
5186 | * the '*' in m{(a+|b)*}' */ | |
5187 | break; | |
5188 | case TAIL: /* placeholder while compiling (A|B|C) */ | |
a0d0e21e | 5189 | break; |
3c0563b9 | 5190 | |
40a82448 DM |
5191 | #undef ST |
5192 | #define ST st->u.eval | |
c277df42 | 5193 | { |
c277df42 | 5194 | SV *ret; |
d2f13c59 | 5195 | REGEXP *re_sv; |
6bda09f9 | 5196 | regexp *re; |
f8fc2ecf | 5197 | regexp_internal *rei; |
1a147d38 YO |
5198 | regnode *startpoint; |
5199 | ||
3c0563b9 | 5200 | case GOSTART: /* (?R) */ |
e7707071 YO |
5201 | case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */ |
5202 | if (cur_eval && cur_eval->locinput==locinput) { | |
24b23f37 | 5203 | if (cur_eval->u.eval.close_paren == (U32)ARG(scan)) |
1a147d38 | 5204 | Perl_croak(aTHX_ "Infinite recursion in regex"); |
4b196cd4 | 5205 | if ( ++nochange_depth > max_nochange_depth ) |
1a147d38 YO |
5206 | Perl_croak(aTHX_ |
5207 | "Pattern subroutine nesting without pos change" | |
5208 | " exceeded limit in regex"); | |
6bda09f9 YO |
5209 | } else { |
5210 | nochange_depth = 0; | |
1a147d38 | 5211 | } |
288b8c02 | 5212 | re_sv = rex_sv; |
6bda09f9 | 5213 | re = rex; |
f8fc2ecf | 5214 | rei = rexi; |
1a147d38 | 5215 | if (OP(scan)==GOSUB) { |
6bda09f9 YO |
5216 | startpoint = scan + ARG2L(scan); |
5217 | ST.close_paren = ARG(scan); | |
5218 | } else { | |
f8fc2ecf | 5219 | startpoint = rei->program+1; |
6bda09f9 YO |
5220 | ST.close_paren = 0; |
5221 | } | |
d1b2014a YO |
5222 | |
5223 | /* Save all the positions seen so far. */ | |
5224 | ST.cp = regcppush(rex, 0, maxopenparen); | |
5225 | REGCP_SET(ST.lastcp); | |
5226 | ||
5227 | /* and then jump to the code we share with EVAL */ | |
6bda09f9 | 5228 | goto eval_recurse_doit; |
a74ff37d | 5229 | /* NOTREACHED */ |
3c0563b9 | 5230 | |
6bda09f9 YO |
5231 | case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */ |
5232 | if (cur_eval && cur_eval->locinput==locinput) { | |
4b196cd4 | 5233 | if ( ++nochange_depth > max_nochange_depth ) |
1a147d38 | 5234 | Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex"); |
6bda09f9 YO |
5235 | } else { |
5236 | nochange_depth = 0; | |
5237 | } | |
8e5e9ebe | 5238 | { |
4aabdb9b | 5239 | /* execute the code in the {...} */ |
81ed78b2 | 5240 | |
4aabdb9b | 5241 | dSP; |
a6dc34f1 | 5242 | IV before; |
1f4d1a1e | 5243 | OP * const oop = PL_op; |
4aabdb9b | 5244 | COP * const ocurcop = PL_curcop; |
81ed78b2 | 5245 | OP *nop; |
81ed78b2 | 5246 | CV *newcv; |
91332126 | 5247 | |
74088413 | 5248 | /* save *all* paren positions */ |
92da3157 | 5249 | regcppush(rex, 0, maxopenparen); |
74088413 DM |
5250 | REGCP_SET(runops_cp); |
5251 | ||
81ed78b2 DM |
5252 | if (!caller_cv) |
5253 | caller_cv = find_runcv(NULL); | |
5254 | ||
4aabdb9b | 5255 | n = ARG(scan); |
81ed78b2 | 5256 | |
b30fcab9 | 5257 | if (rexi->data->what[n] == 'r') { /* code from an external qr */ |
8d919b0a | 5258 | newcv = (ReANY( |
b30fcab9 DM |
5259 | (REGEXP*)(rexi->data->data[n]) |
5260 | ))->qr_anoncv | |
81ed78b2 DM |
5261 | ; |
5262 | nop = (OP*)rexi->data->data[n+1]; | |
b30fcab9 DM |
5263 | } |
5264 | else if (rexi->data->what[n] == 'l') { /* literal code */ | |
81ed78b2 DM |
5265 | newcv = caller_cv; |
5266 | nop = (OP*)rexi->data->data[n]; | |
5267 | assert(CvDEPTH(newcv)); | |
68e2671b DM |
5268 | } |
5269 | else { | |
d24ca0c5 DM |
5270 | /* literal with own CV */ |
5271 | assert(rexi->data->what[n] == 'L'); | |
81ed78b2 DM |
5272 | newcv = rex->qr_anoncv; |
5273 | nop = (OP*)rexi->data->data[n]; | |
68e2671b | 5274 | } |
81ed78b2 | 5275 | |
0e458318 DM |
5276 | /* normally if we're about to execute code from the same |
5277 | * CV that we used previously, we just use the existing | |
5278 | * CX stack entry. However, its possible that in the | |
5279 | * meantime we may have backtracked, popped from the save | |
5280 | * stack, and undone the SAVECOMPPAD(s) associated with | |
5281 | * PUSH_MULTICALL; in which case PL_comppad no longer | |
5282 | * points to newcv's pad. */ | |
5283 | if (newcv != last_pushed_cv || PL_comppad != last_pad) | |
5284 | { | |
b0065247 DM |
5285 | U8 flags = (CXp_SUB_RE | |
5286 | ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0)); | |
0e458318 | 5287 | if (last_pushed_cv) { |
b0065247 | 5288 | CHANGE_MULTICALL_FLAGS(newcv, flags); |
0e458318 DM |
5289 | } |
5290 | else { | |
b0065247 | 5291 | PUSH_MULTICALL_FLAGS(newcv, flags); |
0e458318 DM |
5292 | } |
5293 | last_pushed_cv = newcv; | |
5294 | } | |
c31ee3bb DM |
5295 | else { |
5296 | /* these assignments are just to silence compiler | |
5297 | * warnings */ | |
5298 | multicall_cop = NULL; | |
5299 | newsp = NULL; | |
5300 | } | |
0e458318 DM |
5301 | last_pad = PL_comppad; |
5302 | ||
2e2e3f36 DM |
5303 | /* the initial nextstate you would normally execute |
5304 | * at the start of an eval (which would cause error | |
5305 | * messages to come from the eval), may be optimised | |
5306 | * away from the execution path in the regex code blocks; | |
5307 | * so manually set PL_curcop to it initially */ | |
5308 | { | |
81ed78b2 | 5309 | OP *o = cUNOPx(nop)->op_first; |
2e2e3f36 DM |
5310 | assert(o->op_type == OP_NULL); |
5311 | if (o->op_targ == OP_SCOPE) { | |
5312 | o = cUNOPo->op_first; | |
5313 | } | |
5314 | else { | |
5315 | assert(o->op_targ == OP_LEAVE); | |
5316 | o = cUNOPo->op_first; | |
5317 | assert(o->op_type == OP_ENTER); | |
1ed44841 | 5318 | o = OP_SIBLING(o); |
2e2e3f36 DM |
5319 | } |
5320 | ||
5321 | if (o->op_type != OP_STUB) { | |
5322 | assert( o->op_type == OP_NEXTSTATE | |
5323 | || o->op_type == OP_DBSTATE | |
5324 | || (o->op_type == OP_NULL | |
5325 | && ( o->op_targ == OP_NEXTSTATE | |
5326 | || o->op_targ == OP_DBSTATE | |
5327 | ) | |
5328 | ) | |
5329 | ); | |
5330 | PL_curcop = (COP*)o; | |
5331 | } | |
5332 | } | |
81ed78b2 | 5333 | nop = nop->op_next; |
2e2e3f36 | 5334 | |
24b23f37 | 5335 | DEBUG_STATE_r( PerlIO_printf(Perl_debug_log, |
81ed78b2 DM |
5336 | " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) ); |
5337 | ||
8adc0f72 | 5338 | rex->offs[0].end = locinput - reginfo->strbeg; |
bf2039a9 | 5339 | if (reginfo->info_aux_eval->pos_magic) |
25fdce4a FC |
5340 | MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic, |
5341 | reginfo->sv, reginfo->strbeg, | |
5342 | locinput - reginfo->strbeg); | |
4aabdb9b | 5343 | |
2bf803e2 YO |
5344 | if (sv_yes_mark) { |
5345 | SV *sv_mrk = get_sv("REGMARK", 1); | |
5346 | sv_setsv(sv_mrk, sv_yes_mark); | |
5347 | } | |
5348 | ||
81ed78b2 DM |
5349 | /* we don't use MULTICALL here as we want to call the |
5350 | * first op of the block of interest, rather than the | |
5351 | * first op of the sub */ | |
a6dc34f1 | 5352 | before = (IV)(SP-PL_stack_base); |
81ed78b2 | 5353 | PL_op = nop; |
8e5e9ebe RGS |
5354 | CALLRUNOPS(aTHX); /* Scalar context. */ |
5355 | SPAGAIN; | |
a6dc34f1 | 5356 | if ((IV)(SP-PL_stack_base) == before) |
075aa684 | 5357 | ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */ |
8e5e9ebe RGS |
5358 | else { |
5359 | ret = POPs; | |
5360 | PUTBACK; | |
5361 | } | |
4aabdb9b | 5362 | |
e4bfbed3 DM |
5363 | /* before restoring everything, evaluate the returned |
5364 | * value, so that 'uninit' warnings don't use the wrong | |
497d0a96 DM |
5365 | * PL_op or pad. Also need to process any magic vars |
5366 | * (e.g. $1) *before* parentheses are restored */ | |
e4bfbed3 DM |
5367 | |
5368 | PL_op = NULL; | |
5369 | ||
5e98dac2 | 5370 | re_sv = NULL; |
e4bfbed3 DM |
5371 | if (logical == 0) /* (?{})/ */ |
5372 | sv_setsv(save_scalar(PL_replgv), ret); /* $^R */ | |
5373 | else if (logical == 1) { /* /(?(?{...})X|Y)/ */ | |
5374 | sw = cBOOL(SvTRUE(ret)); | |
5375 | logical = 0; | |
5376 | } | |
5377 | else { /* /(??{}) */ | |
497d0a96 DM |
5378 | /* if its overloaded, let the regex compiler handle |
5379 | * it; otherwise extract regex, or stringify */ | |
237da807 | 5380 | if (SvGMAGICAL(ret)) |
2685dc2d | 5381 | ret = sv_mortalcopy(ret); |
497d0a96 DM |
5382 | if (!SvAMAGIC(ret)) { |
5383 | SV *sv = ret; | |
5384 | if (SvROK(sv)) | |
5385 | sv = SvRV(sv); | |
5386 | if (SvTYPE(sv) == SVt_REGEXP) | |
5387 | re_sv = (REGEXP*) sv; | |
63620942 FC |
5388 | else if (SvSMAGICAL(ret)) { |
5389 | MAGIC *mg = mg_find(ret, PERL_MAGIC_qr); | |
497d0a96 DM |
5390 | if (mg) |
5391 | re_sv = (REGEXP *) mg->mg_obj; | |
5392 | } | |
e4bfbed3 | 5393 | |
2685dc2d | 5394 | /* force any undef warnings here */ |
237da807 FC |
5395 | if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) { |
5396 | ret = sv_mortalcopy(ret); | |
497d0a96 DM |
5397 | (void) SvPV_force_nolen(ret); |
5398 | } | |
e4bfbed3 DM |
5399 | } |
5400 | ||
5401 | } | |
5402 | ||
81ed78b2 DM |
5403 | /* *** Note that at this point we don't restore |
5404 | * PL_comppad, (or pop the CxSUB) on the assumption it may | |
5405 | * be used again soon. This is safe as long as nothing | |
5406 | * in the regexp code uses the pad ! */ | |
4aabdb9b | 5407 | PL_op = oop; |
4aabdb9b | 5408 | PL_curcop = ocurcop; |
92da3157 | 5409 | S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen); |
f5df269c | 5410 | PL_curpm = PL_reg_curpm; |
e4bfbed3 DM |
5411 | |
5412 | if (logical != 2) | |
4aabdb9b | 5413 | break; |
8e5e9ebe | 5414 | } |
e4bfbed3 DM |
5415 | |
5416 | /* only /(??{})/ from now on */ | |
24d3c4a9 | 5417 | logical = 0; |
4aabdb9b | 5418 | { |
4f639d21 DM |
5419 | /* extract RE object from returned value; compiling if |
5420 | * necessary */ | |
5c35adbb | 5421 | |
575c37f6 DM |
5422 | if (re_sv) { |
5423 | re_sv = reg_temp_copy(NULL, re_sv); | |
288b8c02 | 5424 | } |
0f5d15d6 | 5425 | else { |
c737faaf | 5426 | U32 pm_flags = 0; |
0f5d15d6 | 5427 | |
9753d940 DM |
5428 | if (SvUTF8(ret) && IN_BYTES) { |
5429 | /* In use 'bytes': make a copy of the octet | |
5430 | * sequence, but without the flag on */ | |
b9ad30b4 NC |
5431 | STRLEN len; |
5432 | const char *const p = SvPV(ret, len); | |
5433 | ret = newSVpvn_flags(p, len, SVs_TEMP); | |
5434 | } | |
732caac7 DM |
5435 | if (rex->intflags & PREGf_USE_RE_EVAL) |
5436 | pm_flags |= PMf_USE_RE_EVAL; | |
5437 | ||
5438 | /* if we got here, it should be an engine which | |
5439 | * supports compiling code blocks and stuff */ | |
5440 | assert(rex->engine && rex->engine->op_comp); | |
ec841a27 | 5441 | assert(!(scan->flags & ~RXf_PMf_COMPILETIME)); |
575c37f6 | 5442 | re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL, |
ec841a27 DM |
5443 | rex->engine, NULL, NULL, |
5444 | /* copy /msix etc to inner pattern */ | |
13f27704 | 5445 | ARG2L(scan), |
ec841a27 | 5446 | pm_flags); |
732caac7 | 5447 | |
9041c2e3 | 5448 | if (!(SvFLAGS(ret) |
237da807 FC |
5449 | & (SVs_TEMP | SVs_GMG | SVf_ROK)) |
5450 | && (!SvPADTMP(ret) || SvREADONLY(ret))) { | |
a2794585 NC |
5451 | /* This isn't a first class regexp. Instead, it's |
5452 | caching a regexp onto an existing, Perl visible | |
5453 | scalar. */ | |
575c37f6 | 5454 | sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0); |
3ce3ed55 | 5455 | } |
0f5d15d6 | 5456 | } |
e1ff3a88 | 5457 | SAVEFREESV(re_sv); |
8d919b0a | 5458 | re = ReANY(re_sv); |
4aabdb9b | 5459 | } |
07bc277f | 5460 | RXp_MATCH_COPIED_off(re); |
28d8d7f4 YO |
5461 | re->subbeg = rex->subbeg; |
5462 | re->sublen = rex->sublen; | |
6502e081 DM |
5463 | re->suboffset = rex->suboffset; |
5464 | re->subcoffset = rex->subcoffset; | |
d1b2014a YO |
5465 | re->lastparen = 0; |
5466 | re->lastcloseparen = 0; | |
f8fc2ecf | 5467 | rei = RXi_GET(re); |
6bda09f9 | 5468 | DEBUG_EXECUTE_r( |
220db18a DM |
5469 | debug_start_match(re_sv, utf8_target, locinput, |
5470 | reginfo->strend, "Matching embedded"); | |
6bda09f9 | 5471 | ); |
f8fc2ecf | 5472 | startpoint = rei->program + 1; |
1a147d38 | 5473 | ST.close_paren = 0; /* only used for GOSUB */ |
d1b2014a YO |
5474 | /* Save all the seen positions so far. */ |
5475 | ST.cp = regcppush(rex, 0, maxopenparen); | |
5476 | REGCP_SET(ST.lastcp); | |
5477 | /* and set maxopenparen to 0, since we are starting a "fresh" match */ | |
5478 | maxopenparen = 0; | |
5479 | /* run the pattern returned from (??{...}) */ | |
aa283a38 | 5480 | |
d1b2014a YO |
5481 | eval_recurse_doit: /* Share code with GOSUB below this line |
5482 | * At this point we expect the stack context to be | |
5483 | * set up correctly */ | |
4aabdb9b | 5484 | |
1cb95af7 DM |
5485 | /* invalidate the S-L poscache. We're now executing a |
5486 | * different set of WHILEM ops (and their associated | |
5487 | * indexes) against the same string, so the bits in the | |
5488 | * cache are meaningless. Setting maxiter to zero forces | |
5489 | * the cache to be invalidated and zeroed before reuse. | |
5490 | * XXX This is too dramatic a measure. Ideally we should | |
5491 | * save the old cache and restore when running the outer | |
5492 | * pattern again */ | |
1cb48e53 | 5493 | reginfo->poscache_maxiter = 0; |
4aabdb9b | 5494 | |
d1b2014a | 5495 | /* the new regexp might have a different is_utf8_pat than we do */ |
aed7b151 | 5496 | is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv)); |
faec1544 | 5497 | |
288b8c02 | 5498 | ST.prev_rex = rex_sv; |
faec1544 | 5499 | ST.prev_curlyx = cur_curlyx; |
ec43f78b DM |
5500 | rex_sv = re_sv; |
5501 | SET_reg_curpm(rex_sv); | |
288b8c02 | 5502 | rex = re; |
f8fc2ecf | 5503 | rexi = rei; |
faec1544 | 5504 | cur_curlyx = NULL; |
40a82448 | 5505 | ST.B = next; |
faec1544 DM |
5506 | ST.prev_eval = cur_eval; |
5507 | cur_eval = st; | |
faec1544 | 5508 | /* now continue from first node in postoned RE */ |
4d5016e5 | 5509 | PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput); |
a74ff37d | 5510 | /* NOTREACHED */ |
e5964223 | 5511 | NOT_REACHED; |
c277df42 | 5512 | } |
40a82448 | 5513 | |
faec1544 DM |
5514 | case EVAL_AB: /* cleanup after a successful (??{A})B */ |
5515 | /* note: this is called twice; first after popping B, then A */ | |
ec43f78b | 5516 | rex_sv = ST.prev_rex; |
aed7b151 | 5517 | is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv)); |
ec43f78b | 5518 | SET_reg_curpm(rex_sv); |
8d919b0a | 5519 | rex = ReANY(rex_sv); |
f8fc2ecf | 5520 | rexi = RXi_GET(rex); |
4b22688e YO |
5521 | { |
5522 | /* preserve $^R across LEAVE's. See Bug 121070. */ | |
5523 | SV *save_sv= GvSV(PL_replgv); | |
5524 | SvREFCNT_inc(save_sv); | |
5525 | regcpblow(ST.cp); /* LEAVE in disguise */ | |
5526 | sv_setsv(GvSV(PL_replgv), save_sv); | |
5527 | SvREFCNT_dec(save_sv); | |
5528 | } | |
faec1544 DM |
5529 | cur_eval = ST.prev_eval; |
5530 | cur_curlyx = ST.prev_curlyx; | |
34a81e2b | 5531 | |
1cb95af7 | 5532 | /* Invalidate cache. See "invalidate" comment above. */ |
1cb48e53 | 5533 | reginfo->poscache_maxiter = 0; |
e7707071 | 5534 | if ( nochange_depth ) |
4b196cd4 | 5535 | nochange_depth--; |
262b90c4 | 5536 | sayYES; |
40a82448 | 5537 | |
40a82448 | 5538 | |
faec1544 DM |
5539 | case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */ |
5540 | /* note: this is called twice; first after popping B, then A */ | |
ec43f78b | 5541 | rex_sv = ST.prev_rex; |
aed7b151 | 5542 | is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv)); |
ec43f78b | 5543 | SET_reg_curpm(rex_sv); |
8d919b0a | 5544 | rex = ReANY(rex_sv); |
f8fc2ecf | 5545 | rexi = RXi_GET(rex); |
0357f1fd | 5546 | |
40a82448 | 5547 | REGCP_UNWIND(ST.lastcp); |
92da3157 | 5548 | regcppop(rex, &maxopenparen); |
faec1544 DM |
5549 | cur_eval = ST.prev_eval; |
5550 | cur_curlyx = ST.prev_curlyx; | |
1cb95af7 | 5551 | /* Invalidate cache. See "invalidate" comment above. */ |
1cb48e53 | 5552 | reginfo->poscache_maxiter = 0; |
e7707071 | 5553 | if ( nochange_depth ) |
4b196cd4 | 5554 | nochange_depth--; |
40a82448 | 5555 | sayNO_SILENT; |
40a82448 DM |
5556 | #undef ST |
5557 | ||
3c0563b9 | 5558 | case OPEN: /* ( */ |
c277df42 | 5559 | n = ARG(scan); /* which paren pair */ |
9d9163fb | 5560 | rex->offs[n].start_tmp = locinput - reginfo->strbeg; |
92da3157 DM |
5561 | if (n > maxopenparen) |
5562 | maxopenparen = n; | |
495f47a5 | 5563 | DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, |
92da3157 | 5564 | "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n", |
495f47a5 DM |
5565 | PTR2UV(rex), |
5566 | PTR2UV(rex->offs), | |
5567 | (UV)n, | |
5568 | (IV)rex->offs[n].start_tmp, | |
92da3157 | 5569 | (UV)maxopenparen |
495f47a5 | 5570 | )); |
e2e6a0f1 | 5571 | lastopen = n; |
a0d0e21e | 5572 | break; |
495f47a5 DM |
5573 | |
5574 | /* XXX really need to log other places start/end are set too */ | |
5575 | #define CLOSE_CAPTURE \ | |
5576 | rex->offs[n].start = rex->offs[n].start_tmp; \ | |
9d9163fb | 5577 | rex->offs[n].end = locinput - reginfo->strbeg; \ |
495f47a5 DM |
5578 | DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \ |
5579 | "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \ | |
5580 | PTR2UV(rex), \ | |
5581 | PTR2UV(rex->offs), \ | |
5582 | (UV)n, \ | |
5583 | (IV)rex->offs[n].start, \ | |
5584 | (IV)rex->offs[n].end \ | |
5585 | )) | |
5586 | ||
3c0563b9 | 5587 | case CLOSE: /* ) */ |
c277df42 | 5588 | n = ARG(scan); /* which paren pair */ |
495f47a5 | 5589 | CLOSE_CAPTURE; |
b93070ed DM |
5590 | if (n > rex->lastparen) |
5591 | rex->lastparen = n; | |
5592 | rex->lastcloseparen = n; | |
3b6647e0 | 5593 | if (cur_eval && cur_eval->u.eval.close_paren == n) { |
6bda09f9 YO |
5594 | goto fake_end; |
5595 | } | |
a0d0e21e | 5596 | break; |
3c0563b9 DM |
5597 | |
5598 | case ACCEPT: /* (*ACCEPT) */ | |
e2e6a0f1 YO |
5599 | if (ARG(scan)){ |
5600 | regnode *cursor; | |
5601 | for (cursor=scan; | |
5602 | cursor && OP(cursor)!=END; | |
5603 | cursor=regnext(cursor)) | |
5604 | { | |
5605 | if ( OP(cursor)==CLOSE ){ | |
5606 | n = ARG(cursor); | |
5607 | if ( n <= lastopen ) { | |
495f47a5 | 5608 | CLOSE_CAPTURE; |
b93070ed DM |
5609 | if (n > rex->lastparen) |
5610 | rex->lastparen = n; | |
5611 | rex->lastcloseparen = n; | |
3b6647e0 RB |
5612 | if ( n == ARG(scan) || (cur_eval && |
5613 | cur_eval->u.eval.close_paren == n)) | |
e2e6a0f1 YO |
5614 | break; |
5615 | } | |
5616 | } | |
5617 | } | |
5618 | } | |
5619 | goto fake_end; | |
a74ff37d | 5620 | /* NOTREACHED */ |
3c0563b9 DM |
5621 | |
5622 | case GROUPP: /* (?(1)) */ | |
c277df42 | 5623 | n = ARG(scan); /* which paren pair */ |
b93070ed | 5624 | sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1); |
c277df42 | 5625 | break; |
3c0563b9 DM |
5626 | |
5627 | case NGROUPP: /* (?(<name>)) */ | |
0a4db386 | 5628 | /* reg_check_named_buff_matched returns 0 for no match */ |
f2338a2e | 5629 | sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan)); |
0a4db386 | 5630 | break; |
3c0563b9 DM |
5631 | |
5632 | case INSUBP: /* (?(R)) */ | |
0a4db386 | 5633 | n = ARG(scan); |
3b6647e0 | 5634 | sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n)); |
0a4db386 | 5635 | break; |
3c0563b9 DM |
5636 | |
5637 | case DEFINEP: /* (?(DEFINE)) */ | |
0a4db386 YO |
5638 | sw = 0; |
5639 | break; | |
3c0563b9 DM |
5640 | |
5641 | case IFTHEN: /* (?(cond)A|B) */ | |
1cb48e53 | 5642 | reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */ |
24d3c4a9 | 5643 | if (sw) |
c277df42 IZ |
5644 | next = NEXTOPER(NEXTOPER(scan)); |
5645 | else { | |
5646 | next = scan + ARG(scan); | |
5647 | if (OP(next) == IFTHEN) /* Fake one. */ | |
5648 | next = NEXTOPER(NEXTOPER(next)); | |
5649 | } | |
5650 | break; | |
3c0563b9 DM |
5651 | |
5652 | case LOGICAL: /* modifier for EVAL and IFMATCH */ | |
24d3c4a9 | 5653 | logical = scan->flags; |
c277df42 | 5654 | break; |
c476f425 | 5655 | |
2ab05381 | 5656 | /******************************************************************* |
2ab05381 | 5657 | |
c476f425 DM |
5658 | The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/ |
5659 | pattern, where A and B are subpatterns. (For simple A, CURLYM or | |
5660 | STAR/PLUS/CURLY/CURLYN are used instead.) | |
2ab05381 | 5661 | |
c476f425 | 5662 | A*B is compiled as <CURLYX><A><WHILEM><B> |
2ab05381 | 5663 | |
c476f425 DM |
5664 | On entry to the subpattern, CURLYX is called. This pushes a CURLYX |
5665 | state, which contains the current count, initialised to -1. It also sets | |
5666 | cur_curlyx to point to this state, with any previous value saved in the | |
5667 | state block. | |
2ab05381 | 5668 | |
c476f425 DM |
5669 | CURLYX then jumps straight to the WHILEM op, rather than executing A, |
5670 | since the pattern may possibly match zero times (i.e. it's a while {} loop | |
5671 | rather than a do {} while loop). | |
2ab05381 | 5672 | |
c476f425 DM |
5673 | Each entry to WHILEM represents a successful match of A. The count in the |
5674 | CURLYX block is incremented, another WHILEM state is pushed, and execution | |
5675 | passes to A or B depending on greediness and the current count. | |
2ab05381 | 5676 | |
c476f425 DM |
5677 | For example, if matching against the string a1a2a3b (where the aN are |
5678 | substrings that match /A/), then the match progresses as follows: (the | |
5679 | pushed states are interspersed with the bits of strings matched so far): | |
2ab05381 | 5680 | |
c476f425 DM |
5681 | <CURLYX cnt=-1> |
5682 | <CURLYX cnt=0><WHILEM> | |
5683 | <CURLYX cnt=1><WHILEM> a1 <WHILEM> | |
5684 | <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM> | |
5685 | <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> | |
5686 | <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b | |
2ab05381 | 5687 | |
c476f425 DM |
5688 | (Contrast this with something like CURLYM, which maintains only a single |
5689 | backtrack state: | |
2ab05381 | 5690 | |
c476f425 DM |
5691 | <CURLYM cnt=0> a1 |
5692 | a1 <CURLYM cnt=1> a2 | |
5693 | a1 a2 <CURLYM cnt=2> a3 | |
5694 | a1 a2 a3 <CURLYM cnt=3> b | |
5695 | ) | |
2ab05381 | 5696 | |
c476f425 DM |
5697 | Each WHILEM state block marks a point to backtrack to upon partial failure |
5698 | of A or B, and also contains some minor state data related to that | |
5699 | iteration. The CURLYX block, pointed to by cur_curlyx, contains the | |
5700 | overall state, such as the count, and pointers to the A and B ops. | |
2ab05381 | 5701 | |
c476f425 DM |
5702 | This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx |
5703 | must always point to the *current* CURLYX block, the rules are: | |
2ab05381 | 5704 | |
c476f425 DM |
5705 | When executing CURLYX, save the old cur_curlyx in the CURLYX state block, |
5706 | and set cur_curlyx to point the new block. | |
2ab05381 | 5707 | |
c476f425 DM |
5708 | When popping the CURLYX block after a successful or unsuccessful match, |
5709 | restore the previous cur_curlyx. | |
2ab05381 | 5710 | |
c476f425 DM |
5711 | When WHILEM is about to execute B, save the current cur_curlyx, and set it |
5712 | to the outer one saved in the CURLYX block. | |
2ab05381 | 5713 | |
c476f425 DM |
5714 | When popping the WHILEM block after a successful or unsuccessful B match, |
5715 | restore the previous cur_curlyx. | |
2ab05381 | 5716 | |
c476f425 DM |
5717 | Here's an example for the pattern (AI* BI)*BO |
5718 | I and O refer to inner and outer, C and W refer to CURLYX and WHILEM: | |
2ab05381 | 5719 | |
c476f425 DM |
5720 | cur_ |
5721 | curlyx backtrack stack | |
5722 | ------ --------------- | |
5723 | NULL | |
5724 | CO <CO prev=NULL> <WO> | |
5725 | CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai | |
5726 | CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi | |
5727 | NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo | |
2ab05381 | 5728 | |
c476f425 DM |
5729 | At this point the pattern succeeds, and we work back down the stack to |
5730 | clean up, restoring as we go: | |
95b24440 | 5731 | |
c476f425 DM |
5732 | CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi |
5733 | CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai | |
5734 | CO <CO prev=NULL> <WO> | |
5735 | NULL | |
a0374537 | 5736 | |
c476f425 DM |
5737 | *******************************************************************/ |
5738 | ||
5739 | #define ST st->u.curlyx | |
5740 | ||
5741 | case CURLYX: /* start of /A*B/ (for complex A) */ | |
5742 | { | |
5743 | /* No need to save/restore up to this paren */ | |
5744 | I32 parenfloor = scan->flags; | |
5745 | ||
5746 | assert(next); /* keep Coverity happy */ | |
5747 | if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */ | |
5748 | next += ARG(next); | |
5749 | ||
5750 | /* XXXX Probably it is better to teach regpush to support | |
92da3157 | 5751 | parenfloor > maxopenparen ... */ |
b93070ed DM |
5752 | if (parenfloor > (I32)rex->lastparen) |
5753 | parenfloor = rex->lastparen; /* Pessimization... */ | |
c476f425 DM |
5754 | |
5755 | ST.prev_curlyx= cur_curlyx; | |
5756 | cur_curlyx = st; | |
5757 | ST.cp = PL_savestack_ix; | |
5758 | ||
5759 | /* these fields contain the state of the current curly. | |
5760 | * they are accessed by subsequent WHILEMs */ | |
5761 | ST.parenfloor = parenfloor; | |
d02d6d97 | 5762 | ST.me = scan; |
c476f425 | 5763 | ST.B = next; |
24d3c4a9 DM |
5764 | ST.minmod = minmod; |
5765 | minmod = 0; | |
c476f425 DM |
5766 | ST.count = -1; /* this will be updated by WHILEM */ |
5767 | ST.lastloc = NULL; /* this will be updated by WHILEM */ | |
5768 | ||
4d5016e5 | 5769 | PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput); |
a74ff37d | 5770 | /* NOTREACHED */ |
e5964223 | 5771 | NOT_REACHED; |
c476f425 | 5772 | } |
a0d0e21e | 5773 | |
c476f425 | 5774 | case CURLYX_end: /* just finished matching all of A*B */ |
c476f425 DM |
5775 | cur_curlyx = ST.prev_curlyx; |
5776 | sayYES; | |
a74ff37d | 5777 | /* NOTREACHED */ |
e5964223 | 5778 | NOT_REACHED; |
a0d0e21e | 5779 | |
c476f425 DM |
5780 | case CURLYX_end_fail: /* just failed to match all of A*B */ |
5781 | regcpblow(ST.cp); | |
5782 | cur_curlyx = ST.prev_curlyx; | |
5783 | sayNO; | |
a74ff37d | 5784 | /* NOTREACHED */ |
e5964223 | 5785 | NOT_REACHED; |
4633a7c4 | 5786 | |
a0d0e21e | 5787 | |
c476f425 DM |
5788 | #undef ST |
5789 | #define ST st->u.whilem | |
5790 | ||
5791 | case WHILEM: /* just matched an A in /A*B/ (for complex A) */ | |
5792 | { | |
5793 | /* see the discussion above about CURLYX/WHILEM */ | |
c476f425 | 5794 | I32 n; |
1a522d5d JH |
5795 | int min, max; |
5796 | regnode *A; | |
d02d6d97 | 5797 | |
c476f425 | 5798 | assert(cur_curlyx); /* keep Coverity happy */ |
1a522d5d JH |
5799 | |
5800 | min = ARG1(cur_curlyx->u.curlyx.me); | |
5801 | max = ARG2(cur_curlyx->u.curlyx.me); | |
5802 | A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS; | |
c476f425 DM |
5803 | n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */ |
5804 | ST.save_lastloc = cur_curlyx->u.curlyx.lastloc; | |
5805 | ST.cache_offset = 0; | |
5806 | ST.cache_mask = 0; | |
5807 | ||
c476f425 DM |
5808 | |
5809 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, | |
d02d6d97 DM |
5810 | "%*s whilem: matched %ld out of %d..%d\n", |
5811 | REPORT_CODE_OFF+depth*2, "", (long)n, min, max) | |
c476f425 | 5812 | ); |
a0d0e21e | 5813 | |
c476f425 | 5814 | /* First just match a string of min A's. */ |
a0d0e21e | 5815 | |
d02d6d97 | 5816 | if (n < min) { |
92da3157 DM |
5817 | ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, |
5818 | maxopenparen); | |
c476f425 | 5819 | cur_curlyx->u.curlyx.lastloc = locinput; |
92e82afa YO |
5820 | REGCP_SET(ST.lastcp); |
5821 | ||
4d5016e5 | 5822 | PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput); |
a74ff37d | 5823 | /* NOTREACHED */ |
e5964223 | 5824 | NOT_REACHED; |
c476f425 DM |
5825 | } |
5826 | ||
5827 | /* If degenerate A matches "", assume A done. */ | |
5828 | ||
5829 | if (locinput == cur_curlyx->u.curlyx.lastloc) { | |
5830 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, | |
5831 | "%*s whilem: empty match detected, trying continuation...\n", | |
5832 | REPORT_CODE_OFF+depth*2, "") | |
5833 | ); | |
5834 | goto do_whilem_B_max; | |
5835 | } | |
5836 | ||
1cb95af7 DM |
5837 | /* super-linear cache processing. |
5838 | * | |
5839 | * The idea here is that for certain types of CURLYX/WHILEM - | |
5840 | * principally those whose upper bound is infinity (and | |
5841 | * excluding regexes that have things like \1 and other very | |
5842 | * non-regular expresssiony things), then if a pattern like | |
5843 | * /....A*.../ fails and we backtrack to the WHILEM, then we | |
5844 | * make a note that this particular WHILEM op was at string | |
5845 | * position 47 (say) when the rest of pattern failed. Then, if | |
5846 | * we ever find ourselves back at that WHILEM, and at string | |
5847 | * position 47 again, we can just fail immediately rather than | |
5848 | * running the rest of the pattern again. | |
5849 | * | |
5850 | * This is very handy when patterns start to go | |
5851 | * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up | |
5852 | * with a combinatorial explosion of backtracking. | |
5853 | * | |
5854 | * The cache is implemented as a bit array, with one bit per | |
5855 | * string byte position per WHILEM op (up to 16) - so its | |
5856 | * between 0.25 and 2x the string size. | |
5857 | * | |
5858 | * To avoid allocating a poscache buffer every time, we do an | |
5859 | * initially countdown; only after we have executed a WHILEM | |
5860 | * op (string-length x #WHILEMs) times do we allocate the | |
5861 | * cache. | |
5862 | * | |
5863 | * The top 4 bits of scan->flags byte say how many different | |
5864 | * relevant CURLLYX/WHILEM op pairs there are, while the | |
5865 | * bottom 4-bits is the identifying index number of this | |
5866 | * WHILEM. | |
5867 | */ | |
c476f425 DM |
5868 | |
5869 | if (scan->flags) { | |
a0d0e21e | 5870 | |
1cb48e53 | 5871 | if (!reginfo->poscache_maxiter) { |
c476f425 DM |
5872 | /* start the countdown: Postpone detection until we |
5873 | * know the match is not *that* much linear. */ | |
1cb48e53 | 5874 | reginfo->poscache_maxiter |
9d9163fb DM |
5875 | = (reginfo->strend - reginfo->strbeg + 1) |
5876 | * (scan->flags>>4); | |
66bf836d | 5877 | /* possible overflow for long strings and many CURLYX's */ |
1cb48e53 DM |
5878 | if (reginfo->poscache_maxiter < 0) |
5879 | reginfo->poscache_maxiter = I32_MAX; | |
5880 | reginfo->poscache_iter = reginfo->poscache_maxiter; | |
2c2d71f5 | 5881 | } |
c476f425 | 5882 | |
1cb48e53 | 5883 | if (reginfo->poscache_iter-- == 0) { |
c476f425 | 5884 | /* initialise cache */ |
ea3daa5d | 5885 | const SSize_t size = (reginfo->poscache_maxiter + 7)/8; |
2ac8ff4b DM |
5886 | regmatch_info_aux *const aux = reginfo->info_aux; |
5887 | if (aux->poscache) { | |
ea3daa5d | 5888 | if ((SSize_t)reginfo->poscache_size < size) { |
2ac8ff4b DM |
5889 | Renew(aux->poscache, size, char); |
5890 | reginfo->poscache_size = size; | |
2c2d71f5 | 5891 | } |
2ac8ff4b | 5892 | Zero(aux->poscache, size, char); |
2c2d71f5 JH |
5893 | } |
5894 | else { | |
2ac8ff4b DM |
5895 | reginfo->poscache_size = size; |
5896 | Newxz(aux->poscache, size, char); | |
2c2d71f5 | 5897 | } |
c476f425 DM |
5898 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
5899 | "%swhilem: Detected a super-linear match, switching on caching%s...\n", | |
5900 | PL_colors[4], PL_colors[5]) | |
5901 | ); | |
2c2d71f5 | 5902 | } |
c476f425 | 5903 | |
1cb48e53 | 5904 | if (reginfo->poscache_iter < 0) { |
c476f425 | 5905 | /* have we already failed at this position? */ |
ea3daa5d | 5906 | SSize_t offset, mask; |
338e600a DM |
5907 | |
5908 | reginfo->poscache_iter = -1; /* stop eventual underflow */ | |
c476f425 | 5909 | offset = (scan->flags & 0xf) - 1 |
9d9163fb DM |
5910 | + (locinput - reginfo->strbeg) |
5911 | * (scan->flags>>4); | |
c476f425 DM |
5912 | mask = 1 << (offset % 8); |
5913 | offset /= 8; | |
2ac8ff4b | 5914 | if (reginfo->info_aux->poscache[offset] & mask) { |
c476f425 DM |
5915 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
5916 | "%*s whilem: (cache) already tried at this position...\n", | |
5917 | REPORT_CODE_OFF+depth*2, "") | |
2c2d71f5 | 5918 | ); |
3298f257 | 5919 | sayNO; /* cache records failure */ |
2c2d71f5 | 5920 | } |
c476f425 DM |
5921 | ST.cache_offset = offset; |
5922 | ST.cache_mask = mask; | |
2c2d71f5 | 5923 | } |
c476f425 | 5924 | } |
2c2d71f5 | 5925 | |
c476f425 | 5926 | /* Prefer B over A for minimal matching. */ |
a687059c | 5927 | |
c476f425 DM |
5928 | if (cur_curlyx->u.curlyx.minmod) { |
5929 | ST.save_curlyx = cur_curlyx; | |
5930 | cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; | |
92da3157 DM |
5931 | ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor, |
5932 | maxopenparen); | |
c476f425 | 5933 | REGCP_SET(ST.lastcp); |
4d5016e5 DM |
5934 | PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B, |
5935 | locinput); | |
a74ff37d | 5936 | /* NOTREACHED */ |
e5964223 | 5937 | NOT_REACHED; |
c476f425 | 5938 | } |
a0d0e21e | 5939 | |
c476f425 DM |
5940 | /* Prefer A over B for maximal matching. */ |
5941 | ||
d02d6d97 | 5942 | if (n < max) { /* More greed allowed? */ |
92da3157 DM |
5943 | ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, |
5944 | maxopenparen); | |
c476f425 DM |
5945 | cur_curlyx->u.curlyx.lastloc = locinput; |
5946 | REGCP_SET(ST.lastcp); | |
4d5016e5 | 5947 | PUSH_STATE_GOTO(WHILEM_A_max, A, locinput); |
a74ff37d | 5948 | /* NOTREACHED */ |
e5964223 | 5949 | NOT_REACHED; |
c476f425 DM |
5950 | } |
5951 | goto do_whilem_B_max; | |
5952 | } | |
a74ff37d | 5953 | /* NOTREACHED */ |
e5964223 | 5954 | NOT_REACHED; |
c476f425 DM |
5955 | |
5956 | case WHILEM_B_min: /* just matched B in a minimal match */ | |
5957 | case WHILEM_B_max: /* just matched B in a maximal match */ | |
5958 | cur_curlyx = ST.save_curlyx; | |
5959 | sayYES; | |
a74ff37d | 5960 | /* NOTREACHED */ |
e5964223 | 5961 | NOT_REACHED; |
c476f425 DM |
5962 | |
5963 | case WHILEM_B_max_fail: /* just failed to match B in a maximal match */ | |
5964 | cur_curlyx = ST.save_curlyx; | |
5965 | cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; | |
5966 | cur_curlyx->u.curlyx.count--; | |
5967 | CACHEsayNO; | |
a74ff37d | 5968 | /* NOTREACHED */ |
e5964223 | 5969 | NOT_REACHED; |
c476f425 DM |
5970 | |
5971 | case WHILEM_A_min_fail: /* just failed to match A in a minimal match */ | |
924ba076 | 5972 | /* FALLTHROUGH */ |
c476f425 | 5973 | case WHILEM_A_pre_fail: /* just failed to match even minimal A */ |
92e82afa | 5974 | REGCP_UNWIND(ST.lastcp); |
92da3157 | 5975 | regcppop(rex, &maxopenparen); |
c476f425 DM |
5976 | cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; |
5977 | cur_curlyx->u.curlyx.count--; | |
5978 | CACHEsayNO; | |
a74ff37d | 5979 | /* NOTREACHED */ |
e5964223 | 5980 | NOT_REACHED; |
c476f425 DM |
5981 | |
5982 | case WHILEM_A_max_fail: /* just failed to match A in a maximal match */ | |
5983 | REGCP_UNWIND(ST.lastcp); | |
92da3157 | 5984 | regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */ |
c476f425 DM |
5985 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
5986 | "%*s whilem: failed, trying continuation...\n", | |
5987 | REPORT_CODE_OFF+depth*2, "") | |
5988 | ); | |
5989 | do_whilem_B_max: | |
5990 | if (cur_curlyx->u.curlyx.count >= REG_INFTY | |
5991 | && ckWARN(WARN_REGEXP) | |
39819bd9 | 5992 | && !reginfo->warned) |
c476f425 | 5993 | { |
39819bd9 | 5994 | reginfo->warned = TRUE; |
dcbac5bb FC |
5995 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), |
5996 | "Complex regular subexpression recursion limit (%d) " | |
5997 | "exceeded", | |
c476f425 DM |
5998 | REG_INFTY - 1); |
5999 | } | |
6000 | ||
6001 | /* now try B */ | |
6002 | ST.save_curlyx = cur_curlyx; | |
6003 | cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; | |
4d5016e5 DM |
6004 | PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B, |
6005 | locinput); | |
a74ff37d | 6006 | /* NOTREACHED */ |
e5964223 | 6007 | NOT_REACHED; |
c476f425 DM |
6008 | |
6009 | case WHILEM_B_min_fail: /* just failed to match B in a minimal match */ | |
6010 | cur_curlyx = ST.save_curlyx; | |
6011 | REGCP_UNWIND(ST.lastcp); | |
92da3157 | 6012 | regcppop(rex, &maxopenparen); |
c476f425 | 6013 | |
d02d6d97 | 6014 | if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) { |
c476f425 DM |
6015 | /* Maximum greed exceeded */ |
6016 | if (cur_curlyx->u.curlyx.count >= REG_INFTY | |
6017 | && ckWARN(WARN_REGEXP) | |
39819bd9 | 6018 | && !reginfo->warned) |
c476f425 | 6019 | { |
39819bd9 | 6020 | reginfo->warned = TRUE; |
c476f425 | 6021 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), |
dcbac5bb FC |
6022 | "Complex regular subexpression recursion " |
6023 | "limit (%d) exceeded", | |
c476f425 | 6024 | REG_INFTY - 1); |
a0d0e21e | 6025 | } |
c476f425 | 6026 | cur_curlyx->u.curlyx.count--; |
3ab3c9b4 | 6027 | CACHEsayNO; |
a0d0e21e | 6028 | } |
c476f425 DM |
6029 | |
6030 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, | |
6031 | "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "") | |
6032 | ); | |
6033 | /* Try grabbing another A and see if it helps. */ | |
c476f425 | 6034 | cur_curlyx->u.curlyx.lastloc = locinput; |
92da3157 DM |
6035 | ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, |
6036 | maxopenparen); | |
c476f425 | 6037 | REGCP_SET(ST.lastcp); |
d02d6d97 | 6038 | PUSH_STATE_GOTO(WHILEM_A_min, |
4d5016e5 DM |
6039 | /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS, |
6040 | locinput); | |
a74ff37d | 6041 | /* NOTREACHED */ |
e5964223 | 6042 | NOT_REACHED; |
40a82448 DM |
6043 | |
6044 | #undef ST | |
6045 | #define ST st->u.branch | |
6046 | ||
6047 | case BRANCHJ: /* /(...|A|...)/ with long next pointer */ | |
c277df42 IZ |
6048 | next = scan + ARG(scan); |
6049 | if (next == scan) | |
6050 | next = NULL; | |
40a82448 | 6051 | scan = NEXTOPER(scan); |
924ba076 | 6052 | /* FALLTHROUGH */ |
c277df42 | 6053 | |
40a82448 DM |
6054 | case BRANCH: /* /(...|A|...)/ */ |
6055 | scan = NEXTOPER(scan); /* scan now points to inner node */ | |
b93070ed | 6056 | ST.lastparen = rex->lastparen; |
f6033a9d | 6057 | ST.lastcloseparen = rex->lastcloseparen; |
40a82448 DM |
6058 | ST.next_branch = next; |
6059 | REGCP_SET(ST.cp); | |
02db2b7b | 6060 | |
40a82448 | 6061 | /* Now go into the branch */ |
5d458dd8 | 6062 | if (has_cutgroup) { |
4d5016e5 | 6063 | PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput); |
5d458dd8 | 6064 | } else { |
4d5016e5 | 6065 | PUSH_STATE_GOTO(BRANCH_next, scan, locinput); |
5d458dd8 | 6066 | } |
a74ff37d | 6067 | /* NOTREACHED */ |
e5964223 | 6068 | NOT_REACHED; |
3c0563b9 DM |
6069 | |
6070 | case CUTGROUP: /* /(*THEN)/ */ | |
5d458dd8 | 6071 | sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL : |
ad64d0ec | 6072 | MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
4d5016e5 | 6073 | PUSH_STATE_GOTO(CUTGROUP_next, next, locinput); |
a74ff37d | 6074 | /* NOTREACHED */ |
e5964223 | 6075 | NOT_REACHED; |
3c0563b9 | 6076 | |
5d458dd8 YO |
6077 | case CUTGROUP_next_fail: |
6078 | do_cutgroup = 1; | |
6079 | no_final = 1; | |
6080 | if (st->u.mark.mark_name) | |
6081 | sv_commit = st->u.mark.mark_name; | |
6082 | sayNO; | |
a74ff37d | 6083 | /* NOTREACHED */ |
e5964223 | 6084 | NOT_REACHED; |
3c0563b9 | 6085 | |
5d458dd8 YO |
6086 | case BRANCH_next: |
6087 | sayYES; | |
a74ff37d | 6088 | /* NOTREACHED */ |
e5964223 | 6089 | NOT_REACHED; |
3c0563b9 | 6090 | |
40a82448 | 6091 | case BRANCH_next_fail: /* that branch failed; try the next, if any */ |
5d458dd8 YO |
6092 | if (do_cutgroup) { |
6093 | do_cutgroup = 0; | |
6094 | no_final = 0; | |
6095 | } | |
40a82448 | 6096 | REGCP_UNWIND(ST.cp); |
a8d1f4b4 | 6097 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
40a82448 DM |
6098 | scan = ST.next_branch; |
6099 | /* no more branches? */ | |
5d458dd8 YO |
6100 | if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) { |
6101 | DEBUG_EXECUTE_r({ | |
6102 | PerlIO_printf( Perl_debug_log, | |
6103 | "%*s %sBRANCH failed...%s\n", | |
6104 | REPORT_CODE_OFF+depth*2, "", | |
6105 | PL_colors[4], | |
6106 | PL_colors[5] ); | |
6107 | }); | |
6108 | sayNO_SILENT; | |
6109 | } | |
40a82448 | 6110 | continue; /* execute next BRANCH[J] op */ |
a74ff37d | 6111 | /* NOTREACHED */ |
40a82448 | 6112 | |
3c0563b9 | 6113 | case MINMOD: /* next op will be non-greedy, e.g. A*? */ |
24d3c4a9 | 6114 | minmod = 1; |
a0d0e21e | 6115 | break; |
40a82448 DM |
6116 | |
6117 | #undef ST | |
6118 | #define ST st->u.curlym | |
6119 | ||
6120 | case CURLYM: /* /A{m,n}B/ where A is fixed-length */ | |
6121 | ||
6122 | /* This is an optimisation of CURLYX that enables us to push | |
84d2fa14 | 6123 | * only a single backtracking state, no matter how many matches |
40a82448 DM |
6124 | * there are in {m,n}. It relies on the pattern being constant |
6125 | * length, with no parens to influence future backrefs | |
6126 | */ | |
6127 | ||
6128 | ST.me = scan; | |
dc45a647 | 6129 | scan = NEXTOPER(scan) + NODE_STEP_REGNODE; |
40a82448 | 6130 | |
f6033a9d DM |
6131 | ST.lastparen = rex->lastparen; |
6132 | ST.lastcloseparen = rex->lastcloseparen; | |
6133 | ||
40a82448 DM |
6134 | /* if paren positive, emulate an OPEN/CLOSE around A */ |
6135 | if (ST.me->flags) { | |
3b6647e0 | 6136 | U32 paren = ST.me->flags; |
92da3157 DM |
6137 | if (paren > maxopenparen) |
6138 | maxopenparen = paren; | |
c277df42 | 6139 | scan += NEXT_OFF(scan); /* Skip former OPEN. */ |
6407bf3b | 6140 | } |
40a82448 DM |
6141 | ST.A = scan; |
6142 | ST.B = next; | |
6143 | ST.alen = 0; | |
6144 | ST.count = 0; | |
24d3c4a9 DM |
6145 | ST.minmod = minmod; |
6146 | minmod = 0; | |
40a82448 DM |
6147 | ST.c1 = CHRTEST_UNINIT; |
6148 | REGCP_SET(ST.cp); | |
6407bf3b | 6149 | |
40a82448 DM |
6150 | if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */ |
6151 | goto curlym_do_B; | |
6152 | ||
6153 | curlym_do_A: /* execute the A in /A{m,n}B/ */ | |
4d5016e5 | 6154 | PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */ |
a74ff37d | 6155 | /* NOTREACHED */ |
e5964223 | 6156 | NOT_REACHED; |
5f80c4cf | 6157 | |
40a82448 | 6158 | case CURLYM_A: /* we've just matched an A */ |
40a82448 DM |
6159 | ST.count++; |
6160 | /* after first match, determine A's length: u.curlym.alen */ | |
6161 | if (ST.count == 1) { | |
ba44c216 | 6162 | if (reginfo->is_utf8_target) { |
c07e9d7b DM |
6163 | char *s = st->locinput; |
6164 | while (s < locinput) { | |
40a82448 DM |
6165 | ST.alen++; |
6166 | s += UTF8SKIP(s); | |
6167 | } | |
6168 | } | |
6169 | else { | |
c07e9d7b | 6170 | ST.alen = locinput - st->locinput; |
40a82448 DM |
6171 | } |
6172 | if (ST.alen == 0) | |
6173 | ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me); | |
6174 | } | |
0cadcf80 DM |
6175 | DEBUG_EXECUTE_r( |
6176 | PerlIO_printf(Perl_debug_log, | |
40a82448 | 6177 | "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n", |
5bc10b2c | 6178 | (int)(REPORT_CODE_OFF+(depth*2)), "", |
40a82448 | 6179 | (IV) ST.count, (IV)ST.alen) |
0cadcf80 DM |
6180 | ); |
6181 | ||
0a4db386 | 6182 | if (cur_eval && cur_eval->u.eval.close_paren && |
24b23f37 | 6183 | cur_eval->u.eval.close_paren == (U32)ST.me->flags) |
0a4db386 YO |
6184 | goto fake_end; |
6185 | ||
c966426a DM |
6186 | { |
6187 | I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)); | |
6188 | if ( max == REG_INFTY || ST.count < max ) | |
6189 | goto curlym_do_A; /* try to match another A */ | |
6190 | } | |
40a82448 | 6191 | goto curlym_do_B; /* try to match B */ |
5f80c4cf | 6192 | |
40a82448 DM |
6193 | case CURLYM_A_fail: /* just failed to match an A */ |
6194 | REGCP_UNWIND(ST.cp); | |
0a4db386 YO |
6195 | |
6196 | if (ST.minmod || ST.count < ARG1(ST.me) /* min*/ | |
6197 | || (cur_eval && cur_eval->u.eval.close_paren && | |
24b23f37 | 6198 | cur_eval->u.eval.close_paren == (U32)ST.me->flags)) |
40a82448 | 6199 | sayNO; |
0cadcf80 | 6200 | |
40a82448 | 6201 | curlym_do_B: /* execute the B in /A{m,n}B/ */ |
40a82448 DM |
6202 | if (ST.c1 == CHRTEST_UNINIT) { |
6203 | /* calculate c1 and c2 for possible match of 1st char | |
6204 | * following curly */ | |
6205 | ST.c1 = ST.c2 = CHRTEST_VOID; | |
d20a21f4 | 6206 | assert(ST.B); |
40a82448 DM |
6207 | if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) { |
6208 | regnode *text_node = ST.B; | |
6209 | if (! HAS_TEXT(text_node)) | |
6210 | FIND_NEXT_IMPT(text_node); | |
ee9b8eae YO |
6211 | /* this used to be |
6212 | ||
6213 | (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT) | |
6214 | ||
6215 | But the former is redundant in light of the latter. | |
6216 | ||
6217 | if this changes back then the macro for | |
6218 | IS_TEXT and friends need to change. | |
6219 | */ | |
c74f6de9 | 6220 | if (PL_regkind[OP(text_node)] == EXACT) { |
79a2a0e8 | 6221 | if (! S_setup_EXACTISH_ST_c1_c2(aTHX_ |
984e6dd1 | 6222 | text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8, |
aed7b151 | 6223 | reginfo)) |
c74f6de9 KW |
6224 | { |
6225 | sayNO; | |
6226 | } | |
c277df42 | 6227 | } |
c277df42 | 6228 | } |
40a82448 DM |
6229 | } |
6230 | ||
6231 | DEBUG_EXECUTE_r( | |
6232 | PerlIO_printf(Perl_debug_log, | |
6233 | "%*s CURLYM trying tail with matches=%"IVdf"...\n", | |
5bc10b2c | 6234 | (int)(REPORT_CODE_OFF+(depth*2)), |
40a82448 DM |
6235 | "", (IV)ST.count) |
6236 | ); | |
c74f6de9 | 6237 | if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) { |
79a2a0e8 KW |
6238 | if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) { |
6239 | if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)) | |
6240 | && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput))) | |
6241 | { | |
6242 | /* simulate B failing */ | |
6243 | DEBUG_OPTIMISE_r( | |
6244 | PerlIO_printf(Perl_debug_log, | |
33daa3a5 | 6245 | "%*s CURLYM Fast bail next target=0x%"UVXf" c1=0x%"UVXf" c2=0x%"UVXf"\n", |
79a2a0e8 KW |
6246 | (int)(REPORT_CODE_OFF+(depth*2)),"", |
6247 | valid_utf8_to_uvchr((U8 *) locinput, NULL), | |
6248 | valid_utf8_to_uvchr(ST.c1_utf8, NULL), | |
6249 | valid_utf8_to_uvchr(ST.c2_utf8, NULL)) | |
6250 | ); | |
6251 | state_num = CURLYM_B_fail; | |
6252 | goto reenter_switch; | |
6253 | } | |
6254 | } | |
6255 | else if (nextchr != ST.c1 && nextchr != ST.c2) { | |
5400f398 KW |
6256 | /* simulate B failing */ |
6257 | DEBUG_OPTIMISE_r( | |
6258 | PerlIO_printf(Perl_debug_log, | |
33daa3a5 | 6259 | "%*s CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n", |
5400f398 | 6260 | (int)(REPORT_CODE_OFF+(depth*2)),"", |
79a2a0e8 KW |
6261 | (int) nextchr, ST.c1, ST.c2) |
6262 | ); | |
5400f398 KW |
6263 | state_num = CURLYM_B_fail; |
6264 | goto reenter_switch; | |
6265 | } | |
c74f6de9 | 6266 | } |
40a82448 DM |
6267 | |
6268 | if (ST.me->flags) { | |
f6033a9d | 6269 | /* emulate CLOSE: mark current A as captured */ |
40a82448 DM |
6270 | I32 paren = ST.me->flags; |
6271 | if (ST.count) { | |
b93070ed | 6272 | rex->offs[paren].start |
9d9163fb DM |
6273 | = HOPc(locinput, -ST.alen) - reginfo->strbeg; |
6274 | rex->offs[paren].end = locinput - reginfo->strbeg; | |
f6033a9d DM |
6275 | if ((U32)paren > rex->lastparen) |
6276 | rex->lastparen = paren; | |
6277 | rex->lastcloseparen = paren; | |
c277df42 | 6278 | } |
40a82448 | 6279 | else |
b93070ed | 6280 | rex->offs[paren].end = -1; |
0a4db386 | 6281 | if (cur_eval && cur_eval->u.eval.close_paren && |
24b23f37 | 6282 | cur_eval->u.eval.close_paren == (U32)ST.me->flags) |
0a4db386 YO |
6283 | { |
6284 | if (ST.count) | |
6285 | goto fake_end; | |
6286 | else | |
6287 | sayNO; | |
6288 | } | |
c277df42 | 6289 | } |
0a4db386 | 6290 | |
4d5016e5 | 6291 | PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */ |
a74ff37d | 6292 | /* NOTREACHED */ |
e5964223 | 6293 | NOT_REACHED; |
40a82448 DM |
6294 | |
6295 | case CURLYM_B_fail: /* just failed to match a B */ | |
6296 | REGCP_UNWIND(ST.cp); | |
a8d1f4b4 | 6297 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
40a82448 | 6298 | if (ST.minmod) { |
84d2fa14 HS |
6299 | I32 max = ARG2(ST.me); |
6300 | if (max != REG_INFTY && ST.count == max) | |
40a82448 DM |
6301 | sayNO; |
6302 | goto curlym_do_A; /* try to match a further A */ | |
6303 | } | |
6304 | /* backtrack one A */ | |
6305 | if (ST.count == ARG1(ST.me) /* min */) | |
6306 | sayNO; | |
6307 | ST.count--; | |
7016d6eb | 6308 | SET_locinput(HOPc(locinput, -ST.alen)); |
40a82448 DM |
6309 | goto curlym_do_B; /* try to match B */ |
6310 | ||
c255a977 DM |
6311 | #undef ST |
6312 | #define ST st->u.curly | |
40a82448 | 6313 | |
c255a977 DM |
6314 | #define CURLY_SETPAREN(paren, success) \ |
6315 | if (paren) { \ | |
6316 | if (success) { \ | |
9d9163fb DM |
6317 | rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \ |
6318 | rex->offs[paren].end = locinput - reginfo->strbeg; \ | |
f6033a9d DM |
6319 | if (paren > rex->lastparen) \ |
6320 | rex->lastparen = paren; \ | |
b93070ed | 6321 | rex->lastcloseparen = paren; \ |
c255a977 | 6322 | } \ |
f6033a9d | 6323 | else { \ |
b93070ed | 6324 | rex->offs[paren].end = -1; \ |
f6033a9d DM |
6325 | rex->lastparen = ST.lastparen; \ |
6326 | rex->lastcloseparen = ST.lastcloseparen; \ | |
6327 | } \ | |
c255a977 DM |
6328 | } |
6329 | ||
b40a2c17 | 6330 | case STAR: /* /A*B/ where A is width 1 char */ |
c255a977 DM |
6331 | ST.paren = 0; |
6332 | ST.min = 0; | |
6333 | ST.max = REG_INFTY; | |
a0d0e21e LW |
6334 | scan = NEXTOPER(scan); |
6335 | goto repeat; | |
3c0563b9 | 6336 | |
b40a2c17 | 6337 | case PLUS: /* /A+B/ where A is width 1 char */ |
c255a977 DM |
6338 | ST.paren = 0; |
6339 | ST.min = 1; | |
6340 | ST.max = REG_INFTY; | |
c277df42 | 6341 | scan = NEXTOPER(scan); |
c255a977 | 6342 | goto repeat; |
3c0563b9 | 6343 | |
b40a2c17 | 6344 | case CURLYN: /* /(A){m,n}B/ where A is width 1 char */ |
5400f398 KW |
6345 | ST.paren = scan->flags; /* Which paren to set */ |
6346 | ST.lastparen = rex->lastparen; | |
f6033a9d | 6347 | ST.lastcloseparen = rex->lastcloseparen; |
92da3157 DM |
6348 | if (ST.paren > maxopenparen) |
6349 | maxopenparen = ST.paren; | |
c255a977 DM |
6350 | ST.min = ARG1(scan); /* min to match */ |
6351 | ST.max = ARG2(scan); /* max to match */ | |
0a4db386 | 6352 | if (cur_eval && cur_eval->u.eval.close_paren && |
86413ec0 | 6353 | cur_eval->u.eval.close_paren == (U32)ST.paren) { |
0a4db386 YO |
6354 | ST.min=1; |
6355 | ST.max=1; | |
6356 | } | |
c255a977 DM |
6357 | scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE); |
6358 | goto repeat; | |
3c0563b9 | 6359 | |
b40a2c17 | 6360 | case CURLY: /* /A{m,n}B/ where A is width 1 char */ |
c255a977 DM |
6361 | ST.paren = 0; |
6362 | ST.min = ARG1(scan); /* min to match */ | |
6363 | ST.max = ARG2(scan); /* max to match */ | |
6364 | scan = NEXTOPER(scan) + NODE_STEP_REGNODE; | |
c277df42 | 6365 | repeat: |
a0d0e21e LW |
6366 | /* |
6367 | * Lookahead to avoid useless match attempts | |
6368 | * when we know what character comes next. | |
c255a977 | 6369 | * |
5f80c4cf JP |
6370 | * Used to only do .*x and .*?x, but now it allows |
6371 | * for )'s, ('s and (?{ ... })'s to be in the way | |
6372 | * of the quantifier and the EXACT-like node. -- japhy | |
6373 | */ | |
6374 | ||
eb5c1be8 | 6375 | assert(ST.min <= ST.max); |
3337dfe3 KW |
6376 | if (! HAS_TEXT(next) && ! JUMPABLE(next)) { |
6377 | ST.c1 = ST.c2 = CHRTEST_VOID; | |
6378 | } | |
6379 | else { | |
5f80c4cf JP |
6380 | regnode *text_node = next; |
6381 | ||
3dab1dad YO |
6382 | if (! HAS_TEXT(text_node)) |
6383 | FIND_NEXT_IMPT(text_node); | |
5f80c4cf | 6384 | |
9e137952 | 6385 | if (! HAS_TEXT(text_node)) |
c255a977 | 6386 | ST.c1 = ST.c2 = CHRTEST_VOID; |
5f80c4cf | 6387 | else { |
ee9b8eae | 6388 | if ( PL_regkind[OP(text_node)] != EXACT ) { |
c255a977 | 6389 | ST.c1 = ST.c2 = CHRTEST_VOID; |
cca55fe3 | 6390 | } |
c74f6de9 | 6391 | else { |
ee9b8eae YO |
6392 | |
6393 | /* Currently we only get here when | |
6394 | ||
6395 | PL_rekind[OP(text_node)] == EXACT | |
6396 | ||
6397 | if this changes back then the macro for IS_TEXT and | |
6398 | friends need to change. */ | |
79a2a0e8 | 6399 | if (! S_setup_EXACTISH_ST_c1_c2(aTHX_ |
984e6dd1 | 6400 | text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8, |
aed7b151 | 6401 | reginfo)) |
c74f6de9 KW |
6402 | { |
6403 | sayNO; | |
6404 | } | |
6405 | } | |
1aa99e6b | 6406 | } |
bbce6d69 | 6407 | } |
c255a977 DM |
6408 | |
6409 | ST.A = scan; | |
6410 | ST.B = next; | |
24d3c4a9 | 6411 | if (minmod) { |
eb72505d | 6412 | char *li = locinput; |
24d3c4a9 | 6413 | minmod = 0; |
984e6dd1 | 6414 | if (ST.min && |
f9176b44 | 6415 | regrepeat(rex, &li, ST.A, reginfo, ST.min, depth) |
984e6dd1 | 6416 | < ST.min) |
4633a7c4 | 6417 | sayNO; |
7016d6eb | 6418 | SET_locinput(li); |
c255a977 | 6419 | ST.count = ST.min; |
c255a977 DM |
6420 | REGCP_SET(ST.cp); |
6421 | if (ST.c1 == CHRTEST_VOID) | |
6422 | goto curly_try_B_min; | |
6423 | ||
6424 | ST.oldloc = locinput; | |
6425 | ||
6426 | /* set ST.maxpos to the furthest point along the | |
6427 | * string that could possibly match */ | |
6428 | if (ST.max == REG_INFTY) { | |
220db18a | 6429 | ST.maxpos = reginfo->strend - 1; |
f2ed9b32 | 6430 | if (utf8_target) |
c255a977 DM |
6431 | while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos)) |
6432 | ST.maxpos--; | |
6433 | } | |
f2ed9b32 | 6434 | else if (utf8_target) { |
c255a977 DM |
6435 | int m = ST.max - ST.min; |
6436 | for (ST.maxpos = locinput; | |
220db18a | 6437 | m >0 && ST.maxpos < reginfo->strend; m--) |
c255a977 DM |
6438 | ST.maxpos += UTF8SKIP(ST.maxpos); |
6439 | } | |
6440 | else { | |
6441 | ST.maxpos = locinput + ST.max - ST.min; | |
220db18a DM |
6442 | if (ST.maxpos >= reginfo->strend) |
6443 | ST.maxpos = reginfo->strend - 1; | |
c255a977 DM |
6444 | } |
6445 | goto curly_try_B_min_known; | |
6446 | ||
6447 | } | |
6448 | else { | |
eb72505d DM |
6449 | /* avoid taking address of locinput, so it can remain |
6450 | * a register var */ | |
6451 | char *li = locinput; | |
f9176b44 | 6452 | ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth); |
c255a977 DM |
6453 | if (ST.count < ST.min) |
6454 | sayNO; | |
7016d6eb | 6455 | SET_locinput(li); |
c255a977 DM |
6456 | if ((ST.count > ST.min) |
6457 | && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL)) | |
6458 | { | |
6459 | /* A{m,n} must come at the end of the string, there's | |
6460 | * no point in backing off ... */ | |
6461 | ST.min = ST.count; | |
6462 | /* ...except that $ and \Z can match before *and* after | |
6463 | newline at the end. Consider "\n\n" =~ /\n+\Z\n/. | |
6464 | We may back off by one in this case. */ | |
eb72505d | 6465 | if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS) |
c255a977 DM |
6466 | ST.min--; |
6467 | } | |
6468 | REGCP_SET(ST.cp); | |
6469 | goto curly_try_B_max; | |
6470 | } | |
a74ff37d | 6471 | /* NOTREACHED */ |
e5964223 | 6472 | NOT_REACHED; |
c255a977 DM |
6473 | |
6474 | case CURLY_B_min_known_fail: | |
6475 | /* failed to find B in a non-greedy match where c1,c2 valid */ | |
c255a977 | 6476 | |
c255a977 | 6477 | REGCP_UNWIND(ST.cp); |
a8d1f4b4 DM |
6478 | if (ST.paren) { |
6479 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); | |
6480 | } | |
c255a977 DM |
6481 | /* Couldn't or didn't -- move forward. */ |
6482 | ST.oldloc = locinput; | |
f2ed9b32 | 6483 | if (utf8_target) |
c255a977 DM |
6484 | locinput += UTF8SKIP(locinput); |
6485 | else | |
6486 | locinput++; | |
6487 | ST.count++; | |
6488 | curly_try_B_min_known: | |
6489 | /* find the next place where 'B' could work, then call B */ | |
6490 | { | |
6491 | int n; | |
f2ed9b32 | 6492 | if (utf8_target) { |
c255a977 DM |
6493 | n = (ST.oldloc == locinput) ? 0 : 1; |
6494 | if (ST.c1 == ST.c2) { | |
c255a977 | 6495 | /* set n to utf8_distance(oldloc, locinput) */ |
79a2a0e8 KW |
6496 | while (locinput <= ST.maxpos |
6497 | && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))) | |
6498 | { | |
6499 | locinput += UTF8SKIP(locinput); | |
c255a977 DM |
6500 | n++; |
6501 | } | |
1aa99e6b IH |
6502 | } |
6503 | else { | |
c255a977 | 6504 | /* set n to utf8_distance(oldloc, locinput) */ |
79a2a0e8 KW |
6505 | while (locinput <= ST.maxpos |
6506 | && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)) | |
6507 | && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput))) | |
6508 | { | |
6509 | locinput += UTF8SKIP(locinput); | |
c255a977 | 6510 | n++; |
1aa99e6b | 6511 | } |
0fe9bf95 IZ |
6512 | } |
6513 | } | |
5400f398 | 6514 | else { /* Not utf8_target */ |
c255a977 DM |
6515 | if (ST.c1 == ST.c2) { |
6516 | while (locinput <= ST.maxpos && | |
6517 | UCHARAT(locinput) != ST.c1) | |
6518 | locinput++; | |
bbce6d69 | 6519 | } |
c255a977 DM |
6520 | else { |
6521 | while (locinput <= ST.maxpos | |
6522 | && UCHARAT(locinput) != ST.c1 | |
6523 | && UCHARAT(locinput) != ST.c2) | |
6524 | locinput++; | |
a0ed51b3 | 6525 | } |
c255a977 DM |
6526 | n = locinput - ST.oldloc; |
6527 | } | |
6528 | if (locinput > ST.maxpos) | |
6529 | sayNO; | |
c255a977 | 6530 | if (n) { |
eb72505d DM |
6531 | /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is |
6532 | * at b; check that everything between oldloc and | |
6533 | * locinput matches */ | |
6534 | char *li = ST.oldloc; | |
c255a977 | 6535 | ST.count += n; |
f9176b44 | 6536 | if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n) |
4633a7c4 | 6537 | sayNO; |
eb72505d | 6538 | assert(n == REG_INFTY || locinput == li); |
a0d0e21e | 6539 | } |
c255a977 | 6540 | CURLY_SETPAREN(ST.paren, ST.count); |
0a4db386 | 6541 | if (cur_eval && cur_eval->u.eval.close_paren && |
86413ec0 | 6542 | cur_eval->u.eval.close_paren == (U32)ST.paren) { |
0a4db386 YO |
6543 | goto fake_end; |
6544 | } | |
4d5016e5 | 6545 | PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput); |
a0d0e21e | 6546 | } |
a74ff37d | 6547 | /* NOTREACHED */ |
e5964223 | 6548 | NOT_REACHED; |
c255a977 DM |
6549 | |
6550 | case CURLY_B_min_fail: | |
6551 | /* failed to find B in a non-greedy match where c1,c2 invalid */ | |
c255a977 DM |
6552 | |
6553 | REGCP_UNWIND(ST.cp); | |
a8d1f4b4 DM |
6554 | if (ST.paren) { |
6555 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); | |
6556 | } | |
c255a977 | 6557 | /* failed -- move forward one */ |
f73aaa43 | 6558 | { |
eb72505d | 6559 | char *li = locinput; |
f9176b44 | 6560 | if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) { |
f73aaa43 DM |
6561 | sayNO; |
6562 | } | |
eb72505d | 6563 | locinput = li; |
f73aaa43 DM |
6564 | } |
6565 | { | |
c255a977 | 6566 | ST.count++; |
c255a977 DM |
6567 | if (ST.count <= ST.max || (ST.max == REG_INFTY && |
6568 | ST.count > 0)) /* count overflow ? */ | |
15272685 | 6569 | { |
c255a977 DM |
6570 | curly_try_B_min: |
6571 | CURLY_SETPAREN(ST.paren, ST.count); | |
0a4db386 | 6572 | if (cur_eval && cur_eval->u.eval.close_paren && |
86413ec0 | 6573 | cur_eval->u.eval.close_paren == (U32)ST.paren) { |
0a4db386 YO |
6574 | goto fake_end; |
6575 | } | |
4d5016e5 | 6576 | PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput); |
a0d0e21e LW |
6577 | } |
6578 | } | |
c74f6de9 | 6579 | sayNO; |
a74ff37d | 6580 | /* NOTREACHED */ |
e5964223 | 6581 | NOT_REACHED; |
c255a977 DM |
6582 | |
6583 | curly_try_B_max: | |
6584 | /* a successful greedy match: now try to match B */ | |
40d049e4 | 6585 | if (cur_eval && cur_eval->u.eval.close_paren && |
86413ec0 | 6586 | cur_eval->u.eval.close_paren == (U32)ST.paren) { |
40d049e4 YO |
6587 | goto fake_end; |
6588 | } | |
c255a977 | 6589 | { |
220db18a | 6590 | bool could_match = locinput < reginfo->strend; |
79a2a0e8 | 6591 | |
c255a977 | 6592 | /* If it could work, try it. */ |
79a2a0e8 KW |
6593 | if (ST.c1 != CHRTEST_VOID && could_match) { |
6594 | if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target) | |
6595 | { | |
6596 | could_match = memEQ(locinput, | |
6597 | ST.c1_utf8, | |
6598 | UTF8SKIP(locinput)) | |
6599 | || memEQ(locinput, | |
6600 | ST.c2_utf8, | |
6601 | UTF8SKIP(locinput)); | |
6602 | } | |
6603 | else { | |
6604 | could_match = UCHARAT(locinput) == ST.c1 | |
6605 | || UCHARAT(locinput) == ST.c2; | |
6606 | } | |
6607 | } | |
6608 | if (ST.c1 == CHRTEST_VOID || could_match) { | |
c255a977 | 6609 | CURLY_SETPAREN(ST.paren, ST.count); |
4d5016e5 | 6610 | PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput); |
a74ff37d | 6611 | /* NOTREACHED */ |
e5964223 | 6612 | NOT_REACHED; |
c255a977 DM |
6613 | } |
6614 | } | |
924ba076 | 6615 | /* FALLTHROUGH */ |
3c0563b9 | 6616 | |
c255a977 DM |
6617 | case CURLY_B_max_fail: |
6618 | /* failed to find B in a greedy match */ | |
c255a977 DM |
6619 | |
6620 | REGCP_UNWIND(ST.cp); | |
a8d1f4b4 DM |
6621 | if (ST.paren) { |
6622 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); | |
6623 | } | |
c255a977 DM |
6624 | /* back up. */ |
6625 | if (--ST.count < ST.min) | |
6626 | sayNO; | |
eb72505d | 6627 | locinput = HOPc(locinput, -1); |
c255a977 DM |
6628 | goto curly_try_B_max; |
6629 | ||
6630 | #undef ST | |
6631 | ||
3c0563b9 | 6632 | case END: /* last op of main pattern */ |
6bda09f9 | 6633 | fake_end: |
faec1544 DM |
6634 | if (cur_eval) { |
6635 | /* we've just finished A in /(??{A})B/; now continue with B */ | |
faec1544 | 6636 | |
288b8c02 | 6637 | st->u.eval.prev_rex = rex_sv; /* inner */ |
92da3157 DM |
6638 | |
6639 | /* Save *all* the positions. */ | |
6640 | st->u.eval.cp = regcppush(rex, 0, maxopenparen); | |
ec43f78b | 6641 | rex_sv = cur_eval->u.eval.prev_rex; |
aed7b151 | 6642 | is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv)); |
ec43f78b | 6643 | SET_reg_curpm(rex_sv); |
8d919b0a | 6644 | rex = ReANY(rex_sv); |
f8fc2ecf | 6645 | rexi = RXi_GET(rex); |
faec1544 | 6646 | cur_curlyx = cur_eval->u.eval.prev_curlyx; |
34a81e2b | 6647 | |
faec1544 | 6648 | REGCP_SET(st->u.eval.lastcp); |
faec1544 DM |
6649 | |
6650 | /* Restore parens of the outer rex without popping the | |
6651 | * savestack */ | |
92da3157 DM |
6652 | S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp, |
6653 | &maxopenparen); | |
faec1544 DM |
6654 | |
6655 | st->u.eval.prev_eval = cur_eval; | |
6656 | cur_eval = cur_eval->u.eval.prev_eval; | |
6657 | DEBUG_EXECUTE_r( | |
2a49f0f5 JH |
6658 | PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n", |
6659 | REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval));); | |
e7707071 YO |
6660 | if ( nochange_depth ) |
6661 | nochange_depth--; | |
6662 | ||
4d5016e5 DM |
6663 | PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B, |
6664 | locinput); /* match B */ | |
faec1544 DM |
6665 | } |
6666 | ||
3b0527fe | 6667 | if (locinput < reginfo->till) { |
a3621e74 | 6668 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
7821416a IZ |
6669 | "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n", |
6670 | PL_colors[4], | |
6d59b646 DM |
6671 | (long)(locinput - startpos), |
6672 | (long)(reginfo->till - startpos), | |
7821416a | 6673 | PL_colors[5])); |
58e23c8d | 6674 | |
262b90c4 | 6675 | sayNO_SILENT; /* Cannot match: too short. */ |
7821416a | 6676 | } |
262b90c4 | 6677 | sayYES; /* Success! */ |
dad79028 DM |
6678 | |
6679 | case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */ | |
6680 | DEBUG_EXECUTE_r( | |
6681 | PerlIO_printf(Perl_debug_log, | |
6682 | "%*s %ssubpattern success...%s\n", | |
5bc10b2c | 6683 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])); |
262b90c4 | 6684 | sayYES; /* Success! */ |
dad79028 | 6685 | |
40a82448 DM |
6686 | #undef ST |
6687 | #define ST st->u.ifmatch | |
6688 | ||
37f53970 DM |
6689 | { |
6690 | char *newstart; | |
6691 | ||
40a82448 DM |
6692 | case SUSPEND: /* (?>A) */ |
6693 | ST.wanted = 1; | |
37f53970 | 6694 | newstart = locinput; |
9041c2e3 | 6695 | goto do_ifmatch; |
dad79028 | 6696 | |
40a82448 DM |
6697 | case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */ |
6698 | ST.wanted = 0; | |
dad79028 DM |
6699 | goto ifmatch_trivial_fail_test; |
6700 | ||
40a82448 DM |
6701 | case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */ |
6702 | ST.wanted = 1; | |
dad79028 | 6703 | ifmatch_trivial_fail_test: |
a0ed51b3 | 6704 | if (scan->flags) { |
52657f30 | 6705 | char * const s = HOPBACKc(locinput, scan->flags); |
dad79028 DM |
6706 | if (!s) { |
6707 | /* trivial fail */ | |
24d3c4a9 DM |
6708 | if (logical) { |
6709 | logical = 0; | |
f2338a2e | 6710 | sw = 1 - cBOOL(ST.wanted); |
dad79028 | 6711 | } |
40a82448 | 6712 | else if (ST.wanted) |
dad79028 DM |
6713 | sayNO; |
6714 | next = scan + ARG(scan); | |
6715 | if (next == scan) | |
6716 | next = NULL; | |
6717 | break; | |
6718 | } | |
37f53970 | 6719 | newstart = s; |
a0ed51b3 LW |
6720 | } |
6721 | else | |
37f53970 | 6722 | newstart = locinput; |
a0ed51b3 | 6723 | |
c277df42 | 6724 | do_ifmatch: |
40a82448 | 6725 | ST.me = scan; |
24d3c4a9 | 6726 | ST.logical = logical; |
24d786f4 YO |
6727 | logical = 0; /* XXX: reset state of logical once it has been saved into ST */ |
6728 | ||
40a82448 | 6729 | /* execute body of (?...A) */ |
37f53970 | 6730 | PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart); |
a74ff37d | 6731 | /* NOTREACHED */ |
e5964223 | 6732 | NOT_REACHED; |
37f53970 | 6733 | } |
40a82448 DM |
6734 | |
6735 | case IFMATCH_A_fail: /* body of (?...A) failed */ | |
6736 | ST.wanted = !ST.wanted; | |
924ba076 | 6737 | /* FALLTHROUGH */ |
40a82448 DM |
6738 | |
6739 | case IFMATCH_A: /* body of (?...A) succeeded */ | |
24d3c4a9 | 6740 | if (ST.logical) { |
f2338a2e | 6741 | sw = cBOOL(ST.wanted); |
40a82448 DM |
6742 | } |
6743 | else if (!ST.wanted) | |
6744 | sayNO; | |
6745 | ||
37f53970 DM |
6746 | if (OP(ST.me) != SUSPEND) { |
6747 | /* restore old position except for (?>...) */ | |
6748 | locinput = st->locinput; | |
40a82448 DM |
6749 | } |
6750 | scan = ST.me + ARG(ST.me); | |
6751 | if (scan == ST.me) | |
6752 | scan = NULL; | |
6753 | continue; /* execute B */ | |
6754 | ||
6755 | #undef ST | |
dad79028 | 6756 | |
3c0563b9 DM |
6757 | case LONGJMP: /* alternative with many branches compiles to |
6758 | * (BRANCHJ; EXACT ...; LONGJMP ) x N */ | |
c277df42 IZ |
6759 | next = scan + ARG(scan); |
6760 | if (next == scan) | |
6761 | next = NULL; | |
a0d0e21e | 6762 | break; |
3c0563b9 DM |
6763 | |
6764 | case COMMIT: /* (*COMMIT) */ | |
220db18a | 6765 | reginfo->cutpoint = reginfo->strend; |
e2e6a0f1 | 6766 | /* FALLTHROUGH */ |
3c0563b9 DM |
6767 | |
6768 | case PRUNE: /* (*PRUNE) */ | |
e2e6a0f1 | 6769 | if (!scan->flags) |
ad64d0ec | 6770 | sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
4d5016e5 | 6771 | PUSH_STATE_GOTO(COMMIT_next, next, locinput); |
a74ff37d | 6772 | /* NOTREACHED */ |
e5964223 | 6773 | NOT_REACHED; |
3c0563b9 | 6774 | |
54612592 YO |
6775 | case COMMIT_next_fail: |
6776 | no_final = 1; | |
6777 | /* FALLTHROUGH */ | |
3c0563b9 DM |
6778 | |
6779 | case OPFAIL: /* (*FAIL) */ | |
7f69552c | 6780 | sayNO; |
a74ff37d | 6781 | /* NOTREACHED */ |
e5964223 | 6782 | NOT_REACHED; |
e2e6a0f1 YO |
6783 | |
6784 | #define ST st->u.mark | |
3c0563b9 | 6785 | case MARKPOINT: /* (*MARK:foo) */ |
e2e6a0f1 | 6786 | ST.prev_mark = mark_state; |
5d458dd8 | 6787 | ST.mark_name = sv_commit = sv_yes_mark |
ad64d0ec | 6788 | = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
e2e6a0f1 | 6789 | mark_state = st; |
4d5016e5 DM |
6790 | ST.mark_loc = locinput; |
6791 | PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput); | |
a74ff37d | 6792 | /* NOTREACHED */ |
e5964223 | 6793 | NOT_REACHED; |
3c0563b9 | 6794 | |
e2e6a0f1 YO |
6795 | case MARKPOINT_next: |
6796 | mark_state = ST.prev_mark; | |
6797 | sayYES; | |
a74ff37d | 6798 | /* NOTREACHED */ |
e5964223 | 6799 | NOT_REACHED; |
3c0563b9 | 6800 | |
e2e6a0f1 | 6801 | case MARKPOINT_next_fail: |
5d458dd8 | 6802 | if (popmark && sv_eq(ST.mark_name,popmark)) |
e2e6a0f1 YO |
6803 | { |
6804 | if (ST.mark_loc > startpoint) | |
6805 | reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1); | |
6806 | popmark = NULL; /* we found our mark */ | |
6807 | sv_commit = ST.mark_name; | |
6808 | ||
6809 | DEBUG_EXECUTE_r({ | |
5d458dd8 | 6810 | PerlIO_printf(Perl_debug_log, |
e2e6a0f1 YO |
6811 | "%*s %ssetting cutpoint to mark:%"SVf"...%s\n", |
6812 | REPORT_CODE_OFF+depth*2, "", | |
be2597df | 6813 | PL_colors[4], SVfARG(sv_commit), PL_colors[5]); |
e2e6a0f1 YO |
6814 | }); |
6815 | } | |
6816 | mark_state = ST.prev_mark; | |
5d458dd8 YO |
6817 | sv_yes_mark = mark_state ? |
6818 | mark_state->u.mark.mark_name : NULL; | |
e2e6a0f1 | 6819 | sayNO; |
a74ff37d | 6820 | /* NOTREACHED */ |
e5964223 | 6821 | NOT_REACHED; |
3c0563b9 DM |
6822 | |
6823 | case SKIP: /* (*SKIP) */ | |
5d458dd8 | 6824 | if (scan->flags) { |
2bf803e2 | 6825 | /* (*SKIP) : if we fail we cut here*/ |
5d458dd8 | 6826 | ST.mark_name = NULL; |
e2e6a0f1 | 6827 | ST.mark_loc = locinput; |
4d5016e5 | 6828 | PUSH_STATE_GOTO(SKIP_next,next, locinput); |
5d458dd8 | 6829 | } else { |
2bf803e2 | 6830 | /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was, |
5d458dd8 YO |
6831 | otherwise do nothing. Meaning we need to scan |
6832 | */ | |
6833 | regmatch_state *cur = mark_state; | |
ad64d0ec | 6834 | SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
5d458dd8 YO |
6835 | |
6836 | while (cur) { | |
6837 | if ( sv_eq( cur->u.mark.mark_name, | |
6838 | find ) ) | |
6839 | { | |
6840 | ST.mark_name = find; | |
4d5016e5 | 6841 | PUSH_STATE_GOTO( SKIP_next, next, locinput); |
5d458dd8 YO |
6842 | } |
6843 | cur = cur->u.mark.prev_mark; | |
6844 | } | |
e2e6a0f1 | 6845 | } |
2bf803e2 | 6846 | /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */ |
5d458dd8 | 6847 | break; |
3c0563b9 | 6848 | |
5d458dd8 YO |
6849 | case SKIP_next_fail: |
6850 | if (ST.mark_name) { | |
6851 | /* (*CUT:NAME) - Set up to search for the name as we | |
6852 | collapse the stack*/ | |
6853 | popmark = ST.mark_name; | |
6854 | } else { | |
6855 | /* (*CUT) - No name, we cut here.*/ | |
e2e6a0f1 YO |
6856 | if (ST.mark_loc > startpoint) |
6857 | reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1); | |
5d458dd8 YO |
6858 | /* but we set sv_commit to latest mark_name if there |
6859 | is one so they can test to see how things lead to this | |
6860 | cut */ | |
6861 | if (mark_state) | |
6862 | sv_commit=mark_state->u.mark.mark_name; | |
6863 | } | |
e2e6a0f1 YO |
6864 | no_final = 1; |
6865 | sayNO; | |
a74ff37d | 6866 | /* NOTREACHED */ |
e5964223 | 6867 | NOT_REACHED; |
e2e6a0f1 | 6868 | #undef ST |
3c0563b9 DM |
6869 | |
6870 | case LNBREAK: /* \R */ | |
220db18a | 6871 | if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) { |
e1d1eefb | 6872 | locinput += n; |
e1d1eefb YO |
6873 | } else |
6874 | sayNO; | |
6875 | break; | |
6876 | ||
a0d0e21e | 6877 | default: |
b900a521 | 6878 | PerlIO_printf(Perl_error_log, "%"UVxf" %d\n", |
d7d93a81 | 6879 | PTR2UV(scan), OP(scan)); |
cea2e8a9 | 6880 | Perl_croak(aTHX_ "regexp memory corruption"); |
28b98f76 DM |
6881 | |
6882 | /* this is a point to jump to in order to increment | |
6883 | * locinput by one character */ | |
6884 | increment_locinput: | |
e6ca698c | 6885 | assert(!NEXTCHR_IS_EOS); |
28b98f76 DM |
6886 | if (utf8_target) { |
6887 | locinput += PL_utf8skip[nextchr]; | |
7016d6eb | 6888 | /* locinput is allowed to go 1 char off the end, but not 2+ */ |
220db18a | 6889 | if (locinput > reginfo->strend) |
28b98f76 | 6890 | sayNO; |
28b98f76 DM |
6891 | } |
6892 | else | |
3640db6b | 6893 | locinput++; |
28b98f76 | 6894 | break; |
5d458dd8 YO |
6895 | |
6896 | } /* end switch */ | |
95b24440 | 6897 | |
5d458dd8 YO |
6898 | /* switch break jumps here */ |
6899 | scan = next; /* prepare to execute the next op and ... */ | |
6900 | continue; /* ... jump back to the top, reusing st */ | |
a74ff37d | 6901 | /* NOTREACHED */ |
95b24440 | 6902 | |
40a82448 DM |
6903 | push_yes_state: |
6904 | /* push a state that backtracks on success */ | |
6905 | st->u.yes.prev_yes_state = yes_state; | |
6906 | yes_state = st; | |
924ba076 | 6907 | /* FALLTHROUGH */ |
40a82448 DM |
6908 | push_state: |
6909 | /* push a new regex state, then continue at scan */ | |
6910 | { | |
6911 | regmatch_state *newst; | |
6912 | ||
24b23f37 YO |
6913 | DEBUG_STACK_r({ |
6914 | regmatch_state *cur = st; | |
6915 | regmatch_state *curyes = yes_state; | |
6916 | int curd = depth; | |
6917 | regmatch_slab *slab = PL_regmatch_slab; | |
6918 | for (;curd > -1;cur--,curd--) { | |
6919 | if (cur < SLAB_FIRST(slab)) { | |
6920 | slab = slab->prev; | |
6921 | cur = SLAB_LAST(slab); | |
6922 | } | |
6923 | PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n", | |
6924 | REPORT_CODE_OFF + 2 + depth * 2,"", | |
13d6edb4 | 6925 | curd, PL_reg_name[cur->resume_state], |
24b23f37 YO |
6926 | (curyes == cur) ? "yes" : "" |
6927 | ); | |
6928 | if (curyes == cur) | |
6929 | curyes = cur->u.yes.prev_yes_state; | |
6930 | } | |
6931 | } else | |
6932 | DEBUG_STATE_pp("push") | |
6933 | ); | |
40a82448 | 6934 | depth++; |
40a82448 DM |
6935 | st->locinput = locinput; |
6936 | newst = st+1; | |
6937 | if (newst > SLAB_LAST(PL_regmatch_slab)) | |
6938 | newst = S_push_slab(aTHX); | |
6939 | PL_regmatch_state = newst; | |
786e8c11 | 6940 | |
4d5016e5 | 6941 | locinput = pushinput; |
40a82448 DM |
6942 | st = newst; |
6943 | continue; | |
a74ff37d | 6944 | /* NOTREACHED */ |
40a82448 | 6945 | } |
a0d0e21e | 6946 | } |
a687059c | 6947 | |
a0d0e21e LW |
6948 | /* |
6949 | * We get here only if there's trouble -- normally "case END" is | |
6950 | * the terminating point. | |
6951 | */ | |
cea2e8a9 | 6952 | Perl_croak(aTHX_ "corrupted regexp pointers"); |
a74ff37d | 6953 | /* NOTREACHED */ |
4633a7c4 | 6954 | sayNO; |
e5964223 | 6955 | NOT_REACHED; |
4633a7c4 | 6956 | |
262b90c4 | 6957 | yes: |
77cb431f DM |
6958 | if (yes_state) { |
6959 | /* we have successfully completed a subexpression, but we must now | |
6960 | * pop to the state marked by yes_state and continue from there */ | |
77cb431f | 6961 | assert(st != yes_state); |
5bc10b2c DM |
6962 | #ifdef DEBUGGING |
6963 | while (st != yes_state) { | |
6964 | st--; | |
6965 | if (st < SLAB_FIRST(PL_regmatch_slab)) { | |
6966 | PL_regmatch_slab = PL_regmatch_slab->prev; | |
6967 | st = SLAB_LAST(PL_regmatch_slab); | |
6968 | } | |
e2e6a0f1 | 6969 | DEBUG_STATE_r({ |
54612592 YO |
6970 | if (no_final) { |
6971 | DEBUG_STATE_pp("pop (no final)"); | |
6972 | } else { | |
6973 | DEBUG_STATE_pp("pop (yes)"); | |
6974 | } | |
e2e6a0f1 | 6975 | }); |
5bc10b2c DM |
6976 | depth--; |
6977 | } | |
6978 | #else | |
77cb431f DM |
6979 | while (yes_state < SLAB_FIRST(PL_regmatch_slab) |
6980 | || yes_state > SLAB_LAST(PL_regmatch_slab)) | |
6981 | { | |
6982 | /* not in this slab, pop slab */ | |
6983 | depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1); | |
6984 | PL_regmatch_slab = PL_regmatch_slab->prev; | |
6985 | st = SLAB_LAST(PL_regmatch_slab); | |
6986 | } | |
6987 | depth -= (st - yes_state); | |
5bc10b2c | 6988 | #endif |
77cb431f DM |
6989 | st = yes_state; |
6990 | yes_state = st->u.yes.prev_yes_state; | |
6991 | PL_regmatch_state = st; | |
24b23f37 | 6992 | |
3640db6b | 6993 | if (no_final) |
5d458dd8 | 6994 | locinput= st->locinput; |
54612592 | 6995 | state_num = st->resume_state + no_final; |
24d3c4a9 | 6996 | goto reenter_switch; |
77cb431f DM |
6997 | } |
6998 | ||
a3621e74 | 6999 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n", |
e4584336 | 7000 | PL_colors[4], PL_colors[5])); |
02db2b7b | 7001 | |
bf2039a9 | 7002 | if (reginfo->info_aux_eval) { |
19b95bf0 DM |
7003 | /* each successfully executed (?{...}) block does the equivalent of |
7004 | * local $^R = do {...} | |
7005 | * When popping the save stack, all these locals would be undone; | |
7006 | * bypass this by setting the outermost saved $^R to the latest | |
7007 | * value */ | |
4b22688e YO |
7008 | /* I dont know if this is needed or works properly now. |
7009 | * see code related to PL_replgv elsewhere in this file. | |
7010 | * Yves | |
7011 | */ | |
19b95bf0 DM |
7012 | if (oreplsv != GvSV(PL_replgv)) |
7013 | sv_setsv(oreplsv, GvSV(PL_replgv)); | |
7014 | } | |
95b24440 | 7015 | result = 1; |
aa283a38 | 7016 | goto final_exit; |
4633a7c4 LW |
7017 | |
7018 | no: | |
a3621e74 | 7019 | DEBUG_EXECUTE_r( |
7821416a | 7020 | PerlIO_printf(Perl_debug_log, |
786e8c11 | 7021 | "%*s %sfailed...%s\n", |
5bc10b2c | 7022 | REPORT_CODE_OFF+depth*2, "", |
786e8c11 | 7023 | PL_colors[4], PL_colors[5]) |
7821416a | 7024 | ); |
aa283a38 | 7025 | |
262b90c4 | 7026 | no_silent: |
54612592 YO |
7027 | if (no_final) { |
7028 | if (yes_state) { | |
7029 | goto yes; | |
7030 | } else { | |
7031 | goto final_exit; | |
7032 | } | |
7033 | } | |
aa283a38 DM |
7034 | if (depth) { |
7035 | /* there's a previous state to backtrack to */ | |
40a82448 DM |
7036 | st--; |
7037 | if (st < SLAB_FIRST(PL_regmatch_slab)) { | |
7038 | PL_regmatch_slab = PL_regmatch_slab->prev; | |
7039 | st = SLAB_LAST(PL_regmatch_slab); | |
7040 | } | |
7041 | PL_regmatch_state = st; | |
40a82448 | 7042 | locinput= st->locinput; |
40a82448 | 7043 | |
5bc10b2c DM |
7044 | DEBUG_STATE_pp("pop"); |
7045 | depth--; | |
262b90c4 DM |
7046 | if (yes_state == st) |
7047 | yes_state = st->u.yes.prev_yes_state; | |
5bc10b2c | 7048 | |
24d3c4a9 DM |
7049 | state_num = st->resume_state + 1; /* failure = success + 1 */ |
7050 | goto reenter_switch; | |
95b24440 | 7051 | } |
24d3c4a9 | 7052 | result = 0; |
aa283a38 | 7053 | |
262b90c4 | 7054 | final_exit: |
bbe252da | 7055 | if (rex->intflags & PREGf_VERBARG_SEEN) { |
5d458dd8 YO |
7056 | SV *sv_err = get_sv("REGERROR", 1); |
7057 | SV *sv_mrk = get_sv("REGMARK", 1); | |
7058 | if (result) { | |
e2e6a0f1 | 7059 | sv_commit = &PL_sv_no; |
5d458dd8 YO |
7060 | if (!sv_yes_mark) |
7061 | sv_yes_mark = &PL_sv_yes; | |
7062 | } else { | |
7063 | if (!sv_commit) | |
7064 | sv_commit = &PL_sv_yes; | |
7065 | sv_yes_mark = &PL_sv_no; | |
7066 | } | |
316ebaf2 JH |
7067 | assert(sv_err); |
7068 | assert(sv_mrk); | |
5d458dd8 YO |
7069 | sv_setsv(sv_err, sv_commit); |
7070 | sv_setsv(sv_mrk, sv_yes_mark); | |
e2e6a0f1 | 7071 | } |
19b95bf0 | 7072 | |
81ed78b2 DM |
7073 | |
7074 | if (last_pushed_cv) { | |
7075 | dSP; | |
7076 | POP_MULTICALL; | |
4f8dbb2d | 7077 | PERL_UNUSED_VAR(SP); |
81ed78b2 DM |
7078 | } |
7079 | ||
9d9163fb DM |
7080 | assert(!result || locinput - reginfo->strbeg >= 0); |
7081 | return result ? locinput - reginfo->strbeg : -1; | |
a687059c LW |
7082 | } |
7083 | ||
7084 | /* | |
7085 | - regrepeat - repeatedly match something simple, report how many | |
d60de1d1 | 7086 | * |
e64f369d KW |
7087 | * What 'simple' means is a node which can be the operand of a quantifier like |
7088 | * '+', or {1,3} | |
7089 | * | |
d60de1d1 DM |
7090 | * startposp - pointer a pointer to the start position. This is updated |
7091 | * to point to the byte following the highest successful | |
7092 | * match. | |
7093 | * p - the regnode to be repeatedly matched against. | |
220db18a | 7094 | * reginfo - struct holding match state, such as strend |
4063ade8 | 7095 | * max - maximum number of things to match. |
d60de1d1 | 7096 | * depth - (for debugging) backtracking depth. |
a687059c | 7097 | */ |
76e3520e | 7098 | STATIC I32 |
272d35c9 | 7099 | S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p, |
f9176b44 | 7100 | regmatch_info *const reginfo, I32 max, int depth) |
a687059c | 7101 | { |
4063ade8 | 7102 | char *scan; /* Pointer to current position in target string */ |
eb578fdb | 7103 | I32 c; |
220db18a | 7104 | char *loceol = reginfo->strend; /* local version */ |
4063ade8 | 7105 | I32 hardcount = 0; /* How many matches so far */ |
ba44c216 | 7106 | bool utf8_target = reginfo->is_utf8_target; |
3018b823 | 7107 | int to_complement = 0; /* Invert the result? */ |
d513472c | 7108 | UV utf8_flags; |
3018b823 | 7109 | _char_class_number classnum; |
4f55667c SP |
7110 | #ifndef DEBUGGING |
7111 | PERL_UNUSED_ARG(depth); | |
7112 | #endif | |
a0d0e21e | 7113 | |
7918f24d NC |
7114 | PERL_ARGS_ASSERT_REGREPEAT; |
7115 | ||
f73aaa43 | 7116 | scan = *startposp; |
faf11cac HS |
7117 | if (max == REG_INFTY) |
7118 | max = I32_MAX; | |
dfb8f192 | 7119 | else if (! utf8_target && loceol - scan > max) |
7f596f4c | 7120 | loceol = scan + max; |
4063ade8 KW |
7121 | |
7122 | /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down | |
7123 | * to the maximum of how far we should go in it (leaving it set to the real | |
7124 | * end, if the maximum permissible would take us beyond that). This allows | |
7125 | * us to make the loop exit condition that we haven't gone past <loceol> to | |
7126 | * also mean that we haven't exceeded the max permissible count, saving a | |
7127 | * test each time through the loop. But it assumes that the OP matches a | |
7128 | * single byte, which is true for most of the OPs below when applied to a | |
7129 | * non-UTF-8 target. Those relatively few OPs that don't have this | |
7130 | * characteristic will have to compensate. | |
7131 | * | |
7132 | * There is no adjustment for UTF-8 targets, as the number of bytes per | |
7133 | * character varies. OPs will have to test both that the count is less | |
7134 | * than the max permissible (using <hardcount> to keep track), and that we | |
7135 | * are still within the bounds of the string (using <loceol>. A few OPs | |
7136 | * match a single byte no matter what the encoding. They can omit the max | |
7137 | * test if, for the UTF-8 case, they do the adjustment that was skipped | |
7138 | * above. | |
7139 | * | |
7140 | * Thus, the code above sets things up for the common case; and exceptional | |
7141 | * cases need extra work; the common case is to make sure <scan> doesn't | |
7142 | * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the | |
7143 | * count doesn't exceed the maximum permissible */ | |
7144 | ||
a0d0e21e | 7145 | switch (OP(p)) { |
22c35a8c | 7146 | case REG_ANY: |
f2ed9b32 | 7147 | if (utf8_target) { |
1aa99e6b | 7148 | while (scan < loceol && hardcount < max && *scan != '\n') { |
ffc61ed2 JH |
7149 | scan += UTF8SKIP(scan); |
7150 | hardcount++; | |
7151 | } | |
7152 | } else { | |
7153 | while (scan < loceol && *scan != '\n') | |
7154 | scan++; | |
a0ed51b3 LW |
7155 | } |
7156 | break; | |
ffc61ed2 | 7157 | case SANY: |
f2ed9b32 | 7158 | if (utf8_target) { |
a0804c9e | 7159 | while (scan < loceol && hardcount < max) { |
def8e4ea JH |
7160 | scan += UTF8SKIP(scan); |
7161 | hardcount++; | |
7162 | } | |
7163 | } | |
7164 | else | |
7165 | scan = loceol; | |
a0ed51b3 | 7166 | break; |
4063ade8 | 7167 | case CANY: /* Move <scan> forward <max> bytes, unless goes off end */ |
9597860a | 7168 | if (utf8_target && loceol - scan > max) { |
4063ade8 KW |
7169 | |
7170 | /* <loceol> hadn't been adjusted in the UTF-8 case */ | |
7171 | scan += max; | |
7172 | } | |
7173 | else { | |
7174 | scan = loceol; | |
7175 | } | |
f33976b4 | 7176 | break; |
59d32103 | 7177 | case EXACT: |
f9176b44 | 7178 | assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1); |
613a425d | 7179 | |
59d32103 | 7180 | c = (U8)*STRING(p); |
59d32103 | 7181 | |
5e4a1da1 KW |
7182 | /* Can use a simple loop if the pattern char to match on is invariant |
7183 | * under UTF-8, or both target and pattern aren't UTF-8. Note that we | |
7184 | * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's | |
7185 | * true iff it doesn't matter if the argument is in UTF-8 or not */ | |
f9176b44 | 7186 | if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) { |
e9369824 | 7187 | if (utf8_target && loceol - scan > max) { |
4063ade8 KW |
7188 | /* We didn't adjust <loceol> because is UTF-8, but ok to do so, |
7189 | * since here, to match at all, 1 char == 1 byte */ | |
7190 | loceol = scan + max; | |
7191 | } | |
59d32103 KW |
7192 | while (scan < loceol && UCHARAT(scan) == c) { |
7193 | scan++; | |
7194 | } | |
7195 | } | |
f9176b44 | 7196 | else if (reginfo->is_utf8_pat) { |
5e4a1da1 KW |
7197 | if (utf8_target) { |
7198 | STRLEN scan_char_len; | |
5e4a1da1 | 7199 | |
4063ade8 | 7200 | /* When both target and pattern are UTF-8, we have to do |
5e4a1da1 KW |
7201 | * string EQ */ |
7202 | while (hardcount < max | |
9a902117 KW |
7203 | && scan < loceol |
7204 | && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p) | |
5e4a1da1 KW |
7205 | && memEQ(scan, STRING(p), scan_char_len)) |
7206 | { | |
4200a00c | 7207 | scan += scan_char_len; |
5e4a1da1 KW |
7208 | hardcount++; |
7209 | } | |
7210 | } | |
7211 | else if (! UTF8_IS_ABOVE_LATIN1(c)) { | |
b40a2c17 | 7212 | |
5e4a1da1 KW |
7213 | /* Target isn't utf8; convert the character in the UTF-8 |
7214 | * pattern to non-UTF8, and do a simple loop */ | |
94bb8c36 | 7215 | c = TWO_BYTE_UTF8_TO_NATIVE(c, *(STRING(p) + 1)); |
5e4a1da1 KW |
7216 | while (scan < loceol && UCHARAT(scan) == c) { |
7217 | scan++; | |
7218 | } | |
7219 | } /* else pattern char is above Latin1, can't possibly match the | |
7220 | non-UTF-8 target */ | |
b40a2c17 | 7221 | } |
5e4a1da1 | 7222 | else { |
59d32103 | 7223 | |
5e4a1da1 KW |
7224 | /* Here, the string must be utf8; pattern isn't, and <c> is |
7225 | * different in utf8 than not, so can't compare them directly. | |
7226 | * Outside the loop, find the two utf8 bytes that represent c, and | |
7227 | * then look for those in sequence in the utf8 string */ | |
59d32103 KW |
7228 | U8 high = UTF8_TWO_BYTE_HI(c); |
7229 | U8 low = UTF8_TWO_BYTE_LO(c); | |
59d32103 KW |
7230 | |
7231 | while (hardcount < max | |
7232 | && scan + 1 < loceol | |
7233 | && UCHARAT(scan) == high | |
7234 | && UCHARAT(scan + 1) == low) | |
7235 | { | |
7236 | scan += 2; | |
7237 | hardcount++; | |
7238 | } | |
7239 | } | |
7240 | break; | |
5e4a1da1 | 7241 | |
098b07d5 KW |
7242 | case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */ |
7243 | assert(! reginfo->is_utf8_pat); | |
924ba076 | 7244 | /* FALLTHROUGH */ |
2f7f8cb1 | 7245 | case EXACTFA: |
098b07d5 | 7246 | utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
2f7f8cb1 KW |
7247 | goto do_exactf; |
7248 | ||
d4e0b827 | 7249 | case EXACTFL: |
cea315b6 | 7250 | utf8_flags = FOLDEQ_LOCALE; |
17580e7a KW |
7251 | goto do_exactf; |
7252 | ||
2fdb7295 KW |
7253 | case EXACTF: /* This node only generated for non-utf8 patterns */ |
7254 | assert(! reginfo->is_utf8_pat); | |
098b07d5 KW |
7255 | utf8_flags = 0; |
7256 | goto do_exactf; | |
62bf7766 | 7257 | |
3c760661 | 7258 | case EXACTFU_SS: |
9a5a5549 | 7259 | case EXACTFU: |
f9176b44 | 7260 | utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0; |
59d32103 | 7261 | |
613a425d KW |
7262 | do_exactf: { |
7263 | int c1, c2; | |
7264 | U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1]; | |
d4e0b827 | 7265 | |
f9176b44 | 7266 | assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1); |
613a425d | 7267 | |
984e6dd1 | 7268 | if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8, |
aed7b151 | 7269 | reginfo)) |
984e6dd1 | 7270 | { |
613a425d | 7271 | if (c1 == CHRTEST_VOID) { |
49b95fad | 7272 | /* Use full Unicode fold matching */ |
220db18a | 7273 | char *tmpeol = reginfo->strend; |
f9176b44 | 7274 | STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1; |
49b95fad KW |
7275 | while (hardcount < max |
7276 | && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target, | |
7277 | STRING(p), NULL, pat_len, | |
f9176b44 | 7278 | reginfo->is_utf8_pat, utf8_flags)) |
49b95fad KW |
7279 | { |
7280 | scan = tmpeol; | |
220db18a | 7281 | tmpeol = reginfo->strend; |
49b95fad KW |
7282 | hardcount++; |
7283 | } | |
613a425d KW |
7284 | } |
7285 | else if (utf8_target) { | |
7286 | if (c1 == c2) { | |
4063ade8 KW |
7287 | while (scan < loceol |
7288 | && hardcount < max | |
613a425d KW |
7289 | && memEQ(scan, c1_utf8, UTF8SKIP(scan))) |
7290 | { | |
7291 | scan += UTF8SKIP(scan); | |
7292 | hardcount++; | |
7293 | } | |
7294 | } | |
7295 | else { | |
4063ade8 KW |
7296 | while (scan < loceol |
7297 | && hardcount < max | |
613a425d KW |
7298 | && (memEQ(scan, c1_utf8, UTF8SKIP(scan)) |
7299 | || memEQ(scan, c2_utf8, UTF8SKIP(scan)))) | |
7300 | { | |
7301 | scan += UTF8SKIP(scan); | |
7302 | hardcount++; | |
7303 | } | |
7304 | } | |
7305 | } | |
7306 | else if (c1 == c2) { | |
7307 | while (scan < loceol && UCHARAT(scan) == c1) { | |
7308 | scan++; | |
7309 | } | |
7310 | } | |
7311 | else { | |
7312 | while (scan < loceol && | |
7313 | (UCHARAT(scan) == c1 || UCHARAT(scan) == c2)) | |
7314 | { | |
7315 | scan++; | |
7316 | } | |
7317 | } | |
634c83a2 | 7318 | } |
bbce6d69 | 7319 | break; |
613a425d | 7320 | } |
a0d0e21e | 7321 | case ANYOF: |
e0193e47 | 7322 | if (utf8_target) { |
4e8910e0 | 7323 | while (hardcount < max |
9a902117 | 7324 | && scan < loceol |
3db24e1e | 7325 | && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target)) |
4e8910e0 | 7326 | { |
9a902117 | 7327 | scan += UTF8SKIP(scan); |
ffc61ed2 JH |
7328 | hardcount++; |
7329 | } | |
7330 | } else { | |
32fc9b6a | 7331 | while (scan < loceol && REGINCLASS(prog, p, (U8*)scan)) |
ffc61ed2 JH |
7332 | scan++; |
7333 | } | |
a0d0e21e | 7334 | break; |
4063ade8 | 7335 | |
3018b823 | 7336 | /* The argument (FLAGS) to all the POSIX node types is the class number */ |
980866de | 7337 | |
3018b823 KW |
7338 | case NPOSIXL: |
7339 | to_complement = 1; | |
7340 | /* FALLTHROUGH */ | |
980866de | 7341 | |
3018b823 | 7342 | case POSIXL: |
3018b823 KW |
7343 | if (! utf8_target) { |
7344 | while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p), | |
7345 | *scan))) | |
a12cf05f | 7346 | { |
3018b823 KW |
7347 | scan++; |
7348 | } | |
7349 | } else { | |
7350 | while (hardcount < max && scan < loceol | |
7351 | && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p), | |
7352 | (U8 *) scan))) | |
7353 | { | |
7354 | scan += UTF8SKIP(scan); | |
ffc61ed2 JH |
7355 | hardcount++; |
7356 | } | |
a0ed51b3 LW |
7357 | } |
7358 | break; | |
0658cdde | 7359 | |
3018b823 KW |
7360 | case POSIXD: |
7361 | if (utf8_target) { | |
7362 | goto utf8_posix; | |
7363 | } | |
7364 | /* FALLTHROUGH */ | |
7365 | ||
0658cdde | 7366 | case POSIXA: |
0430522f | 7367 | if (utf8_target && loceol - scan > max) { |
4063ade8 | 7368 | |
7aee35ff KW |
7369 | /* We didn't adjust <loceol> at the beginning of this routine |
7370 | * because is UTF-8, but it is actually ok to do so, since here, to | |
7371 | * match, 1 char == 1 byte. */ | |
4063ade8 KW |
7372 | loceol = scan + max; |
7373 | } | |
7374 | while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) { | |
0658cdde KW |
7375 | scan++; |
7376 | } | |
7377 | break; | |
980866de | 7378 | |
3018b823 KW |
7379 | case NPOSIXD: |
7380 | if (utf8_target) { | |
7381 | to_complement = 1; | |
7382 | goto utf8_posix; | |
7383 | } | |
924ba076 | 7384 | /* FALLTHROUGH */ |
980866de | 7385 | |
3018b823 KW |
7386 | case NPOSIXA: |
7387 | if (! utf8_target) { | |
7388 | while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) { | |
a12cf05f KW |
7389 | scan++; |
7390 | } | |
4063ade8 | 7391 | } |
3018b823 | 7392 | else { |
980866de | 7393 | |
3018b823 | 7394 | /* The complement of something that matches only ASCII matches all |
837226c8 | 7395 | * non-ASCII, plus everything in ASCII that isn't in the class. */ |
bedac28b | 7396 | while (hardcount < max && scan < loceol |
837226c8 | 7397 | && (! isASCII_utf8(scan) |
3018b823 | 7398 | || ! _generic_isCC_A((U8) *scan, FLAGS(p)))) |
a12cf05f | 7399 | { |
3018b823 | 7400 | scan += UTF8SKIP(scan); |
ffc61ed2 JH |
7401 | hardcount++; |
7402 | } | |
3018b823 KW |
7403 | } |
7404 | break; | |
980866de | 7405 | |
3018b823 KW |
7406 | case NPOSIXU: |
7407 | to_complement = 1; | |
7408 | /* FALLTHROUGH */ | |
7409 | ||
7410 | case POSIXU: | |
7411 | if (! utf8_target) { | |
7412 | while (scan < loceol && to_complement | |
7413 | ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p)))) | |
4063ade8 | 7414 | { |
3018b823 KW |
7415 | scan++; |
7416 | } | |
cfaf538b KW |
7417 | } |
7418 | else { | |
3018b823 KW |
7419 | utf8_posix: |
7420 | classnum = (_char_class_number) FLAGS(p); | |
7421 | if (classnum < _FIRST_NON_SWASH_CC) { | |
7422 | ||
7423 | /* Here, a swash is needed for above-Latin1 code points. | |
7424 | * Process as many Latin1 code points using the built-in rules. | |
7425 | * Go to another loop to finish processing upon encountering | |
7426 | * the first Latin1 code point. We could do that in this loop | |
7427 | * as well, but the other way saves having to test if the swash | |
7428 | * has been loaded every time through the loop: extra space to | |
7429 | * save a test. */ | |
7430 | while (hardcount < max && scan < loceol) { | |
7431 | if (UTF8_IS_INVARIANT(*scan)) { | |
7432 | if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan, | |
7433 | classnum)))) | |
7434 | { | |
7435 | break; | |
7436 | } | |
7437 | scan++; | |
7438 | } | |
7439 | else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) { | |
7440 | if (! (to_complement | |
94bb8c36 KW |
7441 | ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*scan, |
7442 | *(scan + 1)), | |
3018b823 KW |
7443 | classnum)))) |
7444 | { | |
7445 | break; | |
7446 | } | |
7447 | scan += 2; | |
7448 | } | |
7449 | else { | |
7450 | goto found_above_latin1; | |
7451 | } | |
7452 | ||
7453 | hardcount++; | |
7454 | } | |
7455 | } | |
7456 | else { | |
7457 | /* For these character classes, the knowledge of how to handle | |
7458 | * every code point is compiled in to Perl via a macro. This | |
7459 | * code is written for making the loops as tight as possible. | |
7460 | * It could be refactored to save space instead */ | |
7461 | switch (classnum) { | |
7462 | case _CC_ENUM_SPACE: /* XXX would require separate code | |
7463 | if we revert the change of \v | |
7464 | matching this */ | |
924ba076 | 7465 | /* FALLTHROUGH */ |
3018b823 KW |
7466 | case _CC_ENUM_PSXSPC: |
7467 | while (hardcount < max | |
7468 | && scan < loceol | |
7469 | && (to_complement ^ cBOOL(isSPACE_utf8(scan)))) | |
7470 | { | |
7471 | scan += UTF8SKIP(scan); | |
7472 | hardcount++; | |
7473 | } | |
7474 | break; | |
7475 | case _CC_ENUM_BLANK: | |
7476 | while (hardcount < max | |
7477 | && scan < loceol | |
7478 | && (to_complement ^ cBOOL(isBLANK_utf8(scan)))) | |
7479 | { | |
7480 | scan += UTF8SKIP(scan); | |
7481 | hardcount++; | |
7482 | } | |
7483 | break; | |
7484 | case _CC_ENUM_XDIGIT: | |
7485 | while (hardcount < max | |
7486 | && scan < loceol | |
7487 | && (to_complement ^ cBOOL(isXDIGIT_utf8(scan)))) | |
7488 | { | |
7489 | scan += UTF8SKIP(scan); | |
7490 | hardcount++; | |
7491 | } | |
7492 | break; | |
7493 | case _CC_ENUM_VERTSPACE: | |
7494 | while (hardcount < max | |
7495 | && scan < loceol | |
7496 | && (to_complement ^ cBOOL(isVERTWS_utf8(scan)))) | |
7497 | { | |
7498 | scan += UTF8SKIP(scan); | |
7499 | hardcount++; | |
7500 | } | |
7501 | break; | |
7502 | case _CC_ENUM_CNTRL: | |
7503 | while (hardcount < max | |
7504 | && scan < loceol | |
7505 | && (to_complement ^ cBOOL(isCNTRL_utf8(scan)))) | |
7506 | { | |
7507 | scan += UTF8SKIP(scan); | |
7508 | hardcount++; | |
7509 | } | |
7510 | break; | |
7511 | default: | |
7512 | Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum); | |
7513 | } | |
7514 | } | |
a0ed51b3 | 7515 | } |
3018b823 | 7516 | break; |
4063ade8 | 7517 | |
3018b823 KW |
7518 | found_above_latin1: /* Continuation of POSIXU and NPOSIXU */ |
7519 | ||
7520 | /* Load the swash if not already present */ | |
7521 | if (! PL_utf8_swash_ptrs[classnum]) { | |
7522 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; | |
7523 | PL_utf8_swash_ptrs[classnum] = _core_swash_init( | |
2a16ac92 KW |
7524 | "utf8", |
7525 | "", | |
7526 | &PL_sv_undef, 1, 0, | |
7527 | PL_XPosix_ptrs[classnum], &flags); | |
4063ade8 | 7528 | } |
3018b823 KW |
7529 | |
7530 | while (hardcount < max && scan < loceol | |
7531 | && to_complement ^ cBOOL(_generic_utf8( | |
7532 | classnum, | |
7533 | scan, | |
7534 | swash_fetch(PL_utf8_swash_ptrs[classnum], | |
7535 | (U8 *) scan, | |
7536 | TRUE)))) | |
7537 | { | |
7538 | scan += UTF8SKIP(scan); | |
7539 | hardcount++; | |
7540 | } | |
7541 | break; | |
7542 | ||
e1d1eefb | 7543 | case LNBREAK: |
e64f369d KW |
7544 | if (utf8_target) { |
7545 | while (hardcount < max && scan < loceol && | |
7546 | (c=is_LNBREAK_utf8_safe(scan, loceol))) { | |
7547 | scan += c; | |
7548 | hardcount++; | |
7549 | } | |
7550 | } else { | |
7551 | /* LNBREAK can match one or two latin chars, which is ok, but we | |
7552 | * have to use hardcount in this situation, and throw away the | |
7553 | * adjustment to <loceol> done before the switch statement */ | |
220db18a | 7554 | loceol = reginfo->strend; |
e64f369d KW |
7555 | while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) { |
7556 | scan+=c; | |
7557 | hardcount++; | |
7558 | } | |
7559 | } | |
7560 | break; | |
e1d1eefb | 7561 | |
584b1f02 KW |
7562 | case BOUND: |
7563 | case BOUNDA: | |
7564 | case BOUNDL: | |
7565 | case BOUNDU: | |
7566 | case EOS: | |
7567 | case GPOS: | |
7568 | case KEEPS: | |
7569 | case NBOUND: | |
7570 | case NBOUNDA: | |
7571 | case NBOUNDL: | |
7572 | case NBOUNDU: | |
7573 | case OPFAIL: | |
7574 | case SBOL: | |
7575 | case SEOL: | |
7576 | /* These are all 0 width, so match right here or not at all. */ | |
7577 | break; | |
7578 | ||
7579 | default: | |
7580 | Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]); | |
a74ff37d | 7581 | /* NOTREACHED */ |
e5964223 | 7582 | NOT_REACHED; |
584b1f02 | 7583 | |
a0d0e21e | 7584 | } |
a687059c | 7585 | |
a0ed51b3 LW |
7586 | if (hardcount) |
7587 | c = hardcount; | |
7588 | else | |
f73aaa43 DM |
7589 | c = scan - *startposp; |
7590 | *startposp = scan; | |
a687059c | 7591 | |
a3621e74 | 7592 | DEBUG_r({ |
e68ec53f | 7593 | GET_RE_DEBUG_FLAGS_DECL; |
be8e71aa | 7594 | DEBUG_EXECUTE_r({ |
e68ec53f | 7595 | SV * const prop = sv_newmortal(); |
8b9781c9 | 7596 | regprop(prog, prop, p, reginfo, NULL); |
e68ec53f | 7597 | PerlIO_printf(Perl_debug_log, |
be8e71aa | 7598 | "%*s %s can match %"IVdf" times out of %"IVdf"...\n", |
e2e6a0f1 | 7599 | REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max); |
a3621e74 | 7600 | }); |
be8e71aa | 7601 | }); |
9041c2e3 | 7602 | |
a0d0e21e | 7603 | return(c); |
a687059c LW |
7604 | } |
7605 | ||
c277df42 | 7606 | |
be8e71aa | 7607 | #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) |
c277df42 | 7608 | /* |
6c6525b8 | 7609 | - regclass_swash - prepare the utf8 swash. Wraps the shared core version to |
e0193e47 KW |
7610 | create a copy so that changes the caller makes won't change the shared one. |
7611 | If <altsvp> is non-null, will return NULL in it, for back-compat. | |
6c6525b8 | 7612 | */ |
ffc61ed2 | 7613 | SV * |
5aaab254 | 7614 | Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp) |
ffc61ed2 | 7615 | { |
6c6525b8 | 7616 | PERL_ARGS_ASSERT_REGCLASS_SWASH; |
e0193e47 KW |
7617 | |
7618 | if (altsvp) { | |
7619 | *altsvp = NULL; | |
7620 | } | |
7621 | ||
ef9bc832 | 7622 | return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL, NULL)); |
6c6525b8 | 7623 | } |
6c6525b8 | 7624 | |
3e63bed3 | 7625 | #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */ |
ffc61ed2 JH |
7626 | |
7627 | /* | |
ba7b4546 | 7628 | - reginclass - determine if a character falls into a character class |
832705d4 | 7629 | |
6698fab5 KW |
7630 | n is the ANYOF regnode |
7631 | p is the target string | |
3db24e1e | 7632 | p_end points to one byte beyond the end of the target string |
6698fab5 | 7633 | utf8_target tells whether p is in UTF-8. |
832705d4 | 7634 | |
635cd5d4 | 7635 | Returns true if matched; false otherwise. |
eba1359e | 7636 | |
d5788240 KW |
7637 | Note that this can be a synthetic start class, a combination of various |
7638 | nodes, so things you think might be mutually exclusive, such as locale, | |
7639 | aren't. It can match both locale and non-locale | |
7640 | ||
bbce6d69 | 7641 | */ |
7642 | ||
76e3520e | 7643 | STATIC bool |
3db24e1e | 7644 | S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target) |
bbce6d69 | 7645 | { |
27da23d5 | 7646 | dVAR; |
a3b680e6 | 7647 | const char flags = ANYOF_FLAGS(n); |
bbce6d69 | 7648 | bool match = FALSE; |
cc07378b | 7649 | UV c = *p; |
1aa99e6b | 7650 | |
7918f24d NC |
7651 | PERL_ARGS_ASSERT_REGINCLASS; |
7652 | ||
afd2eb18 KW |
7653 | /* If c is not already the code point, get it. Note that |
7654 | * UTF8_IS_INVARIANT() works even if not in UTF-8 */ | |
7655 | if (! UTF8_IS_INVARIANT(c) && utf8_target) { | |
635cd5d4 | 7656 | STRLEN c_len = 0; |
3db24e1e | 7657 | c = utf8n_to_uvchr(p, p_end - p, &c_len, |
6182169b KW |
7658 | (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) |
7659 | | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY); | |
7660 | /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for | |
7661 | * UTF8_ALLOW_FFFF */ | |
f7ab54c6 | 7662 | if (c_len == (STRLEN)-1) |
e8a70c6f | 7663 | Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)"); |
19f67299 | 7664 | } |
4b3cda86 | 7665 | |
7cdde544 | 7666 | /* If this character is potentially in the bitmap, check it */ |
dcb20b36 | 7667 | if (c < NUM_ANYOF_CODE_POINTS) { |
ffc61ed2 JH |
7668 | if (ANYOF_BITMAP_TEST(n, c)) |
7669 | match = TRUE; | |
93e92956 KW |
7670 | else if ((flags & ANYOF_MATCHES_ALL_NON_UTF8_NON_ASCII) |
7671 | && ! utf8_target | |
7672 | && ! isASCII(c)) | |
11454c59 KW |
7673 | { |
7674 | match = TRUE; | |
7675 | } | |
1462525b | 7676 | else if (flags & ANYOF_LOCALE_FLAGS) { |
6942fd9a | 7677 | if ((flags & ANYOF_LOC_FOLD) |
e0a1ff7a | 7678 | && c < 256 |
6942fd9a KW |
7679 | && ANYOF_BITMAP_TEST(n, PL_fold_locale[c])) |
7680 | { | |
7681 | match = TRUE; | |
b99851e1 | 7682 | } |
e0a1ff7a KW |
7683 | else if (ANYOF_POSIXL_TEST_ANY_SET(n) |
7684 | && c < 256 | |
7685 | ) { | |
31c7f561 KW |
7686 | |
7687 | /* The data structure is arranged so bits 0, 2, 4, ... are set | |
7688 | * if the class includes the Posix character class given by | |
7689 | * bit/2; and 1, 3, 5, ... are set if the class includes the | |
7690 | * complemented Posix class given by int(bit/2). So we loop | |
7691 | * through the bits, each time changing whether we complement | |
7692 | * the result or not. Suppose for the sake of illustration | |
7693 | * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0 | |
7694 | * is set, it means there is a match for this ANYOF node if the | |
7695 | * character is in the class given by the expression (0 / 2 = 0 | |
7696 | * = \w). If it is in that class, isFOO_lc() will return 1, | |
7697 | * and since 'to_complement' is 0, the result will stay TRUE, | |
7698 | * and we exit the loop. Suppose instead that bit 0 is 0, but | |
7699 | * bit 1 is 1. That means there is a match if the character | |
7700 | * matches \W. We won't bother to call isFOO_lc() on bit 0, | |
7701 | * but will on bit 1. On the second iteration 'to_complement' | |
7702 | * will be 1, so the exclusive or will reverse things, so we | |
7703 | * are testing for \W. On the third iteration, 'to_complement' | |
7704 | * will be 0, and we would be testing for \s; the fourth | |
b0d691b2 KW |
7705 | * iteration would test for \S, etc. |
7706 | * | |
7707 | * Note that this code assumes that all the classes are closed | |
7708 | * under folding. For example, if a character matches \w, then | |
7709 | * its fold does too; and vice versa. This should be true for | |
7710 | * any well-behaved locale for all the currently defined Posix | |
7711 | * classes, except for :lower: and :upper:, which are handled | |
7712 | * by the pseudo-class :cased: which matches if either of the | |
7713 | * other two does. To get rid of this assumption, an outer | |
7714 | * loop could be used below to iterate over both the source | |
7715 | * character, and its fold (if different) */ | |
31c7f561 KW |
7716 | |
7717 | int count = 0; | |
7718 | int to_complement = 0; | |
522b3c1e | 7719 | |
31c7f561 | 7720 | while (count < ANYOF_MAX) { |
8efd3f97 | 7721 | if (ANYOF_POSIXL_TEST(n, count) |
31c7f561 KW |
7722 | && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c))) |
7723 | { | |
7724 | match = TRUE; | |
7725 | break; | |
7726 | } | |
7727 | count++; | |
7728 | to_complement ^= 1; | |
7729 | } | |
ffc61ed2 | 7730 | } |
a0ed51b3 | 7731 | } |
a0ed51b3 LW |
7732 | } |
7733 | ||
31f05a37 | 7734 | |
7cdde544 | 7735 | /* If the bitmap didn't (or couldn't) match, and something outside the |
3b04b210 | 7736 | * bitmap could match, try that. */ |
ef87b810 | 7737 | if (!match) { |
93e92956 KW |
7738 | if (c >= NUM_ANYOF_CODE_POINTS |
7739 | && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP)) | |
7740 | { | |
7741 | match = TRUE; /* Everything above the bitmap matches */ | |
e051a21d | 7742 | } |
93e92956 KW |
7743 | else if ((flags & ANYOF_HAS_NONBITMAP_NON_UTF8_MATCHES) |
7744 | || (utf8_target && (flags & ANYOF_HAS_UTF8_NONBITMAP_MATCHES)) | |
1ee208c4 KW |
7745 | || ((flags & ANYOF_LOC_FOLD) |
7746 | && IN_UTF8_CTYPE_LOCALE | |
93e92956 | 7747 | && ARG(n) != ANYOF_ONLY_HAS_BITMAP)) |
3b04b210 | 7748 | { |
1ee208c4 KW |
7749 | SV* only_utf8_locale = NULL; |
7750 | SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0, | |
ef9bc832 | 7751 | &only_utf8_locale, NULL); |
7cdde544 | 7752 | if (sw) { |
893ef8be | 7753 | U8 utf8_buffer[2]; |
7cdde544 KW |
7754 | U8 * utf8_p; |
7755 | if (utf8_target) { | |
7756 | utf8_p = (U8 *) p; | |
e0193e47 | 7757 | } else { /* Convert to utf8 */ |
893ef8be KW |
7758 | utf8_p = utf8_buffer; |
7759 | append_utf8_from_native_byte(*p, &utf8_p); | |
7760 | utf8_p = utf8_buffer; | |
7cdde544 | 7761 | } |
f56b6394 | 7762 | |
e0193e47 | 7763 | if (swash_fetch(sw, utf8_p, TRUE)) { |
7cdde544 | 7764 | match = TRUE; |
e0193e47 | 7765 | } |
7cdde544 | 7766 | } |
1ee208c4 KW |
7767 | if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) { |
7768 | match = _invlist_contains_cp(only_utf8_locale, c); | |
7769 | } | |
7cdde544 | 7770 | } |
5073ffbd KW |
7771 | |
7772 | if (UNICODE_IS_SUPER(c) | |
ae986089 | 7773 | && (flags & ANYOF_WARN_SUPER) |
5073ffbd KW |
7774 | && ckWARN_d(WARN_NON_UNICODE)) |
7775 | { | |
7776 | Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE), | |
2d88a86a | 7777 | "Matched non-Unicode code point 0x%04"UVXf" against Unicode property; may not be portable", c); |
5073ffbd | 7778 | } |
7cdde544 KW |
7779 | } |
7780 | ||
5dbb0c08 KW |
7781 | #if ANYOF_INVERT != 1 |
7782 | /* Depending on compiler optimization cBOOL takes time, so if don't have to | |
7783 | * use it, don't */ | |
7784 | # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below, | |
7785 | #endif | |
7786 | ||
f0fdc1c9 | 7787 | /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */ |
5dbb0c08 | 7788 | return (flags & ANYOF_INVERT) ^ match; |
a0ed51b3 | 7789 | } |
161b471a | 7790 | |
dfe13c55 | 7791 | STATIC U8 * |
ea3daa5d | 7792 | S_reghop3(U8 *s, SSize_t off, const U8* lim) |
9041c2e3 | 7793 | { |
6af86488 KW |
7794 | /* return the position 'off' UTF-8 characters away from 's', forward if |
7795 | * 'off' >= 0, backwards if negative. But don't go outside of position | |
7796 | * 'lim', which better be < s if off < 0 */ | |
7797 | ||
7918f24d NC |
7798 | PERL_ARGS_ASSERT_REGHOP3; |
7799 | ||
a0ed51b3 | 7800 | if (off >= 0) { |
1aa99e6b | 7801 | while (off-- && s < lim) { |
ffc61ed2 | 7802 | /* XXX could check well-formedness here */ |
a0ed51b3 | 7803 | s += UTF8SKIP(s); |
ffc61ed2 | 7804 | } |
a0ed51b3 LW |
7805 | } |
7806 | else { | |
1de06328 YO |
7807 | while (off++ && s > lim) { |
7808 | s--; | |
7809 | if (UTF8_IS_CONTINUED(*s)) { | |
7810 | while (s > lim && UTF8_IS_CONTINUATION(*s)) | |
7811 | s--; | |
a0ed51b3 | 7812 | } |
1de06328 | 7813 | /* XXX could check well-formedness here */ |
a0ed51b3 LW |
7814 | } |
7815 | } | |
7816 | return s; | |
7817 | } | |
161b471a | 7818 | |
dfe13c55 | 7819 | STATIC U8 * |
ea3daa5d | 7820 | S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim) |
1de06328 | 7821 | { |
7918f24d NC |
7822 | PERL_ARGS_ASSERT_REGHOP4; |
7823 | ||
1de06328 YO |
7824 | if (off >= 0) { |
7825 | while (off-- && s < rlim) { | |
7826 | /* XXX could check well-formedness here */ | |
7827 | s += UTF8SKIP(s); | |
7828 | } | |
7829 | } | |
7830 | else { | |
7831 | while (off++ && s > llim) { | |
7832 | s--; | |
7833 | if (UTF8_IS_CONTINUED(*s)) { | |
7834 | while (s > llim && UTF8_IS_CONTINUATION(*s)) | |
7835 | s--; | |
7836 | } | |
7837 | /* XXX could check well-formedness here */ | |
7838 | } | |
7839 | } | |
7840 | return s; | |
7841 | } | |
1de06328 | 7842 | |
557f47af DM |
7843 | /* like reghop3, but returns NULL on overrun, rather than returning last |
7844 | * char pos */ | |
7845 | ||
1de06328 | 7846 | STATIC U8 * |
ea3daa5d | 7847 | S_reghopmaybe3(U8* s, SSize_t off, const U8* lim) |
a0ed51b3 | 7848 | { |
7918f24d NC |
7849 | PERL_ARGS_ASSERT_REGHOPMAYBE3; |
7850 | ||
a0ed51b3 | 7851 | if (off >= 0) { |
1aa99e6b | 7852 | while (off-- && s < lim) { |
ffc61ed2 | 7853 | /* XXX could check well-formedness here */ |
a0ed51b3 | 7854 | s += UTF8SKIP(s); |
ffc61ed2 | 7855 | } |
a0ed51b3 | 7856 | if (off >= 0) |
3dab1dad | 7857 | return NULL; |
a0ed51b3 LW |
7858 | } |
7859 | else { | |
1de06328 YO |
7860 | while (off++ && s > lim) { |
7861 | s--; | |
7862 | if (UTF8_IS_CONTINUED(*s)) { | |
7863 | while (s > lim && UTF8_IS_CONTINUATION(*s)) | |
7864 | s--; | |
a0ed51b3 | 7865 | } |
1de06328 | 7866 | /* XXX could check well-formedness here */ |
a0ed51b3 LW |
7867 | } |
7868 | if (off <= 0) | |
3dab1dad | 7869 | return NULL; |
a0ed51b3 LW |
7870 | } |
7871 | return s; | |
7872 | } | |
51371543 | 7873 | |
a75351a1 DM |
7874 | |
7875 | /* when executing a regex that may have (?{}), extra stuff needs setting | |
7876 | up that will be visible to the called code, even before the current | |
7877 | match has finished. In particular: | |
7878 | ||
7879 | * $_ is localised to the SV currently being matched; | |
7880 | * pos($_) is created if necessary, ready to be updated on each call-out | |
7881 | to code; | |
7882 | * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm | |
7883 | isn't set until the current pattern is successfully finished), so that | |
7884 | $1 etc of the match-so-far can be seen; | |
7885 | * save the old values of subbeg etc of the current regex, and set then | |
7886 | to the current string (again, this is normally only done at the end | |
7887 | of execution) | |
a75351a1 DM |
7888 | */ |
7889 | ||
7890 | static void | |
7891 | S_setup_eval_state(pTHX_ regmatch_info *const reginfo) | |
7892 | { | |
7893 | MAGIC *mg; | |
7894 | regexp *const rex = ReANY(reginfo->prog); | |
bf2039a9 | 7895 | regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval; |
8adc0f72 | 7896 | |
8adc0f72 | 7897 | eval_state->rex = rex; |
a75351a1 | 7898 | |
a75351a1 DM |
7899 | if (reginfo->sv) { |
7900 | /* Make $_ available to executed code. */ | |
7901 | if (reginfo->sv != DEFSV) { | |
7902 | SAVE_DEFSV; | |
7903 | DEFSV_set(reginfo->sv); | |
7904 | } | |
7905 | ||
96c2a8ff | 7906 | if (!(mg = mg_find_mglob(reginfo->sv))) { |
a75351a1 | 7907 | /* prepare for quick setting of pos */ |
96c2a8ff | 7908 | mg = sv_magicext_mglob(reginfo->sv); |
a75351a1 DM |
7909 | mg->mg_len = -1; |
7910 | } | |
8adc0f72 DM |
7911 | eval_state->pos_magic = mg; |
7912 | eval_state->pos = mg->mg_len; | |
25fdce4a | 7913 | eval_state->pos_flags = mg->mg_flags; |
a75351a1 | 7914 | } |
8adc0f72 DM |
7915 | else |
7916 | eval_state->pos_magic = NULL; | |
7917 | ||
a75351a1 | 7918 | if (!PL_reg_curpm) { |
f65e70f5 DM |
7919 | /* PL_reg_curpm is a fake PMOP that we can attach the current |
7920 | * regex to and point PL_curpm at, so that $1 et al are visible | |
7921 | * within a /(?{})/. It's just allocated once per interpreter the | |
7922 | * first time its needed */ | |
a75351a1 DM |
7923 | Newxz(PL_reg_curpm, 1, PMOP); |
7924 | #ifdef USE_ITHREADS | |
7925 | { | |
7926 | SV* const repointer = &PL_sv_undef; | |
7927 | /* this regexp is also owned by the new PL_reg_curpm, which | |
7928 | will try to free it. */ | |
7929 | av_push(PL_regex_padav, repointer); | |
b9f2b683 | 7930 | PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav); |
a75351a1 DM |
7931 | PL_regex_pad = AvARRAY(PL_regex_padav); |
7932 | } | |
7933 | #endif | |
7934 | } | |
7935 | SET_reg_curpm(reginfo->prog); | |
8adc0f72 | 7936 | eval_state->curpm = PL_curpm; |
a75351a1 DM |
7937 | PL_curpm = PL_reg_curpm; |
7938 | if (RXp_MATCH_COPIED(rex)) { | |
7939 | /* Here is a serious problem: we cannot rewrite subbeg, | |
7940 | since it may be needed if this match fails. Thus | |
7941 | $` inside (?{}) could fail... */ | |
8adc0f72 DM |
7942 | eval_state->subbeg = rex->subbeg; |
7943 | eval_state->sublen = rex->sublen; | |
7944 | eval_state->suboffset = rex->suboffset; | |
a8ee055f | 7945 | eval_state->subcoffset = rex->subcoffset; |
a75351a1 | 7946 | #ifdef PERL_ANY_COW |
8adc0f72 | 7947 | eval_state->saved_copy = rex->saved_copy; |
a75351a1 DM |
7948 | #endif |
7949 | RXp_MATCH_COPIED_off(rex); | |
7950 | } | |
7951 | else | |
8adc0f72 | 7952 | eval_state->subbeg = NULL; |
a75351a1 DM |
7953 | rex->subbeg = (char *)reginfo->strbeg; |
7954 | rex->suboffset = 0; | |
7955 | rex->subcoffset = 0; | |
7956 | rex->sublen = reginfo->strend - reginfo->strbeg; | |
7957 | } | |
7958 | ||
bf2039a9 DM |
7959 | |
7960 | /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */ | |
a75351a1 | 7961 | |
51371543 | 7962 | static void |
bf2039a9 | 7963 | S_cleanup_regmatch_info_aux(pTHX_ void *arg) |
51371543 | 7964 | { |
bf2039a9 DM |
7965 | regmatch_info_aux *aux = (regmatch_info_aux *) arg; |
7966 | regmatch_info_aux_eval *eval_state = aux->info_aux_eval; | |
331b2dcc | 7967 | regmatch_slab *s; |
bf2039a9 | 7968 | |
2ac8ff4b DM |
7969 | Safefree(aux->poscache); |
7970 | ||
331b2dcc | 7971 | if (eval_state) { |
bf2039a9 | 7972 | |
331b2dcc | 7973 | /* undo the effects of S_setup_eval_state() */ |
bf2039a9 | 7974 | |
331b2dcc DM |
7975 | if (eval_state->subbeg) { |
7976 | regexp * const rex = eval_state->rex; | |
7977 | rex->subbeg = eval_state->subbeg; | |
7978 | rex->sublen = eval_state->sublen; | |
7979 | rex->suboffset = eval_state->suboffset; | |
7980 | rex->subcoffset = eval_state->subcoffset; | |
db2c6cb3 | 7981 | #ifdef PERL_ANY_COW |
331b2dcc | 7982 | rex->saved_copy = eval_state->saved_copy; |
ed252734 | 7983 | #endif |
331b2dcc DM |
7984 | RXp_MATCH_COPIED_on(rex); |
7985 | } | |
7986 | if (eval_state->pos_magic) | |
25fdce4a | 7987 | { |
331b2dcc | 7988 | eval_state->pos_magic->mg_len = eval_state->pos; |
25fdce4a FC |
7989 | eval_state->pos_magic->mg_flags = |
7990 | (eval_state->pos_magic->mg_flags & ~MGf_BYTES) | |
7991 | | (eval_state->pos_flags & MGf_BYTES); | |
7992 | } | |
331b2dcc DM |
7993 | |
7994 | PL_curpm = eval_state->curpm; | |
8adc0f72 | 7995 | } |
bf2039a9 | 7996 | |
331b2dcc DM |
7997 | PL_regmatch_state = aux->old_regmatch_state; |
7998 | PL_regmatch_slab = aux->old_regmatch_slab; | |
7999 | ||
8000 | /* free all slabs above current one - this must be the last action | |
8001 | * of this function, as aux and eval_state are allocated within | |
8002 | * slabs and may be freed here */ | |
8003 | ||
8004 | s = PL_regmatch_slab->next; | |
8005 | if (s) { | |
8006 | PL_regmatch_slab->next = NULL; | |
8007 | while (s) { | |
8008 | regmatch_slab * const osl = s; | |
8009 | s = s->next; | |
8010 | Safefree(osl); | |
8011 | } | |
8012 | } | |
51371543 | 8013 | } |
33b8afdf | 8014 | |
8adc0f72 | 8015 | |
33b8afdf | 8016 | STATIC void |
5aaab254 | 8017 | S_to_utf8_substr(pTHX_ regexp *prog) |
33b8afdf | 8018 | { |
7e0d5ad7 KW |
8019 | /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile |
8020 | * on the converted value */ | |
8021 | ||
a1cac82e | 8022 | int i = 1; |
7918f24d NC |
8023 | |
8024 | PERL_ARGS_ASSERT_TO_UTF8_SUBSTR; | |
8025 | ||
a1cac82e NC |
8026 | do { |
8027 | if (prog->substrs->data[i].substr | |
8028 | && !prog->substrs->data[i].utf8_substr) { | |
8029 | SV* const sv = newSVsv(prog->substrs->data[i].substr); | |
8030 | prog->substrs->data[i].utf8_substr = sv; | |
8031 | sv_utf8_upgrade(sv); | |
610460f9 | 8032 | if (SvVALID(prog->substrs->data[i].substr)) { |
cffe132d | 8033 | if (SvTAIL(prog->substrs->data[i].substr)) { |
610460f9 NC |
8034 | /* Trim the trailing \n that fbm_compile added last |
8035 | time. */ | |
8036 | SvCUR_set(sv, SvCUR(sv) - 1); | |
8037 | /* Whilst this makes the SV technically "invalid" (as its | |
8038 | buffer is no longer followed by "\0") when fbm_compile() | |
8039 | adds the "\n" back, a "\0" is restored. */ | |
cffe132d NC |
8040 | fbm_compile(sv, FBMcf_TAIL); |
8041 | } else | |
8042 | fbm_compile(sv, 0); | |
610460f9 | 8043 | } |
a1cac82e NC |
8044 | if (prog->substrs->data[i].substr == prog->check_substr) |
8045 | prog->check_utf8 = sv; | |
8046 | } | |
8047 | } while (i--); | |
33b8afdf JH |
8048 | } |
8049 | ||
7e0d5ad7 | 8050 | STATIC bool |
5aaab254 | 8051 | S_to_byte_substr(pTHX_ regexp *prog) |
33b8afdf | 8052 | { |
7e0d5ad7 KW |
8053 | /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile |
8054 | * on the converted value; returns FALSE if can't be converted. */ | |
8055 | ||
a1cac82e | 8056 | int i = 1; |
7918f24d NC |
8057 | |
8058 | PERL_ARGS_ASSERT_TO_BYTE_SUBSTR; | |
8059 | ||
a1cac82e NC |
8060 | do { |
8061 | if (prog->substrs->data[i].utf8_substr | |
8062 | && !prog->substrs->data[i].substr) { | |
8063 | SV* sv = newSVsv(prog->substrs->data[i].utf8_substr); | |
7e0d5ad7 KW |
8064 | if (! sv_utf8_downgrade(sv, TRUE)) { |
8065 | return FALSE; | |
8066 | } | |
5400f398 KW |
8067 | if (SvVALID(prog->substrs->data[i].utf8_substr)) { |
8068 | if (SvTAIL(prog->substrs->data[i].utf8_substr)) { | |
8069 | /* Trim the trailing \n that fbm_compile added last | |
8070 | time. */ | |
8071 | SvCUR_set(sv, SvCUR(sv) - 1); | |
8072 | fbm_compile(sv, FBMcf_TAIL); | |
8073 | } else | |
8074 | fbm_compile(sv, 0); | |
8075 | } | |
a1cac82e NC |
8076 | prog->substrs->data[i].substr = sv; |
8077 | if (prog->substrs->data[i].utf8_substr == prog->check_utf8) | |
8078 | prog->check_substr = sv; | |
33b8afdf | 8079 | } |
a1cac82e | 8080 | } while (i--); |
7e0d5ad7 KW |
8081 | |
8082 | return TRUE; | |
33b8afdf | 8083 | } |
66610fdd RGS |
8084 | |
8085 | /* | |
8086 | * Local variables: | |
8087 | * c-indentation-style: bsd | |
8088 | * c-basic-offset: 4 | |
14d04a33 | 8089 | * indent-tabs-mode: nil |
66610fdd RGS |
8090 | * End: |
8091 | * | |
14d04a33 | 8092 | * ex: set ts=8 sts=4 sw=4 et: |
37442d52 | 8093 | */ |