| 1 | /* regexec.c |
| 2 | */ |
| 3 | |
| 4 | /* |
| 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"] |
| 10 | */ |
| 11 | |
| 12 | /* This file contains functions for executing a regular expression. See |
| 13 | * also regcomp.c which funnily enough, contains functions for compiling |
| 14 | * a regular expression. |
| 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. |
| 20 | */ |
| 21 | |
| 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 | |
| 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 | |
| 36 | #ifdef PERL_EXT_RE_BUILD |
| 37 | #include "re_top.h" |
| 38 | #endif |
| 39 | |
| 40 | /* |
| 41 | * pregcomp and pregexec -- regsub and regerror are not used in perl |
| 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 | **** |
| 62 | **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, |
| 63 | **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 |
| 64 | **** by Larry Wall and others |
| 65 | **** |
| 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. |
| 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" |
| 74 | #define PERL_IN_REGEXEC_C |
| 75 | #include "perl.h" |
| 76 | |
| 77 | #ifdef PERL_IN_XSUB_RE |
| 78 | # include "re_comp.h" |
| 79 | #else |
| 80 | # include "regcomp.h" |
| 81 | #endif |
| 82 | |
| 83 | #define RF_tainted 1 /* tainted information used? e.g. locale */ |
| 84 | #define RF_warned 2 /* warned about big count? */ |
| 85 | |
| 86 | #define RF_utf8 8 /* Pattern contains multibyte chars? */ |
| 87 | |
| 88 | #define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0) |
| 89 | |
| 90 | #define RS_init 1 /* eval environment created */ |
| 91 | #define RS_set 2 /* replsv value is set */ |
| 92 | |
| 93 | #ifndef STATIC |
| 94 | #define STATIC static |
| 95 | #endif |
| 96 | |
| 97 | /* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass |
| 98 | * call if there are no complications: i.e., if everything matchable is |
| 99 | * straight forward in the bitmap */ |
| 100 | #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \ |
| 101 | : ANYOF_BITMAP_TEST(p,*(c))) |
| 102 | |
| 103 | /* |
| 104 | * Forwards. |
| 105 | */ |
| 106 | |
| 107 | #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv)) |
| 108 | #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b) |
| 109 | |
| 110 | #define HOPc(pos,off) \ |
| 111 | (char *)(PL_reg_match_utf8 \ |
| 112 | ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \ |
| 113 | : (U8*)(pos + off)) |
| 114 | #define HOPBACKc(pos, off) \ |
| 115 | (char*)(PL_reg_match_utf8\ |
| 116 | ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \ |
| 117 | : (pos - off >= PL_bostr) \ |
| 118 | ? (U8*)pos - off \ |
| 119 | : NULL) |
| 120 | |
| 121 | #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off)) |
| 122 | #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim)) |
| 123 | |
| 124 | /* these are unrolled below in the CCC_TRY_XXX defined */ |
| 125 | #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \ |
| 126 | if (!CAT2(PL_utf8_,class)) { \ |
| 127 | bool ok; \ |
| 128 | ENTER; save_re_context(); \ |
| 129 | ok=CAT2(is_utf8_,class)((const U8*)str); \ |
| 130 | assert(ok); LEAVE; } } STMT_END |
| 131 | |
| 132 | /* Doesn't do an assert to verify that is correct */ |
| 133 | #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \ |
| 134 | if (!CAT2(PL_utf8_,class)) { \ |
| 135 | bool throw_away __attribute__unused__; \ |
| 136 | ENTER; save_re_context(); \ |
| 137 | throw_away = CAT2(is_utf8_,class)((const U8*)" "); \ |
| 138 | LEAVE; } } STMT_END |
| 139 | |
| 140 | #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a") |
| 141 | #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0") |
| 142 | #define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ") |
| 143 | |
| 144 | #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \ |
| 145 | LOAD_UTF8_CHARCLASS(X_begin, " "); \ |
| 146 | LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \ |
| 147 | /* These are utf8 constants, and not utf-ebcdic constants, so the \ |
| 148 | * assert should likely and hopefully fail on an EBCDIC machine */ \ |
| 149 | LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \ |
| 150 | \ |
| 151 | /* No asserts are done for these, in case called on an early \ |
| 152 | * Unicode version in which they map to nothing */ \ |
| 153 | LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \ |
| 154 | LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \ |
| 155 | LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \ |
| 156 | LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \ |
| 157 | LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\ |
| 158 | LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \ |
| 159 | LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */ |
| 160 | |
| 161 | #define PLACEHOLDER /* Something for the preprocessor to grab onto */ |
| 162 | |
| 163 | /* The actual code for CCC_TRY, which uses several variables from the routine |
| 164 | * it's callable from. It is designed to be the bulk of a case statement. |
| 165 | * FUNC is the macro or function to call on non-utf8 targets that indicate if |
| 166 | * nextchr matches the class. |
| 167 | * UTF8_TEST is the whole test string to use for utf8 targets |
| 168 | * LOAD is what to use to test, and if not present to load in the swash for the |
| 169 | * class |
| 170 | * POS_OR_NEG is either empty or ! to complement the results of FUNC or |
| 171 | * UTF8_TEST test. |
| 172 | * The logic is: Fail if we're at the end-of-string; otherwise if the target is |
| 173 | * utf8 and a variant, load the swash if necessary and test using the utf8 |
| 174 | * test. Advance to the next character if test is ok, otherwise fail; If not |
| 175 | * utf8 or an invariant under utf8, use the non-utf8 test, and fail if it |
| 176 | * fails, or advance to the next character */ |
| 177 | |
| 178 | #define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \ |
| 179 | if (locinput >= PL_regeol) { \ |
| 180 | sayNO; \ |
| 181 | } \ |
| 182 | if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \ |
| 183 | LOAD_UTF8_CHARCLASS(CLASS, STR); \ |
| 184 | if (POS_OR_NEG (UTF8_TEST)) { \ |
| 185 | sayNO; \ |
| 186 | } \ |
| 187 | locinput += PL_utf8skip[nextchr]; \ |
| 188 | nextchr = UCHARAT(locinput); \ |
| 189 | break; \ |
| 190 | } \ |
| 191 | if (POS_OR_NEG (FUNC(nextchr))) { \ |
| 192 | sayNO; \ |
| 193 | } \ |
| 194 | nextchr = UCHARAT(++locinput); \ |
| 195 | break; |
| 196 | |
| 197 | /* Handle the non-locale cases for a character class and its complement. It |
| 198 | * calls _CCC_TRY_CODE with a ! to complement the test for the character class. |
| 199 | * This is because that code fails when the test succeeds, so we want to have |
| 200 | * the test fail so that the code succeeds. The swash is stored in a |
| 201 | * predictable PL_ place */ |
| 202 | #define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \ |
| 203 | CLASS, STR) \ |
| 204 | case NAME: \ |
| 205 | _CCC_TRY_CODE( !, FUNC, \ |
| 206 | cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \ |
| 207 | (U8*)locinput, TRUE)), \ |
| 208 | CLASS, STR) \ |
| 209 | case NNAME: \ |
| 210 | _CCC_TRY_CODE( PLACEHOLDER , FUNC, \ |
| 211 | cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \ |
| 212 | (U8*)locinput, TRUE)), \ |
| 213 | CLASS, STR) \ |
| 214 | |
| 215 | /* Generate the case statements for both locale and non-locale character |
| 216 | * classes in regmatch for classes that don't have special unicode semantics. |
| 217 | * Locales don't use an immediate swash, but an intermediary special locale |
| 218 | * function that is called on the pointer to the current place in the input |
| 219 | * string. That function will resolve to needing the same swash. One might |
| 220 | * think that because we don't know what the locale will match, we shouldn't |
| 221 | * check with the swash loading function that it loaded properly; ie, that we |
| 222 | * should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the |
| 223 | * regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is |
| 224 | * irrelevant here */ |
| 225 | #define CCC_TRY(NAME, NNAME, FUNC, \ |
| 226 | NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \ |
| 227 | NAMEA, NNAMEA, FUNCA, \ |
| 228 | CLASS, STR) \ |
| 229 | case NAMEL: \ |
| 230 | PL_reg_flags |= RF_tainted; \ |
| 231 | _CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \ |
| 232 | case NNAMEL: \ |
| 233 | PL_reg_flags |= RF_tainted; \ |
| 234 | _CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \ |
| 235 | CLASS, STR) \ |
| 236 | case NAMEA: \ |
| 237 | if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \ |
| 238 | sayNO; \ |
| 239 | } \ |
| 240 | /* Matched a utf8-invariant, so don't have to worry about utf8 */ \ |
| 241 | nextchr = UCHARAT(++locinput); \ |
| 242 | break; \ |
| 243 | case NNAMEA: \ |
| 244 | if (locinput >= PL_regeol || FUNCA(nextchr)) { \ |
| 245 | sayNO; \ |
| 246 | } \ |
| 247 | if (utf8_target) { \ |
| 248 | locinput += PL_utf8skip[nextchr]; \ |
| 249 | nextchr = UCHARAT(locinput); \ |
| 250 | } \ |
| 251 | else { \ |
| 252 | nextchr = UCHARAT(++locinput); \ |
| 253 | } \ |
| 254 | break; \ |
| 255 | /* Generate the non-locale cases */ \ |
| 256 | _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR) |
| 257 | |
| 258 | /* This is like CCC_TRY, but has an extra set of parameters for generating case |
| 259 | * statements to handle separate Unicode semantics nodes */ |
| 260 | #define CCC_TRY_U(NAME, NNAME, FUNC, \ |
| 261 | NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \ |
| 262 | NAMEU, NNAMEU, FUNCU, \ |
| 263 | NAMEA, NNAMEA, FUNCA, \ |
| 264 | CLASS, STR) \ |
| 265 | CCC_TRY(NAME, NNAME, FUNC, \ |
| 266 | NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \ |
| 267 | NAMEA, NNAMEA, FUNCA, \ |
| 268 | CLASS, STR) \ |
| 269 | _CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR) |
| 270 | |
| 271 | /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */ |
| 272 | |
| 273 | /* for use after a quantifier and before an EXACT-like node -- japhy */ |
| 274 | /* it would be nice to rework regcomp.sym to generate this stuff. sigh |
| 275 | * |
| 276 | * NOTE that *nothing* that affects backtracking should be in here, specifically |
| 277 | * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a |
| 278 | * node that is in between two EXACT like nodes when ascertaining what the required |
| 279 | * "follow" character is. This should probably be moved to regex compile time |
| 280 | * although it may be done at run time beause of the REF possibility - more |
| 281 | * investigation required. -- demerphq |
| 282 | */ |
| 283 | #define JUMPABLE(rn) ( \ |
| 284 | OP(rn) == OPEN || \ |
| 285 | (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \ |
| 286 | OP(rn) == EVAL || \ |
| 287 | OP(rn) == SUSPEND || OP(rn) == IFMATCH || \ |
| 288 | OP(rn) == PLUS || OP(rn) == MINMOD || \ |
| 289 | OP(rn) == KEEPS || \ |
| 290 | (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \ |
| 291 | ) |
| 292 | #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT) |
| 293 | |
| 294 | #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF ) |
| 295 | |
| 296 | #if 0 |
| 297 | /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so |
| 298 | we don't need this definition. */ |
| 299 | #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF ) |
| 300 | #define IS_TEXTF(rn) ( (OP(rn)==EXACTFU || OP(rn)==EXACTFA || OP(rn)==EXACTF) || OP(rn)==REFF || OP(rn)==NREFF ) |
| 301 | #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL ) |
| 302 | |
| 303 | #else |
| 304 | /* ... so we use this as its faster. */ |
| 305 | #define IS_TEXT(rn) ( OP(rn)==EXACT ) |
| 306 | #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn) == EXACTFA) |
| 307 | #define IS_TEXTF(rn) ( OP(rn)==EXACTF ) |
| 308 | #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL ) |
| 309 | |
| 310 | #endif |
| 311 | |
| 312 | /* |
| 313 | Search for mandatory following text node; for lookahead, the text must |
| 314 | follow but for lookbehind (rn->flags != 0) we skip to the next step. |
| 315 | */ |
| 316 | #define FIND_NEXT_IMPT(rn) STMT_START { \ |
| 317 | while (JUMPABLE(rn)) { \ |
| 318 | const OPCODE type = OP(rn); \ |
| 319 | if (type == SUSPEND || PL_regkind[type] == CURLY) \ |
| 320 | rn = NEXTOPER(NEXTOPER(rn)); \ |
| 321 | else if (type == PLUS) \ |
| 322 | rn = NEXTOPER(rn); \ |
| 323 | else if (type == IFMATCH) \ |
| 324 | rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \ |
| 325 | else rn += NEXT_OFF(rn); \ |
| 326 | } \ |
| 327 | } STMT_END |
| 328 | |
| 329 | |
| 330 | static void restore_pos(pTHX_ void *arg); |
| 331 | |
| 332 | #define REGCP_PAREN_ELEMS 4 |
| 333 | #define REGCP_OTHER_ELEMS 5 |
| 334 | #define REGCP_FRAME_ELEMS 1 |
| 335 | /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and |
| 336 | * are needed for the regexp context stack bookkeeping. */ |
| 337 | |
| 338 | STATIC CHECKPOINT |
| 339 | S_regcppush(pTHX_ I32 parenfloor) |
| 340 | { |
| 341 | dVAR; |
| 342 | const int retval = PL_savestack_ix; |
| 343 | const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS; |
| 344 | const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS; |
| 345 | const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT; |
| 346 | int p; |
| 347 | GET_RE_DEBUG_FLAGS_DECL; |
| 348 | |
| 349 | if (paren_elems_to_push < 0) |
| 350 | Perl_croak(aTHX_ "panic: paren_elems_to_push < 0"); |
| 351 | |
| 352 | if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems) |
| 353 | Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf |
| 354 | " out of range (%lu-%ld)", |
| 355 | total_elems, (unsigned long)PL_regsize, (long)parenfloor); |
| 356 | |
| 357 | SSGROW(total_elems + REGCP_FRAME_ELEMS); |
| 358 | |
| 359 | for (p = PL_regsize; p > parenfloor; p--) { |
| 360 | /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */ |
| 361 | SSPUSHINT(PL_regoffs[p].end); |
| 362 | SSPUSHINT(PL_regoffs[p].start); |
| 363 | SSPUSHPTR(PL_reg_start_tmp[p]); |
| 364 | SSPUSHINT(p); |
| 365 | DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, |
| 366 | " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n", |
| 367 | (UV)p, (IV)PL_regoffs[p].start, |
| 368 | (IV)(PL_reg_start_tmp[p] - PL_bostr), |
| 369 | (IV)PL_regoffs[p].end |
| 370 | )); |
| 371 | } |
| 372 | /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */ |
| 373 | SSPUSHPTR(PL_regoffs); |
| 374 | SSPUSHINT(PL_regsize); |
| 375 | SSPUSHINT(*PL_reglastparen); |
| 376 | SSPUSHINT(*PL_reglastcloseparen); |
| 377 | SSPUSHPTR(PL_reginput); |
| 378 | SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */ |
| 379 | |
| 380 | return retval; |
| 381 | } |
| 382 | |
| 383 | /* These are needed since we do not localize EVAL nodes: */ |
| 384 | #define REGCP_SET(cp) \ |
| 385 | DEBUG_STATE_r( \ |
| 386 | PerlIO_printf(Perl_debug_log, \ |
| 387 | " Setting an EVAL scope, savestack=%"IVdf"\n", \ |
| 388 | (IV)PL_savestack_ix)); \ |
| 389 | cp = PL_savestack_ix |
| 390 | |
| 391 | #define REGCP_UNWIND(cp) \ |
| 392 | DEBUG_STATE_r( \ |
| 393 | if (cp != PL_savestack_ix) \ |
| 394 | PerlIO_printf(Perl_debug_log, \ |
| 395 | " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \ |
| 396 | (IV)(cp), (IV)PL_savestack_ix)); \ |
| 397 | regcpblow(cp) |
| 398 | |
| 399 | STATIC char * |
| 400 | S_regcppop(pTHX_ const regexp *rex) |
| 401 | { |
| 402 | dVAR; |
| 403 | UV i; |
| 404 | char *input; |
| 405 | GET_RE_DEBUG_FLAGS_DECL; |
| 406 | |
| 407 | PERL_ARGS_ASSERT_REGCPPOP; |
| 408 | |
| 409 | /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */ |
| 410 | i = SSPOPUV; |
| 411 | assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */ |
| 412 | i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */ |
| 413 | input = (char *) SSPOPPTR; |
| 414 | *PL_reglastcloseparen = SSPOPINT; |
| 415 | *PL_reglastparen = SSPOPINT; |
| 416 | PL_regsize = SSPOPINT; |
| 417 | PL_regoffs=(regexp_paren_pair *) SSPOPPTR; |
| 418 | |
| 419 | i -= REGCP_OTHER_ELEMS; |
| 420 | /* Now restore the parentheses context. */ |
| 421 | for ( ; i > 0; i -= REGCP_PAREN_ELEMS) { |
| 422 | I32 tmps; |
| 423 | U32 paren = (U32)SSPOPINT; |
| 424 | PL_reg_start_tmp[paren] = (char *) SSPOPPTR; |
| 425 | PL_regoffs[paren].start = SSPOPINT; |
| 426 | tmps = SSPOPINT; |
| 427 | if (paren <= *PL_reglastparen) |
| 428 | PL_regoffs[paren].end = tmps; |
| 429 | DEBUG_BUFFERS_r( |
| 430 | PerlIO_printf(Perl_debug_log, |
| 431 | " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n", |
| 432 | (UV)paren, (IV)PL_regoffs[paren].start, |
| 433 | (IV)(PL_reg_start_tmp[paren] - PL_bostr), |
| 434 | (IV)PL_regoffs[paren].end, |
| 435 | (paren > *PL_reglastparen ? "(no)" : "")); |
| 436 | ); |
| 437 | } |
| 438 | DEBUG_BUFFERS_r( |
| 439 | if (*PL_reglastparen + 1 <= rex->nparens) { |
| 440 | PerlIO_printf(Perl_debug_log, |
| 441 | " restoring \\%"IVdf"..\\%"IVdf" to undef\n", |
| 442 | (IV)(*PL_reglastparen + 1), (IV)rex->nparens); |
| 443 | } |
| 444 | ); |
| 445 | #if 1 |
| 446 | /* It would seem that the similar code in regtry() |
| 447 | * already takes care of this, and in fact it is in |
| 448 | * a better location to since this code can #if 0-ed out |
| 449 | * but the code in regtry() is needed or otherwise tests |
| 450 | * requiring null fields (pat.t#187 and split.t#{13,14} |
| 451 | * (as of patchlevel 7877) will fail. Then again, |
| 452 | * this code seems to be necessary or otherwise |
| 453 | * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/ |
| 454 | * --jhi updated by dapm */ |
| 455 | for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) { |
| 456 | if (i > PL_regsize) |
| 457 | PL_regoffs[i].start = -1; |
| 458 | PL_regoffs[i].end = -1; |
| 459 | } |
| 460 | #endif |
| 461 | return input; |
| 462 | } |
| 463 | |
| 464 | #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */ |
| 465 | |
| 466 | /* |
| 467 | * pregexec and friends |
| 468 | */ |
| 469 | |
| 470 | #ifndef PERL_IN_XSUB_RE |
| 471 | /* |
| 472 | - pregexec - match a regexp against a string |
| 473 | */ |
| 474 | I32 |
| 475 | Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend, |
| 476 | char *strbeg, I32 minend, SV *screamer, U32 nosave) |
| 477 | /* strend: pointer to null at end of string */ |
| 478 | /* strbeg: real beginning of string */ |
| 479 | /* minend: end of match must be >=minend after stringarg. */ |
| 480 | /* nosave: For optimizations. */ |
| 481 | { |
| 482 | PERL_ARGS_ASSERT_PREGEXEC; |
| 483 | |
| 484 | return |
| 485 | regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL, |
| 486 | nosave ? 0 : REXEC_COPY_STR); |
| 487 | } |
| 488 | #endif |
| 489 | |
| 490 | /* |
| 491 | * Need to implement the following flags for reg_anch: |
| 492 | * |
| 493 | * USE_INTUIT_NOML - Useful to call re_intuit_start() first |
| 494 | * USE_INTUIT_ML |
| 495 | * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer |
| 496 | * INTUIT_AUTORITATIVE_ML |
| 497 | * INTUIT_ONCE_NOML - Intuit can match in one location only. |
| 498 | * INTUIT_ONCE_ML |
| 499 | * |
| 500 | * Another flag for this function: SECOND_TIME (so that float substrs |
| 501 | * with giant delta may be not rechecked). |
| 502 | */ |
| 503 | |
| 504 | /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */ |
| 505 | |
| 506 | /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend. |
| 507 | Otherwise, only SvCUR(sv) is used to get strbeg. */ |
| 508 | |
| 509 | /* XXXX We assume that strpos is strbeg unless sv. */ |
| 510 | |
| 511 | /* XXXX Some places assume that there is a fixed substring. |
| 512 | An update may be needed if optimizer marks as "INTUITable" |
| 513 | RExen without fixed substrings. Similarly, it is assumed that |
| 514 | lengths of all the strings are no more than minlen, thus they |
| 515 | cannot come from lookahead. |
| 516 | (Or minlen should take into account lookahead.) |
| 517 | NOTE: Some of this comment is not correct. minlen does now take account |
| 518 | of lookahead/behind. Further research is required. -- demerphq |
| 519 | |
| 520 | */ |
| 521 | |
| 522 | /* A failure to find a constant substring means that there is no need to make |
| 523 | an expensive call to REx engine, thus we celebrate a failure. Similarly, |
| 524 | finding a substring too deep into the string means that less calls to |
| 525 | regtry() should be needed. |
| 526 | |
| 527 | REx compiler's optimizer found 4 possible hints: |
| 528 | a) Anchored substring; |
| 529 | b) Fixed substring; |
| 530 | c) Whether we are anchored (beginning-of-line or \G); |
| 531 | d) First node (of those at offset 0) which may distinguish positions; |
| 532 | We use a)b)d) and multiline-part of c), and try to find a position in the |
| 533 | string which does not contradict any of them. |
| 534 | */ |
| 535 | |
| 536 | /* Most of decisions we do here should have been done at compile time. |
| 537 | The nodes of the REx which we used for the search should have been |
| 538 | deleted from the finite automaton. */ |
| 539 | |
| 540 | char * |
| 541 | Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos, |
| 542 | char *strend, const U32 flags, re_scream_pos_data *data) |
| 543 | { |
| 544 | dVAR; |
| 545 | struct regexp *const prog = (struct regexp *)SvANY(rx); |
| 546 | register I32 start_shift = 0; |
| 547 | /* Should be nonnegative! */ |
| 548 | register I32 end_shift = 0; |
| 549 | register char *s; |
| 550 | register SV *check; |
| 551 | char *strbeg; |
| 552 | char *t; |
| 553 | const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */ |
| 554 | I32 ml_anch; |
| 555 | register char *other_last = NULL; /* other substr checked before this */ |
| 556 | char *check_at = NULL; /* check substr found at this pos */ |
| 557 | const I32 multiline = prog->extflags & RXf_PMf_MULTILINE; |
| 558 | RXi_GET_DECL(prog,progi); |
| 559 | #ifdef DEBUGGING |
| 560 | const char * const i_strpos = strpos; |
| 561 | #endif |
| 562 | GET_RE_DEBUG_FLAGS_DECL; |
| 563 | |
| 564 | PERL_ARGS_ASSERT_RE_INTUIT_START; |
| 565 | |
| 566 | RX_MATCH_UTF8_set(rx,utf8_target); |
| 567 | |
| 568 | if (RX_UTF8(rx)) { |
| 569 | PL_reg_flags |= RF_utf8; |
| 570 | } |
| 571 | DEBUG_EXECUTE_r( |
| 572 | debug_start_match(rx, utf8_target, strpos, strend, |
| 573 | sv ? "Guessing start of match in sv for" |
| 574 | : "Guessing start of match in string for"); |
| 575 | ); |
| 576 | |
| 577 | /* CHR_DIST() would be more correct here but it makes things slow. */ |
| 578 | if (prog->minlen > strend - strpos) { |
| 579 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 580 | "String too short... [re_intuit_start]\n")); |
| 581 | goto fail; |
| 582 | } |
| 583 | |
| 584 | strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos; |
| 585 | PL_regeol = strend; |
| 586 | if (utf8_target) { |
| 587 | if (!prog->check_utf8 && prog->check_substr) |
| 588 | to_utf8_substr(prog); |
| 589 | check = prog->check_utf8; |
| 590 | } else { |
| 591 | if (!prog->check_substr && prog->check_utf8) |
| 592 | to_byte_substr(prog); |
| 593 | check = prog->check_substr; |
| 594 | } |
| 595 | if (check == &PL_sv_undef) { |
| 596 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 597 | "Non-utf8 string cannot match utf8 check string\n")); |
| 598 | goto fail; |
| 599 | } |
| 600 | if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */ |
| 601 | ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE) |
| 602 | || ( (prog->extflags & RXf_ANCH_BOL) |
| 603 | && !multiline ) ); /* Check after \n? */ |
| 604 | |
| 605 | if (!ml_anch) { |
| 606 | if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */ |
| 607 | && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */ |
| 608 | /* SvCUR is not set on references: SvRV and SvPVX_const overlap */ |
| 609 | && sv && !SvROK(sv) |
| 610 | && (strpos != strbeg)) { |
| 611 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n")); |
| 612 | goto fail; |
| 613 | } |
| 614 | if (prog->check_offset_min == prog->check_offset_max && |
| 615 | !(prog->extflags & RXf_CANY_SEEN)) { |
| 616 | /* Substring at constant offset from beg-of-str... */ |
| 617 | I32 slen; |
| 618 | |
| 619 | s = HOP3c(strpos, prog->check_offset_min, strend); |
| 620 | |
| 621 | if (SvTAIL(check)) { |
| 622 | slen = SvCUR(check); /* >= 1 */ |
| 623 | |
| 624 | if ( strend - s > slen || strend - s < slen - 1 |
| 625 | || (strend - s == slen && strend[-1] != '\n')) { |
| 626 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n")); |
| 627 | goto fail_finish; |
| 628 | } |
| 629 | /* Now should match s[0..slen-2] */ |
| 630 | slen--; |
| 631 | if (slen && (*SvPVX_const(check) != *s |
| 632 | || (slen > 1 |
| 633 | && memNE(SvPVX_const(check), s, slen)))) { |
| 634 | report_neq: |
| 635 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n")); |
| 636 | goto fail_finish; |
| 637 | } |
| 638 | } |
| 639 | else if (*SvPVX_const(check) != *s |
| 640 | || ((slen = SvCUR(check)) > 1 |
| 641 | && memNE(SvPVX_const(check), s, slen))) |
| 642 | goto report_neq; |
| 643 | check_at = s; |
| 644 | goto success_at_start; |
| 645 | } |
| 646 | } |
| 647 | /* Match is anchored, but substr is not anchored wrt beg-of-str. */ |
| 648 | s = strpos; |
| 649 | start_shift = prog->check_offset_min; /* okay to underestimate on CC */ |
| 650 | end_shift = prog->check_end_shift; |
| 651 | |
| 652 | if (!ml_anch) { |
| 653 | const I32 end = prog->check_offset_max + CHR_SVLEN(check) |
| 654 | - (SvTAIL(check) != 0); |
| 655 | const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end; |
| 656 | |
| 657 | if (end_shift < eshift) |
| 658 | end_shift = eshift; |
| 659 | } |
| 660 | } |
| 661 | else { /* Can match at random position */ |
| 662 | ml_anch = 0; |
| 663 | s = strpos; |
| 664 | start_shift = prog->check_offset_min; /* okay to underestimate on CC */ |
| 665 | end_shift = prog->check_end_shift; |
| 666 | |
| 667 | /* end shift should be non negative here */ |
| 668 | } |
| 669 | |
| 670 | #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */ |
| 671 | if (end_shift < 0) |
| 672 | Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ", |
| 673 | (IV)end_shift, RX_PRECOMP(prog)); |
| 674 | #endif |
| 675 | |
| 676 | restart: |
| 677 | /* Find a possible match in the region s..strend by looking for |
| 678 | the "check" substring in the region corrected by start/end_shift. */ |
| 679 | |
| 680 | { |
| 681 | I32 srch_start_shift = start_shift; |
| 682 | I32 srch_end_shift = end_shift; |
| 683 | if (srch_start_shift < 0 && strbeg - s > srch_start_shift) { |
| 684 | srch_end_shift -= ((strbeg - s) - srch_start_shift); |
| 685 | srch_start_shift = strbeg - s; |
| 686 | } |
| 687 | DEBUG_OPTIMISE_MORE_r({ |
| 688 | PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n", |
| 689 | (IV)prog->check_offset_min, |
| 690 | (IV)srch_start_shift, |
| 691 | (IV)srch_end_shift, |
| 692 | (IV)prog->check_end_shift); |
| 693 | }); |
| 694 | |
| 695 | if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) { |
| 696 | I32 p = -1; /* Internal iterator of scream. */ |
| 697 | I32 * const pp = data ? data->scream_pos : &p; |
| 698 | const MAGIC *mg; |
| 699 | bool found = FALSE; |
| 700 | |
| 701 | assert(SvMAGICAL(sv)); |
| 702 | mg = mg_find(sv, PERL_MAGIC_study); |
| 703 | assert(mg); |
| 704 | |
| 705 | if (mg->mg_private == 1) { |
| 706 | found = ((U8 *)mg->mg_ptr)[BmRARE(check)] != (U8)~0; |
| 707 | } else if (mg->mg_private == 2) { |
| 708 | found = ((U16 *)mg->mg_ptr)[BmRARE(check)] != (U16)~0; |
| 709 | } else { |
| 710 | assert (mg->mg_private == 4); |
| 711 | found = ((U32 *)mg->mg_ptr)[BmRARE(check)] != (U32)~0; |
| 712 | } |
| 713 | |
| 714 | if (found |
| 715 | || ( BmRARE(check) == '\n' |
| 716 | && (BmPREVIOUS(check) == SvCUR(check) - 1) |
| 717 | && SvTAIL(check) )) |
| 718 | s = screaminstr(sv, check, |
| 719 | srch_start_shift + (s - strbeg), srch_end_shift, pp, 0); |
| 720 | else |
| 721 | goto fail_finish; |
| 722 | /* we may be pointing at the wrong string */ |
| 723 | if (s && RXp_MATCH_COPIED(prog)) |
| 724 | s = strbeg + (s - SvPVX_const(sv)); |
| 725 | if (data) |
| 726 | *data->scream_olds = s; |
| 727 | } |
| 728 | else { |
| 729 | U8* start_point; |
| 730 | U8* end_point; |
| 731 | if (prog->extflags & RXf_CANY_SEEN) { |
| 732 | start_point= (U8*)(s + srch_start_shift); |
| 733 | end_point= (U8*)(strend - srch_end_shift); |
| 734 | } else { |
| 735 | start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend); |
| 736 | end_point= HOP3(strend, -srch_end_shift, strbeg); |
| 737 | } |
| 738 | DEBUG_OPTIMISE_MORE_r({ |
| 739 | PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n", |
| 740 | (int)(end_point - start_point), |
| 741 | (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point), |
| 742 | start_point); |
| 743 | }); |
| 744 | |
| 745 | s = fbm_instr( start_point, end_point, |
| 746 | check, multiline ? FBMrf_MULTILINE : 0); |
| 747 | } |
| 748 | } |
| 749 | /* Update the count-of-usability, remove useless subpatterns, |
| 750 | unshift s. */ |
| 751 | |
| 752 | DEBUG_EXECUTE_r({ |
| 753 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), |
| 754 | SvPVX_const(check), RE_SV_DUMPLEN(check), 30); |
| 755 | PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s", |
| 756 | (s ? "Found" : "Did not find"), |
| 757 | (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) |
| 758 | ? "anchored" : "floating"), |
| 759 | quoted, |
| 760 | RE_SV_TAIL(check), |
| 761 | (s ? " at offset " : "...\n") ); |
| 762 | }); |
| 763 | |
| 764 | if (!s) |
| 765 | goto fail_finish; |
| 766 | /* Finish the diagnostic message */ |
| 767 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) ); |
| 768 | |
| 769 | /* XXX dmq: first branch is for positive lookbehind... |
| 770 | Our check string is offset from the beginning of the pattern. |
| 771 | So we need to do any stclass tests offset forward from that |
| 772 | point. I think. :-( |
| 773 | */ |
| 774 | |
| 775 | |
| 776 | |
| 777 | check_at=s; |
| 778 | |
| 779 | |
| 780 | /* Got a candidate. Check MBOL anchoring, and the *other* substr. |
| 781 | Start with the other substr. |
| 782 | XXXX no SCREAM optimization yet - and a very coarse implementation |
| 783 | XXXX /ttx+/ results in anchored="ttx", floating="x". floating will |
| 784 | *always* match. Probably should be marked during compile... |
| 785 | Probably it is right to do no SCREAM here... |
| 786 | */ |
| 787 | |
| 788 | if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8) |
| 789 | : (prog->float_substr && prog->anchored_substr)) |
| 790 | { |
| 791 | /* Take into account the "other" substring. */ |
| 792 | /* XXXX May be hopelessly wrong for UTF... */ |
| 793 | if (!other_last) |
| 794 | other_last = strpos; |
| 795 | if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) { |
| 796 | do_other_anchored: |
| 797 | { |
| 798 | char * const last = HOP3c(s, -start_shift, strbeg); |
| 799 | char *last1, *last2; |
| 800 | char * const saved_s = s; |
| 801 | SV* must; |
| 802 | |
| 803 | t = s - prog->check_offset_max; |
| 804 | if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */ |
| 805 | && (!utf8_target |
| 806 | || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos)) |
| 807 | && t > strpos))) |
| 808 | NOOP; |
| 809 | else |
| 810 | t = strpos; |
| 811 | t = HOP3c(t, prog->anchored_offset, strend); |
| 812 | if (t < other_last) /* These positions already checked */ |
| 813 | t = other_last; |
| 814 | last2 = last1 = HOP3c(strend, -prog->minlen, strbeg); |
| 815 | if (last < last1) |
| 816 | last1 = last; |
| 817 | /* XXXX It is not documented what units *_offsets are in. |
| 818 | We assume bytes, but this is clearly wrong. |
| 819 | Meaning this code needs to be carefully reviewed for errors. |
| 820 | dmq. |
| 821 | */ |
| 822 | |
| 823 | /* On end-of-str: see comment below. */ |
| 824 | must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr; |
| 825 | if (must == &PL_sv_undef) { |
| 826 | s = (char*)NULL; |
| 827 | DEBUG_r(must = prog->anchored_utf8); /* for debug */ |
| 828 | } |
| 829 | else |
| 830 | s = fbm_instr( |
| 831 | (unsigned char*)t, |
| 832 | HOP3(HOP3(last1, prog->anchored_offset, strend) |
| 833 | + SvCUR(must), -(SvTAIL(must)!=0), strbeg), |
| 834 | must, |
| 835 | multiline ? FBMrf_MULTILINE : 0 |
| 836 | ); |
| 837 | DEBUG_EXECUTE_r({ |
| 838 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), |
| 839 | SvPVX_const(must), RE_SV_DUMPLEN(must), 30); |
| 840 | PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s", |
| 841 | (s ? "Found" : "Contradicts"), |
| 842 | quoted, RE_SV_TAIL(must)); |
| 843 | }); |
| 844 | |
| 845 | |
| 846 | if (!s) { |
| 847 | if (last1 >= last2) { |
| 848 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 849 | ", giving up...\n")); |
| 850 | goto fail_finish; |
| 851 | } |
| 852 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 853 | ", trying floating at offset %ld...\n", |
| 854 | (long)(HOP3c(saved_s, 1, strend) - i_strpos))); |
| 855 | other_last = HOP3c(last1, prog->anchored_offset+1, strend); |
| 856 | s = HOP3c(last, 1, strend); |
| 857 | goto restart; |
| 858 | } |
| 859 | else { |
| 860 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n", |
| 861 | (long)(s - i_strpos))); |
| 862 | t = HOP3c(s, -prog->anchored_offset, strbeg); |
| 863 | other_last = HOP3c(s, 1, strend); |
| 864 | s = saved_s; |
| 865 | if (t == strpos) |
| 866 | goto try_at_start; |
| 867 | goto try_at_offset; |
| 868 | } |
| 869 | } |
| 870 | } |
| 871 | else { /* Take into account the floating substring. */ |
| 872 | char *last, *last1; |
| 873 | char * const saved_s = s; |
| 874 | SV* must; |
| 875 | |
| 876 | t = HOP3c(s, -start_shift, strbeg); |
| 877 | last1 = last = |
| 878 | HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg); |
| 879 | if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset) |
| 880 | last = HOP3c(t, prog->float_max_offset, strend); |
| 881 | s = HOP3c(t, prog->float_min_offset, strend); |
| 882 | if (s < other_last) |
| 883 | s = other_last; |
| 884 | /* XXXX It is not documented what units *_offsets are in. Assume bytes. */ |
| 885 | must = utf8_target ? prog->float_utf8 : prog->float_substr; |
| 886 | /* fbm_instr() takes into account exact value of end-of-str |
| 887 | if the check is SvTAIL(ed). Since false positives are OK, |
| 888 | and end-of-str is not later than strend we are OK. */ |
| 889 | if (must == &PL_sv_undef) { |
| 890 | s = (char*)NULL; |
| 891 | DEBUG_r(must = prog->float_utf8); /* for debug message */ |
| 892 | } |
| 893 | else |
| 894 | s = fbm_instr((unsigned char*)s, |
| 895 | (unsigned char*)last + SvCUR(must) |
| 896 | - (SvTAIL(must)!=0), |
| 897 | must, multiline ? FBMrf_MULTILINE : 0); |
| 898 | DEBUG_EXECUTE_r({ |
| 899 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), |
| 900 | SvPVX_const(must), RE_SV_DUMPLEN(must), 30); |
| 901 | PerlIO_printf(Perl_debug_log, "%s floating substr %s%s", |
| 902 | (s ? "Found" : "Contradicts"), |
| 903 | quoted, RE_SV_TAIL(must)); |
| 904 | }); |
| 905 | if (!s) { |
| 906 | if (last1 == last) { |
| 907 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 908 | ", giving up...\n")); |
| 909 | goto fail_finish; |
| 910 | } |
| 911 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 912 | ", trying anchored starting at offset %ld...\n", |
| 913 | (long)(saved_s + 1 - i_strpos))); |
| 914 | other_last = last; |
| 915 | s = HOP3c(t, 1, strend); |
| 916 | goto restart; |
| 917 | } |
| 918 | else { |
| 919 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n", |
| 920 | (long)(s - i_strpos))); |
| 921 | other_last = s; /* Fix this later. --Hugo */ |
| 922 | s = saved_s; |
| 923 | if (t == strpos) |
| 924 | goto try_at_start; |
| 925 | goto try_at_offset; |
| 926 | } |
| 927 | } |
| 928 | } |
| 929 | |
| 930 | |
| 931 | t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos); |
| 932 | |
| 933 | DEBUG_OPTIMISE_MORE_r( |
| 934 | PerlIO_printf(Perl_debug_log, |
| 935 | "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n", |
| 936 | (IV)prog->check_offset_min, |
| 937 | (IV)prog->check_offset_max, |
| 938 | (IV)(s-strpos), |
| 939 | (IV)(t-strpos), |
| 940 | (IV)(t-s), |
| 941 | (IV)(strend-strpos) |
| 942 | ) |
| 943 | ); |
| 944 | |
| 945 | if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */ |
| 946 | && (!utf8_target |
| 947 | || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos))) |
| 948 | && t > strpos))) |
| 949 | { |
| 950 | /* Fixed substring is found far enough so that the match |
| 951 | cannot start at strpos. */ |
| 952 | try_at_offset: |
| 953 | if (ml_anch && t[-1] != '\n') { |
| 954 | /* Eventually fbm_*() should handle this, but often |
| 955 | anchored_offset is not 0, so this check will not be wasted. */ |
| 956 | /* XXXX In the code below we prefer to look for "^" even in |
| 957 | presence of anchored substrings. And we search even |
| 958 | beyond the found float position. These pessimizations |
| 959 | are historical artefacts only. */ |
| 960 | find_anchor: |
| 961 | while (t < strend - prog->minlen) { |
| 962 | if (*t == '\n') { |
| 963 | if (t < check_at - prog->check_offset_min) { |
| 964 | if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) { |
| 965 | /* Since we moved from the found position, |
| 966 | we definitely contradict the found anchored |
| 967 | substr. Due to the above check we do not |
| 968 | contradict "check" substr. |
| 969 | Thus we can arrive here only if check substr |
| 970 | is float. Redo checking for "other"=="fixed". |
| 971 | */ |
| 972 | strpos = t + 1; |
| 973 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n", |
| 974 | PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset))); |
| 975 | goto do_other_anchored; |
| 976 | } |
| 977 | /* We don't contradict the found floating substring. */ |
| 978 | /* XXXX Why not check for STCLASS? */ |
| 979 | s = t + 1; |
| 980 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n", |
| 981 | PL_colors[0], PL_colors[1], (long)(s - i_strpos))); |
| 982 | goto set_useful; |
| 983 | } |
| 984 | /* Position contradicts check-string */ |
| 985 | /* XXXX probably better to look for check-string |
| 986 | than for "\n", so one should lower the limit for t? */ |
| 987 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n", |
| 988 | PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos))); |
| 989 | other_last = strpos = s = t + 1; |
| 990 | goto restart; |
| 991 | } |
| 992 | t++; |
| 993 | } |
| 994 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n", |
| 995 | PL_colors[0], PL_colors[1])); |
| 996 | goto fail_finish; |
| 997 | } |
| 998 | else { |
| 999 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n", |
| 1000 | PL_colors[0], PL_colors[1])); |
| 1001 | } |
| 1002 | s = t; |
| 1003 | set_useful: |
| 1004 | ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */ |
| 1005 | } |
| 1006 | else { |
| 1007 | /* The found string does not prohibit matching at strpos, |
| 1008 | - no optimization of calling REx engine can be performed, |
| 1009 | unless it was an MBOL and we are not after MBOL, |
| 1010 | or a future STCLASS check will fail this. */ |
| 1011 | try_at_start: |
| 1012 | /* Even in this situation we may use MBOL flag if strpos is offset |
| 1013 | wrt the start of the string. */ |
| 1014 | if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */ |
| 1015 | && (strpos != strbeg) && strpos[-1] != '\n' |
| 1016 | /* May be due to an implicit anchor of m{.*foo} */ |
| 1017 | && !(prog->intflags & PREGf_IMPLICIT)) |
| 1018 | { |
| 1019 | t = strpos; |
| 1020 | goto find_anchor; |
| 1021 | } |
| 1022 | DEBUG_EXECUTE_r( if (ml_anch) |
| 1023 | PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n", |
| 1024 | (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]); |
| 1025 | ); |
| 1026 | success_at_start: |
| 1027 | if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */ |
| 1028 | && (utf8_target ? ( |
| 1029 | prog->check_utf8 /* Could be deleted already */ |
| 1030 | && --BmUSEFUL(prog->check_utf8) < 0 |
| 1031 | && (prog->check_utf8 == prog->float_utf8) |
| 1032 | ) : ( |
| 1033 | prog->check_substr /* Could be deleted already */ |
| 1034 | && --BmUSEFUL(prog->check_substr) < 0 |
| 1035 | && (prog->check_substr == prog->float_substr) |
| 1036 | ))) |
| 1037 | { |
| 1038 | /* If flags & SOMETHING - do not do it many times on the same match */ |
| 1039 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n")); |
| 1040 | /* XXX Does the destruction order has to change with utf8_target? */ |
| 1041 | SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr); |
| 1042 | SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8); |
| 1043 | prog->check_substr = prog->check_utf8 = NULL; /* disable */ |
| 1044 | prog->float_substr = prog->float_utf8 = NULL; /* clear */ |
| 1045 | check = NULL; /* abort */ |
| 1046 | s = strpos; |
| 1047 | /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag |
| 1048 | see http://bugs.activestate.com/show_bug.cgi?id=87173 */ |
| 1049 | if (prog->intflags & PREGf_IMPLICIT) |
| 1050 | prog->extflags &= ~RXf_ANCH_MBOL; |
| 1051 | /* XXXX This is a remnant of the old implementation. It |
| 1052 | looks wasteful, since now INTUIT can use many |
| 1053 | other heuristics. */ |
| 1054 | prog->extflags &= ~RXf_USE_INTUIT; |
| 1055 | /* XXXX What other flags might need to be cleared in this branch? */ |
| 1056 | } |
| 1057 | else |
| 1058 | s = strpos; |
| 1059 | } |
| 1060 | |
| 1061 | /* Last resort... */ |
| 1062 | /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */ |
| 1063 | /* trie stclasses are too expensive to use here, we are better off to |
| 1064 | leave it to regmatch itself */ |
| 1065 | if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) { |
| 1066 | /* minlen == 0 is possible if regstclass is \b or \B, |
| 1067 | and the fixed substr is ''$. |
| 1068 | Since minlen is already taken into account, s+1 is before strend; |
| 1069 | accidentally, minlen >= 1 guaranties no false positives at s + 1 |
| 1070 | even for \b or \B. But (minlen? 1 : 0) below assumes that |
| 1071 | regstclass does not come from lookahead... */ |
| 1072 | /* If regstclass takes bytelength more than 1: If charlength==1, OK. |
| 1073 | This leaves EXACTF-ish only, which are dealt with in find_byclass(). */ |
| 1074 | const U8* const str = (U8*)STRING(progi->regstclass); |
| 1075 | const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT |
| 1076 | ? CHR_DIST(str+STR_LEN(progi->regstclass), str) |
| 1077 | : 1); |
| 1078 | char * endpos; |
| 1079 | if (prog->anchored_substr || prog->anchored_utf8 || ml_anch) |
| 1080 | endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend); |
| 1081 | else if (prog->float_substr || prog->float_utf8) |
| 1082 | endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend); |
| 1083 | else |
| 1084 | endpos= strend; |
| 1085 | |
| 1086 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf"\n", |
| 1087 | (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg))); |
| 1088 | |
| 1089 | t = s; |
| 1090 | s = find_byclass(prog, progi->regstclass, s, endpos, NULL); |
| 1091 | if (!s) { |
| 1092 | #ifdef DEBUGGING |
| 1093 | const char *what = NULL; |
| 1094 | #endif |
| 1095 | if (endpos == strend) { |
| 1096 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 1097 | "Could not match STCLASS...\n") ); |
| 1098 | goto fail; |
| 1099 | } |
| 1100 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 1101 | "This position contradicts STCLASS...\n") ); |
| 1102 | if ((prog->extflags & RXf_ANCH) && !ml_anch) |
| 1103 | goto fail; |
| 1104 | /* Contradict one of substrings */ |
| 1105 | if (prog->anchored_substr || prog->anchored_utf8) { |
| 1106 | if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) { |
| 1107 | DEBUG_EXECUTE_r( what = "anchored" ); |
| 1108 | hop_and_restart: |
| 1109 | s = HOP3c(t, 1, strend); |
| 1110 | if (s + start_shift + end_shift > strend) { |
| 1111 | /* XXXX Should be taken into account earlier? */ |
| 1112 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 1113 | "Could not match STCLASS...\n") ); |
| 1114 | goto fail; |
| 1115 | } |
| 1116 | if (!check) |
| 1117 | goto giveup; |
| 1118 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 1119 | "Looking for %s substr starting at offset %ld...\n", |
| 1120 | what, (long)(s + start_shift - i_strpos)) ); |
| 1121 | goto restart; |
| 1122 | } |
| 1123 | /* Have both, check_string is floating */ |
| 1124 | if (t + start_shift >= check_at) /* Contradicts floating=check */ |
| 1125 | goto retry_floating_check; |
| 1126 | /* Recheck anchored substring, but not floating... */ |
| 1127 | s = check_at; |
| 1128 | if (!check) |
| 1129 | goto giveup; |
| 1130 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 1131 | "Looking for anchored substr starting at offset %ld...\n", |
| 1132 | (long)(other_last - i_strpos)) ); |
| 1133 | goto do_other_anchored; |
| 1134 | } |
| 1135 | /* Another way we could have checked stclass at the |
| 1136 | current position only: */ |
| 1137 | if (ml_anch) { |
| 1138 | s = t = t + 1; |
| 1139 | if (!check) |
| 1140 | goto giveup; |
| 1141 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 1142 | "Looking for /%s^%s/m starting at offset %ld...\n", |
| 1143 | PL_colors[0], PL_colors[1], (long)(t - i_strpos)) ); |
| 1144 | goto try_at_offset; |
| 1145 | } |
| 1146 | if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */ |
| 1147 | goto fail; |
| 1148 | /* Check is floating substring. */ |
| 1149 | retry_floating_check: |
| 1150 | t = check_at - start_shift; |
| 1151 | DEBUG_EXECUTE_r( what = "floating" ); |
| 1152 | goto hop_and_restart; |
| 1153 | } |
| 1154 | if (t != s) { |
| 1155 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 1156 | "By STCLASS: moving %ld --> %ld\n", |
| 1157 | (long)(t - i_strpos), (long)(s - i_strpos)) |
| 1158 | ); |
| 1159 | } |
| 1160 | else { |
| 1161 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 1162 | "Does not contradict STCLASS...\n"); |
| 1163 | ); |
| 1164 | } |
| 1165 | } |
| 1166 | giveup: |
| 1167 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n", |
| 1168 | PL_colors[4], (check ? "Guessed" : "Giving up"), |
| 1169 | PL_colors[5], (long)(s - i_strpos)) ); |
| 1170 | return s; |
| 1171 | |
| 1172 | fail_finish: /* Substring not found */ |
| 1173 | if (prog->check_substr || prog->check_utf8) /* could be removed already */ |
| 1174 | BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */ |
| 1175 | fail: |
| 1176 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n", |
| 1177 | PL_colors[4], PL_colors[5])); |
| 1178 | return NULL; |
| 1179 | } |
| 1180 | |
| 1181 | #define DECL_TRIE_TYPE(scan) \ |
| 1182 | const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \ |
| 1183 | trie_type = (scan->flags != EXACT) \ |
| 1184 | ? (utf8_target ? trie_utf8_fold : (UTF_PATTERN ? trie_latin_utf8_fold : trie_plain)) \ |
| 1185 | : (utf8_target ? trie_utf8 : trie_plain) |
| 1186 | |
| 1187 | #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \ |
| 1188 | uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \ |
| 1189 | switch (trie_type) { \ |
| 1190 | case trie_utf8_fold: \ |
| 1191 | if ( foldlen>0 ) { \ |
| 1192 | uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \ |
| 1193 | foldlen -= len; \ |
| 1194 | uscan += len; \ |
| 1195 | len=0; \ |
| 1196 | } else { \ |
| 1197 | uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \ |
| 1198 | uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \ |
| 1199 | foldlen -= UNISKIP( uvc ); \ |
| 1200 | uscan = foldbuf + UNISKIP( uvc ); \ |
| 1201 | } \ |
| 1202 | break; \ |
| 1203 | case trie_latin_utf8_fold: \ |
| 1204 | if ( foldlen>0 ) { \ |
| 1205 | uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \ |
| 1206 | foldlen -= len; \ |
| 1207 | uscan += len; \ |
| 1208 | len=0; \ |
| 1209 | } else { \ |
| 1210 | len = 1; \ |
| 1211 | uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \ |
| 1212 | foldlen -= UNISKIP( uvc ); \ |
| 1213 | uscan = foldbuf + UNISKIP( uvc ); \ |
| 1214 | } \ |
| 1215 | break; \ |
| 1216 | case trie_utf8: \ |
| 1217 | uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \ |
| 1218 | break; \ |
| 1219 | case trie_plain: \ |
| 1220 | uvc = (UV)*uc; \ |
| 1221 | len = 1; \ |
| 1222 | } \ |
| 1223 | if (uvc < 256) { \ |
| 1224 | charid = trie->charmap[ uvc ]; \ |
| 1225 | } \ |
| 1226 | else { \ |
| 1227 | charid = 0; \ |
| 1228 | if (widecharmap) { \ |
| 1229 | SV** const svpp = hv_fetch(widecharmap, \ |
| 1230 | (char*)&uvc, sizeof(UV), 0); \ |
| 1231 | if (svpp) \ |
| 1232 | charid = (U16)SvIV(*svpp); \ |
| 1233 | } \ |
| 1234 | } \ |
| 1235 | } STMT_END |
| 1236 | |
| 1237 | #define REXEC_FBC_EXACTISH_SCAN(CoNd) \ |
| 1238 | STMT_START { \ |
| 1239 | while (s <= e) { \ |
| 1240 | if ( (CoNd) \ |
| 1241 | && (ln == 1 || folder(s, pat_string, ln)) \ |
| 1242 | && (!reginfo || regtry(reginfo, &s)) ) \ |
| 1243 | goto got_it; \ |
| 1244 | s++; \ |
| 1245 | } \ |
| 1246 | } STMT_END |
| 1247 | |
| 1248 | #define REXEC_FBC_UTF8_SCAN(CoDe) \ |
| 1249 | STMT_START { \ |
| 1250 | while (s + (uskip = UTF8SKIP(s)) <= strend) { \ |
| 1251 | CoDe \ |
| 1252 | s += uskip; \ |
| 1253 | } \ |
| 1254 | } STMT_END |
| 1255 | |
| 1256 | #define REXEC_FBC_SCAN(CoDe) \ |
| 1257 | STMT_START { \ |
| 1258 | while (s < strend) { \ |
| 1259 | CoDe \ |
| 1260 | s++; \ |
| 1261 | } \ |
| 1262 | } STMT_END |
| 1263 | |
| 1264 | #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \ |
| 1265 | REXEC_FBC_UTF8_SCAN( \ |
| 1266 | if (CoNd) { \ |
| 1267 | if (tmp && (!reginfo || regtry(reginfo, &s))) \ |
| 1268 | goto got_it; \ |
| 1269 | else \ |
| 1270 | tmp = doevery; \ |
| 1271 | } \ |
| 1272 | else \ |
| 1273 | tmp = 1; \ |
| 1274 | ) |
| 1275 | |
| 1276 | #define REXEC_FBC_CLASS_SCAN(CoNd) \ |
| 1277 | REXEC_FBC_SCAN( \ |
| 1278 | if (CoNd) { \ |
| 1279 | if (tmp && (!reginfo || regtry(reginfo, &s))) \ |
| 1280 | goto got_it; \ |
| 1281 | else \ |
| 1282 | tmp = doevery; \ |
| 1283 | } \ |
| 1284 | else \ |
| 1285 | tmp = 1; \ |
| 1286 | ) |
| 1287 | |
| 1288 | #define REXEC_FBC_TRYIT \ |
| 1289 | if ((!reginfo || regtry(reginfo, &s))) \ |
| 1290 | goto got_it |
| 1291 | |
| 1292 | #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \ |
| 1293 | if (utf8_target) { \ |
| 1294 | REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \ |
| 1295 | } \ |
| 1296 | else { \ |
| 1297 | REXEC_FBC_CLASS_SCAN(CoNd); \ |
| 1298 | } |
| 1299 | |
| 1300 | #define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \ |
| 1301 | if (utf8_target) { \ |
| 1302 | UtFpReLoAd; \ |
| 1303 | REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \ |
| 1304 | } \ |
| 1305 | else { \ |
| 1306 | REXEC_FBC_CLASS_SCAN(CoNd); \ |
| 1307 | } |
| 1308 | |
| 1309 | #define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \ |
| 1310 | PL_reg_flags |= RF_tainted; \ |
| 1311 | if (utf8_target) { \ |
| 1312 | REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \ |
| 1313 | } \ |
| 1314 | else { \ |
| 1315 | REXEC_FBC_CLASS_SCAN(CoNd); \ |
| 1316 | } |
| 1317 | |
| 1318 | #define DUMP_EXEC_POS(li,s,doutf8) \ |
| 1319 | dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8) |
| 1320 | |
| 1321 | |
| 1322 | #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \ |
| 1323 | tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \ |
| 1324 | tmp = TEST_NON_UTF8(tmp); \ |
| 1325 | REXEC_FBC_UTF8_SCAN( \ |
| 1326 | if (tmp == ! TEST_NON_UTF8((U8) *s)) { \ |
| 1327 | tmp = !tmp; \ |
| 1328 | IF_SUCCESS; \ |
| 1329 | } \ |
| 1330 | else { \ |
| 1331 | IF_FAIL; \ |
| 1332 | } \ |
| 1333 | ); \ |
| 1334 | |
| 1335 | #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \ |
| 1336 | if (s == PL_bostr) { \ |
| 1337 | tmp = '\n'; \ |
| 1338 | } \ |
| 1339 | else { \ |
| 1340 | U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \ |
| 1341 | tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \ |
| 1342 | } \ |
| 1343 | tmp = TeSt1_UtF8; \ |
| 1344 | LOAD_UTF8_CHARCLASS_ALNUM(); \ |
| 1345 | REXEC_FBC_UTF8_SCAN( \ |
| 1346 | if (tmp == ! (TeSt2_UtF8)) { \ |
| 1347 | tmp = !tmp; \ |
| 1348 | IF_SUCCESS; \ |
| 1349 | } \ |
| 1350 | else { \ |
| 1351 | IF_FAIL; \ |
| 1352 | } \ |
| 1353 | ); \ |
| 1354 | |
| 1355 | /* The only difference between the BOUND and NBOUND cases is that |
| 1356 | * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in |
| 1357 | * NBOUND. This is accomplished by passing it in either the if or else clause, |
| 1358 | * with the other one being empty */ |
| 1359 | #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \ |
| 1360 | FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER) |
| 1361 | |
| 1362 | #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \ |
| 1363 | FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER) |
| 1364 | |
| 1365 | #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \ |
| 1366 | FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT) |
| 1367 | |
| 1368 | #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \ |
| 1369 | FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT) |
| 1370 | |
| 1371 | |
| 1372 | /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to |
| 1373 | * be passed in completely with the variable name being tested, which isn't |
| 1374 | * such a clean interface, but this is easier to read than it was before. We |
| 1375 | * are looking for the boundary (or non-boundary between a word and non-word |
| 1376 | * character. The utf8 and non-utf8 cases have the same logic, but the details |
| 1377 | * must be different. Find the "wordness" of the character just prior to this |
| 1378 | * one, and compare it with the wordness of this one. If they differ, we have |
| 1379 | * a boundary. At the beginning of the string, pretend that the previous |
| 1380 | * character was a new-line */ |
| 1381 | #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \ |
| 1382 | if (utf8_target) { \ |
| 1383 | UTF8_CODE \ |
| 1384 | } \ |
| 1385 | else { /* Not utf8 */ \ |
| 1386 | tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \ |
| 1387 | tmp = TEST_NON_UTF8(tmp); \ |
| 1388 | REXEC_FBC_SCAN( \ |
| 1389 | if (tmp == ! TEST_NON_UTF8((U8) *s)) { \ |
| 1390 | tmp = !tmp; \ |
| 1391 | IF_SUCCESS; \ |
| 1392 | } \ |
| 1393 | else { \ |
| 1394 | IF_FAIL; \ |
| 1395 | } \ |
| 1396 | ); \ |
| 1397 | } \ |
| 1398 | if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \ |
| 1399 | goto got_it; |
| 1400 | |
| 1401 | /* We know what class REx starts with. Try to find this position... */ |
| 1402 | /* if reginfo is NULL, its a dryrun */ |
| 1403 | /* annoyingly all the vars in this routine have different names from their counterparts |
| 1404 | in regmatch. /grrr */ |
| 1405 | |
| 1406 | STATIC char * |
| 1407 | S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, |
| 1408 | const char *strend, regmatch_info *reginfo) |
| 1409 | { |
| 1410 | dVAR; |
| 1411 | const I32 doevery = (prog->intflags & PREGf_SKIP) == 0; |
| 1412 | char *pat_string; /* The pattern's exactish string */ |
| 1413 | char *pat_end; /* ptr to end char of pat_string */ |
| 1414 | re_fold_t folder; /* Function for computing non-utf8 folds */ |
| 1415 | const U8 *fold_array; /* array for folding ords < 256 */ |
| 1416 | STRLEN ln; |
| 1417 | STRLEN lnc; |
| 1418 | register STRLEN uskip; |
| 1419 | U8 c1; |
| 1420 | U8 c2; |
| 1421 | char *e; |
| 1422 | register I32 tmp = 1; /* Scratch variable? */ |
| 1423 | register const bool utf8_target = PL_reg_match_utf8; |
| 1424 | UV utf8_fold_flags = 0; |
| 1425 | RXi_GET_DECL(prog,progi); |
| 1426 | |
| 1427 | PERL_ARGS_ASSERT_FIND_BYCLASS; |
| 1428 | |
| 1429 | /* We know what class it must start with. */ |
| 1430 | switch (OP(c)) { |
| 1431 | case ANYOFV: |
| 1432 | case ANYOF: |
| 1433 | if (utf8_target || OP(c) == ANYOFV) { |
| 1434 | STRLEN inclasslen = strend - s; |
| 1435 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 1436 | reginclass(prog, c, (U8*)s, &inclasslen, utf8_target)); |
| 1437 | } |
| 1438 | else { |
| 1439 | REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s)); |
| 1440 | } |
| 1441 | break; |
| 1442 | case CANY: |
| 1443 | REXEC_FBC_SCAN( |
| 1444 | if (tmp && (!reginfo || regtry(reginfo, &s))) |
| 1445 | goto got_it; |
| 1446 | else |
| 1447 | tmp = doevery; |
| 1448 | ); |
| 1449 | break; |
| 1450 | |
| 1451 | case EXACTFA: |
| 1452 | if (UTF_PATTERN || utf8_target) { |
| 1453 | utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 1454 | goto do_exactf_utf8; |
| 1455 | } |
| 1456 | fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */ |
| 1457 | folder = foldEQ_latin1; /* /a, except the sharp s one which */ |
| 1458 | goto do_exactf_non_utf8; /* isn't dealt with by these */ |
| 1459 | |
| 1460 | case EXACTFU: |
| 1461 | if (UTF_PATTERN || utf8_target) { |
| 1462 | utf8_fold_flags = 0; |
| 1463 | goto do_exactf_utf8; |
| 1464 | } |
| 1465 | |
| 1466 | /* Any 'ss' in the pattern should have been replaced by regcomp, |
| 1467 | * so we don't have to worry here about this single special case |
| 1468 | * in the Latin1 range */ |
| 1469 | fold_array = PL_fold_latin1; |
| 1470 | folder = foldEQ_latin1; |
| 1471 | goto do_exactf_non_utf8; |
| 1472 | |
| 1473 | case EXACTF: |
| 1474 | if (UTF_PATTERN || utf8_target) { |
| 1475 | utf8_fold_flags = 0; |
| 1476 | goto do_exactf_utf8; |
| 1477 | } |
| 1478 | fold_array = PL_fold; |
| 1479 | folder = foldEQ; |
| 1480 | goto do_exactf_non_utf8; |
| 1481 | |
| 1482 | case EXACTFL: |
| 1483 | if (UTF_PATTERN || utf8_target) { |
| 1484 | utf8_fold_flags = FOLDEQ_UTF8_LOCALE; |
| 1485 | goto do_exactf_utf8; |
| 1486 | } |
| 1487 | fold_array = PL_fold_locale; |
| 1488 | folder = foldEQ_locale; |
| 1489 | |
| 1490 | /* FALL THROUGH */ |
| 1491 | |
| 1492 | do_exactf_non_utf8: /* Neither pattern nor string are UTF8 */ |
| 1493 | |
| 1494 | /* The idea in the non-utf8 EXACTF* cases is to first find the |
| 1495 | * first character of the EXACTF* node and then, if necessary, |
| 1496 | * case-insensitively compare the full text of the node. c1 is the |
| 1497 | * first character. c2 is its fold. This logic will not work for |
| 1498 | * Unicode semantics and the german sharp ss, which hence should |
| 1499 | * not be compiled into a node that gets here. */ |
| 1500 | pat_string = STRING(c); |
| 1501 | ln = STR_LEN(c); /* length to match in octets/bytes */ |
| 1502 | |
| 1503 | e = HOP3c(strend, -((I32)ln), s); |
| 1504 | |
| 1505 | if (!reginfo && e < s) { |
| 1506 | e = s; /* Due to minlen logic of intuit() */ |
| 1507 | } |
| 1508 | |
| 1509 | c1 = *pat_string; |
| 1510 | c2 = fold_array[c1]; |
| 1511 | if (c1 == c2) { /* If char and fold are the same */ |
| 1512 | REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1); |
| 1513 | } |
| 1514 | else { |
| 1515 | REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2); |
| 1516 | } |
| 1517 | break; |
| 1518 | |
| 1519 | do_exactf_utf8: |
| 1520 | |
| 1521 | /* If one of the operands is in utf8, we can't use the simpler |
| 1522 | * folding above, due to the fact that many different characters |
| 1523 | * can have the same fold, or portion of a fold, or different- |
| 1524 | * length fold */ |
| 1525 | pat_string = STRING(c); |
| 1526 | ln = STR_LEN(c); /* length to match in octets/bytes */ |
| 1527 | pat_end = pat_string + ln; |
| 1528 | lnc = (UTF_PATTERN) /* length to match in characters */ |
| 1529 | ? utf8_length((U8 *) pat_string, (U8 *) pat_end) |
| 1530 | : ln; |
| 1531 | |
| 1532 | e = HOP3c(strend, -((I32)lnc), s); |
| 1533 | |
| 1534 | if (!reginfo && e < s) { |
| 1535 | e = s; /* Due to minlen logic of intuit() */ |
| 1536 | } |
| 1537 | |
| 1538 | while (s <= e) { |
| 1539 | char *my_strend= (char *)strend; |
| 1540 | if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target, |
| 1541 | pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags) |
| 1542 | && (!reginfo || regtry(reginfo, &s)) ) |
| 1543 | { |
| 1544 | goto got_it; |
| 1545 | } |
| 1546 | s += UTF8SKIP(s); |
| 1547 | } |
| 1548 | break; |
| 1549 | case BOUNDL: |
| 1550 | PL_reg_flags |= RF_tainted; |
| 1551 | FBC_BOUND(isALNUM_LC, |
| 1552 | isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)), |
| 1553 | isALNUM_LC_utf8((U8*)s)); |
| 1554 | break; |
| 1555 | case NBOUNDL: |
| 1556 | PL_reg_flags |= RF_tainted; |
| 1557 | FBC_NBOUND(isALNUM_LC, |
| 1558 | isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)), |
| 1559 | isALNUM_LC_utf8((U8*)s)); |
| 1560 | break; |
| 1561 | case BOUND: |
| 1562 | FBC_BOUND(isWORDCHAR, |
| 1563 | isALNUM_uni(tmp), |
| 1564 | cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target))); |
| 1565 | break; |
| 1566 | case BOUNDA: |
| 1567 | FBC_BOUND_NOLOAD(isWORDCHAR_A, |
| 1568 | isWORDCHAR_A(tmp), |
| 1569 | isWORDCHAR_A((U8*)s)); |
| 1570 | break; |
| 1571 | case NBOUND: |
| 1572 | FBC_NBOUND(isWORDCHAR, |
| 1573 | isALNUM_uni(tmp), |
| 1574 | cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target))); |
| 1575 | break; |
| 1576 | case NBOUNDA: |
| 1577 | FBC_NBOUND_NOLOAD(isWORDCHAR_A, |
| 1578 | isWORDCHAR_A(tmp), |
| 1579 | isWORDCHAR_A((U8*)s)); |
| 1580 | break; |
| 1581 | case BOUNDU: |
| 1582 | FBC_BOUND(isWORDCHAR_L1, |
| 1583 | isALNUM_uni(tmp), |
| 1584 | cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target))); |
| 1585 | break; |
| 1586 | case NBOUNDU: |
| 1587 | FBC_NBOUND(isWORDCHAR_L1, |
| 1588 | isALNUM_uni(tmp), |
| 1589 | cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target))); |
| 1590 | break; |
| 1591 | case ALNUML: |
| 1592 | REXEC_FBC_CSCAN_TAINT( |
| 1593 | isALNUM_LC_utf8((U8*)s), |
| 1594 | isALNUM_LC(*s) |
| 1595 | ); |
| 1596 | break; |
| 1597 | case ALNUMU: |
| 1598 | REXEC_FBC_CSCAN_PRELOAD( |
| 1599 | LOAD_UTF8_CHARCLASS_ALNUM(), |
| 1600 | swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target), |
| 1601 | isWORDCHAR_L1((U8) *s) |
| 1602 | ); |
| 1603 | break; |
| 1604 | case ALNUM: |
| 1605 | REXEC_FBC_CSCAN_PRELOAD( |
| 1606 | LOAD_UTF8_CHARCLASS_ALNUM(), |
| 1607 | swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target), |
| 1608 | isWORDCHAR((U8) *s) |
| 1609 | ); |
| 1610 | break; |
| 1611 | case ALNUMA: |
| 1612 | /* Don't need to worry about utf8, as it can match only a single |
| 1613 | * byte invariant character */ |
| 1614 | REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s)); |
| 1615 | break; |
| 1616 | case NALNUMU: |
| 1617 | REXEC_FBC_CSCAN_PRELOAD( |
| 1618 | LOAD_UTF8_CHARCLASS_ALNUM(), |
| 1619 | !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target), |
| 1620 | ! isWORDCHAR_L1((U8) *s) |
| 1621 | ); |
| 1622 | break; |
| 1623 | case NALNUM: |
| 1624 | REXEC_FBC_CSCAN_PRELOAD( |
| 1625 | LOAD_UTF8_CHARCLASS_ALNUM(), |
| 1626 | !swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target), |
| 1627 | ! isALNUM(*s) |
| 1628 | ); |
| 1629 | break; |
| 1630 | case NALNUMA: |
| 1631 | REXEC_FBC_CSCAN( |
| 1632 | !isWORDCHAR_A(*s), |
| 1633 | !isWORDCHAR_A(*s) |
| 1634 | ); |
| 1635 | break; |
| 1636 | case NALNUML: |
| 1637 | REXEC_FBC_CSCAN_TAINT( |
| 1638 | !isALNUM_LC_utf8((U8*)s), |
| 1639 | !isALNUM_LC(*s) |
| 1640 | ); |
| 1641 | break; |
| 1642 | case SPACEU: |
| 1643 | REXEC_FBC_CSCAN_PRELOAD( |
| 1644 | LOAD_UTF8_CHARCLASS_SPACE(), |
| 1645 | *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target), |
| 1646 | isSPACE_L1((U8) *s) |
| 1647 | ); |
| 1648 | break; |
| 1649 | case SPACE: |
| 1650 | REXEC_FBC_CSCAN_PRELOAD( |
| 1651 | LOAD_UTF8_CHARCLASS_SPACE(), |
| 1652 | *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target), |
| 1653 | isSPACE((U8) *s) |
| 1654 | ); |
| 1655 | break; |
| 1656 | case SPACEA: |
| 1657 | /* Don't need to worry about utf8, as it can match only a single |
| 1658 | * byte invariant character */ |
| 1659 | REXEC_FBC_CLASS_SCAN( isSPACE_A(*s)); |
| 1660 | break; |
| 1661 | case SPACEL: |
| 1662 | REXEC_FBC_CSCAN_TAINT( |
| 1663 | isSPACE_LC_utf8((U8*)s), |
| 1664 | isSPACE_LC(*s) |
| 1665 | ); |
| 1666 | break; |
| 1667 | case NSPACEU: |
| 1668 | REXEC_FBC_CSCAN_PRELOAD( |
| 1669 | LOAD_UTF8_CHARCLASS_SPACE(), |
| 1670 | !( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)), |
| 1671 | ! isSPACE_L1((U8) *s) |
| 1672 | ); |
| 1673 | break; |
| 1674 | case NSPACE: |
| 1675 | REXEC_FBC_CSCAN_PRELOAD( |
| 1676 | LOAD_UTF8_CHARCLASS_SPACE(), |
| 1677 | !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)), |
| 1678 | ! isSPACE((U8) *s) |
| 1679 | ); |
| 1680 | break; |
| 1681 | case NSPACEA: |
| 1682 | REXEC_FBC_CSCAN( |
| 1683 | !isSPACE_A(*s), |
| 1684 | !isSPACE_A(*s) |
| 1685 | ); |
| 1686 | break; |
| 1687 | case NSPACEL: |
| 1688 | REXEC_FBC_CSCAN_TAINT( |
| 1689 | !isSPACE_LC_utf8((U8*)s), |
| 1690 | !isSPACE_LC(*s) |
| 1691 | ); |
| 1692 | break; |
| 1693 | case DIGIT: |
| 1694 | REXEC_FBC_CSCAN_PRELOAD( |
| 1695 | LOAD_UTF8_CHARCLASS_DIGIT(), |
| 1696 | swash_fetch(PL_utf8_digit,(U8*)s, utf8_target), |
| 1697 | isDIGIT(*s) |
| 1698 | ); |
| 1699 | break; |
| 1700 | case DIGITA: |
| 1701 | /* Don't need to worry about utf8, as it can match only a single |
| 1702 | * byte invariant character */ |
| 1703 | REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s)); |
| 1704 | break; |
| 1705 | case DIGITL: |
| 1706 | REXEC_FBC_CSCAN_TAINT( |
| 1707 | isDIGIT_LC_utf8((U8*)s), |
| 1708 | isDIGIT_LC(*s) |
| 1709 | ); |
| 1710 | break; |
| 1711 | case NDIGIT: |
| 1712 | REXEC_FBC_CSCAN_PRELOAD( |
| 1713 | LOAD_UTF8_CHARCLASS_DIGIT(), |
| 1714 | !swash_fetch(PL_utf8_digit,(U8*)s, utf8_target), |
| 1715 | !isDIGIT(*s) |
| 1716 | ); |
| 1717 | break; |
| 1718 | case NDIGITA: |
| 1719 | REXEC_FBC_CSCAN( |
| 1720 | !isDIGIT_A(*s), |
| 1721 | !isDIGIT_A(*s) |
| 1722 | ); |
| 1723 | break; |
| 1724 | case NDIGITL: |
| 1725 | REXEC_FBC_CSCAN_TAINT( |
| 1726 | !isDIGIT_LC_utf8((U8*)s), |
| 1727 | !isDIGIT_LC(*s) |
| 1728 | ); |
| 1729 | break; |
| 1730 | case LNBREAK: |
| 1731 | REXEC_FBC_CSCAN( |
| 1732 | is_LNBREAK_utf8(s), |
| 1733 | is_LNBREAK_latin1(s) |
| 1734 | ); |
| 1735 | break; |
| 1736 | case VERTWS: |
| 1737 | REXEC_FBC_CSCAN( |
| 1738 | is_VERTWS_utf8(s), |
| 1739 | is_VERTWS_latin1(s) |
| 1740 | ); |
| 1741 | break; |
| 1742 | case NVERTWS: |
| 1743 | REXEC_FBC_CSCAN( |
| 1744 | !is_VERTWS_utf8(s), |
| 1745 | !is_VERTWS_latin1(s) |
| 1746 | ); |
| 1747 | break; |
| 1748 | case HORIZWS: |
| 1749 | REXEC_FBC_CSCAN( |
| 1750 | is_HORIZWS_utf8(s), |
| 1751 | is_HORIZWS_latin1(s) |
| 1752 | ); |
| 1753 | break; |
| 1754 | case NHORIZWS: |
| 1755 | REXEC_FBC_CSCAN( |
| 1756 | !is_HORIZWS_utf8(s), |
| 1757 | !is_HORIZWS_latin1(s) |
| 1758 | ); |
| 1759 | break; |
| 1760 | case AHOCORASICKC: |
| 1761 | case AHOCORASICK: |
| 1762 | { |
| 1763 | DECL_TRIE_TYPE(c); |
| 1764 | /* what trie are we using right now */ |
| 1765 | reg_ac_data *aho |
| 1766 | = (reg_ac_data*)progi->data->data[ ARG( c ) ]; |
| 1767 | reg_trie_data *trie |
| 1768 | = (reg_trie_data*)progi->data->data[ aho->trie ]; |
| 1769 | HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]); |
| 1770 | |
| 1771 | const char *last_start = strend - trie->minlen; |
| 1772 | #ifdef DEBUGGING |
| 1773 | const char *real_start = s; |
| 1774 | #endif |
| 1775 | STRLEN maxlen = trie->maxlen; |
| 1776 | SV *sv_points; |
| 1777 | U8 **points; /* map of where we were in the input string |
| 1778 | when reading a given char. For ASCII this |
| 1779 | is unnecessary overhead as the relationship |
| 1780 | is always 1:1, but for Unicode, especially |
| 1781 | case folded Unicode this is not true. */ |
| 1782 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
| 1783 | U8 *bitmap=NULL; |
| 1784 | |
| 1785 | |
| 1786 | GET_RE_DEBUG_FLAGS_DECL; |
| 1787 | |
| 1788 | /* We can't just allocate points here. We need to wrap it in |
| 1789 | * an SV so it gets freed properly if there is a croak while |
| 1790 | * running the match */ |
| 1791 | ENTER; |
| 1792 | SAVETMPS; |
| 1793 | sv_points=newSV(maxlen * sizeof(U8 *)); |
| 1794 | SvCUR_set(sv_points, |
| 1795 | maxlen * sizeof(U8 *)); |
| 1796 | SvPOK_on(sv_points); |
| 1797 | sv_2mortal(sv_points); |
| 1798 | points=(U8**)SvPV_nolen(sv_points ); |
| 1799 | if ( trie_type != trie_utf8_fold |
| 1800 | && (trie->bitmap || OP(c)==AHOCORASICKC) ) |
| 1801 | { |
| 1802 | if (trie->bitmap) |
| 1803 | bitmap=(U8*)trie->bitmap; |
| 1804 | else |
| 1805 | bitmap=(U8*)ANYOF_BITMAP(c); |
| 1806 | } |
| 1807 | /* this is the Aho-Corasick algorithm modified a touch |
| 1808 | to include special handling for long "unknown char" |
| 1809 | sequences. The basic idea being that we use AC as long |
| 1810 | as we are dealing with a possible matching char, when |
| 1811 | we encounter an unknown char (and we have not encountered |
| 1812 | an accepting state) we scan forward until we find a legal |
| 1813 | starting char. |
| 1814 | AC matching is basically that of trie matching, except |
| 1815 | that when we encounter a failing transition, we fall back |
| 1816 | to the current states "fail state", and try the current char |
| 1817 | again, a process we repeat until we reach the root state, |
| 1818 | state 1, or a legal transition. If we fail on the root state |
| 1819 | then we can either terminate if we have reached an accepting |
| 1820 | state previously, or restart the entire process from the beginning |
| 1821 | if we have not. |
| 1822 | |
| 1823 | */ |
| 1824 | while (s <= last_start) { |
| 1825 | const U32 uniflags = UTF8_ALLOW_DEFAULT; |
| 1826 | U8 *uc = (U8*)s; |
| 1827 | U16 charid = 0; |
| 1828 | U32 base = 1; |
| 1829 | U32 state = 1; |
| 1830 | UV uvc = 0; |
| 1831 | STRLEN len = 0; |
| 1832 | STRLEN foldlen = 0; |
| 1833 | U8 *uscan = (U8*)NULL; |
| 1834 | U8 *leftmost = NULL; |
| 1835 | #ifdef DEBUGGING |
| 1836 | U32 accepted_word= 0; |
| 1837 | #endif |
| 1838 | U32 pointpos = 0; |
| 1839 | |
| 1840 | while ( state && uc <= (U8*)strend ) { |
| 1841 | int failed=0; |
| 1842 | U32 word = aho->states[ state ].wordnum; |
| 1843 | |
| 1844 | if( state==1 ) { |
| 1845 | if ( bitmap ) { |
| 1846 | DEBUG_TRIE_EXECUTE_r( |
| 1847 | if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { |
| 1848 | dump_exec_pos( (char *)uc, c, strend, real_start, |
| 1849 | (char *)uc, utf8_target ); |
| 1850 | PerlIO_printf( Perl_debug_log, |
| 1851 | " Scanning for legal start char...\n"); |
| 1852 | } |
| 1853 | ); |
| 1854 | if (utf8_target) { |
| 1855 | while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { |
| 1856 | uc += UTF8SKIP(uc); |
| 1857 | } |
| 1858 | } else { |
| 1859 | while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { |
| 1860 | uc++; |
| 1861 | } |
| 1862 | } |
| 1863 | s= (char *)uc; |
| 1864 | } |
| 1865 | if (uc >(U8*)last_start) break; |
| 1866 | } |
| 1867 | |
| 1868 | if ( word ) { |
| 1869 | U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ]; |
| 1870 | if (!leftmost || lpos < leftmost) { |
| 1871 | DEBUG_r(accepted_word=word); |
| 1872 | leftmost= lpos; |
| 1873 | } |
| 1874 | if (base==0) break; |
| 1875 | |
| 1876 | } |
| 1877 | points[pointpos++ % maxlen]= uc; |
| 1878 | REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, |
| 1879 | uscan, len, uvc, charid, foldlen, |
| 1880 | foldbuf, uniflags); |
| 1881 | DEBUG_TRIE_EXECUTE_r({ |
| 1882 | dump_exec_pos( (char *)uc, c, strend, real_start, |
| 1883 | s, utf8_target ); |
| 1884 | PerlIO_printf(Perl_debug_log, |
| 1885 | " Charid:%3u CP:%4"UVxf" ", |
| 1886 | charid, uvc); |
| 1887 | }); |
| 1888 | |
| 1889 | do { |
| 1890 | #ifdef DEBUGGING |
| 1891 | word = aho->states[ state ].wordnum; |
| 1892 | #endif |
| 1893 | base = aho->states[ state ].trans.base; |
| 1894 | |
| 1895 | DEBUG_TRIE_EXECUTE_r({ |
| 1896 | if (failed) |
| 1897 | dump_exec_pos( (char *)uc, c, strend, real_start, |
| 1898 | s, utf8_target ); |
| 1899 | PerlIO_printf( Perl_debug_log, |
| 1900 | "%sState: %4"UVxf", word=%"UVxf, |
| 1901 | failed ? " Fail transition to " : "", |
| 1902 | (UV)state, (UV)word); |
| 1903 | }); |
| 1904 | if ( base ) { |
| 1905 | U32 tmp; |
| 1906 | I32 offset; |
| 1907 | if (charid && |
| 1908 | ( ((offset = base + charid |
| 1909 | - 1 - trie->uniquecharcount)) >= 0) |
| 1910 | && ((U32)offset < trie->lasttrans) |
| 1911 | && trie->trans[offset].check == state |
| 1912 | && (tmp=trie->trans[offset].next)) |
| 1913 | { |
| 1914 | DEBUG_TRIE_EXECUTE_r( |
| 1915 | PerlIO_printf( Perl_debug_log," - legal\n")); |
| 1916 | state = tmp; |
| 1917 | break; |
| 1918 | } |
| 1919 | else { |
| 1920 | DEBUG_TRIE_EXECUTE_r( |
| 1921 | PerlIO_printf( Perl_debug_log," - fail\n")); |
| 1922 | failed = 1; |
| 1923 | state = aho->fail[state]; |
| 1924 | } |
| 1925 | } |
| 1926 | else { |
| 1927 | /* we must be accepting here */ |
| 1928 | DEBUG_TRIE_EXECUTE_r( |
| 1929 | PerlIO_printf( Perl_debug_log," - accepting\n")); |
| 1930 | failed = 1; |
| 1931 | break; |
| 1932 | } |
| 1933 | } while(state); |
| 1934 | uc += len; |
| 1935 | if (failed) { |
| 1936 | if (leftmost) |
| 1937 | break; |
| 1938 | if (!state) state = 1; |
| 1939 | } |
| 1940 | } |
| 1941 | if ( aho->states[ state ].wordnum ) { |
| 1942 | U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ]; |
| 1943 | if (!leftmost || lpos < leftmost) { |
| 1944 | DEBUG_r(accepted_word=aho->states[ state ].wordnum); |
| 1945 | leftmost = lpos; |
| 1946 | } |
| 1947 | } |
| 1948 | if (leftmost) { |
| 1949 | s = (char*)leftmost; |
| 1950 | DEBUG_TRIE_EXECUTE_r({ |
| 1951 | PerlIO_printf( |
| 1952 | Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n", |
| 1953 | (UV)accepted_word, (IV)(s - real_start) |
| 1954 | ); |
| 1955 | }); |
| 1956 | if (!reginfo || regtry(reginfo, &s)) { |
| 1957 | FREETMPS; |
| 1958 | LEAVE; |
| 1959 | goto got_it; |
| 1960 | } |
| 1961 | s = HOPc(s,1); |
| 1962 | DEBUG_TRIE_EXECUTE_r({ |
| 1963 | PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n"); |
| 1964 | }); |
| 1965 | } else { |
| 1966 | DEBUG_TRIE_EXECUTE_r( |
| 1967 | PerlIO_printf( Perl_debug_log,"No match.\n")); |
| 1968 | break; |
| 1969 | } |
| 1970 | } |
| 1971 | FREETMPS; |
| 1972 | LEAVE; |
| 1973 | } |
| 1974 | break; |
| 1975 | default: |
| 1976 | Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c)); |
| 1977 | break; |
| 1978 | } |
| 1979 | return 0; |
| 1980 | got_it: |
| 1981 | return s; |
| 1982 | } |
| 1983 | |
| 1984 | |
| 1985 | /* |
| 1986 | - regexec_flags - match a regexp against a string |
| 1987 | */ |
| 1988 | I32 |
| 1989 | Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend, |
| 1990 | char *strbeg, I32 minend, SV *sv, void *data, U32 flags) |
| 1991 | /* strend: pointer to null at end of string */ |
| 1992 | /* strbeg: real beginning of string */ |
| 1993 | /* minend: end of match must be >=minend after stringarg. */ |
| 1994 | /* data: May be used for some additional optimizations. |
| 1995 | Currently its only used, with a U32 cast, for transmitting |
| 1996 | the ganch offset when doing a /g match. This will change */ |
| 1997 | /* nosave: For optimizations. */ |
| 1998 | { |
| 1999 | dVAR; |
| 2000 | struct regexp *const prog = (struct regexp *)SvANY(rx); |
| 2001 | /*register*/ char *s; |
| 2002 | register regnode *c; |
| 2003 | /*register*/ char *startpos = stringarg; |
| 2004 | I32 minlen; /* must match at least this many chars */ |
| 2005 | I32 dontbother = 0; /* how many characters not to try at end */ |
| 2006 | I32 end_shift = 0; /* Same for the end. */ /* CC */ |
| 2007 | I32 scream_pos = -1; /* Internal iterator of scream. */ |
| 2008 | char *scream_olds = NULL; |
| 2009 | const bool utf8_target = cBOOL(DO_UTF8(sv)); |
| 2010 | I32 multiline; |
| 2011 | RXi_GET_DECL(prog,progi); |
| 2012 | regmatch_info reginfo; /* create some info to pass to regtry etc */ |
| 2013 | regexp_paren_pair *swap = NULL; |
| 2014 | GET_RE_DEBUG_FLAGS_DECL; |
| 2015 | |
| 2016 | PERL_ARGS_ASSERT_REGEXEC_FLAGS; |
| 2017 | PERL_UNUSED_ARG(data); |
| 2018 | |
| 2019 | /* Be paranoid... */ |
| 2020 | if (prog == NULL || startpos == NULL) { |
| 2021 | Perl_croak(aTHX_ "NULL regexp parameter"); |
| 2022 | return 0; |
| 2023 | } |
| 2024 | |
| 2025 | multiline = prog->extflags & RXf_PMf_MULTILINE; |
| 2026 | reginfo.prog = rx; /* Yes, sorry that this is confusing. */ |
| 2027 | |
| 2028 | RX_MATCH_UTF8_set(rx, utf8_target); |
| 2029 | DEBUG_EXECUTE_r( |
| 2030 | debug_start_match(rx, utf8_target, startpos, strend, |
| 2031 | "Matching"); |
| 2032 | ); |
| 2033 | |
| 2034 | minlen = prog->minlen; |
| 2035 | |
| 2036 | if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) { |
| 2037 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 2038 | "String too short [regexec_flags]...\n")); |
| 2039 | goto phooey; |
| 2040 | } |
| 2041 | |
| 2042 | |
| 2043 | /* Check validity of program. */ |
| 2044 | if (UCHARAT(progi->program) != REG_MAGIC) { |
| 2045 | Perl_croak(aTHX_ "corrupted regexp program"); |
| 2046 | } |
| 2047 | |
| 2048 | PL_reg_flags = 0; |
| 2049 | PL_reg_eval_set = 0; |
| 2050 | PL_reg_maxiter = 0; |
| 2051 | |
| 2052 | if (RX_UTF8(rx)) |
| 2053 | PL_reg_flags |= RF_utf8; |
| 2054 | |
| 2055 | /* Mark beginning of line for ^ and lookbehind. */ |
| 2056 | reginfo.bol = startpos; /* XXX not used ??? */ |
| 2057 | PL_bostr = strbeg; |
| 2058 | reginfo.sv = sv; |
| 2059 | |
| 2060 | /* Mark end of line for $ (and such) */ |
| 2061 | PL_regeol = strend; |
| 2062 | |
| 2063 | /* see how far we have to get to not match where we matched before */ |
| 2064 | reginfo.till = startpos+minend; |
| 2065 | |
| 2066 | /* If there is a "must appear" string, look for it. */ |
| 2067 | s = startpos; |
| 2068 | |
| 2069 | if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */ |
| 2070 | MAGIC *mg; |
| 2071 | if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */ |
| 2072 | reginfo.ganch = startpos + prog->gofs; |
| 2073 | DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log, |
| 2074 | "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs)); |
| 2075 | } else if (sv && SvTYPE(sv) >= SVt_PVMG |
| 2076 | && SvMAGIC(sv) |
| 2077 | && (mg = mg_find(sv, PERL_MAGIC_regex_global)) |
| 2078 | && mg->mg_len >= 0) { |
| 2079 | reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */ |
| 2080 | DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log, |
| 2081 | "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len)); |
| 2082 | |
| 2083 | if (prog->extflags & RXf_ANCH_GPOS) { |
| 2084 | if (s > reginfo.ganch) |
| 2085 | goto phooey; |
| 2086 | s = reginfo.ganch - prog->gofs; |
| 2087 | DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log, |
| 2088 | "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs)); |
| 2089 | if (s < strbeg) |
| 2090 | goto phooey; |
| 2091 | } |
| 2092 | } |
| 2093 | else if (data) { |
| 2094 | reginfo.ganch = strbeg + PTR2UV(data); |
| 2095 | DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log, |
| 2096 | "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data))); |
| 2097 | |
| 2098 | } else { /* pos() not defined */ |
| 2099 | reginfo.ganch = strbeg; |
| 2100 | DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log, |
| 2101 | "GPOS: reginfo.ganch = strbeg\n")); |
| 2102 | } |
| 2103 | } |
| 2104 | if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) { |
| 2105 | /* We have to be careful. If the previous successful match |
| 2106 | was from this regex we don't want a subsequent partially |
| 2107 | successful match to clobber the old results. |
| 2108 | So when we detect this possibility we add a swap buffer |
| 2109 | to the re, and switch the buffer each match. If we fail |
| 2110 | we switch it back, otherwise we leave it swapped. |
| 2111 | */ |
| 2112 | swap = prog->offs; |
| 2113 | /* do we need a save destructor here for eval dies? */ |
| 2114 | Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair); |
| 2115 | } |
| 2116 | if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) { |
| 2117 | re_scream_pos_data d; |
| 2118 | |
| 2119 | d.scream_olds = &scream_olds; |
| 2120 | d.scream_pos = &scream_pos; |
| 2121 | s = re_intuit_start(rx, sv, s, strend, flags, &d); |
| 2122 | if (!s) { |
| 2123 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n")); |
| 2124 | goto phooey; /* not present */ |
| 2125 | } |
| 2126 | } |
| 2127 | |
| 2128 | |
| 2129 | |
| 2130 | /* Simplest case: anchored match need be tried only once. */ |
| 2131 | /* [unless only anchor is BOL and multiline is set] */ |
| 2132 | if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) { |
| 2133 | if (s == startpos && regtry(®info, &startpos)) |
| 2134 | goto got_it; |
| 2135 | else if (multiline || (prog->intflags & PREGf_IMPLICIT) |
| 2136 | || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */ |
| 2137 | { |
| 2138 | char *end; |
| 2139 | |
| 2140 | if (minlen) |
| 2141 | dontbother = minlen - 1; |
| 2142 | end = HOP3c(strend, -dontbother, strbeg) - 1; |
| 2143 | /* for multiline we only have to try after newlines */ |
| 2144 | if (prog->check_substr || prog->check_utf8) { |
| 2145 | /* because of the goto we can not easily reuse the macros for bifurcating the |
| 2146 | unicode/non-unicode match modes here like we do elsewhere - demerphq */ |
| 2147 | if (utf8_target) { |
| 2148 | if (s == startpos) |
| 2149 | goto after_try_utf8; |
| 2150 | while (1) { |
| 2151 | if (regtry(®info, &s)) { |
| 2152 | goto got_it; |
| 2153 | } |
| 2154 | after_try_utf8: |
| 2155 | if (s > end) { |
| 2156 | goto phooey; |
| 2157 | } |
| 2158 | if (prog->extflags & RXf_USE_INTUIT) { |
| 2159 | s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL); |
| 2160 | if (!s) { |
| 2161 | goto phooey; |
| 2162 | } |
| 2163 | } |
| 2164 | else { |
| 2165 | s += UTF8SKIP(s); |
| 2166 | } |
| 2167 | } |
| 2168 | } /* end search for check string in unicode */ |
| 2169 | else { |
| 2170 | if (s == startpos) { |
| 2171 | goto after_try_latin; |
| 2172 | } |
| 2173 | while (1) { |
| 2174 | if (regtry(®info, &s)) { |
| 2175 | goto got_it; |
| 2176 | } |
| 2177 | after_try_latin: |
| 2178 | if (s > end) { |
| 2179 | goto phooey; |
| 2180 | } |
| 2181 | if (prog->extflags & RXf_USE_INTUIT) { |
| 2182 | s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL); |
| 2183 | if (!s) { |
| 2184 | goto phooey; |
| 2185 | } |
| 2186 | } |
| 2187 | else { |
| 2188 | s++; |
| 2189 | } |
| 2190 | } |
| 2191 | } /* end search for check string in latin*/ |
| 2192 | } /* end search for check string */ |
| 2193 | else { /* search for newline */ |
| 2194 | if (s > startpos) { |
| 2195 | /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/ |
| 2196 | s--; |
| 2197 | } |
| 2198 | /* We can use a more efficient search as newlines are the same in unicode as they are in latin */ |
| 2199 | while (s < end) { |
| 2200 | if (*s++ == '\n') { /* don't need PL_utf8skip here */ |
| 2201 | if (regtry(®info, &s)) |
| 2202 | goto got_it; |
| 2203 | } |
| 2204 | } |
| 2205 | } /* end search for newline */ |
| 2206 | } /* end anchored/multiline check string search */ |
| 2207 | goto phooey; |
| 2208 | } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK)) |
| 2209 | { |
| 2210 | /* the warning about reginfo.ganch being used without initialization |
| 2211 | is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN |
| 2212 | and we only enter this block when the same bit is set. */ |
| 2213 | char *tmp_s = reginfo.ganch - prog->gofs; |
| 2214 | |
| 2215 | if (tmp_s >= strbeg && regtry(®info, &tmp_s)) |
| 2216 | goto got_it; |
| 2217 | goto phooey; |
| 2218 | } |
| 2219 | |
| 2220 | /* Messy cases: unanchored match. */ |
| 2221 | if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) { |
| 2222 | /* we have /x+whatever/ */ |
| 2223 | /* it must be a one character string (XXXX Except UTF_PATTERN?) */ |
| 2224 | char ch; |
| 2225 | #ifdef DEBUGGING |
| 2226 | int did_match = 0; |
| 2227 | #endif |
| 2228 | if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)) |
| 2229 | utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog); |
| 2230 | ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0]; |
| 2231 | |
| 2232 | if (utf8_target) { |
| 2233 | REXEC_FBC_SCAN( |
| 2234 | if (*s == ch) { |
| 2235 | DEBUG_EXECUTE_r( did_match = 1 ); |
| 2236 | if (regtry(®info, &s)) goto got_it; |
| 2237 | s += UTF8SKIP(s); |
| 2238 | while (s < strend && *s == ch) |
| 2239 | s += UTF8SKIP(s); |
| 2240 | } |
| 2241 | ); |
| 2242 | } |
| 2243 | else { |
| 2244 | REXEC_FBC_SCAN( |
| 2245 | if (*s == ch) { |
| 2246 | DEBUG_EXECUTE_r( did_match = 1 ); |
| 2247 | if (regtry(®info, &s)) goto got_it; |
| 2248 | s++; |
| 2249 | while (s < strend && *s == ch) |
| 2250 | s++; |
| 2251 | } |
| 2252 | ); |
| 2253 | } |
| 2254 | DEBUG_EXECUTE_r(if (!did_match) |
| 2255 | PerlIO_printf(Perl_debug_log, |
| 2256 | "Did not find anchored character...\n") |
| 2257 | ); |
| 2258 | } |
| 2259 | else if (prog->anchored_substr != NULL |
| 2260 | || prog->anchored_utf8 != NULL |
| 2261 | || ((prog->float_substr != NULL || prog->float_utf8 != NULL) |
| 2262 | && prog->float_max_offset < strend - s)) { |
| 2263 | SV *must; |
| 2264 | I32 back_max; |
| 2265 | I32 back_min; |
| 2266 | char *last; |
| 2267 | char *last1; /* Last position checked before */ |
| 2268 | #ifdef DEBUGGING |
| 2269 | int did_match = 0; |
| 2270 | #endif |
| 2271 | if (prog->anchored_substr || prog->anchored_utf8) { |
| 2272 | if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)) |
| 2273 | utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog); |
| 2274 | must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr; |
| 2275 | back_max = back_min = prog->anchored_offset; |
| 2276 | } else { |
| 2277 | if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) |
| 2278 | utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog); |
| 2279 | must = utf8_target ? prog->float_utf8 : prog->float_substr; |
| 2280 | back_max = prog->float_max_offset; |
| 2281 | back_min = prog->float_min_offset; |
| 2282 | } |
| 2283 | |
| 2284 | |
| 2285 | if (must == &PL_sv_undef) |
| 2286 | /* could not downgrade utf8 check substring, so must fail */ |
| 2287 | goto phooey; |
| 2288 | |
| 2289 | if (back_min<0) { |
| 2290 | last = strend; |
| 2291 | } else { |
| 2292 | last = HOP3c(strend, /* Cannot start after this */ |
| 2293 | -(I32)(CHR_SVLEN(must) |
| 2294 | - (SvTAIL(must) != 0) + back_min), strbeg); |
| 2295 | } |
| 2296 | if (s > PL_bostr) |
| 2297 | last1 = HOPc(s, -1); |
| 2298 | else |
| 2299 | last1 = s - 1; /* bogus */ |
| 2300 | |
| 2301 | /* XXXX check_substr already used to find "s", can optimize if |
| 2302 | check_substr==must. */ |
| 2303 | scream_pos = -1; |
| 2304 | dontbother = end_shift; |
| 2305 | strend = HOPc(strend, -dontbother); |
| 2306 | while ( (s <= last) && |
| 2307 | ((flags & REXEC_SCREAM) && SvSCREAM(sv) |
| 2308 | ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg, |
| 2309 | end_shift, &scream_pos, 0)) |
| 2310 | : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)), |
| 2311 | (unsigned char*)strend, must, |
| 2312 | multiline ? FBMrf_MULTILINE : 0))) ) { |
| 2313 | /* we may be pointing at the wrong string */ |
| 2314 | if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog)) |
| 2315 | s = strbeg + (s - SvPVX_const(sv)); |
| 2316 | DEBUG_EXECUTE_r( did_match = 1 ); |
| 2317 | if (HOPc(s, -back_max) > last1) { |
| 2318 | last1 = HOPc(s, -back_min); |
| 2319 | s = HOPc(s, -back_max); |
| 2320 | } |
| 2321 | else { |
| 2322 | char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1; |
| 2323 | |
| 2324 | last1 = HOPc(s, -back_min); |
| 2325 | s = t; |
| 2326 | } |
| 2327 | if (utf8_target) { |
| 2328 | while (s <= last1) { |
| 2329 | if (regtry(®info, &s)) |
| 2330 | goto got_it; |
| 2331 | s += UTF8SKIP(s); |
| 2332 | } |
| 2333 | } |
| 2334 | else { |
| 2335 | while (s <= last1) { |
| 2336 | if (regtry(®info, &s)) |
| 2337 | goto got_it; |
| 2338 | s++; |
| 2339 | } |
| 2340 | } |
| 2341 | } |
| 2342 | DEBUG_EXECUTE_r(if (!did_match) { |
| 2343 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), |
| 2344 | SvPVX_const(must), RE_SV_DUMPLEN(must), 30); |
| 2345 | PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n", |
| 2346 | ((must == prog->anchored_substr || must == prog->anchored_utf8) |
| 2347 | ? "anchored" : "floating"), |
| 2348 | quoted, RE_SV_TAIL(must)); |
| 2349 | }); |
| 2350 | goto phooey; |
| 2351 | } |
| 2352 | else if ( (c = progi->regstclass) ) { |
| 2353 | if (minlen) { |
| 2354 | const OPCODE op = OP(progi->regstclass); |
| 2355 | /* don't bother with what can't match */ |
| 2356 | if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE) |
| 2357 | strend = HOPc(strend, -(minlen - 1)); |
| 2358 | } |
| 2359 | DEBUG_EXECUTE_r({ |
| 2360 | SV * const prop = sv_newmortal(); |
| 2361 | regprop(prog, prop, c); |
| 2362 | { |
| 2363 | RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1), |
| 2364 | s,strend-s,60); |
| 2365 | PerlIO_printf(Perl_debug_log, |
| 2366 | "Matching stclass %.*s against %s (%d bytes)\n", |
| 2367 | (int)SvCUR(prop), SvPVX_const(prop), |
| 2368 | quoted, (int)(strend - s)); |
| 2369 | } |
| 2370 | }); |
| 2371 | if (find_byclass(prog, c, s, strend, ®info)) |
| 2372 | goto got_it; |
| 2373 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n")); |
| 2374 | } |
| 2375 | else { |
| 2376 | dontbother = 0; |
| 2377 | if (prog->float_substr != NULL || prog->float_utf8 != NULL) { |
| 2378 | /* Trim the end. */ |
| 2379 | char *last; |
| 2380 | SV* float_real; |
| 2381 | |
| 2382 | if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) |
| 2383 | utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog); |
| 2384 | float_real = utf8_target ? prog->float_utf8 : prog->float_substr; |
| 2385 | |
| 2386 | if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) { |
| 2387 | last = screaminstr(sv, float_real, s - strbeg, |
| 2388 | end_shift, &scream_pos, 1); /* last one */ |
| 2389 | if (!last) |
| 2390 | last = scream_olds; /* Only one occurrence. */ |
| 2391 | /* we may be pointing at the wrong string */ |
| 2392 | else if (RXp_MATCH_COPIED(prog)) |
| 2393 | s = strbeg + (s - SvPVX_const(sv)); |
| 2394 | } |
| 2395 | else { |
| 2396 | STRLEN len; |
| 2397 | const char * const little = SvPV_const(float_real, len); |
| 2398 | |
| 2399 | if (SvTAIL(float_real)) { |
| 2400 | if (memEQ(strend - len + 1, little, len - 1)) |
| 2401 | last = strend - len + 1; |
| 2402 | else if (!multiline) |
| 2403 | last = memEQ(strend - len, little, len) |
| 2404 | ? strend - len : NULL; |
| 2405 | else |
| 2406 | goto find_last; |
| 2407 | } else { |
| 2408 | find_last: |
| 2409 | if (len) |
| 2410 | last = rninstr(s, strend, little, little + len); |
| 2411 | else |
| 2412 | last = strend; /* matching "$" */ |
| 2413 | } |
| 2414 | } |
| 2415 | if (last == NULL) { |
| 2416 | DEBUG_EXECUTE_r( |
| 2417 | PerlIO_printf(Perl_debug_log, |
| 2418 | "%sCan't trim the tail, match fails (should not happen)%s\n", |
| 2419 | PL_colors[4], PL_colors[5])); |
| 2420 | goto phooey; /* Should not happen! */ |
| 2421 | } |
| 2422 | dontbother = strend - last + prog->float_min_offset; |
| 2423 | } |
| 2424 | if (minlen && (dontbother < minlen)) |
| 2425 | dontbother = minlen - 1; |
| 2426 | strend -= dontbother; /* this one's always in bytes! */ |
| 2427 | /* We don't know much -- general case. */ |
| 2428 | if (utf8_target) { |
| 2429 | for (;;) { |
| 2430 | if (regtry(®info, &s)) |
| 2431 | goto got_it; |
| 2432 | if (s >= strend) |
| 2433 | break; |
| 2434 | s += UTF8SKIP(s); |
| 2435 | }; |
| 2436 | } |
| 2437 | else { |
| 2438 | do { |
| 2439 | if (regtry(®info, &s)) |
| 2440 | goto got_it; |
| 2441 | } while (s++ < strend); |
| 2442 | } |
| 2443 | } |
| 2444 | |
| 2445 | /* Failure. */ |
| 2446 | goto phooey; |
| 2447 | |
| 2448 | got_it: |
| 2449 | Safefree(swap); |
| 2450 | RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted); |
| 2451 | |
| 2452 | if (PL_reg_eval_set) |
| 2453 | restore_pos(aTHX_ prog); |
| 2454 | if (RXp_PAREN_NAMES(prog)) |
| 2455 | (void)hv_iterinit(RXp_PAREN_NAMES(prog)); |
| 2456 | |
| 2457 | /* make sure $`, $&, $', and $digit will work later */ |
| 2458 | if ( !(flags & REXEC_NOT_FIRST) ) { |
| 2459 | RX_MATCH_COPY_FREE(rx); |
| 2460 | if (flags & REXEC_COPY_STR) { |
| 2461 | const I32 i = PL_regeol - startpos + (stringarg - strbeg); |
| 2462 | #ifdef PERL_OLD_COPY_ON_WRITE |
| 2463 | if ((SvIsCOW(sv) |
| 2464 | || (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) { |
| 2465 | if (DEBUG_C_TEST) { |
| 2466 | PerlIO_printf(Perl_debug_log, |
| 2467 | "Copy on write: regexp capture, type %d\n", |
| 2468 | (int) SvTYPE(sv)); |
| 2469 | } |
| 2470 | prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv); |
| 2471 | prog->subbeg = (char *)SvPVX_const(prog->saved_copy); |
| 2472 | assert (SvPOKp(prog->saved_copy)); |
| 2473 | } else |
| 2474 | #endif |
| 2475 | { |
| 2476 | RX_MATCH_COPIED_on(rx); |
| 2477 | s = savepvn(strbeg, i); |
| 2478 | prog->subbeg = s; |
| 2479 | } |
| 2480 | prog->sublen = i; |
| 2481 | } |
| 2482 | else { |
| 2483 | prog->subbeg = strbeg; |
| 2484 | prog->sublen = PL_regeol - strbeg; /* strend may have been modified */ |
| 2485 | } |
| 2486 | } |
| 2487 | |
| 2488 | return 1; |
| 2489 | |
| 2490 | phooey: |
| 2491 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n", |
| 2492 | PL_colors[4], PL_colors[5])); |
| 2493 | if (PL_reg_eval_set) |
| 2494 | restore_pos(aTHX_ prog); |
| 2495 | if (swap) { |
| 2496 | /* we failed :-( roll it back */ |
| 2497 | Safefree(prog->offs); |
| 2498 | prog->offs = swap; |
| 2499 | } |
| 2500 | |
| 2501 | return 0; |
| 2502 | } |
| 2503 | |
| 2504 | |
| 2505 | /* |
| 2506 | - regtry - try match at specific point |
| 2507 | */ |
| 2508 | STATIC I32 /* 0 failure, 1 success */ |
| 2509 | S_regtry(pTHX_ regmatch_info *reginfo, char **startpos) |
| 2510 | { |
| 2511 | dVAR; |
| 2512 | CHECKPOINT lastcp; |
| 2513 | REGEXP *const rx = reginfo->prog; |
| 2514 | regexp *const prog = (struct regexp *)SvANY(rx); |
| 2515 | RXi_GET_DECL(prog,progi); |
| 2516 | GET_RE_DEBUG_FLAGS_DECL; |
| 2517 | |
| 2518 | PERL_ARGS_ASSERT_REGTRY; |
| 2519 | |
| 2520 | reginfo->cutpoint=NULL; |
| 2521 | |
| 2522 | if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) { |
| 2523 | MAGIC *mg; |
| 2524 | |
| 2525 | PL_reg_eval_set = RS_init; |
| 2526 | DEBUG_EXECUTE_r(DEBUG_s( |
| 2527 | PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n", |
| 2528 | (IV)(PL_stack_sp - PL_stack_base)); |
| 2529 | )); |
| 2530 | SAVESTACK_CXPOS(); |
| 2531 | cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base; |
| 2532 | /* Otherwise OP_NEXTSTATE will free whatever on stack now. */ |
| 2533 | SAVETMPS; |
| 2534 | /* Apparently this is not needed, judging by wantarray. */ |
| 2535 | /* SAVEI8(cxstack[cxstack_ix].blk_gimme); |
| 2536 | cxstack[cxstack_ix].blk_gimme = G_SCALAR; */ |
| 2537 | |
| 2538 | if (reginfo->sv) { |
| 2539 | /* Make $_ available to executed code. */ |
| 2540 | if (reginfo->sv != DEFSV) { |
| 2541 | SAVE_DEFSV; |
| 2542 | DEFSV_set(reginfo->sv); |
| 2543 | } |
| 2544 | |
| 2545 | if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv) |
| 2546 | && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) { |
| 2547 | /* prepare for quick setting of pos */ |
| 2548 | #ifdef PERL_OLD_COPY_ON_WRITE |
| 2549 | if (SvIsCOW(reginfo->sv)) |
| 2550 | sv_force_normal_flags(reginfo->sv, 0); |
| 2551 | #endif |
| 2552 | mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global, |
| 2553 | &PL_vtbl_mglob, NULL, 0); |
| 2554 | mg->mg_len = -1; |
| 2555 | } |
| 2556 | PL_reg_magic = mg; |
| 2557 | PL_reg_oldpos = mg->mg_len; |
| 2558 | SAVEDESTRUCTOR_X(restore_pos, prog); |
| 2559 | } |
| 2560 | if (!PL_reg_curpm) { |
| 2561 | Newxz(PL_reg_curpm, 1, PMOP); |
| 2562 | #ifdef USE_ITHREADS |
| 2563 | { |
| 2564 | SV* const repointer = &PL_sv_undef; |
| 2565 | /* this regexp is also owned by the new PL_reg_curpm, which |
| 2566 | will try to free it. */ |
| 2567 | av_push(PL_regex_padav, repointer); |
| 2568 | PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav); |
| 2569 | PL_regex_pad = AvARRAY(PL_regex_padav); |
| 2570 | } |
| 2571 | #endif |
| 2572 | } |
| 2573 | #ifdef USE_ITHREADS |
| 2574 | /* It seems that non-ithreads works both with and without this code. |
| 2575 | So for efficiency reasons it seems best not to have the code |
| 2576 | compiled when it is not needed. */ |
| 2577 | /* This is safe against NULLs: */ |
| 2578 | ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); |
| 2579 | /* PM_reg_curpm owns a reference to this regexp. */ |
| 2580 | (void)ReREFCNT_inc(rx); |
| 2581 | #endif |
| 2582 | PM_SETRE(PL_reg_curpm, rx); |
| 2583 | PL_reg_oldcurpm = PL_curpm; |
| 2584 | PL_curpm = PL_reg_curpm; |
| 2585 | if (RXp_MATCH_COPIED(prog)) { |
| 2586 | /* Here is a serious problem: we cannot rewrite subbeg, |
| 2587 | since it may be needed if this match fails. Thus |
| 2588 | $` inside (?{}) could fail... */ |
| 2589 | PL_reg_oldsaved = prog->subbeg; |
| 2590 | PL_reg_oldsavedlen = prog->sublen; |
| 2591 | #ifdef PERL_OLD_COPY_ON_WRITE |
| 2592 | PL_nrs = prog->saved_copy; |
| 2593 | #endif |
| 2594 | RXp_MATCH_COPIED_off(prog); |
| 2595 | } |
| 2596 | else |
| 2597 | PL_reg_oldsaved = NULL; |
| 2598 | prog->subbeg = PL_bostr; |
| 2599 | prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */ |
| 2600 | } |
| 2601 | DEBUG_EXECUTE_r(PL_reg_starttry = *startpos); |
| 2602 | prog->offs[0].start = *startpos - PL_bostr; |
| 2603 | PL_reginput = *startpos; |
| 2604 | PL_reglastparen = &prog->lastparen; |
| 2605 | PL_reglastcloseparen = &prog->lastcloseparen; |
| 2606 | prog->lastparen = 0; |
| 2607 | prog->lastcloseparen = 0; |
| 2608 | PL_regsize = 0; |
| 2609 | PL_regoffs = prog->offs; |
| 2610 | if (PL_reg_start_tmpl <= prog->nparens) { |
| 2611 | PL_reg_start_tmpl = prog->nparens*3/2 + 3; |
| 2612 | if(PL_reg_start_tmp) |
| 2613 | Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*); |
| 2614 | else |
| 2615 | Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*); |
| 2616 | } |
| 2617 | |
| 2618 | /* XXXX What this code is doing here?!!! There should be no need |
| 2619 | to do this again and again, PL_reglastparen should take care of |
| 2620 | this! --ilya*/ |
| 2621 | |
| 2622 | /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code. |
| 2623 | * Actually, the code in regcppop() (which Ilya may be meaning by |
| 2624 | * PL_reglastparen), is not needed at all by the test suite |
| 2625 | * (op/regexp, op/pat, op/split), but that code is needed otherwise |
| 2626 | * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/ |
| 2627 | * Meanwhile, this code *is* needed for the |
| 2628 | * above-mentioned test suite tests to succeed. The common theme |
| 2629 | * on those tests seems to be returning null fields from matches. |
| 2630 | * --jhi updated by dapm */ |
| 2631 | #if 1 |
| 2632 | if (prog->nparens) { |
| 2633 | regexp_paren_pair *pp = PL_regoffs; |
| 2634 | register I32 i; |
| 2635 | for (i = prog->nparens; i > (I32)*PL_reglastparen; i--) { |
| 2636 | ++pp; |
| 2637 | pp->start = -1; |
| 2638 | pp->end = -1; |
| 2639 | } |
| 2640 | } |
| 2641 | #endif |
| 2642 | REGCP_SET(lastcp); |
| 2643 | if (regmatch(reginfo, progi->program + 1)) { |
| 2644 | PL_regoffs[0].end = PL_reginput - PL_bostr; |
| 2645 | return 1; |
| 2646 | } |
| 2647 | if (reginfo->cutpoint) |
| 2648 | *startpos= reginfo->cutpoint; |
| 2649 | REGCP_UNWIND(lastcp); |
| 2650 | return 0; |
| 2651 | } |
| 2652 | |
| 2653 | |
| 2654 | #define sayYES goto yes |
| 2655 | #define sayNO goto no |
| 2656 | #define sayNO_SILENT goto no_silent |
| 2657 | |
| 2658 | /* we dont use STMT_START/END here because it leads to |
| 2659 | "unreachable code" warnings, which are bogus, but distracting. */ |
| 2660 | #define CACHEsayNO \ |
| 2661 | if (ST.cache_mask) \ |
| 2662 | PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \ |
| 2663 | sayNO |
| 2664 | |
| 2665 | /* this is used to determine how far from the left messages like |
| 2666 | 'failed...' are printed. It should be set such that messages |
| 2667 | are inline with the regop output that created them. |
| 2668 | */ |
| 2669 | #define REPORT_CODE_OFF 32 |
| 2670 | |
| 2671 | |
| 2672 | #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */ |
| 2673 | #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */ |
| 2674 | |
| 2675 | #define SLAB_FIRST(s) (&(s)->states[0]) |
| 2676 | #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1]) |
| 2677 | |
| 2678 | /* grab a new slab and return the first slot in it */ |
| 2679 | |
| 2680 | STATIC regmatch_state * |
| 2681 | S_push_slab(pTHX) |
| 2682 | { |
| 2683 | #if PERL_VERSION < 9 && !defined(PERL_CORE) |
| 2684 | dMY_CXT; |
| 2685 | #endif |
| 2686 | regmatch_slab *s = PL_regmatch_slab->next; |
| 2687 | if (!s) { |
| 2688 | Newx(s, 1, regmatch_slab); |
| 2689 | s->prev = PL_regmatch_slab; |
| 2690 | s->next = NULL; |
| 2691 | PL_regmatch_slab->next = s; |
| 2692 | } |
| 2693 | PL_regmatch_slab = s; |
| 2694 | return SLAB_FIRST(s); |
| 2695 | } |
| 2696 | |
| 2697 | |
| 2698 | /* push a new state then goto it */ |
| 2699 | |
| 2700 | #define PUSH_STATE_GOTO(state, node) \ |
| 2701 | scan = node; \ |
| 2702 | st->resume_state = state; \ |
| 2703 | goto push_state; |
| 2704 | |
| 2705 | /* push a new state with success backtracking, then goto it */ |
| 2706 | |
| 2707 | #define PUSH_YES_STATE_GOTO(state, node) \ |
| 2708 | scan = node; \ |
| 2709 | st->resume_state = state; \ |
| 2710 | goto push_yes_state; |
| 2711 | |
| 2712 | |
| 2713 | |
| 2714 | /* |
| 2715 | |
| 2716 | regmatch() - main matching routine |
| 2717 | |
| 2718 | This is basically one big switch statement in a loop. We execute an op, |
| 2719 | set 'next' to point the next op, and continue. If we come to a point which |
| 2720 | we may need to backtrack to on failure such as (A|B|C), we push a |
| 2721 | backtrack state onto the backtrack stack. On failure, we pop the top |
| 2722 | state, and re-enter the loop at the state indicated. If there are no more |
| 2723 | states to pop, we return failure. |
| 2724 | |
| 2725 | Sometimes we also need to backtrack on success; for example /A+/, where |
| 2726 | after successfully matching one A, we need to go back and try to |
| 2727 | match another one; similarly for lookahead assertions: if the assertion |
| 2728 | completes successfully, we backtrack to the state just before the assertion |
| 2729 | and then carry on. In these cases, the pushed state is marked as |
| 2730 | 'backtrack on success too'. This marking is in fact done by a chain of |
| 2731 | pointers, each pointing to the previous 'yes' state. On success, we pop to |
| 2732 | the nearest yes state, discarding any intermediate failure-only states. |
| 2733 | Sometimes a yes state is pushed just to force some cleanup code to be |
| 2734 | called at the end of a successful match or submatch; e.g. (??{$re}) uses |
| 2735 | it to free the inner regex. |
| 2736 | |
| 2737 | Note that failure backtracking rewinds the cursor position, while |
| 2738 | success backtracking leaves it alone. |
| 2739 | |
| 2740 | A pattern is complete when the END op is executed, while a subpattern |
| 2741 | such as (?=foo) is complete when the SUCCESS op is executed. Both of these |
| 2742 | ops trigger the "pop to last yes state if any, otherwise return true" |
| 2743 | behaviour. |
| 2744 | |
| 2745 | A common convention in this function is to use A and B to refer to the two |
| 2746 | subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is |
| 2747 | the subpattern to be matched possibly multiple times, while B is the entire |
| 2748 | rest of the pattern. Variable and state names reflect this convention. |
| 2749 | |
| 2750 | The states in the main switch are the union of ops and failure/success of |
| 2751 | substates associated with with that op. For example, IFMATCH is the op |
| 2752 | that does lookahead assertions /(?=A)B/ and so the IFMATCH state means |
| 2753 | 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just |
| 2754 | successfully matched A and IFMATCH_A_fail is a state saying that we have |
| 2755 | just failed to match A. Resume states always come in pairs. The backtrack |
| 2756 | state we push is marked as 'IFMATCH_A', but when that is popped, we resume |
| 2757 | at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking |
| 2758 | on success or failure. |
| 2759 | |
| 2760 | The struct that holds a backtracking state is actually a big union, with |
| 2761 | one variant for each major type of op. The variable st points to the |
| 2762 | top-most backtrack struct. To make the code clearer, within each |
| 2763 | block of code we #define ST to alias the relevant union. |
| 2764 | |
| 2765 | Here's a concrete example of a (vastly oversimplified) IFMATCH |
| 2766 | implementation: |
| 2767 | |
| 2768 | switch (state) { |
| 2769 | .... |
| 2770 | |
| 2771 | #define ST st->u.ifmatch |
| 2772 | |
| 2773 | case IFMATCH: // we are executing the IFMATCH op, (?=A)B |
| 2774 | ST.foo = ...; // some state we wish to save |
| 2775 | ... |
| 2776 | // push a yes backtrack state with a resume value of |
| 2777 | // IFMATCH_A/IFMATCH_A_fail, then continue execution at the |
| 2778 | // first node of A: |
| 2779 | PUSH_YES_STATE_GOTO(IFMATCH_A, A); |
| 2780 | // NOTREACHED |
| 2781 | |
| 2782 | case IFMATCH_A: // we have successfully executed A; now continue with B |
| 2783 | next = B; |
| 2784 | bar = ST.foo; // do something with the preserved value |
| 2785 | break; |
| 2786 | |
| 2787 | case IFMATCH_A_fail: // A failed, so the assertion failed |
| 2788 | ...; // do some housekeeping, then ... |
| 2789 | sayNO; // propagate the failure |
| 2790 | |
| 2791 | #undef ST |
| 2792 | |
| 2793 | ... |
| 2794 | } |
| 2795 | |
| 2796 | For any old-timers reading this who are familiar with the old recursive |
| 2797 | approach, the code above is equivalent to: |
| 2798 | |
| 2799 | case IFMATCH: // we are executing the IFMATCH op, (?=A)B |
| 2800 | { |
| 2801 | int foo = ... |
| 2802 | ... |
| 2803 | if (regmatch(A)) { |
| 2804 | next = B; |
| 2805 | bar = foo; |
| 2806 | break; |
| 2807 | } |
| 2808 | ...; // do some housekeeping, then ... |
| 2809 | sayNO; // propagate the failure |
| 2810 | } |
| 2811 | |
| 2812 | The topmost backtrack state, pointed to by st, is usually free. If you |
| 2813 | want to claim it, populate any ST.foo fields in it with values you wish to |
| 2814 | save, then do one of |
| 2815 | |
| 2816 | PUSH_STATE_GOTO(resume_state, node); |
| 2817 | PUSH_YES_STATE_GOTO(resume_state, node); |
| 2818 | |
| 2819 | which sets that backtrack state's resume value to 'resume_state', pushes a |
| 2820 | new free entry to the top of the backtrack stack, then goes to 'node'. |
| 2821 | On backtracking, the free slot is popped, and the saved state becomes the |
| 2822 | new free state. An ST.foo field in this new top state can be temporarily |
| 2823 | accessed to retrieve values, but once the main loop is re-entered, it |
| 2824 | becomes available for reuse. |
| 2825 | |
| 2826 | Note that the depth of the backtrack stack constantly increases during the |
| 2827 | left-to-right execution of the pattern, rather than going up and down with |
| 2828 | the pattern nesting. For example the stack is at its maximum at Z at the |
| 2829 | end of the pattern, rather than at X in the following: |
| 2830 | |
| 2831 | /(((X)+)+)+....(Y)+....Z/ |
| 2832 | |
| 2833 | The only exceptions to this are lookahead/behind assertions and the cut, |
| 2834 | (?>A), which pop all the backtrack states associated with A before |
| 2835 | continuing. |
| 2836 | |
| 2837 | Backtrack state structs are allocated in slabs of about 4K in size. |
| 2838 | PL_regmatch_state and st always point to the currently active state, |
| 2839 | and PL_regmatch_slab points to the slab currently containing |
| 2840 | PL_regmatch_state. The first time regmatch() is called, the first slab is |
| 2841 | allocated, and is never freed until interpreter destruction. When the slab |
| 2842 | is full, a new one is allocated and chained to the end. At exit from |
| 2843 | regmatch(), slabs allocated since entry are freed. |
| 2844 | |
| 2845 | */ |
| 2846 | |
| 2847 | |
| 2848 | #define DEBUG_STATE_pp(pp) \ |
| 2849 | DEBUG_STATE_r({ \ |
| 2850 | DUMP_EXEC_POS(locinput, scan, utf8_target); \ |
| 2851 | PerlIO_printf(Perl_debug_log, \ |
| 2852 | " %*s"pp" %s%s%s%s%s\n", \ |
| 2853 | depth*2, "", \ |
| 2854 | PL_reg_name[st->resume_state], \ |
| 2855 | ((st==yes_state||st==mark_state) ? "[" : ""), \ |
| 2856 | ((st==yes_state) ? "Y" : ""), \ |
| 2857 | ((st==mark_state) ? "M" : ""), \ |
| 2858 | ((st==yes_state||st==mark_state) ? "]" : "") \ |
| 2859 | ); \ |
| 2860 | }); |
| 2861 | |
| 2862 | |
| 2863 | #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1) |
| 2864 | |
| 2865 | #ifdef DEBUGGING |
| 2866 | |
| 2867 | STATIC void |
| 2868 | S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target, |
| 2869 | const char *start, const char *end, const char *blurb) |
| 2870 | { |
| 2871 | const bool utf8_pat = RX_UTF8(prog) ? 1 : 0; |
| 2872 | |
| 2873 | PERL_ARGS_ASSERT_DEBUG_START_MATCH; |
| 2874 | |
| 2875 | if (!PL_colorset) |
| 2876 | reginitcolors(); |
| 2877 | { |
| 2878 | RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0), |
| 2879 | RX_PRECOMP_const(prog), RX_PRELEN(prog), 60); |
| 2880 | |
| 2881 | RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1), |
| 2882 | start, end - start, 60); |
| 2883 | |
| 2884 | PerlIO_printf(Perl_debug_log, |
| 2885 | "%s%s REx%s %s against %s\n", |
| 2886 | PL_colors[4], blurb, PL_colors[5], s0, s1); |
| 2887 | |
| 2888 | if (utf8_target||utf8_pat) |
| 2889 | PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n", |
| 2890 | utf8_pat ? "pattern" : "", |
| 2891 | utf8_pat && utf8_target ? " and " : "", |
| 2892 | utf8_target ? "string" : "" |
| 2893 | ); |
| 2894 | } |
| 2895 | } |
| 2896 | |
| 2897 | STATIC void |
| 2898 | S_dump_exec_pos(pTHX_ const char *locinput, |
| 2899 | const regnode *scan, |
| 2900 | const char *loc_regeol, |
| 2901 | const char *loc_bostr, |
| 2902 | const char *loc_reg_starttry, |
| 2903 | const bool utf8_target) |
| 2904 | { |
| 2905 | const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4]; |
| 2906 | const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */ |
| 2907 | int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput); |
| 2908 | /* The part of the string before starttry has one color |
| 2909 | (pref0_len chars), between starttry and current |
| 2910 | position another one (pref_len - pref0_len chars), |
| 2911 | after the current position the third one. |
| 2912 | We assume that pref0_len <= pref_len, otherwise we |
| 2913 | decrease pref0_len. */ |
| 2914 | int pref_len = (locinput - loc_bostr) > (5 + taill) - l |
| 2915 | ? (5 + taill) - l : locinput - loc_bostr; |
| 2916 | int pref0_len; |
| 2917 | |
| 2918 | PERL_ARGS_ASSERT_DUMP_EXEC_POS; |
| 2919 | |
| 2920 | while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len))) |
| 2921 | pref_len++; |
| 2922 | pref0_len = pref_len - (locinput - loc_reg_starttry); |
| 2923 | if (l + pref_len < (5 + taill) && l < loc_regeol - locinput) |
| 2924 | l = ( loc_regeol - locinput > (5 + taill) - pref_len |
| 2925 | ? (5 + taill) - pref_len : loc_regeol - locinput); |
| 2926 | while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l))) |
| 2927 | l--; |
| 2928 | if (pref0_len < 0) |
| 2929 | pref0_len = 0; |
| 2930 | if (pref0_len > pref_len) |
| 2931 | pref0_len = pref_len; |
| 2932 | { |
| 2933 | const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0; |
| 2934 | |
| 2935 | RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0), |
| 2936 | (locinput - pref_len),pref0_len, 60, 4, 5); |
| 2937 | |
| 2938 | RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1), |
| 2939 | (locinput - pref_len + pref0_len), |
| 2940 | pref_len - pref0_len, 60, 2, 3); |
| 2941 | |
| 2942 | RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2), |
| 2943 | locinput, loc_regeol - locinput, 10, 0, 1); |
| 2944 | |
| 2945 | const STRLEN tlen=len0+len1+len2; |
| 2946 | PerlIO_printf(Perl_debug_log, |
| 2947 | "%4"IVdf" <%.*s%.*s%s%.*s>%*s|", |
| 2948 | (IV)(locinput - loc_bostr), |
| 2949 | len0, s0, |
| 2950 | len1, s1, |
| 2951 | (docolor ? "" : "> <"), |
| 2952 | len2, s2, |
| 2953 | (int)(tlen > 19 ? 0 : 19 - tlen), |
| 2954 | ""); |
| 2955 | } |
| 2956 | } |
| 2957 | |
| 2958 | #endif |
| 2959 | |
| 2960 | /* reg_check_named_buff_matched() |
| 2961 | * Checks to see if a named buffer has matched. The data array of |
| 2962 | * buffer numbers corresponding to the buffer is expected to reside |
| 2963 | * in the regexp->data->data array in the slot stored in the ARG() of |
| 2964 | * node involved. Note that this routine doesn't actually care about the |
| 2965 | * name, that information is not preserved from compilation to execution. |
| 2966 | * Returns the index of the leftmost defined buffer with the given name |
| 2967 | * or 0 if non of the buffers matched. |
| 2968 | */ |
| 2969 | STATIC I32 |
| 2970 | S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) |
| 2971 | { |
| 2972 | I32 n; |
| 2973 | RXi_GET_DECL(rex,rexi); |
| 2974 | SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 2975 | I32 *nums=(I32*)SvPVX(sv_dat); |
| 2976 | |
| 2977 | PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED; |
| 2978 | |
| 2979 | for ( n=0; n<SvIVX(sv_dat); n++ ) { |
| 2980 | if ((I32)*PL_reglastparen >= nums[n] && |
| 2981 | PL_regoffs[nums[n]].end != -1) |
| 2982 | { |
| 2983 | return nums[n]; |
| 2984 | } |
| 2985 | } |
| 2986 | return 0; |
| 2987 | } |
| 2988 | |
| 2989 | |
| 2990 | /* free all slabs above current one - called during LEAVE_SCOPE */ |
| 2991 | |
| 2992 | STATIC void |
| 2993 | S_clear_backtrack_stack(pTHX_ void *p) |
| 2994 | { |
| 2995 | regmatch_slab *s = PL_regmatch_slab->next; |
| 2996 | PERL_UNUSED_ARG(p); |
| 2997 | |
| 2998 | if (!s) |
| 2999 | return; |
| 3000 | PL_regmatch_slab->next = NULL; |
| 3001 | while (s) { |
| 3002 | regmatch_slab * const osl = s; |
| 3003 | s = s->next; |
| 3004 | Safefree(osl); |
| 3005 | } |
| 3006 | } |
| 3007 | |
| 3008 | |
| 3009 | #define SETREX(Re1,Re2) \ |
| 3010 | if (PL_reg_eval_set) PM_SETRE((PL_reg_curpm), (Re2)); \ |
| 3011 | Re1 = (Re2) |
| 3012 | |
| 3013 | STATIC I32 /* 0 failure, 1 success */ |
| 3014 | S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog) |
| 3015 | { |
| 3016 | #if PERL_VERSION < 9 && !defined(PERL_CORE) |
| 3017 | dMY_CXT; |
| 3018 | #endif |
| 3019 | dVAR; |
| 3020 | register const bool utf8_target = PL_reg_match_utf8; |
| 3021 | const U32 uniflags = UTF8_ALLOW_DEFAULT; |
| 3022 | REGEXP *rex_sv = reginfo->prog; |
| 3023 | regexp *rex = (struct regexp *)SvANY(rex_sv); |
| 3024 | RXi_GET_DECL(rex,rexi); |
| 3025 | I32 oldsave; |
| 3026 | /* the current state. This is a cached copy of PL_regmatch_state */ |
| 3027 | register regmatch_state *st; |
| 3028 | /* cache heavy used fields of st in registers */ |
| 3029 | register regnode *scan; |
| 3030 | register regnode *next; |
| 3031 | register U32 n = 0; /* general value; init to avoid compiler warning */ |
| 3032 | register I32 ln = 0; /* len or last; init to avoid compiler warning */ |
| 3033 | register char *locinput = PL_reginput; |
| 3034 | register I32 nextchr; /* is always set to UCHARAT(locinput) */ |
| 3035 | |
| 3036 | bool result = 0; /* return value of S_regmatch */ |
| 3037 | int depth = 0; /* depth of backtrack stack */ |
| 3038 | U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */ |
| 3039 | const U32 max_nochange_depth = |
| 3040 | (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ? |
| 3041 | 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH; |
| 3042 | regmatch_state *yes_state = NULL; /* state to pop to on success of |
| 3043 | subpattern */ |
| 3044 | /* mark_state piggy backs on the yes_state logic so that when we unwind |
| 3045 | the stack on success we can update the mark_state as we go */ |
| 3046 | regmatch_state *mark_state = NULL; /* last mark state we have seen */ |
| 3047 | regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */ |
| 3048 | struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */ |
| 3049 | U32 state_num; |
| 3050 | bool no_final = 0; /* prevent failure from backtracking? */ |
| 3051 | bool do_cutgroup = 0; /* no_final only until next branch/trie entry */ |
| 3052 | char *startpoint = PL_reginput; |
| 3053 | SV *popmark = NULL; /* are we looking for a mark? */ |
| 3054 | SV *sv_commit = NULL; /* last mark name seen in failure */ |
| 3055 | SV *sv_yes_mark = NULL; /* last mark name we have seen |
| 3056 | during a successful match */ |
| 3057 | U32 lastopen = 0; /* last open we saw */ |
| 3058 | bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0; |
| 3059 | SV* const oreplsv = GvSV(PL_replgv); |
| 3060 | /* these three flags are set by various ops to signal information to |
| 3061 | * the very next op. They have a useful lifetime of exactly one loop |
| 3062 | * iteration, and are not preserved or restored by state pushes/pops |
| 3063 | */ |
| 3064 | bool sw = 0; /* the condition value in (?(cond)a|b) */ |
| 3065 | bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */ |
| 3066 | int logical = 0; /* the following EVAL is: |
| 3067 | 0: (?{...}) |
| 3068 | 1: (?(?{...})X|Y) |
| 3069 | 2: (??{...}) |
| 3070 | or the following IFMATCH/UNLESSM is: |
| 3071 | false: plain (?=foo) |
| 3072 | true: used as a condition: (?(?=foo)) |
| 3073 | */ |
| 3074 | #ifdef DEBUGGING |
| 3075 | GET_RE_DEBUG_FLAGS_DECL; |
| 3076 | #endif |
| 3077 | |
| 3078 | PERL_ARGS_ASSERT_REGMATCH; |
| 3079 | |
| 3080 | DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({ |
| 3081 | PerlIO_printf(Perl_debug_log,"regmatch start\n"); |
| 3082 | })); |
| 3083 | /* on first ever call to regmatch, allocate first slab */ |
| 3084 | if (!PL_regmatch_slab) { |
| 3085 | Newx(PL_regmatch_slab, 1, regmatch_slab); |
| 3086 | PL_regmatch_slab->prev = NULL; |
| 3087 | PL_regmatch_slab->next = NULL; |
| 3088 | PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab); |
| 3089 | } |
| 3090 | |
| 3091 | oldsave = PL_savestack_ix; |
| 3092 | SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL); |
| 3093 | SAVEVPTR(PL_regmatch_slab); |
| 3094 | SAVEVPTR(PL_regmatch_state); |
| 3095 | |
| 3096 | /* grab next free state slot */ |
| 3097 | st = ++PL_regmatch_state; |
| 3098 | if (st > SLAB_LAST(PL_regmatch_slab)) |
| 3099 | st = PL_regmatch_state = S_push_slab(aTHX); |
| 3100 | |
| 3101 | /* Note that nextchr is a byte even in UTF */ |
| 3102 | nextchr = UCHARAT(locinput); |
| 3103 | scan = prog; |
| 3104 | while (scan != NULL) { |
| 3105 | |
| 3106 | DEBUG_EXECUTE_r( { |
| 3107 | SV * const prop = sv_newmortal(); |
| 3108 | regnode *rnext=regnext(scan); |
| 3109 | DUMP_EXEC_POS( locinput, scan, utf8_target ); |
| 3110 | regprop(rex, prop, scan); |
| 3111 | |
| 3112 | PerlIO_printf(Perl_debug_log, |
| 3113 | "%3"IVdf":%*s%s(%"IVdf")\n", |
| 3114 | (IV)(scan - rexi->program), depth*2, "", |
| 3115 | SvPVX_const(prop), |
| 3116 | (PL_regkind[OP(scan)] == END || !rnext) ? |
| 3117 | 0 : (IV)(rnext - rexi->program)); |
| 3118 | }); |
| 3119 | |
| 3120 | next = scan + NEXT_OFF(scan); |
| 3121 | if (next == scan) |
| 3122 | next = NULL; |
| 3123 | state_num = OP(scan); |
| 3124 | |
| 3125 | reenter_switch: |
| 3126 | |
| 3127 | assert(PL_reglastparen == &rex->lastparen); |
| 3128 | assert(PL_reglastcloseparen == &rex->lastcloseparen); |
| 3129 | assert(PL_regoffs == rex->offs); |
| 3130 | |
| 3131 | switch (state_num) { |
| 3132 | case BOL: |
| 3133 | if (locinput == PL_bostr) |
| 3134 | { |
| 3135 | /* reginfo->till = reginfo->bol; */ |
| 3136 | break; |
| 3137 | } |
| 3138 | sayNO; |
| 3139 | case MBOL: |
| 3140 | if (locinput == PL_bostr || |
| 3141 | ((nextchr || locinput < PL_regeol) && locinput[-1] == '\n')) |
| 3142 | { |
| 3143 | break; |
| 3144 | } |
| 3145 | sayNO; |
| 3146 | case SBOL: |
| 3147 | if (locinput == PL_bostr) |
| 3148 | break; |
| 3149 | sayNO; |
| 3150 | case GPOS: |
| 3151 | if (locinput == reginfo->ganch) |
| 3152 | break; |
| 3153 | sayNO; |
| 3154 | |
| 3155 | case KEEPS: |
| 3156 | /* update the startpoint */ |
| 3157 | st->u.keeper.val = PL_regoffs[0].start; |
| 3158 | PL_reginput = locinput; |
| 3159 | PL_regoffs[0].start = locinput - PL_bostr; |
| 3160 | PUSH_STATE_GOTO(KEEPS_next, next); |
| 3161 | /*NOT-REACHED*/ |
| 3162 | case KEEPS_next_fail: |
| 3163 | /* rollback the start point change */ |
| 3164 | PL_regoffs[0].start = st->u.keeper.val; |
| 3165 | sayNO_SILENT; |
| 3166 | /*NOT-REACHED*/ |
| 3167 | case EOL: |
| 3168 | goto seol; |
| 3169 | case MEOL: |
| 3170 | if ((nextchr || locinput < PL_regeol) && nextchr != '\n') |
| 3171 | sayNO; |
| 3172 | break; |
| 3173 | case SEOL: |
| 3174 | seol: |
| 3175 | if ((nextchr || locinput < PL_regeol) && nextchr != '\n') |
| 3176 | sayNO; |
| 3177 | if (PL_regeol - locinput > 1) |
| 3178 | sayNO; |
| 3179 | break; |
| 3180 | case EOS: |
| 3181 | if (PL_regeol != locinput) |
| 3182 | sayNO; |
| 3183 | break; |
| 3184 | case SANY: |
| 3185 | if (!nextchr && locinput >= PL_regeol) |
| 3186 | sayNO; |
| 3187 | if (utf8_target) { |
| 3188 | locinput += PL_utf8skip[nextchr]; |
| 3189 | if (locinput > PL_regeol) |
| 3190 | sayNO; |
| 3191 | nextchr = UCHARAT(locinput); |
| 3192 | } |
| 3193 | else |
| 3194 | nextchr = UCHARAT(++locinput); |
| 3195 | break; |
| 3196 | case CANY: |
| 3197 | if (!nextchr && locinput >= PL_regeol) |
| 3198 | sayNO; |
| 3199 | nextchr = UCHARAT(++locinput); |
| 3200 | break; |
| 3201 | case REG_ANY: |
| 3202 | if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n') |
| 3203 | sayNO; |
| 3204 | if (utf8_target) { |
| 3205 | locinput += PL_utf8skip[nextchr]; |
| 3206 | if (locinput > PL_regeol) |
| 3207 | sayNO; |
| 3208 | nextchr = UCHARAT(locinput); |
| 3209 | } |
| 3210 | else |
| 3211 | nextchr = UCHARAT(++locinput); |
| 3212 | break; |
| 3213 | |
| 3214 | #undef ST |
| 3215 | #define ST st->u.trie |
| 3216 | case TRIEC: |
| 3217 | /* In this case the charclass data is available inline so |
| 3218 | we can fail fast without a lot of extra overhead. |
| 3219 | */ |
| 3220 | if (scan->flags == EXACT || !utf8_target) { |
| 3221 | if(!ANYOF_BITMAP_TEST(scan, *locinput)) { |
| 3222 | DEBUG_EXECUTE_r( |
| 3223 | PerlIO_printf(Perl_debug_log, |
| 3224 | "%*s %sfailed to match trie start class...%s\n", |
| 3225 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) |
| 3226 | ); |
| 3227 | sayNO_SILENT; |
| 3228 | /* NOTREACHED */ |
| 3229 | } |
| 3230 | } |
| 3231 | /* FALL THROUGH */ |
| 3232 | case TRIE: |
| 3233 | /* the basic plan of execution of the trie is: |
| 3234 | * At the beginning, run though all the states, and |
| 3235 | * find the longest-matching word. Also remember the position |
| 3236 | * of the shortest matching word. For example, this pattern: |
| 3237 | * 1 2 3 4 5 |
| 3238 | * ab|a|x|abcd|abc |
| 3239 | * when matched against the string "abcde", will generate |
| 3240 | * accept states for all words except 3, with the longest |
| 3241 | * matching word being 4, and the shortest being 1 (with |
| 3242 | * the position being after char 1 of the string). |
| 3243 | * |
| 3244 | * Then for each matching word, in word order (i.e. 1,2,4,5), |
| 3245 | * we run the remainder of the pattern; on each try setting |
| 3246 | * the current position to the character following the word, |
| 3247 | * returning to try the next word on failure. |
| 3248 | * |
| 3249 | * We avoid having to build a list of words at runtime by |
| 3250 | * using a compile-time structure, wordinfo[].prev, which |
| 3251 | * gives, for each word, the previous accepting word (if any). |
| 3252 | * In the case above it would contain the mappings 1->2, 2->0, |
| 3253 | * 3->0, 4->5, 5->1. We can use this table to generate, from |
| 3254 | * the longest word (4 above), a list of all words, by |
| 3255 | * following the list of prev pointers; this gives us the |
| 3256 | * unordered list 4,5,1,2. Then given the current word we have |
| 3257 | * just tried, we can go through the list and find the |
| 3258 | * next-biggest word to try (so if we just failed on word 2, |
| 3259 | * the next in the list is 4). |
| 3260 | * |
| 3261 | * Since at runtime we don't record the matching position in |
| 3262 | * the string for each word, we have to work that out for |
| 3263 | * each word we're about to process. The wordinfo table holds |
| 3264 | * the character length of each word; given that we recorded |
| 3265 | * at the start: the position of the shortest word and its |
| 3266 | * length in chars, we just need to move the pointer the |
| 3267 | * difference between the two char lengths. Depending on |
| 3268 | * Unicode status and folding, that's cheap or expensive. |
| 3269 | * |
| 3270 | * This algorithm is optimised for the case where are only a |
| 3271 | * small number of accept states, i.e. 0,1, or maybe 2. |
| 3272 | * With lots of accepts states, and having to try all of them, |
| 3273 | * it becomes quadratic on number of accept states to find all |
| 3274 | * the next words. |
| 3275 | */ |
| 3276 | |
| 3277 | { |
| 3278 | /* what type of TRIE am I? (utf8 makes this contextual) */ |
| 3279 | DECL_TRIE_TYPE(scan); |
| 3280 | |
| 3281 | /* what trie are we using right now */ |
| 3282 | reg_trie_data * const trie |
| 3283 | = (reg_trie_data*)rexi->data->data[ ARG( scan ) ]; |
| 3284 | HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]); |
| 3285 | U32 state = trie->startstate; |
| 3286 | |
| 3287 | if (trie->bitmap && trie_type != trie_utf8_fold && |
| 3288 | !TRIE_BITMAP_TEST(trie,*locinput) |
| 3289 | ) { |
| 3290 | if (trie->states[ state ].wordnum) { |
| 3291 | DEBUG_EXECUTE_r( |
| 3292 | PerlIO_printf(Perl_debug_log, |
| 3293 | "%*s %smatched empty string...%s\n", |
| 3294 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) |
| 3295 | ); |
| 3296 | if (!trie->jump) |
| 3297 | break; |
| 3298 | } else { |
| 3299 | DEBUG_EXECUTE_r( |
| 3300 | PerlIO_printf(Perl_debug_log, |
| 3301 | "%*s %sfailed to match trie start class...%s\n", |
| 3302 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) |
| 3303 | ); |
| 3304 | sayNO_SILENT; |
| 3305 | } |
| 3306 | } |
| 3307 | |
| 3308 | { |
| 3309 | U8 *uc = ( U8* )locinput; |
| 3310 | |
| 3311 | STRLEN len = 0; |
| 3312 | STRLEN foldlen = 0; |
| 3313 | U8 *uscan = (U8*)NULL; |
| 3314 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
| 3315 | U32 charcount = 0; /* how many input chars we have matched */ |
| 3316 | U32 accepted = 0; /* have we seen any accepting states? */ |
| 3317 | |
| 3318 | ST.B = next; |
| 3319 | ST.jump = trie->jump; |
| 3320 | ST.me = scan; |
| 3321 | ST.firstpos = NULL; |
| 3322 | ST.longfold = FALSE; /* char longer if folded => it's harder */ |
| 3323 | ST.nextword = 0; |
| 3324 | |
| 3325 | /* fully traverse the TRIE; note the position of the |
| 3326 | shortest accept state and the wordnum of the longest |
| 3327 | accept state */ |
| 3328 | |
| 3329 | while ( state && uc <= (U8*)PL_regeol ) { |
| 3330 | U32 base = trie->states[ state ].trans.base; |
| 3331 | UV uvc = 0; |
| 3332 | U16 charid = 0; |
| 3333 | U16 wordnum; |
| 3334 | wordnum = trie->states[ state ].wordnum; |
| 3335 | |
| 3336 | if (wordnum) { /* it's an accept state */ |
| 3337 | if (!accepted) { |
| 3338 | accepted = 1; |
| 3339 | /* record first match position */ |
| 3340 | if (ST.longfold) { |
| 3341 | ST.firstpos = (U8*)locinput; |
| 3342 | ST.firstchars = 0; |
| 3343 | } |
| 3344 | else { |
| 3345 | ST.firstpos = uc; |
| 3346 | ST.firstchars = charcount; |
| 3347 | } |
| 3348 | } |
| 3349 | if (!ST.nextword || wordnum < ST.nextword) |
| 3350 | ST.nextword = wordnum; |
| 3351 | ST.topword = wordnum; |
| 3352 | } |
| 3353 | |
| 3354 | DEBUG_TRIE_EXECUTE_r({ |
| 3355 | DUMP_EXEC_POS( (char *)uc, scan, utf8_target ); |
| 3356 | PerlIO_printf( Perl_debug_log, |
| 3357 | "%*s %sState: %4"UVxf" Accepted: %c ", |
| 3358 | 2+depth * 2, "", PL_colors[4], |
| 3359 | (UV)state, (accepted ? 'Y' : 'N')); |
| 3360 | }); |
| 3361 | |
| 3362 | /* read a char and goto next state */ |
| 3363 | if ( base ) { |
| 3364 | I32 offset; |
| 3365 | REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, |
| 3366 | uscan, len, uvc, charid, foldlen, |
| 3367 | foldbuf, uniflags); |
| 3368 | charcount++; |
| 3369 | if (foldlen>0) |
| 3370 | ST.longfold = TRUE; |
| 3371 | if (charid && |
| 3372 | ( ((offset = |
| 3373 | base + charid - 1 - trie->uniquecharcount)) >= 0) |
| 3374 | |
| 3375 | && ((U32)offset < trie->lasttrans) |
| 3376 | && trie->trans[offset].check == state) |
| 3377 | { |
| 3378 | state = trie->trans[offset].next; |
| 3379 | } |
| 3380 | else { |
| 3381 | state = 0; |
| 3382 | } |
| 3383 | uc += len; |
| 3384 | |
| 3385 | } |
| 3386 | else { |
| 3387 | state = 0; |
| 3388 | } |
| 3389 | DEBUG_TRIE_EXECUTE_r( |
| 3390 | PerlIO_printf( Perl_debug_log, |
| 3391 | "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n", |
| 3392 | charid, uvc, (UV)state, PL_colors[5] ); |
| 3393 | ); |
| 3394 | } |
| 3395 | if (!accepted) |
| 3396 | sayNO; |
| 3397 | |
| 3398 | /* calculate total number of accept states */ |
| 3399 | { |
| 3400 | U16 w = ST.topword; |
| 3401 | accepted = 0; |
| 3402 | while (w) { |
| 3403 | w = trie->wordinfo[w].prev; |
| 3404 | accepted++; |
| 3405 | } |
| 3406 | ST.accepted = accepted; |
| 3407 | } |
| 3408 | |
| 3409 | DEBUG_EXECUTE_r( |
| 3410 | PerlIO_printf( Perl_debug_log, |
| 3411 | "%*s %sgot %"IVdf" possible matches%s\n", |
| 3412 | REPORT_CODE_OFF + depth * 2, "", |
| 3413 | PL_colors[4], (IV)ST.accepted, PL_colors[5] ); |
| 3414 | ); |
| 3415 | goto trie_first_try; /* jump into the fail handler */ |
| 3416 | }} |
| 3417 | /* NOTREACHED */ |
| 3418 | |
| 3419 | case TRIE_next_fail: /* we failed - try next alternative */ |
| 3420 | if ( ST.jump) { |
| 3421 | REGCP_UNWIND(ST.cp); |
| 3422 | for (n = *PL_reglastparen; n > ST.lastparen; n--) |
| 3423 | PL_regoffs[n].end = -1; |
| 3424 | *PL_reglastparen = n; |
| 3425 | } |
| 3426 | if (!--ST.accepted) { |
| 3427 | DEBUG_EXECUTE_r({ |
| 3428 | PerlIO_printf( Perl_debug_log, |
| 3429 | "%*s %sTRIE failed...%s\n", |
| 3430 | REPORT_CODE_OFF+depth*2, "", |
| 3431 | PL_colors[4], |
| 3432 | PL_colors[5] ); |
| 3433 | }); |
| 3434 | sayNO_SILENT; |
| 3435 | } |
| 3436 | { |
| 3437 | /* Find next-highest word to process. Note that this code |
| 3438 | * is O(N^2) per trie run (O(N) per branch), so keep tight */ |
| 3439 | register U16 min = 0; |
| 3440 | register U16 word; |
| 3441 | register U16 const nextword = ST.nextword; |
| 3442 | register reg_trie_wordinfo * const wordinfo |
| 3443 | = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo; |
| 3444 | for (word=ST.topword; word; word=wordinfo[word].prev) { |
| 3445 | if (word > nextword && (!min || word < min)) |
| 3446 | min = word; |
| 3447 | } |
| 3448 | ST.nextword = min; |
| 3449 | } |
| 3450 | |
| 3451 | trie_first_try: |
| 3452 | if (do_cutgroup) { |
| 3453 | do_cutgroup = 0; |
| 3454 | no_final = 0; |
| 3455 | } |
| 3456 | |
| 3457 | if ( ST.jump) { |
| 3458 | ST.lastparen = *PL_reglastparen; |
| 3459 | REGCP_SET(ST.cp); |
| 3460 | } |
| 3461 | |
| 3462 | /* find start char of end of current word */ |
| 3463 | { |
| 3464 | U32 chars; /* how many chars to skip */ |
| 3465 | U8 *uc = ST.firstpos; |
| 3466 | reg_trie_data * const trie |
| 3467 | = (reg_trie_data*)rexi->data->data[ARG(ST.me)]; |
| 3468 | |
| 3469 | assert((trie->wordinfo[ST.nextword].len - trie->prefixlen) |
| 3470 | >= ST.firstchars); |
| 3471 | chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen) |
| 3472 | - ST.firstchars; |
| 3473 | |
| 3474 | if (ST.longfold) { |
| 3475 | /* the hard option - fold each char in turn and find |
| 3476 | * its folded length (which may be different */ |
| 3477 | U8 foldbuf[UTF8_MAXBYTES_CASE + 1]; |
| 3478 | STRLEN foldlen; |
| 3479 | STRLEN len; |
| 3480 | UV uvc; |
| 3481 | U8 *uscan; |
| 3482 | |
| 3483 | while (chars) { |
| 3484 | if (utf8_target) { |
| 3485 | uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len, |
| 3486 | uniflags); |
| 3487 | uc += len; |
| 3488 | } |
| 3489 | else { |
| 3490 | uvc = *uc; |
| 3491 | uc++; |
| 3492 | } |
| 3493 | uvc = to_uni_fold(uvc, foldbuf, &foldlen); |
| 3494 | uscan = foldbuf; |
| 3495 | while (foldlen) { |
| 3496 | if (!--chars) |
| 3497 | break; |
| 3498 | uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len, |
| 3499 | uniflags); |
| 3500 | uscan += len; |
| 3501 | foldlen -= len; |
| 3502 | } |
| 3503 | } |
| 3504 | } |
| 3505 | else { |
| 3506 | if (utf8_target) |
| 3507 | while (chars--) |
| 3508 | uc += UTF8SKIP(uc); |
| 3509 | else |
| 3510 | uc += chars; |
| 3511 | } |
| 3512 | PL_reginput = (char *)uc; |
| 3513 | } |
| 3514 | |
| 3515 | scan = (ST.jump && ST.jump[ST.nextword]) |
| 3516 | ? ST.me + ST.jump[ST.nextword] |
| 3517 | : ST.B; |
| 3518 | |
| 3519 | DEBUG_EXECUTE_r({ |
| 3520 | PerlIO_printf( Perl_debug_log, |
| 3521 | "%*s %sTRIE matched word #%d, continuing%s\n", |
| 3522 | REPORT_CODE_OFF+depth*2, "", |
| 3523 | PL_colors[4], |
| 3524 | ST.nextword, |
| 3525 | PL_colors[5] |
| 3526 | ); |
| 3527 | }); |
| 3528 | |
| 3529 | if (ST.accepted > 1 || has_cutgroup) { |
| 3530 | PUSH_STATE_GOTO(TRIE_next, scan); |
| 3531 | /* NOTREACHED */ |
| 3532 | } |
| 3533 | /* only one choice left - just continue */ |
| 3534 | DEBUG_EXECUTE_r({ |
| 3535 | AV *const trie_words |
| 3536 | = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]); |
| 3537 | SV ** const tmp = av_fetch( trie_words, |
| 3538 | ST.nextword-1, 0 ); |
| 3539 | SV *sv= tmp ? sv_newmortal() : NULL; |
| 3540 | |
| 3541 | PerlIO_printf( Perl_debug_log, |
| 3542 | "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n", |
| 3543 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], |
| 3544 | ST.nextword, |
| 3545 | tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0, |
| 3546 | PL_colors[0], PL_colors[1], |
| 3547 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII |
| 3548 | ) |
| 3549 | : "not compiled under -Dr", |
| 3550 | PL_colors[5] ); |
| 3551 | }); |
| 3552 | |
| 3553 | locinput = PL_reginput; |
| 3554 | nextchr = UCHARAT(locinput); |
| 3555 | continue; /* execute rest of RE */ |
| 3556 | /* NOTREACHED */ |
| 3557 | #undef ST |
| 3558 | |
| 3559 | case EXACT: { |
| 3560 | char *s = STRING(scan); |
| 3561 | ln = STR_LEN(scan); |
| 3562 | if (utf8_target != UTF_PATTERN) { |
| 3563 | /* The target and the pattern have differing utf8ness. */ |
| 3564 | char *l = locinput; |
| 3565 | const char * const e = s + ln; |
| 3566 | |
| 3567 | if (utf8_target) { |
| 3568 | /* The target is utf8, the pattern is not utf8. */ |
| 3569 | while (s < e) { |
| 3570 | STRLEN ulen; |
| 3571 | if (l >= PL_regeol) |
| 3572 | sayNO; |
| 3573 | if (NATIVE_TO_UNI(*(U8*)s) != |
| 3574 | utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen, |
| 3575 | uniflags)) |
| 3576 | sayNO; |
| 3577 | l += ulen; |
| 3578 | s ++; |
| 3579 | } |
| 3580 | } |
| 3581 | else { |
| 3582 | /* The target is not utf8, the pattern is utf8. */ |
| 3583 | while (s < e) { |
| 3584 | STRLEN ulen; |
| 3585 | if (l >= PL_regeol) |
| 3586 | sayNO; |
| 3587 | if (NATIVE_TO_UNI(*((U8*)l)) != |
| 3588 | utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen, |
| 3589 | uniflags)) |
| 3590 | sayNO; |
| 3591 | s += ulen; |
| 3592 | l ++; |
| 3593 | } |
| 3594 | } |
| 3595 | locinput = l; |
| 3596 | nextchr = UCHARAT(locinput); |
| 3597 | break; |
| 3598 | } |
| 3599 | /* The target and the pattern have the same utf8ness. */ |
| 3600 | /* Inline the first character, for speed. */ |
| 3601 | if (UCHARAT(s) != nextchr) |
| 3602 | sayNO; |
| 3603 | if (PL_regeol - locinput < ln) |
| 3604 | sayNO; |
| 3605 | if (ln > 1 && memNE(s, locinput, ln)) |
| 3606 | sayNO; |
| 3607 | locinput += ln; |
| 3608 | nextchr = UCHARAT(locinput); |
| 3609 | break; |
| 3610 | } |
| 3611 | case EXACTFL: { |
| 3612 | re_fold_t folder; |
| 3613 | const U8 * fold_array; |
| 3614 | const char * s; |
| 3615 | U32 fold_utf8_flags; |
| 3616 | |
| 3617 | PL_reg_flags |= RF_tainted; |
| 3618 | folder = foldEQ_locale; |
| 3619 | fold_array = PL_fold_locale; |
| 3620 | fold_utf8_flags = FOLDEQ_UTF8_LOCALE; |
| 3621 | goto do_exactf; |
| 3622 | |
| 3623 | case EXACTFU: |
| 3624 | folder = foldEQ_latin1; |
| 3625 | fold_array = PL_fold_latin1; |
| 3626 | fold_utf8_flags = 0; |
| 3627 | goto do_exactf; |
| 3628 | |
| 3629 | case EXACTFA: |
| 3630 | folder = foldEQ_latin1; |
| 3631 | fold_array = PL_fold_latin1; |
| 3632 | fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 3633 | goto do_exactf; |
| 3634 | |
| 3635 | case EXACTF: |
| 3636 | folder = foldEQ; |
| 3637 | fold_array = PL_fold; |
| 3638 | fold_utf8_flags = 0; |
| 3639 | |
| 3640 | do_exactf: |
| 3641 | s = STRING(scan); |
| 3642 | ln = STR_LEN(scan); |
| 3643 | |
| 3644 | if (utf8_target || UTF_PATTERN) { |
| 3645 | /* Either target or the pattern are utf8. */ |
| 3646 | const char * const l = locinput; |
| 3647 | char *e = PL_regeol; |
| 3648 | |
| 3649 | if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN), |
| 3650 | l, &e, 0, utf8_target, fold_utf8_flags)) |
| 3651 | { |
| 3652 | sayNO; |
| 3653 | } |
| 3654 | locinput = e; |
| 3655 | nextchr = UCHARAT(locinput); |
| 3656 | break; |
| 3657 | } |
| 3658 | |
| 3659 | /* Neither the target nor the pattern are utf8 */ |
| 3660 | if (UCHARAT(s) != nextchr && |
| 3661 | UCHARAT(s) != fold_array[nextchr]) |
| 3662 | { |
| 3663 | sayNO; |
| 3664 | } |
| 3665 | if (PL_regeol - locinput < ln) |
| 3666 | sayNO; |
| 3667 | if (ln > 1 && ! folder(s, locinput, ln)) |
| 3668 | sayNO; |
| 3669 | locinput += ln; |
| 3670 | nextchr = UCHARAT(locinput); |
| 3671 | break; |
| 3672 | } |
| 3673 | |
| 3674 | /* XXX Could improve efficiency by separating these all out using a |
| 3675 | * macro or in-line function. At that point regcomp.c would no longer |
| 3676 | * have to set the FLAGS fields of these */ |
| 3677 | case BOUNDL: |
| 3678 | case NBOUNDL: |
| 3679 | PL_reg_flags |= RF_tainted; |
| 3680 | /* FALL THROUGH */ |
| 3681 | case BOUND: |
| 3682 | case BOUNDU: |
| 3683 | case BOUNDA: |
| 3684 | case NBOUND: |
| 3685 | case NBOUNDU: |
| 3686 | case NBOUNDA: |
| 3687 | /* was last char in word? */ |
| 3688 | if (utf8_target && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET) { |
| 3689 | if (locinput == PL_bostr) |
| 3690 | ln = '\n'; |
| 3691 | else { |
| 3692 | const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr); |
| 3693 | |
| 3694 | ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags); |
| 3695 | } |
| 3696 | if (FLAGS(scan) != REGEX_LOCALE_CHARSET) { |
| 3697 | ln = isALNUM_uni(ln); |
| 3698 | LOAD_UTF8_CHARCLASS_ALNUM(); |
| 3699 | n = swash_fetch(PL_utf8_alnum, (U8*)locinput, utf8_target); |
| 3700 | } |
| 3701 | else { |
| 3702 | ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln)); |
| 3703 | n = isALNUM_LC_utf8((U8*)locinput); |
| 3704 | } |
| 3705 | } |
| 3706 | else { |
| 3707 | |
| 3708 | /* Here the string isn't utf8, or is utf8 and only ascii |
| 3709 | * characters are to match \w. In the latter case looking at |
| 3710 | * the byte just prior to the current one may be just the final |
| 3711 | * byte of a multi-byte character. This is ok. There are two |
| 3712 | * cases: |
| 3713 | * 1) it is a single byte character, and then the test is doing |
| 3714 | * just what it's supposed to. |
| 3715 | * 2) it is a multi-byte character, in which case the final |
| 3716 | * byte is never mistakable for ASCII, and so the test |
| 3717 | * will say it is not a word character, which is the |
| 3718 | * correct answer. */ |
| 3719 | ln = (locinput != PL_bostr) ? |
| 3720 | UCHARAT(locinput - 1) : '\n'; |
| 3721 | switch (FLAGS(scan)) { |
| 3722 | case REGEX_UNICODE_CHARSET: |
| 3723 | ln = isWORDCHAR_L1(ln); |
| 3724 | n = isWORDCHAR_L1(nextchr); |
| 3725 | break; |
| 3726 | case REGEX_LOCALE_CHARSET: |
| 3727 | ln = isALNUM_LC(ln); |
| 3728 | n = isALNUM_LC(nextchr); |
| 3729 | break; |
| 3730 | case REGEX_DEPENDS_CHARSET: |
| 3731 | ln = isALNUM(ln); |
| 3732 | n = isALNUM(nextchr); |
| 3733 | break; |
| 3734 | case REGEX_ASCII_RESTRICTED_CHARSET: |
| 3735 | ln = isWORDCHAR_A(ln); |
| 3736 | n = isWORDCHAR_A(nextchr); |
| 3737 | break; |
| 3738 | default: |
| 3739 | Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan)); |
| 3740 | break; |
| 3741 | } |
| 3742 | } |
| 3743 | /* Note requires that all BOUNDs be lower than all NBOUNDs in |
| 3744 | * regcomp.sym */ |
| 3745 | if (((!ln) == (!n)) == (OP(scan) < NBOUND)) |
| 3746 | sayNO; |
| 3747 | break; |
| 3748 | case ANYOFV: |
| 3749 | case ANYOF: |
| 3750 | if (utf8_target || state_num == ANYOFV) { |
| 3751 | STRLEN inclasslen = PL_regeol - locinput; |
| 3752 | if (locinput >= PL_regeol) |
| 3753 | sayNO; |
| 3754 | |
| 3755 | if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target)) |
| 3756 | sayNO; |
| 3757 | locinput += inclasslen; |
| 3758 | nextchr = UCHARAT(locinput); |
| 3759 | break; |
| 3760 | } |
| 3761 | else { |
| 3762 | if (nextchr < 0) |
| 3763 | nextchr = UCHARAT(locinput); |
| 3764 | if (!nextchr && locinput >= PL_regeol) |
| 3765 | sayNO; |
| 3766 | if (!REGINCLASS(rex, scan, (U8*)locinput)) |
| 3767 | sayNO; |
| 3768 | nextchr = UCHARAT(++locinput); |
| 3769 | break; |
| 3770 | } |
| 3771 | break; |
| 3772 | /* Special char classes - The defines start on line 129 or so */ |
| 3773 | CCC_TRY_U(ALNUM, NALNUM, isWORDCHAR, |
| 3774 | ALNUML, NALNUML, isALNUM_LC, isALNUM_LC_utf8, |
| 3775 | ALNUMU, NALNUMU, isWORDCHAR_L1, |
| 3776 | ALNUMA, NALNUMA, isWORDCHAR_A, |
| 3777 | alnum, "a"); |
| 3778 | |
| 3779 | CCC_TRY_U(SPACE, NSPACE, isSPACE, |
| 3780 | SPACEL, NSPACEL, isSPACE_LC, isSPACE_LC_utf8, |
| 3781 | SPACEU, NSPACEU, isSPACE_L1, |
| 3782 | SPACEA, NSPACEA, isSPACE_A, |
| 3783 | space, " "); |
| 3784 | |
| 3785 | CCC_TRY(DIGIT, NDIGIT, isDIGIT, |
| 3786 | DIGITL, NDIGITL, isDIGIT_LC, isDIGIT_LC_utf8, |
| 3787 | DIGITA, NDIGITA, isDIGIT_A, |
| 3788 | digit, "0"); |
| 3789 | |
| 3790 | case CLUMP: /* Match \X: logical Unicode character. This is defined as |
| 3791 | a Unicode extended Grapheme Cluster */ |
| 3792 | /* From http://www.unicode.org/reports/tr29 (5.2 version). An |
| 3793 | extended Grapheme Cluster is: |
| 3794 | |
| 3795 | CR LF |
| 3796 | | Prepend* Begin Extend* |
| 3797 | | . |
| 3798 | |
| 3799 | Begin is (Hangul-syllable | ! Control) |
| 3800 | Extend is (Grapheme_Extend | Spacing_Mark) |
| 3801 | Control is [ GCB_Control CR LF ] |
| 3802 | |
| 3803 | The discussion below shows how the code for CLUMP is derived |
| 3804 | from this regex. Note that most of these concepts are from |
| 3805 | property values of the Grapheme Cluster Boundary (GCB) property. |
| 3806 | No code point can have multiple property values for a given |
| 3807 | property. Thus a code point in Prepend can't be in Control, but |
| 3808 | it must be in !Control. This is why Control above includes |
| 3809 | GCB_Control plus CR plus LF. The latter two are used in the GCB |
| 3810 | property separately, and so can't be in GCB_Control, even though |
| 3811 | they logically are controls. Control is not the same as gc=cc, |
| 3812 | but includes format and other characters as well. |
| 3813 | |
| 3814 | The Unicode definition of Hangul-syllable is: |
| 3815 | L+ |
| 3816 | | (L* ( ( V | LV ) V* | LVT ) T*) |
| 3817 | | T+ |
| 3818 | ) |
| 3819 | Each of these is a value for the GCB property, and hence must be |
| 3820 | disjoint, so the order they are tested is immaterial, so the |
| 3821 | above can safely be changed to |
| 3822 | T+ |
| 3823 | | L+ |
| 3824 | | (L* ( LVT | ( V | LV ) V*) T*) |
| 3825 | |
| 3826 | The last two terms can be combined like this: |
| 3827 | L* ( L |
| 3828 | | (( LVT | ( V | LV ) V*) T*)) |
| 3829 | |
| 3830 | And refactored into this: |
| 3831 | L* (L | LVT T* | V V* T* | LV V* T*) |
| 3832 | |
| 3833 | That means that if we have seen any L's at all we can quit |
| 3834 | there, but if the next character is an LVT, a V, or an LV we |
| 3835 | should keep going. |
| 3836 | |
| 3837 | There is a subtlety with Prepend* which showed up in testing. |
| 3838 | Note that the Begin, and only the Begin is required in: |
| 3839 | | Prepend* Begin Extend* |
| 3840 | Also, Begin contains '! Control'. A Prepend must be a |
| 3841 | '! Control', which means it must also be a Begin. What it |
| 3842 | comes down to is that if we match Prepend* and then find no |
| 3843 | suitable Begin afterwards, that if we backtrack the last |
| 3844 | Prepend, that one will be a suitable Begin. |
| 3845 | */ |
| 3846 | |
| 3847 | if (locinput >= PL_regeol) |
| 3848 | sayNO; |
| 3849 | if (! utf8_target) { |
| 3850 | |
| 3851 | /* Match either CR LF or '.', as all the other possibilities |
| 3852 | * require utf8 */ |
| 3853 | locinput++; /* Match the . or CR */ |
| 3854 | if (nextchr == '\r' /* And if it was CR, and the next is LF, |
| 3855 | match the LF */ |
| 3856 | && locinput < PL_regeol |
| 3857 | && UCHARAT(locinput) == '\n') locinput++; |
| 3858 | } |
| 3859 | else { |
| 3860 | |
| 3861 | /* Utf8: See if is ( CR LF ); already know that locinput < |
| 3862 | * PL_regeol, so locinput+1 is in bounds */ |
| 3863 | if (nextchr == '\r' && UCHARAT(locinput + 1) == '\n') { |
| 3864 | locinput += 2; |
| 3865 | } |
| 3866 | else { |
| 3867 | /* In case have to backtrack to beginning, then match '.' */ |
| 3868 | char *starting = locinput; |
| 3869 | |
| 3870 | /* In case have to backtrack the last prepend */ |
| 3871 | char *previous_prepend = 0; |
| 3872 | |
| 3873 | LOAD_UTF8_CHARCLASS_GCB(); |
| 3874 | |
| 3875 | /* Match (prepend)* */ |
| 3876 | while (locinput < PL_regeol |
| 3877 | && swash_fetch(PL_utf8_X_prepend, |
| 3878 | (U8*)locinput, utf8_target)) |
| 3879 | { |
| 3880 | previous_prepend = locinput; |
| 3881 | locinput += UTF8SKIP(locinput); |
| 3882 | } |
| 3883 | |
| 3884 | /* As noted above, if we matched a prepend character, but |
| 3885 | * the next thing won't match, back off the last prepend we |
| 3886 | * matched, as it is guaranteed to match the begin */ |
| 3887 | if (previous_prepend |
| 3888 | && (locinput >= PL_regeol |
| 3889 | || ! swash_fetch(PL_utf8_X_begin, |
| 3890 | (U8*)locinput, utf8_target))) |
| 3891 | { |
| 3892 | locinput = previous_prepend; |
| 3893 | } |
| 3894 | |
| 3895 | /* Note that here we know PL_regeol > locinput, as we |
| 3896 | * tested that upon input to this switch case, and if we |
| 3897 | * moved locinput forward, we tested the result just above |
| 3898 | * and it either passed, or we backed off so that it will |
| 3899 | * now pass */ |
| 3900 | if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) { |
| 3901 | |
| 3902 | /* Here did not match the required 'Begin' in the |
| 3903 | * second term. So just match the very first |
| 3904 | * character, the '.' of the final term of the regex */ |
| 3905 | locinput = starting + UTF8SKIP(starting); |
| 3906 | } else { |
| 3907 | |
| 3908 | /* Here is the beginning of a character that can have |
| 3909 | * an extender. It is either a hangul syllable, or a |
| 3910 | * non-control */ |
| 3911 | if (swash_fetch(PL_utf8_X_non_hangul, |
| 3912 | (U8*)locinput, utf8_target)) |
| 3913 | { |
| 3914 | |
| 3915 | /* Here not a Hangul syllable, must be a |
| 3916 | * ('! * Control') */ |
| 3917 | locinput += UTF8SKIP(locinput); |
| 3918 | } else { |
| 3919 | |
| 3920 | /* Here is a Hangul syllable. It can be composed |
| 3921 | * of several individual characters. One |
| 3922 | * possibility is T+ */ |
| 3923 | if (swash_fetch(PL_utf8_X_T, |
| 3924 | (U8*)locinput, utf8_target)) |
| 3925 | { |
| 3926 | while (locinput < PL_regeol |
| 3927 | && swash_fetch(PL_utf8_X_T, |
| 3928 | (U8*)locinput, utf8_target)) |
| 3929 | { |
| 3930 | locinput += UTF8SKIP(locinput); |
| 3931 | } |
| 3932 | } else { |
| 3933 | |
| 3934 | /* Here, not T+, but is a Hangul. That means |
| 3935 | * it is one of the others: L, LV, LVT or V, |
| 3936 | * and matches: |
| 3937 | * L* (L | LVT T* | V V* T* | LV V* T*) */ |
| 3938 | |
| 3939 | /* Match L* */ |
| 3940 | while (locinput < PL_regeol |
| 3941 | && swash_fetch(PL_utf8_X_L, |
| 3942 | (U8*)locinput, utf8_target)) |
| 3943 | { |
| 3944 | locinput += UTF8SKIP(locinput); |
| 3945 | } |
| 3946 | |
| 3947 | /* Here, have exhausted L*. If the next |
| 3948 | * character is not an LV, LVT nor V, it means |
| 3949 | * we had to have at least one L, so matches L+ |
| 3950 | * in the original equation, we have a complete |
| 3951 | * hangul syllable. Are done. */ |
| 3952 | |
| 3953 | if (locinput < PL_regeol |
| 3954 | && swash_fetch(PL_utf8_X_LV_LVT_V, |
| 3955 | (U8*)locinput, utf8_target)) |
| 3956 | { |
| 3957 | |
| 3958 | /* Otherwise keep going. Must be LV, LVT |
| 3959 | * or V. See if LVT */ |
| 3960 | if (swash_fetch(PL_utf8_X_LVT, |
| 3961 | (U8*)locinput, utf8_target)) |
| 3962 | { |
| 3963 | locinput += UTF8SKIP(locinput); |
| 3964 | } else { |
| 3965 | |
| 3966 | /* Must be V or LV. Take it, then |
| 3967 | * match V* */ |
| 3968 | locinput += UTF8SKIP(locinput); |
| 3969 | while (locinput < PL_regeol |
| 3970 | && swash_fetch(PL_utf8_X_V, |
| 3971 | (U8*)locinput, utf8_target)) |
| 3972 | { |
| 3973 | locinput += UTF8SKIP(locinput); |
| 3974 | } |
| 3975 | } |
| 3976 | |
| 3977 | /* And any of LV, LVT, or V can be followed |
| 3978 | * by T* */ |
| 3979 | while (locinput < PL_regeol |
| 3980 | && swash_fetch(PL_utf8_X_T, |
| 3981 | (U8*)locinput, |
| 3982 | utf8_target)) |
| 3983 | { |
| 3984 | locinput += UTF8SKIP(locinput); |
| 3985 | } |
| 3986 | } |
| 3987 | } |
| 3988 | } |
| 3989 | |
| 3990 | /* Match any extender */ |
| 3991 | while (locinput < PL_regeol |
| 3992 | && swash_fetch(PL_utf8_X_extend, |
| 3993 | (U8*)locinput, utf8_target)) |
| 3994 | { |
| 3995 | locinput += UTF8SKIP(locinput); |
| 3996 | } |
| 3997 | } |
| 3998 | } |
| 3999 | if (locinput > PL_regeol) sayNO; |
| 4000 | } |
| 4001 | nextchr = UCHARAT(locinput); |
| 4002 | break; |
| 4003 | |
| 4004 | case NREFFL: |
| 4005 | { /* The capture buffer cases. The ones beginning with N for the |
| 4006 | named buffers just convert to the equivalent numbered and |
| 4007 | pretend they were called as the corresponding numbered buffer |
| 4008 | op. */ |
| 4009 | /* don't initialize these in the declaration, it makes C++ |
| 4010 | unhappy */ |
| 4011 | char *s; |
| 4012 | char type; |
| 4013 | re_fold_t folder; |
| 4014 | const U8 *fold_array; |
| 4015 | UV utf8_fold_flags; |
| 4016 | |
| 4017 | PL_reg_flags |= RF_tainted; |
| 4018 | folder = foldEQ_locale; |
| 4019 | fold_array = PL_fold_locale; |
| 4020 | type = REFFL; |
| 4021 | utf8_fold_flags = FOLDEQ_UTF8_LOCALE; |
| 4022 | goto do_nref; |
| 4023 | |
| 4024 | case NREFFA: |
| 4025 | folder = foldEQ_latin1; |
| 4026 | fold_array = PL_fold_latin1; |
| 4027 | type = REFFA; |
| 4028 | utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 4029 | goto do_nref; |
| 4030 | |
| 4031 | case NREFFU: |
| 4032 | folder = foldEQ_latin1; |
| 4033 | fold_array = PL_fold_latin1; |
| 4034 | type = REFFU; |
| 4035 | utf8_fold_flags = 0; |
| 4036 | goto do_nref; |
| 4037 | |
| 4038 | case NREFF: |
| 4039 | folder = foldEQ; |
| 4040 | fold_array = PL_fold; |
| 4041 | type = REFF; |
| 4042 | utf8_fold_flags = 0; |
| 4043 | goto do_nref; |
| 4044 | |
| 4045 | case NREF: |
| 4046 | type = REF; |
| 4047 | folder = NULL; |
| 4048 | fold_array = NULL; |
| 4049 | utf8_fold_flags = 0; |
| 4050 | do_nref: |
| 4051 | |
| 4052 | /* For the named back references, find the corresponding buffer |
| 4053 | * number */ |
| 4054 | n = reg_check_named_buff_matched(rex,scan); |
| 4055 | |
| 4056 | if ( ! n ) { |
| 4057 | sayNO; |
| 4058 | } |
| 4059 | goto do_nref_ref_common; |
| 4060 | |
| 4061 | case REFFL: |
| 4062 | PL_reg_flags |= RF_tainted; |
| 4063 | folder = foldEQ_locale; |
| 4064 | fold_array = PL_fold_locale; |
| 4065 | utf8_fold_flags = FOLDEQ_UTF8_LOCALE; |
| 4066 | goto do_ref; |
| 4067 | |
| 4068 | case REFFA: |
| 4069 | folder = foldEQ_latin1; |
| 4070 | fold_array = PL_fold_latin1; |
| 4071 | utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 4072 | goto do_ref; |
| 4073 | |
| 4074 | case REFFU: |
| 4075 | folder = foldEQ_latin1; |
| 4076 | fold_array = PL_fold_latin1; |
| 4077 | utf8_fold_flags = 0; |
| 4078 | goto do_ref; |
| 4079 | |
| 4080 | case REFF: |
| 4081 | folder = foldEQ; |
| 4082 | fold_array = PL_fold; |
| 4083 | utf8_fold_flags = 0; |
| 4084 | goto do_ref; |
| 4085 | |
| 4086 | case REF: |
| 4087 | folder = NULL; |
| 4088 | fold_array = NULL; |
| 4089 | utf8_fold_flags = 0; |
| 4090 | |
| 4091 | do_ref: |
| 4092 | type = OP(scan); |
| 4093 | n = ARG(scan); /* which paren pair */ |
| 4094 | |
| 4095 | do_nref_ref_common: |
| 4096 | ln = PL_regoffs[n].start; |
| 4097 | PL_reg_leftiter = PL_reg_maxiter; /* Void cache */ |
| 4098 | if (*PL_reglastparen < n || ln == -1) |
| 4099 | sayNO; /* Do not match unless seen CLOSEn. */ |
| 4100 | if (ln == PL_regoffs[n].end) |
| 4101 | break; |
| 4102 | |
| 4103 | s = PL_bostr + ln; |
| 4104 | if (type != REF /* REF can do byte comparison */ |
| 4105 | && (utf8_target || type == REFFU)) |
| 4106 | { /* XXX handle REFFL better */ |
| 4107 | char * limit = PL_regeol; |
| 4108 | |
| 4109 | /* This call case insensitively compares the entire buffer |
| 4110 | * at s, with the current input starting at locinput, but |
| 4111 | * not going off the end given by PL_regeol, and returns in |
| 4112 | * limit upon success, how much of the current input was |
| 4113 | * matched */ |
| 4114 | if (! foldEQ_utf8_flags(s, NULL, PL_regoffs[n].end - ln, utf8_target, |
| 4115 | locinput, &limit, 0, utf8_target, utf8_fold_flags)) |
| 4116 | { |
| 4117 | sayNO; |
| 4118 | } |
| 4119 | locinput = limit; |
| 4120 | nextchr = UCHARAT(locinput); |
| 4121 | break; |
| 4122 | } |
| 4123 | |
| 4124 | /* Not utf8: Inline the first character, for speed. */ |
| 4125 | if (UCHARAT(s) != nextchr && |
| 4126 | (type == REF || |
| 4127 | UCHARAT(s) != fold_array[nextchr])) |
| 4128 | sayNO; |
| 4129 | ln = PL_regoffs[n].end - ln; |
| 4130 | if (locinput + ln > PL_regeol) |
| 4131 | sayNO; |
| 4132 | if (ln > 1 && (type == REF |
| 4133 | ? memNE(s, locinput, ln) |
| 4134 | : ! folder(s, locinput, ln))) |
| 4135 | sayNO; |
| 4136 | locinput += ln; |
| 4137 | nextchr = UCHARAT(locinput); |
| 4138 | break; |
| 4139 | } |
| 4140 | case NOTHING: |
| 4141 | case TAIL: |
| 4142 | break; |
| 4143 | case BACK: |
| 4144 | break; |
| 4145 | |
| 4146 | #undef ST |
| 4147 | #define ST st->u.eval |
| 4148 | { |
| 4149 | SV *ret; |
| 4150 | REGEXP *re_sv; |
| 4151 | regexp *re; |
| 4152 | regexp_internal *rei; |
| 4153 | regnode *startpoint; |
| 4154 | |
| 4155 | case GOSTART: |
| 4156 | case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */ |
| 4157 | if (cur_eval && cur_eval->locinput==locinput) { |
| 4158 | if (cur_eval->u.eval.close_paren == (U32)ARG(scan)) |
| 4159 | Perl_croak(aTHX_ "Infinite recursion in regex"); |
| 4160 | if ( ++nochange_depth > max_nochange_depth ) |
| 4161 | Perl_croak(aTHX_ |
| 4162 | "Pattern subroutine nesting without pos change" |
| 4163 | " exceeded limit in regex"); |
| 4164 | } else { |
| 4165 | nochange_depth = 0; |
| 4166 | } |
| 4167 | re_sv = rex_sv; |
| 4168 | re = rex; |
| 4169 | rei = rexi; |
| 4170 | (void)ReREFCNT_inc(rex_sv); |
| 4171 | if (OP(scan)==GOSUB) { |
| 4172 | startpoint = scan + ARG2L(scan); |
| 4173 | ST.close_paren = ARG(scan); |
| 4174 | } else { |
| 4175 | startpoint = rei->program+1; |
| 4176 | ST.close_paren = 0; |
| 4177 | } |
| 4178 | goto eval_recurse_doit; |
| 4179 | /* NOTREACHED */ |
| 4180 | case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */ |
| 4181 | if (cur_eval && cur_eval->locinput==locinput) { |
| 4182 | if ( ++nochange_depth > max_nochange_depth ) |
| 4183 | Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex"); |
| 4184 | } else { |
| 4185 | nochange_depth = 0; |
| 4186 | } |
| 4187 | { |
| 4188 | /* execute the code in the {...} */ |
| 4189 | dSP; |
| 4190 | SV ** const before = SP; |
| 4191 | OP_4tree * const oop = PL_op; |
| 4192 | COP * const ocurcop = PL_curcop; |
| 4193 | PAD *old_comppad; |
| 4194 | char *saved_regeol = PL_regeol; |
| 4195 | struct re_save_state saved_state; |
| 4196 | |
| 4197 | /* To not corrupt the existing regex state while executing the |
| 4198 | * eval we would normally put it on the save stack, like with |
| 4199 | * save_re_context. However, re-evals have a weird scoping so we |
| 4200 | * can't just add ENTER/LEAVE here. With that, things like |
| 4201 | * |
| 4202 | * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a}) |
| 4203 | * |
| 4204 | * would break, as they expect the localisation to be unwound |
| 4205 | * only when the re-engine backtracks through the bit that |
| 4206 | * localised it. |
| 4207 | * |
| 4208 | * What we do instead is just saving the state in a local c |
| 4209 | * variable. |
| 4210 | */ |
| 4211 | Copy(&PL_reg_state, &saved_state, 1, struct re_save_state); |
| 4212 | |
| 4213 | n = ARG(scan); |
| 4214 | PL_op = (OP_4tree*)rexi->data->data[n]; |
| 4215 | DEBUG_STATE_r( PerlIO_printf(Perl_debug_log, |
| 4216 | " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) ); |
| 4217 | /* wrap the call in two SAVECOMPPADs. This ensures that |
| 4218 | * when the save stack is eventually unwound, all the |
| 4219 | * accumulated SAVEt_CLEARSV's will be processed with |
| 4220 | * interspersed SAVEt_COMPPAD's to ensure that lexicals |
| 4221 | * are cleared in the right pad */ |
| 4222 | SAVECOMPPAD(); |
| 4223 | PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]); |
| 4224 | PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr; |
| 4225 | |
| 4226 | if (sv_yes_mark) { |
| 4227 | SV *sv_mrk = get_sv("REGMARK", 1); |
| 4228 | sv_setsv(sv_mrk, sv_yes_mark); |
| 4229 | } |
| 4230 | |
| 4231 | CALLRUNOPS(aTHX); /* Scalar context. */ |
| 4232 | SPAGAIN; |
| 4233 | if (SP == before) |
| 4234 | ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */ |
| 4235 | else { |
| 4236 | ret = POPs; |
| 4237 | PUTBACK; |
| 4238 | } |
| 4239 | |
| 4240 | Copy(&saved_state, &PL_reg_state, 1, struct re_save_state); |
| 4241 | |
| 4242 | PL_op = oop; |
| 4243 | SAVECOMPPAD(); |
| 4244 | PAD_RESTORE_LOCAL(old_comppad); |
| 4245 | PL_curcop = ocurcop; |
| 4246 | PL_regeol = saved_regeol; |
| 4247 | if (!logical) { |
| 4248 | /* /(?{...})/ */ |
| 4249 | sv_setsv(save_scalar(PL_replgv), ret); |
| 4250 | break; |
| 4251 | } |
| 4252 | } |
| 4253 | if (logical == 2) { /* Postponed subexpression: /(??{...})/ */ |
| 4254 | logical = 0; |
| 4255 | { |
| 4256 | /* extract RE object from returned value; compiling if |
| 4257 | * necessary */ |
| 4258 | MAGIC *mg = NULL; |
| 4259 | REGEXP *rx = NULL; |
| 4260 | |
| 4261 | if (SvROK(ret)) { |
| 4262 | SV *const sv = SvRV(ret); |
| 4263 | |
| 4264 | if (SvTYPE(sv) == SVt_REGEXP) { |
| 4265 | rx = (REGEXP*) sv; |
| 4266 | } else if (SvSMAGICAL(sv)) { |
| 4267 | mg = mg_find(sv, PERL_MAGIC_qr); |
| 4268 | assert(mg); |
| 4269 | } |
| 4270 | } else if (SvTYPE(ret) == SVt_REGEXP) { |
| 4271 | rx = (REGEXP*) ret; |
| 4272 | } else if (SvSMAGICAL(ret)) { |
| 4273 | if (SvGMAGICAL(ret)) { |
| 4274 | /* I don't believe that there is ever qr magic |
| 4275 | here. */ |
| 4276 | assert(!mg_find(ret, PERL_MAGIC_qr)); |
| 4277 | sv_unmagic(ret, PERL_MAGIC_qr); |
| 4278 | } |
| 4279 | else { |
| 4280 | mg = mg_find(ret, PERL_MAGIC_qr); |
| 4281 | /* testing suggests mg only ends up non-NULL for |
| 4282 | scalars who were upgraded and compiled in the |
| 4283 | else block below. In turn, this is only |
| 4284 | triggered in the "postponed utf8 string" tests |
| 4285 | in t/op/pat.t */ |
| 4286 | } |
| 4287 | } |
| 4288 | |
| 4289 | if (mg) { |
| 4290 | rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/ |
| 4291 | assert(rx); |
| 4292 | } |
| 4293 | if (rx) { |
| 4294 | rx = reg_temp_copy(NULL, rx); |
| 4295 | } |
| 4296 | else { |
| 4297 | U32 pm_flags = 0; |
| 4298 | const I32 osize = PL_regsize; |
| 4299 | |
| 4300 | if (DO_UTF8(ret)) { |
| 4301 | assert (SvUTF8(ret)); |
| 4302 | } else if (SvUTF8(ret)) { |
| 4303 | /* Not doing UTF-8, despite what the SV says. Is |
| 4304 | this only if we're trapped in use 'bytes'? */ |
| 4305 | /* Make a copy of the octet sequence, but without |
| 4306 | the flag on, as the compiler now honours the |
| 4307 | SvUTF8 flag on ret. */ |
| 4308 | STRLEN len; |
| 4309 | const char *const p = SvPV(ret, len); |
| 4310 | ret = newSVpvn_flags(p, len, SVs_TEMP); |
| 4311 | } |
| 4312 | rx = CALLREGCOMP(ret, pm_flags); |
| 4313 | if (!(SvFLAGS(ret) |
| 4314 | & (SVs_TEMP | SVs_PADTMP | SVf_READONLY |
| 4315 | | SVs_GMG))) { |
| 4316 | /* This isn't a first class regexp. Instead, it's |
| 4317 | caching a regexp onto an existing, Perl visible |
| 4318 | scalar. */ |
| 4319 | sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0); |
| 4320 | } |
| 4321 | PL_regsize = osize; |
| 4322 | } |
| 4323 | re_sv = rx; |
| 4324 | re = (struct regexp *)SvANY(rx); |
| 4325 | } |
| 4326 | RXp_MATCH_COPIED_off(re); |
| 4327 | re->subbeg = rex->subbeg; |
| 4328 | re->sublen = rex->sublen; |
| 4329 | rei = RXi_GET(re); |
| 4330 | DEBUG_EXECUTE_r( |
| 4331 | debug_start_match(re_sv, utf8_target, locinput, PL_regeol, |
| 4332 | "Matching embedded"); |
| 4333 | ); |
| 4334 | startpoint = rei->program + 1; |
| 4335 | ST.close_paren = 0; /* only used for GOSUB */ |
| 4336 | /* borrowed from regtry */ |
| 4337 | if (PL_reg_start_tmpl <= re->nparens) { |
| 4338 | PL_reg_start_tmpl = re->nparens*3/2 + 3; |
| 4339 | if(PL_reg_start_tmp) |
| 4340 | Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*); |
| 4341 | else |
| 4342 | Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*); |
| 4343 | } |
| 4344 | |
| 4345 | eval_recurse_doit: /* Share code with GOSUB below this line */ |
| 4346 | /* run the pattern returned from (??{...}) */ |
| 4347 | ST.cp = regcppush(0); /* Save *all* the positions. */ |
| 4348 | REGCP_SET(ST.lastcp); |
| 4349 | |
| 4350 | PL_regoffs = re->offs; /* essentially NOOP on GOSUB */ |
| 4351 | |
| 4352 | /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */ |
| 4353 | PL_reglastparen = &re->lastparen; |
| 4354 | PL_reglastcloseparen = &re->lastcloseparen; |
| 4355 | re->lastparen = 0; |
| 4356 | re->lastcloseparen = 0; |
| 4357 | |
| 4358 | PL_reginput = locinput; |
| 4359 | PL_regsize = 0; |
| 4360 | |
| 4361 | /* XXXX This is too dramatic a measure... */ |
| 4362 | PL_reg_maxiter = 0; |
| 4363 | |
| 4364 | ST.toggle_reg_flags = PL_reg_flags; |
| 4365 | if (RX_UTF8(re_sv)) |
| 4366 | PL_reg_flags |= RF_utf8; |
| 4367 | else |
| 4368 | PL_reg_flags &= ~RF_utf8; |
| 4369 | ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */ |
| 4370 | |
| 4371 | ST.prev_rex = rex_sv; |
| 4372 | ST.prev_curlyx = cur_curlyx; |
| 4373 | SETREX(rex_sv,re_sv); |
| 4374 | rex = re; |
| 4375 | rexi = rei; |
| 4376 | cur_curlyx = NULL; |
| 4377 | ST.B = next; |
| 4378 | ST.prev_eval = cur_eval; |
| 4379 | cur_eval = st; |
| 4380 | /* now continue from first node in postoned RE */ |
| 4381 | PUSH_YES_STATE_GOTO(EVAL_AB, startpoint); |
| 4382 | /* NOTREACHED */ |
| 4383 | } |
| 4384 | /* logical is 1, /(?(?{...})X|Y)/ */ |
| 4385 | sw = cBOOL(SvTRUE(ret)); |
| 4386 | logical = 0; |
| 4387 | break; |
| 4388 | } |
| 4389 | |
| 4390 | case EVAL_AB: /* cleanup after a successful (??{A})B */ |
| 4391 | /* note: this is called twice; first after popping B, then A */ |
| 4392 | PL_reg_flags ^= ST.toggle_reg_flags; |
| 4393 | ReREFCNT_dec(rex_sv); |
| 4394 | SETREX(rex_sv,ST.prev_rex); |
| 4395 | rex = (struct regexp *)SvANY(rex_sv); |
| 4396 | rexi = RXi_GET(rex); |
| 4397 | regcpblow(ST.cp); |
| 4398 | cur_eval = ST.prev_eval; |
| 4399 | cur_curlyx = ST.prev_curlyx; |
| 4400 | |
| 4401 | /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */ |
| 4402 | PL_reglastparen = &rex->lastparen; |
| 4403 | PL_reglastcloseparen = &rex->lastcloseparen; |
| 4404 | /* also update PL_regoffs */ |
| 4405 | PL_regoffs = rex->offs; |
| 4406 | |
| 4407 | /* XXXX This is too dramatic a measure... */ |
| 4408 | PL_reg_maxiter = 0; |
| 4409 | if ( nochange_depth ) |
| 4410 | nochange_depth--; |
| 4411 | sayYES; |
| 4412 | |
| 4413 | |
| 4414 | case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */ |
| 4415 | /* note: this is called twice; first after popping B, then A */ |
| 4416 | PL_reg_flags ^= ST.toggle_reg_flags; |
| 4417 | ReREFCNT_dec(rex_sv); |
| 4418 | SETREX(rex_sv,ST.prev_rex); |
| 4419 | rex = (struct regexp *)SvANY(rex_sv); |
| 4420 | rexi = RXi_GET(rex); |
| 4421 | /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */ |
| 4422 | PL_reglastparen = &rex->lastparen; |
| 4423 | PL_reglastcloseparen = &rex->lastcloseparen; |
| 4424 | |
| 4425 | PL_reginput = locinput; |
| 4426 | REGCP_UNWIND(ST.lastcp); |
| 4427 | regcppop(rex); |
| 4428 | cur_eval = ST.prev_eval; |
| 4429 | cur_curlyx = ST.prev_curlyx; |
| 4430 | /* XXXX This is too dramatic a measure... */ |
| 4431 | PL_reg_maxiter = 0; |
| 4432 | if ( nochange_depth ) |
| 4433 | nochange_depth--; |
| 4434 | sayNO_SILENT; |
| 4435 | #undef ST |
| 4436 | |
| 4437 | case OPEN: |
| 4438 | n = ARG(scan); /* which paren pair */ |
| 4439 | PL_reg_start_tmp[n] = locinput; |
| 4440 | if (n > PL_regsize) |
| 4441 | PL_regsize = n; |
| 4442 | lastopen = n; |
| 4443 | break; |
| 4444 | case CLOSE: |
| 4445 | n = ARG(scan); /* which paren pair */ |
| 4446 | PL_regoffs[n].start = PL_reg_start_tmp[n] - PL_bostr; |
| 4447 | PL_regoffs[n].end = locinput - PL_bostr; |
| 4448 | /*if (n > PL_regsize) |
| 4449 | PL_regsize = n;*/ |
| 4450 | if (n > *PL_reglastparen) |
| 4451 | *PL_reglastparen = n; |
| 4452 | *PL_reglastcloseparen = n; |
| 4453 | if (cur_eval && cur_eval->u.eval.close_paren == n) { |
| 4454 | goto fake_end; |
| 4455 | } |
| 4456 | break; |
| 4457 | case ACCEPT: |
| 4458 | if (ARG(scan)){ |
| 4459 | regnode *cursor; |
| 4460 | for (cursor=scan; |
| 4461 | cursor && OP(cursor)!=END; |
| 4462 | cursor=regnext(cursor)) |
| 4463 | { |
| 4464 | if ( OP(cursor)==CLOSE ){ |
| 4465 | n = ARG(cursor); |
| 4466 | if ( n <= lastopen ) { |
| 4467 | PL_regoffs[n].start |
| 4468 | = PL_reg_start_tmp[n] - PL_bostr; |
| 4469 | PL_regoffs[n].end = locinput - PL_bostr; |
| 4470 | /*if (n > PL_regsize) |
| 4471 | PL_regsize = n;*/ |
| 4472 | if (n > *PL_reglastparen) |
| 4473 | *PL_reglastparen = n; |
| 4474 | *PL_reglastcloseparen = n; |
| 4475 | if ( n == ARG(scan) || (cur_eval && |
| 4476 | cur_eval->u.eval.close_paren == n)) |
| 4477 | break; |
| 4478 | } |
| 4479 | } |
| 4480 | } |
| 4481 | } |
| 4482 | goto fake_end; |
| 4483 | /*NOTREACHED*/ |
| 4484 | case GROUPP: |
| 4485 | n = ARG(scan); /* which paren pair */ |
| 4486 | sw = cBOOL(*PL_reglastparen >= n && PL_regoffs[n].end != -1); |
| 4487 | break; |
| 4488 | case NGROUPP: |
| 4489 | /* reg_check_named_buff_matched returns 0 for no match */ |
| 4490 | sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan)); |
| 4491 | break; |
| 4492 | case INSUBP: |
| 4493 | n = ARG(scan); |
| 4494 | sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n)); |
| 4495 | break; |
| 4496 | case DEFINEP: |
| 4497 | sw = 0; |
| 4498 | break; |
| 4499 | case IFTHEN: |
| 4500 | PL_reg_leftiter = PL_reg_maxiter; /* Void cache */ |
| 4501 | if (sw) |
| 4502 | next = NEXTOPER(NEXTOPER(scan)); |
| 4503 | else { |
| 4504 | next = scan + ARG(scan); |
| 4505 | if (OP(next) == IFTHEN) /* Fake one. */ |
| 4506 | next = NEXTOPER(NEXTOPER(next)); |
| 4507 | } |
| 4508 | break; |
| 4509 | case LOGICAL: |
| 4510 | logical = scan->flags; |
| 4511 | break; |
| 4512 | |
| 4513 | /******************************************************************* |
| 4514 | |
| 4515 | The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/ |
| 4516 | pattern, where A and B are subpatterns. (For simple A, CURLYM or |
| 4517 | STAR/PLUS/CURLY/CURLYN are used instead.) |
| 4518 | |
| 4519 | A*B is compiled as <CURLYX><A><WHILEM><B> |
| 4520 | |
| 4521 | On entry to the subpattern, CURLYX is called. This pushes a CURLYX |
| 4522 | state, which contains the current count, initialised to -1. It also sets |
| 4523 | cur_curlyx to point to this state, with any previous value saved in the |
| 4524 | state block. |
| 4525 | |
| 4526 | CURLYX then jumps straight to the WHILEM op, rather than executing A, |
| 4527 | since the pattern may possibly match zero times (i.e. it's a while {} loop |
| 4528 | rather than a do {} while loop). |
| 4529 | |
| 4530 | Each entry to WHILEM represents a successful match of A. The count in the |
| 4531 | CURLYX block is incremented, another WHILEM state is pushed, and execution |
| 4532 | passes to A or B depending on greediness and the current count. |
| 4533 | |
| 4534 | For example, if matching against the string a1a2a3b (where the aN are |
| 4535 | substrings that match /A/), then the match progresses as follows: (the |
| 4536 | pushed states are interspersed with the bits of strings matched so far): |
| 4537 | |
| 4538 | <CURLYX cnt=-1> |
| 4539 | <CURLYX cnt=0><WHILEM> |
| 4540 | <CURLYX cnt=1><WHILEM> a1 <WHILEM> |
| 4541 | <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM> |
| 4542 | <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> |
| 4543 | <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b |
| 4544 | |
| 4545 | (Contrast this with something like CURLYM, which maintains only a single |
| 4546 | backtrack state: |
| 4547 | |
| 4548 | <CURLYM cnt=0> a1 |
| 4549 | a1 <CURLYM cnt=1> a2 |
| 4550 | a1 a2 <CURLYM cnt=2> a3 |
| 4551 | a1 a2 a3 <CURLYM cnt=3> b |
| 4552 | ) |
| 4553 | |
| 4554 | Each WHILEM state block marks a point to backtrack to upon partial failure |
| 4555 | of A or B, and also contains some minor state data related to that |
| 4556 | iteration. The CURLYX block, pointed to by cur_curlyx, contains the |
| 4557 | overall state, such as the count, and pointers to the A and B ops. |
| 4558 | |
| 4559 | This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx |
| 4560 | must always point to the *current* CURLYX block, the rules are: |
| 4561 | |
| 4562 | When executing CURLYX, save the old cur_curlyx in the CURLYX state block, |
| 4563 | and set cur_curlyx to point the new block. |
| 4564 | |
| 4565 | When popping the CURLYX block after a successful or unsuccessful match, |
| 4566 | restore the previous cur_curlyx. |
| 4567 | |
| 4568 | When WHILEM is about to execute B, save the current cur_curlyx, and set it |
| 4569 | to the outer one saved in the CURLYX block. |
| 4570 | |
| 4571 | When popping the WHILEM block after a successful or unsuccessful B match, |
| 4572 | restore the previous cur_curlyx. |
| 4573 | |
| 4574 | Here's an example for the pattern (AI* BI)*BO |
| 4575 | I and O refer to inner and outer, C and W refer to CURLYX and WHILEM: |
| 4576 | |
| 4577 | cur_ |
| 4578 | curlyx backtrack stack |
| 4579 | ------ --------------- |
| 4580 | NULL |
| 4581 | CO <CO prev=NULL> <WO> |
| 4582 | CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai |
| 4583 | CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi |
| 4584 | NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo |
| 4585 | |
| 4586 | At this point the pattern succeeds, and we work back down the stack to |
| 4587 | clean up, restoring as we go: |
| 4588 | |
| 4589 | CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi |
| 4590 | CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai |
| 4591 | CO <CO prev=NULL> <WO> |
| 4592 | NULL |
| 4593 | |
| 4594 | *******************************************************************/ |
| 4595 | |
| 4596 | #define ST st->u.curlyx |
| 4597 | |
| 4598 | case CURLYX: /* start of /A*B/ (for complex A) */ |
| 4599 | { |
| 4600 | /* No need to save/restore up to this paren */ |
| 4601 | I32 parenfloor = scan->flags; |
| 4602 | |
| 4603 | assert(next); /* keep Coverity happy */ |
| 4604 | if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */ |
| 4605 | next += ARG(next); |
| 4606 | |
| 4607 | /* XXXX Probably it is better to teach regpush to support |
| 4608 | parenfloor > PL_regsize... */ |
| 4609 | if (parenfloor > (I32)*PL_reglastparen) |
| 4610 | parenfloor = *PL_reglastparen; /* Pessimization... */ |
| 4611 | |
| 4612 | ST.prev_curlyx= cur_curlyx; |
| 4613 | cur_curlyx = st; |
| 4614 | ST.cp = PL_savestack_ix; |
| 4615 | |
| 4616 | /* these fields contain the state of the current curly. |
| 4617 | * they are accessed by subsequent WHILEMs */ |
| 4618 | ST.parenfloor = parenfloor; |
| 4619 | ST.me = scan; |
| 4620 | ST.B = next; |
| 4621 | ST.minmod = minmod; |
| 4622 | minmod = 0; |
| 4623 | ST.count = -1; /* this will be updated by WHILEM */ |
| 4624 | ST.lastloc = NULL; /* this will be updated by WHILEM */ |
| 4625 | |
| 4626 | PL_reginput = locinput; |
| 4627 | PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next)); |
| 4628 | /* NOTREACHED */ |
| 4629 | } |
| 4630 | |
| 4631 | case CURLYX_end: /* just finished matching all of A*B */ |
| 4632 | cur_curlyx = ST.prev_curlyx; |
| 4633 | sayYES; |
| 4634 | /* NOTREACHED */ |
| 4635 | |
| 4636 | case CURLYX_end_fail: /* just failed to match all of A*B */ |
| 4637 | regcpblow(ST.cp); |
| 4638 | cur_curlyx = ST.prev_curlyx; |
| 4639 | sayNO; |
| 4640 | /* NOTREACHED */ |
| 4641 | |
| 4642 | |
| 4643 | #undef ST |
| 4644 | #define ST st->u.whilem |
| 4645 | |
| 4646 | case WHILEM: /* just matched an A in /A*B/ (for complex A) */ |
| 4647 | { |
| 4648 | /* see the discussion above about CURLYX/WHILEM */ |
| 4649 | I32 n; |
| 4650 | int min = ARG1(cur_curlyx->u.curlyx.me); |
| 4651 | int max = ARG2(cur_curlyx->u.curlyx.me); |
| 4652 | regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS; |
| 4653 | |
| 4654 | assert(cur_curlyx); /* keep Coverity happy */ |
| 4655 | n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */ |
| 4656 | ST.save_lastloc = cur_curlyx->u.curlyx.lastloc; |
| 4657 | ST.cache_offset = 0; |
| 4658 | ST.cache_mask = 0; |
| 4659 | |
| 4660 | PL_reginput = locinput; |
| 4661 | |
| 4662 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 4663 | "%*s whilem: matched %ld out of %d..%d\n", |
| 4664 | REPORT_CODE_OFF+depth*2, "", (long)n, min, max) |
| 4665 | ); |
| 4666 | |
| 4667 | /* First just match a string of min A's. */ |
| 4668 | |
| 4669 | if (n < min) { |
| 4670 | ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor); |
| 4671 | cur_curlyx->u.curlyx.lastloc = locinput; |
| 4672 | REGCP_SET(ST.lastcp); |
| 4673 | |
| 4674 | PUSH_STATE_GOTO(WHILEM_A_pre, A); |
| 4675 | /* NOTREACHED */ |
| 4676 | } |
| 4677 | |
| 4678 | /* If degenerate A matches "", assume A done. */ |
| 4679 | |
| 4680 | if (locinput == cur_curlyx->u.curlyx.lastloc) { |
| 4681 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 4682 | "%*s whilem: empty match detected, trying continuation...\n", |
| 4683 | REPORT_CODE_OFF+depth*2, "") |
| 4684 | ); |
| 4685 | goto do_whilem_B_max; |
| 4686 | } |
| 4687 | |
| 4688 | /* super-linear cache processing */ |
| 4689 | |
| 4690 | if (scan->flags) { |
| 4691 | |
| 4692 | if (!PL_reg_maxiter) { |
| 4693 | /* start the countdown: Postpone detection until we |
| 4694 | * know the match is not *that* much linear. */ |
| 4695 | PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4); |
| 4696 | /* possible overflow for long strings and many CURLYX's */ |
| 4697 | if (PL_reg_maxiter < 0) |
| 4698 | PL_reg_maxiter = I32_MAX; |
| 4699 | PL_reg_leftiter = PL_reg_maxiter; |
| 4700 | } |
| 4701 | |
| 4702 | if (PL_reg_leftiter-- == 0) { |
| 4703 | /* initialise cache */ |
| 4704 | const I32 size = (PL_reg_maxiter + 7)/8; |
| 4705 | if (PL_reg_poscache) { |
| 4706 | if ((I32)PL_reg_poscache_size < size) { |
| 4707 | Renew(PL_reg_poscache, size, char); |
| 4708 | PL_reg_poscache_size = size; |
| 4709 | } |
| 4710 | Zero(PL_reg_poscache, size, char); |
| 4711 | } |
| 4712 | else { |
| 4713 | PL_reg_poscache_size = size; |
| 4714 | Newxz(PL_reg_poscache, size, char); |
| 4715 | } |
| 4716 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 4717 | "%swhilem: Detected a super-linear match, switching on caching%s...\n", |
| 4718 | PL_colors[4], PL_colors[5]) |
| 4719 | ); |
| 4720 | } |
| 4721 | |
| 4722 | if (PL_reg_leftiter < 0) { |
| 4723 | /* have we already failed at this position? */ |
| 4724 | I32 offset, mask; |
| 4725 | offset = (scan->flags & 0xf) - 1 |
| 4726 | + (locinput - PL_bostr) * (scan->flags>>4); |
| 4727 | mask = 1 << (offset % 8); |
| 4728 | offset /= 8; |
| 4729 | if (PL_reg_poscache[offset] & mask) { |
| 4730 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 4731 | "%*s whilem: (cache) already tried at this position...\n", |
| 4732 | REPORT_CODE_OFF+depth*2, "") |
| 4733 | ); |
| 4734 | sayNO; /* cache records failure */ |
| 4735 | } |
| 4736 | ST.cache_offset = offset; |
| 4737 | ST.cache_mask = mask; |
| 4738 | } |
| 4739 | } |
| 4740 | |
| 4741 | /* Prefer B over A for minimal matching. */ |
| 4742 | |
| 4743 | if (cur_curlyx->u.curlyx.minmod) { |
| 4744 | ST.save_curlyx = cur_curlyx; |
| 4745 | cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; |
| 4746 | ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor); |
| 4747 | REGCP_SET(ST.lastcp); |
| 4748 | PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B); |
| 4749 | /* NOTREACHED */ |
| 4750 | } |
| 4751 | |
| 4752 | /* Prefer A over B for maximal matching. */ |
| 4753 | |
| 4754 | if (n < max) { /* More greed allowed? */ |
| 4755 | ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor); |
| 4756 | cur_curlyx->u.curlyx.lastloc = locinput; |
| 4757 | REGCP_SET(ST.lastcp); |
| 4758 | PUSH_STATE_GOTO(WHILEM_A_max, A); |
| 4759 | /* NOTREACHED */ |
| 4760 | } |
| 4761 | goto do_whilem_B_max; |
| 4762 | } |
| 4763 | /* NOTREACHED */ |
| 4764 | |
| 4765 | case WHILEM_B_min: /* just matched B in a minimal match */ |
| 4766 | case WHILEM_B_max: /* just matched B in a maximal match */ |
| 4767 | cur_curlyx = ST.save_curlyx; |
| 4768 | sayYES; |
| 4769 | /* NOTREACHED */ |
| 4770 | |
| 4771 | case WHILEM_B_max_fail: /* just failed to match B in a maximal match */ |
| 4772 | cur_curlyx = ST.save_curlyx; |
| 4773 | cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; |
| 4774 | cur_curlyx->u.curlyx.count--; |
| 4775 | CACHEsayNO; |
| 4776 | /* NOTREACHED */ |
| 4777 | |
| 4778 | case WHILEM_A_min_fail: /* just failed to match A in a minimal match */ |
| 4779 | /* FALL THROUGH */ |
| 4780 | case WHILEM_A_pre_fail: /* just failed to match even minimal A */ |
| 4781 | REGCP_UNWIND(ST.lastcp); |
| 4782 | regcppop(rex); |
| 4783 | cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; |
| 4784 | cur_curlyx->u.curlyx.count--; |
| 4785 | CACHEsayNO; |
| 4786 | /* NOTREACHED */ |
| 4787 | |
| 4788 | case WHILEM_A_max_fail: /* just failed to match A in a maximal match */ |
| 4789 | REGCP_UNWIND(ST.lastcp); |
| 4790 | regcppop(rex); /* Restore some previous $<digit>s? */ |
| 4791 | PL_reginput = locinput; |
| 4792 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 4793 | "%*s whilem: failed, trying continuation...\n", |
| 4794 | REPORT_CODE_OFF+depth*2, "") |
| 4795 | ); |
| 4796 | do_whilem_B_max: |
| 4797 | if (cur_curlyx->u.curlyx.count >= REG_INFTY |
| 4798 | && ckWARN(WARN_REGEXP) |
| 4799 | && !(PL_reg_flags & RF_warned)) |
| 4800 | { |
| 4801 | PL_reg_flags |= RF_warned; |
| 4802 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded", |
| 4803 | "Complex regular subexpression recursion", |
| 4804 | REG_INFTY - 1); |
| 4805 | } |
| 4806 | |
| 4807 | /* now try B */ |
| 4808 | ST.save_curlyx = cur_curlyx; |
| 4809 | cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; |
| 4810 | PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B); |
| 4811 | /* NOTREACHED */ |
| 4812 | |
| 4813 | case WHILEM_B_min_fail: /* just failed to match B in a minimal match */ |
| 4814 | cur_curlyx = ST.save_curlyx; |
| 4815 | REGCP_UNWIND(ST.lastcp); |
| 4816 | regcppop(rex); |
| 4817 | |
| 4818 | if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) { |
| 4819 | /* Maximum greed exceeded */ |
| 4820 | if (cur_curlyx->u.curlyx.count >= REG_INFTY |
| 4821 | && ckWARN(WARN_REGEXP) |
| 4822 | && !(PL_reg_flags & RF_warned)) |
| 4823 | { |
| 4824 | PL_reg_flags |= RF_warned; |
| 4825 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), |
| 4826 | "%s limit (%d) exceeded", |
| 4827 | "Complex regular subexpression recursion", |
| 4828 | REG_INFTY - 1); |
| 4829 | } |
| 4830 | cur_curlyx->u.curlyx.count--; |
| 4831 | CACHEsayNO; |
| 4832 | } |
| 4833 | |
| 4834 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 4835 | "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "") |
| 4836 | ); |
| 4837 | /* Try grabbing another A and see if it helps. */ |
| 4838 | PL_reginput = locinput; |
| 4839 | cur_curlyx->u.curlyx.lastloc = locinput; |
| 4840 | ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor); |
| 4841 | REGCP_SET(ST.lastcp); |
| 4842 | PUSH_STATE_GOTO(WHILEM_A_min, |
| 4843 | /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS); |
| 4844 | /* NOTREACHED */ |
| 4845 | |
| 4846 | #undef ST |
| 4847 | #define ST st->u.branch |
| 4848 | |
| 4849 | case BRANCHJ: /* /(...|A|...)/ with long next pointer */ |
| 4850 | next = scan + ARG(scan); |
| 4851 | if (next == scan) |
| 4852 | next = NULL; |
| 4853 | scan = NEXTOPER(scan); |
| 4854 | /* FALL THROUGH */ |
| 4855 | |
| 4856 | case BRANCH: /* /(...|A|...)/ */ |
| 4857 | scan = NEXTOPER(scan); /* scan now points to inner node */ |
| 4858 | ST.lastparen = *PL_reglastparen; |
| 4859 | ST.next_branch = next; |
| 4860 | REGCP_SET(ST.cp); |
| 4861 | PL_reginput = locinput; |
| 4862 | |
| 4863 | /* Now go into the branch */ |
| 4864 | if (has_cutgroup) { |
| 4865 | PUSH_YES_STATE_GOTO(BRANCH_next, scan); |
| 4866 | } else { |
| 4867 | PUSH_STATE_GOTO(BRANCH_next, scan); |
| 4868 | } |
| 4869 | /* NOTREACHED */ |
| 4870 | case CUTGROUP: |
| 4871 | PL_reginput = locinput; |
| 4872 | sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL : |
| 4873 | MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 4874 | PUSH_STATE_GOTO(CUTGROUP_next,next); |
| 4875 | /* NOTREACHED */ |
| 4876 | case CUTGROUP_next_fail: |
| 4877 | do_cutgroup = 1; |
| 4878 | no_final = 1; |
| 4879 | if (st->u.mark.mark_name) |
| 4880 | sv_commit = st->u.mark.mark_name; |
| 4881 | sayNO; |
| 4882 | /* NOTREACHED */ |
| 4883 | case BRANCH_next: |
| 4884 | sayYES; |
| 4885 | /* NOTREACHED */ |
| 4886 | case BRANCH_next_fail: /* that branch failed; try the next, if any */ |
| 4887 | if (do_cutgroup) { |
| 4888 | do_cutgroup = 0; |
| 4889 | no_final = 0; |
| 4890 | } |
| 4891 | REGCP_UNWIND(ST.cp); |
| 4892 | for (n = *PL_reglastparen; n > ST.lastparen; n--) |
| 4893 | PL_regoffs[n].end = -1; |
| 4894 | *PL_reglastparen = n; |
| 4895 | /*dmq: *PL_reglastcloseparen = n; */ |
| 4896 | scan = ST.next_branch; |
| 4897 | /* no more branches? */ |
| 4898 | if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) { |
| 4899 | DEBUG_EXECUTE_r({ |
| 4900 | PerlIO_printf( Perl_debug_log, |
| 4901 | "%*s %sBRANCH failed...%s\n", |
| 4902 | REPORT_CODE_OFF+depth*2, "", |
| 4903 | PL_colors[4], |
| 4904 | PL_colors[5] ); |
| 4905 | }); |
| 4906 | sayNO_SILENT; |
| 4907 | } |
| 4908 | continue; /* execute next BRANCH[J] op */ |
| 4909 | /* NOTREACHED */ |
| 4910 | |
| 4911 | case MINMOD: |
| 4912 | minmod = 1; |
| 4913 | break; |
| 4914 | |
| 4915 | #undef ST |
| 4916 | #define ST st->u.curlym |
| 4917 | |
| 4918 | case CURLYM: /* /A{m,n}B/ where A is fixed-length */ |
| 4919 | |
| 4920 | /* This is an optimisation of CURLYX that enables us to push |
| 4921 | * only a single backtracking state, no matter how many matches |
| 4922 | * there are in {m,n}. It relies on the pattern being constant |
| 4923 | * length, with no parens to influence future backrefs |
| 4924 | */ |
| 4925 | |
| 4926 | ST.me = scan; |
| 4927 | scan = NEXTOPER(scan) + NODE_STEP_REGNODE; |
| 4928 | |
| 4929 | /* if paren positive, emulate an OPEN/CLOSE around A */ |
| 4930 | if (ST.me->flags) { |
| 4931 | U32 paren = ST.me->flags; |
| 4932 | if (paren > PL_regsize) |
| 4933 | PL_regsize = paren; |
| 4934 | if (paren > *PL_reglastparen) |
| 4935 | *PL_reglastparen = paren; |
| 4936 | scan += NEXT_OFF(scan); /* Skip former OPEN. */ |
| 4937 | } |
| 4938 | ST.A = scan; |
| 4939 | ST.B = next; |
| 4940 | ST.alen = 0; |
| 4941 | ST.count = 0; |
| 4942 | ST.minmod = minmod; |
| 4943 | minmod = 0; |
| 4944 | ST.c1 = CHRTEST_UNINIT; |
| 4945 | REGCP_SET(ST.cp); |
| 4946 | |
| 4947 | if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */ |
| 4948 | goto curlym_do_B; |
| 4949 | |
| 4950 | curlym_do_A: /* execute the A in /A{m,n}B/ */ |
| 4951 | PL_reginput = locinput; |
| 4952 | PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */ |
| 4953 | /* NOTREACHED */ |
| 4954 | |
| 4955 | case CURLYM_A: /* we've just matched an A */ |
| 4956 | locinput = st->locinput; |
| 4957 | nextchr = UCHARAT(locinput); |
| 4958 | |
| 4959 | ST.count++; |
| 4960 | /* after first match, determine A's length: u.curlym.alen */ |
| 4961 | if (ST.count == 1) { |
| 4962 | if (PL_reg_match_utf8) { |
| 4963 | char *s = locinput; |
| 4964 | while (s < PL_reginput) { |
| 4965 | ST.alen++; |
| 4966 | s += UTF8SKIP(s); |
| 4967 | } |
| 4968 | } |
| 4969 | else { |
| 4970 | ST.alen = PL_reginput - locinput; |
| 4971 | } |
| 4972 | if (ST.alen == 0) |
| 4973 | ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me); |
| 4974 | } |
| 4975 | DEBUG_EXECUTE_r( |
| 4976 | PerlIO_printf(Perl_debug_log, |
| 4977 | "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n", |
| 4978 | (int)(REPORT_CODE_OFF+(depth*2)), "", |
| 4979 | (IV) ST.count, (IV)ST.alen) |
| 4980 | ); |
| 4981 | |
| 4982 | locinput = PL_reginput; |
| 4983 | |
| 4984 | if (cur_eval && cur_eval->u.eval.close_paren && |
| 4985 | cur_eval->u.eval.close_paren == (U32)ST.me->flags) |
| 4986 | goto fake_end; |
| 4987 | |
| 4988 | { |
| 4989 | I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)); |
| 4990 | if ( max == REG_INFTY || ST.count < max ) |
| 4991 | goto curlym_do_A; /* try to match another A */ |
| 4992 | } |
| 4993 | goto curlym_do_B; /* try to match B */ |
| 4994 | |
| 4995 | case CURLYM_A_fail: /* just failed to match an A */ |
| 4996 | REGCP_UNWIND(ST.cp); |
| 4997 | |
| 4998 | if (ST.minmod || ST.count < ARG1(ST.me) /* min*/ |
| 4999 | || (cur_eval && cur_eval->u.eval.close_paren && |
| 5000 | cur_eval->u.eval.close_paren == (U32)ST.me->flags)) |
| 5001 | sayNO; |
| 5002 | |
| 5003 | curlym_do_B: /* execute the B in /A{m,n}B/ */ |
| 5004 | PL_reginput = locinput; |
| 5005 | if (ST.c1 == CHRTEST_UNINIT) { |
| 5006 | /* calculate c1 and c2 for possible match of 1st char |
| 5007 | * following curly */ |
| 5008 | ST.c1 = ST.c2 = CHRTEST_VOID; |
| 5009 | if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) { |
| 5010 | regnode *text_node = ST.B; |
| 5011 | if (! HAS_TEXT(text_node)) |
| 5012 | FIND_NEXT_IMPT(text_node); |
| 5013 | /* this used to be |
| 5014 | |
| 5015 | (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT) |
| 5016 | |
| 5017 | But the former is redundant in light of the latter. |
| 5018 | |
| 5019 | if this changes back then the macro for |
| 5020 | IS_TEXT and friends need to change. |
| 5021 | */ |
| 5022 | if (PL_regkind[OP(text_node)] == EXACT) |
| 5023 | { |
| 5024 | |
| 5025 | ST.c1 = (U8)*STRING(text_node); |
| 5026 | switch (OP(text_node)) { |
| 5027 | case EXACTF: ST.c2 = PL_fold[ST.c1]; break; |
| 5028 | case EXACTFA: |
| 5029 | case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break; |
| 5030 | case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break; |
| 5031 | default: ST.c2 = ST.c1; |
| 5032 | } |
| 5033 | } |
| 5034 | } |
| 5035 | } |
| 5036 | |
| 5037 | DEBUG_EXECUTE_r( |
| 5038 | PerlIO_printf(Perl_debug_log, |
| 5039 | "%*s CURLYM trying tail with matches=%"IVdf"...\n", |
| 5040 | (int)(REPORT_CODE_OFF+(depth*2)), |
| 5041 | "", (IV)ST.count) |
| 5042 | ); |
| 5043 | if (ST.c1 != CHRTEST_VOID |
| 5044 | && UCHARAT(PL_reginput) != ST.c1 |
| 5045 | && UCHARAT(PL_reginput) != ST.c2) |
| 5046 | { |
| 5047 | /* simulate B failing */ |
| 5048 | DEBUG_OPTIMISE_r( |
| 5049 | PerlIO_printf(Perl_debug_log, |
| 5050 | "%*s CURLYM Fast bail c1=%"IVdf" c2=%"IVdf"\n", |
| 5051 | (int)(REPORT_CODE_OFF+(depth*2)),"", |
| 5052 | (IV)ST.c1,(IV)ST.c2 |
| 5053 | )); |
| 5054 | state_num = CURLYM_B_fail; |
| 5055 | goto reenter_switch; |
| 5056 | } |
| 5057 | |
| 5058 | if (ST.me->flags) { |
| 5059 | /* mark current A as captured */ |
| 5060 | I32 paren = ST.me->flags; |
| 5061 | if (ST.count) { |
| 5062 | PL_regoffs[paren].start |
| 5063 | = HOPc(PL_reginput, -ST.alen) - PL_bostr; |
| 5064 | PL_regoffs[paren].end = PL_reginput - PL_bostr; |
| 5065 | /*dmq: *PL_reglastcloseparen = paren; */ |
| 5066 | } |
| 5067 | else |
| 5068 | PL_regoffs[paren].end = -1; |
| 5069 | if (cur_eval && cur_eval->u.eval.close_paren && |
| 5070 | cur_eval->u.eval.close_paren == (U32)ST.me->flags) |
| 5071 | { |
| 5072 | if (ST.count) |
| 5073 | goto fake_end; |
| 5074 | else |
| 5075 | sayNO; |
| 5076 | } |
| 5077 | } |
| 5078 | |
| 5079 | PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */ |
| 5080 | /* NOTREACHED */ |
| 5081 | |
| 5082 | case CURLYM_B_fail: /* just failed to match a B */ |
| 5083 | REGCP_UNWIND(ST.cp); |
| 5084 | if (ST.minmod) { |
| 5085 | I32 max = ARG2(ST.me); |
| 5086 | if (max != REG_INFTY && ST.count == max) |
| 5087 | sayNO; |
| 5088 | goto curlym_do_A; /* try to match a further A */ |
| 5089 | } |
| 5090 | /* backtrack one A */ |
| 5091 | if (ST.count == ARG1(ST.me) /* min */) |
| 5092 | sayNO; |
| 5093 | ST.count--; |
| 5094 | locinput = HOPc(locinput, -ST.alen); |
| 5095 | goto curlym_do_B; /* try to match B */ |
| 5096 | |
| 5097 | #undef ST |
| 5098 | #define ST st->u.curly |
| 5099 | |
| 5100 | #define CURLY_SETPAREN(paren, success) \ |
| 5101 | if (paren) { \ |
| 5102 | if (success) { \ |
| 5103 | PL_regoffs[paren].start = HOPc(locinput, -1) - PL_bostr; \ |
| 5104 | PL_regoffs[paren].end = locinput - PL_bostr; \ |
| 5105 | *PL_reglastcloseparen = paren; \ |
| 5106 | } \ |
| 5107 | else \ |
| 5108 | PL_regoffs[paren].end = -1; \ |
| 5109 | } |
| 5110 | |
| 5111 | case STAR: /* /A*B/ where A is width 1 */ |
| 5112 | ST.paren = 0; |
| 5113 | ST.min = 0; |
| 5114 | ST.max = REG_INFTY; |
| 5115 | scan = NEXTOPER(scan); |
| 5116 | goto repeat; |
| 5117 | case PLUS: /* /A+B/ where A is width 1 */ |
| 5118 | ST.paren = 0; |
| 5119 | ST.min = 1; |
| 5120 | ST.max = REG_INFTY; |
| 5121 | scan = NEXTOPER(scan); |
| 5122 | goto repeat; |
| 5123 | case CURLYN: /* /(A){m,n}B/ where A is width 1 */ |
| 5124 | ST.paren = scan->flags; /* Which paren to set */ |
| 5125 | if (ST.paren > PL_regsize) |
| 5126 | PL_regsize = ST.paren; |
| 5127 | if (ST.paren > *PL_reglastparen) |
| 5128 | *PL_reglastparen = ST.paren; |
| 5129 | ST.min = ARG1(scan); /* min to match */ |
| 5130 | ST.max = ARG2(scan); /* max to match */ |
| 5131 | if (cur_eval && cur_eval->u.eval.close_paren && |
| 5132 | cur_eval->u.eval.close_paren == (U32)ST.paren) { |
| 5133 | ST.min=1; |
| 5134 | ST.max=1; |
| 5135 | } |
| 5136 | scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE); |
| 5137 | goto repeat; |
| 5138 | case CURLY: /* /A{m,n}B/ where A is width 1 */ |
| 5139 | ST.paren = 0; |
| 5140 | ST.min = ARG1(scan); /* min to match */ |
| 5141 | ST.max = ARG2(scan); /* max to match */ |
| 5142 | scan = NEXTOPER(scan) + NODE_STEP_REGNODE; |
| 5143 | repeat: |
| 5144 | /* |
| 5145 | * Lookahead to avoid useless match attempts |
| 5146 | * when we know what character comes next. |
| 5147 | * |
| 5148 | * Used to only do .*x and .*?x, but now it allows |
| 5149 | * for )'s, ('s and (?{ ... })'s to be in the way |
| 5150 | * of the quantifier and the EXACT-like node. -- japhy |
| 5151 | */ |
| 5152 | |
| 5153 | if (ST.min > ST.max) /* XXX make this a compile-time check? */ |
| 5154 | sayNO; |
| 5155 | if (HAS_TEXT(next) || JUMPABLE(next)) { |
| 5156 | U8 *s; |
| 5157 | regnode *text_node = next; |
| 5158 | |
| 5159 | if (! HAS_TEXT(text_node)) |
| 5160 | FIND_NEXT_IMPT(text_node); |
| 5161 | |
| 5162 | if (! HAS_TEXT(text_node)) |
| 5163 | ST.c1 = ST.c2 = CHRTEST_VOID; |
| 5164 | else { |
| 5165 | if ( PL_regkind[OP(text_node)] != EXACT ) { |
| 5166 | ST.c1 = ST.c2 = CHRTEST_VOID; |
| 5167 | goto assume_ok_easy; |
| 5168 | } |
| 5169 | else |
| 5170 | s = (U8*)STRING(text_node); |
| 5171 | |
| 5172 | /* Currently we only get here when |
| 5173 | |
| 5174 | PL_rekind[OP(text_node)] == EXACT |
| 5175 | |
| 5176 | if this changes back then the macro for IS_TEXT and |
| 5177 | friends need to change. */ |
| 5178 | if (!UTF_PATTERN) { |
| 5179 | ST.c1 = *s; |
| 5180 | switch (OP(text_node)) { |
| 5181 | case EXACTF: ST.c2 = PL_fold[ST.c1]; break; |
| 5182 | case EXACTFA: |
| 5183 | case EXACTFU: ST.c2 = PL_fold_latin1[ST.c1]; break; |
| 5184 | case EXACTFL: ST.c2 = PL_fold_locale[ST.c1]; break; |
| 5185 | default: ST.c2 = ST.c1; break; |
| 5186 | } |
| 5187 | } |
| 5188 | else { /* UTF_PATTERN */ |
| 5189 | if (IS_TEXTFU(text_node) || IS_TEXTF(text_node)) { |
| 5190 | STRLEN ulen1, ulen2; |
| 5191 | U8 tmpbuf1[UTF8_MAXBYTES_CASE+1]; |
| 5192 | U8 tmpbuf2[UTF8_MAXBYTES_CASE+1]; |
| 5193 | |
| 5194 | to_utf8_lower((U8*)s, tmpbuf1, &ulen1); |
| 5195 | to_utf8_upper((U8*)s, tmpbuf2, &ulen2); |
| 5196 | #ifdef EBCDIC |
| 5197 | ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0, |
| 5198 | ckWARN(WARN_UTF8) ? |
| 5199 | 0 : UTF8_ALLOW_ANY); |
| 5200 | ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0, |
| 5201 | ckWARN(WARN_UTF8) ? |
| 5202 | 0 : UTF8_ALLOW_ANY); |
| 5203 | #else |
| 5204 | ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0, |
| 5205 | uniflags); |
| 5206 | ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0, |
| 5207 | uniflags); |
| 5208 | #endif |
| 5209 | } |
| 5210 | else { |
| 5211 | ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0, |
| 5212 | uniflags); |
| 5213 | } |
| 5214 | } |
| 5215 | } |
| 5216 | } |
| 5217 | else |
| 5218 | ST.c1 = ST.c2 = CHRTEST_VOID; |
| 5219 | assume_ok_easy: |
| 5220 | |
| 5221 | ST.A = scan; |
| 5222 | ST.B = next; |
| 5223 | PL_reginput = locinput; |
| 5224 | if (minmod) { |
| 5225 | minmod = 0; |
| 5226 | if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min) |
| 5227 | sayNO; |
| 5228 | ST.count = ST.min; |
| 5229 | locinput = PL_reginput; |
| 5230 | REGCP_SET(ST.cp); |
| 5231 | if (ST.c1 == CHRTEST_VOID) |
| 5232 | goto curly_try_B_min; |
| 5233 | |
| 5234 | ST.oldloc = locinput; |
| 5235 | |
| 5236 | /* set ST.maxpos to the furthest point along the |
| 5237 | * string that could possibly match */ |
| 5238 | if (ST.max == REG_INFTY) { |
| 5239 | ST.maxpos = PL_regeol - 1; |
| 5240 | if (utf8_target) |
| 5241 | while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos)) |
| 5242 | ST.maxpos--; |
| 5243 | } |
| 5244 | else if (utf8_target) { |
| 5245 | int m = ST.max - ST.min; |
| 5246 | for (ST.maxpos = locinput; |
| 5247 | m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--) |
| 5248 | ST.maxpos += UTF8SKIP(ST.maxpos); |
| 5249 | } |
| 5250 | else { |
| 5251 | ST.maxpos = locinput + ST.max - ST.min; |
| 5252 | if (ST.maxpos >= PL_regeol) |
| 5253 | ST.maxpos = PL_regeol - 1; |
| 5254 | } |
| 5255 | goto curly_try_B_min_known; |
| 5256 | |
| 5257 | } |
| 5258 | else { |
| 5259 | ST.count = regrepeat(rex, ST.A, ST.max, depth); |
| 5260 | locinput = PL_reginput; |
| 5261 | if (ST.count < ST.min) |
| 5262 | sayNO; |
| 5263 | if ((ST.count > ST.min) |
| 5264 | && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL)) |
| 5265 | { |
| 5266 | /* A{m,n} must come at the end of the string, there's |
| 5267 | * no point in backing off ... */ |
| 5268 | ST.min = ST.count; |
| 5269 | /* ...except that $ and \Z can match before *and* after |
| 5270 | newline at the end. Consider "\n\n" =~ /\n+\Z\n/. |
| 5271 | We may back off by one in this case. */ |
| 5272 | if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS) |
| 5273 | ST.min--; |
| 5274 | } |
| 5275 | REGCP_SET(ST.cp); |
| 5276 | goto curly_try_B_max; |
| 5277 | } |
| 5278 | /* NOTREACHED */ |
| 5279 | |
| 5280 | |
| 5281 | case CURLY_B_min_known_fail: |
| 5282 | /* failed to find B in a non-greedy match where c1,c2 valid */ |
| 5283 | if (ST.paren && ST.count) |
| 5284 | PL_regoffs[ST.paren].end = -1; |
| 5285 | |
| 5286 | PL_reginput = locinput; /* Could be reset... */ |
| 5287 | REGCP_UNWIND(ST.cp); |
| 5288 | /* Couldn't or didn't -- move forward. */ |
| 5289 | ST.oldloc = locinput; |
| 5290 | if (utf8_target) |
| 5291 | locinput += UTF8SKIP(locinput); |
| 5292 | else |
| 5293 | locinput++; |
| 5294 | ST.count++; |
| 5295 | curly_try_B_min_known: |
| 5296 | /* find the next place where 'B' could work, then call B */ |
| 5297 | { |
| 5298 | int n; |
| 5299 | if (utf8_target) { |
| 5300 | n = (ST.oldloc == locinput) ? 0 : 1; |
| 5301 | if (ST.c1 == ST.c2) { |
| 5302 | STRLEN len; |
| 5303 | /* set n to utf8_distance(oldloc, locinput) */ |
| 5304 | while (locinput <= ST.maxpos && |
| 5305 | utf8n_to_uvchr((U8*)locinput, |
| 5306 | UTF8_MAXBYTES, &len, |
| 5307 | uniflags) != (UV)ST.c1) { |
| 5308 | locinput += len; |
| 5309 | n++; |
| 5310 | } |
| 5311 | } |
| 5312 | else { |
| 5313 | /* set n to utf8_distance(oldloc, locinput) */ |
| 5314 | while (locinput <= ST.maxpos) { |
| 5315 | STRLEN len; |
| 5316 | const UV c = utf8n_to_uvchr((U8*)locinput, |
| 5317 | UTF8_MAXBYTES, &len, |
| 5318 | uniflags); |
| 5319 | if (c == (UV)ST.c1 || c == (UV)ST.c2) |
| 5320 | break; |
| 5321 | locinput += len; |
| 5322 | n++; |
| 5323 | } |
| 5324 | } |
| 5325 | } |
| 5326 | else { |
| 5327 | if (ST.c1 == ST.c2) { |
| 5328 | while (locinput <= ST.maxpos && |
| 5329 | UCHARAT(locinput) != ST.c1) |
| 5330 | locinput++; |
| 5331 | } |
| 5332 | else { |
| 5333 | while (locinput <= ST.maxpos |
| 5334 | && UCHARAT(locinput) != ST.c1 |
| 5335 | && UCHARAT(locinput) != ST.c2) |
| 5336 | locinput++; |
| 5337 | } |
| 5338 | n = locinput - ST.oldloc; |
| 5339 | } |
| 5340 | if (locinput > ST.maxpos) |
| 5341 | sayNO; |
| 5342 | /* PL_reginput == oldloc now */ |
| 5343 | if (n) { |
| 5344 | ST.count += n; |
| 5345 | if (regrepeat(rex, ST.A, n, depth) < n) |
| 5346 | sayNO; |
| 5347 | } |
| 5348 | PL_reginput = locinput; |
| 5349 | CURLY_SETPAREN(ST.paren, ST.count); |
| 5350 | if (cur_eval && cur_eval->u.eval.close_paren && |
| 5351 | cur_eval->u.eval.close_paren == (U32)ST.paren) { |
| 5352 | goto fake_end; |
| 5353 | } |
| 5354 | PUSH_STATE_GOTO(CURLY_B_min_known, ST.B); |
| 5355 | } |
| 5356 | /* NOTREACHED */ |
| 5357 | |
| 5358 | |
| 5359 | case CURLY_B_min_fail: |
| 5360 | /* failed to find B in a non-greedy match where c1,c2 invalid */ |
| 5361 | if (ST.paren && ST.count) |
| 5362 | PL_regoffs[ST.paren].end = -1; |
| 5363 | |
| 5364 | REGCP_UNWIND(ST.cp); |
| 5365 | /* failed -- move forward one */ |
| 5366 | PL_reginput = locinput; |
| 5367 | if (regrepeat(rex, ST.A, 1, depth)) { |
| 5368 | ST.count++; |
| 5369 | locinput = PL_reginput; |
| 5370 | if (ST.count <= ST.max || (ST.max == REG_INFTY && |
| 5371 | ST.count > 0)) /* count overflow ? */ |
| 5372 | { |
| 5373 | curly_try_B_min: |
| 5374 | CURLY_SETPAREN(ST.paren, ST.count); |
| 5375 | if (cur_eval && cur_eval->u.eval.close_paren && |
| 5376 | cur_eval->u.eval.close_paren == (U32)ST.paren) { |
| 5377 | goto fake_end; |
| 5378 | } |
| 5379 | PUSH_STATE_GOTO(CURLY_B_min, ST.B); |
| 5380 | } |
| 5381 | } |
| 5382 | sayNO; |
| 5383 | /* NOTREACHED */ |
| 5384 | |
| 5385 | |
| 5386 | curly_try_B_max: |
| 5387 | /* a successful greedy match: now try to match B */ |
| 5388 | if (cur_eval && cur_eval->u.eval.close_paren && |
| 5389 | cur_eval->u.eval.close_paren == (U32)ST.paren) { |
| 5390 | goto fake_end; |
| 5391 | } |
| 5392 | { |
| 5393 | UV c = 0; |
| 5394 | if (ST.c1 != CHRTEST_VOID) |
| 5395 | c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput, |
| 5396 | UTF8_MAXBYTES, 0, uniflags) |
| 5397 | : (UV) UCHARAT(PL_reginput); |
| 5398 | /* If it could work, try it. */ |
| 5399 | if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) { |
| 5400 | CURLY_SETPAREN(ST.paren, ST.count); |
| 5401 | PUSH_STATE_GOTO(CURLY_B_max, ST.B); |
| 5402 | /* NOTREACHED */ |
| 5403 | } |
| 5404 | } |
| 5405 | /* FALL THROUGH */ |
| 5406 | case CURLY_B_max_fail: |
| 5407 | /* failed to find B in a greedy match */ |
| 5408 | if (ST.paren && ST.count) |
| 5409 | PL_regoffs[ST.paren].end = -1; |
| 5410 | |
| 5411 | REGCP_UNWIND(ST.cp); |
| 5412 | /* back up. */ |
| 5413 | if (--ST.count < ST.min) |
| 5414 | sayNO; |
| 5415 | PL_reginput = locinput = HOPc(locinput, -1); |
| 5416 | goto curly_try_B_max; |
| 5417 | |
| 5418 | #undef ST |
| 5419 | |
| 5420 | case END: |
| 5421 | fake_end: |
| 5422 | if (cur_eval) { |
| 5423 | /* we've just finished A in /(??{A})B/; now continue with B */ |
| 5424 | I32 tmpix; |
| 5425 | st->u.eval.toggle_reg_flags |
| 5426 | = cur_eval->u.eval.toggle_reg_flags; |
| 5427 | PL_reg_flags ^= st->u.eval.toggle_reg_flags; |
| 5428 | |
| 5429 | st->u.eval.prev_rex = rex_sv; /* inner */ |
| 5430 | SETREX(rex_sv,cur_eval->u.eval.prev_rex); |
| 5431 | rex = (struct regexp *)SvANY(rex_sv); |
| 5432 | rexi = RXi_GET(rex); |
| 5433 | cur_curlyx = cur_eval->u.eval.prev_curlyx; |
| 5434 | (void)ReREFCNT_inc(rex_sv); |
| 5435 | st->u.eval.cp = regcppush(0); /* Save *all* the positions. */ |
| 5436 | |
| 5437 | /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */ |
| 5438 | PL_reglastparen = &rex->lastparen; |
| 5439 | PL_reglastcloseparen = &rex->lastcloseparen; |
| 5440 | |
| 5441 | REGCP_SET(st->u.eval.lastcp); |
| 5442 | PL_reginput = locinput; |
| 5443 | |
| 5444 | /* Restore parens of the outer rex without popping the |
| 5445 | * savestack */ |
| 5446 | tmpix = PL_savestack_ix; |
| 5447 | PL_savestack_ix = cur_eval->u.eval.lastcp; |
| 5448 | regcppop(rex); |
| 5449 | PL_savestack_ix = tmpix; |
| 5450 | |
| 5451 | st->u.eval.prev_eval = cur_eval; |
| 5452 | cur_eval = cur_eval->u.eval.prev_eval; |
| 5453 | DEBUG_EXECUTE_r( |
| 5454 | PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n", |
| 5455 | REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval));); |
| 5456 | if ( nochange_depth ) |
| 5457 | nochange_depth--; |
| 5458 | |
| 5459 | PUSH_YES_STATE_GOTO(EVAL_AB, |
| 5460 | st->u.eval.prev_eval->u.eval.B); /* match B */ |
| 5461 | } |
| 5462 | |
| 5463 | if (locinput < reginfo->till) { |
| 5464 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 5465 | "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n", |
| 5466 | PL_colors[4], |
| 5467 | (long)(locinput - PL_reg_starttry), |
| 5468 | (long)(reginfo->till - PL_reg_starttry), |
| 5469 | PL_colors[5])); |
| 5470 | |
| 5471 | sayNO_SILENT; /* Cannot match: too short. */ |
| 5472 | } |
| 5473 | PL_reginput = locinput; /* put where regtry can find it */ |
| 5474 | sayYES; /* Success! */ |
| 5475 | |
| 5476 | case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */ |
| 5477 | DEBUG_EXECUTE_r( |
| 5478 | PerlIO_printf(Perl_debug_log, |
| 5479 | "%*s %ssubpattern success...%s\n", |
| 5480 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])); |
| 5481 | PL_reginput = locinput; /* put where regtry can find it */ |
| 5482 | sayYES; /* Success! */ |
| 5483 | |
| 5484 | #undef ST |
| 5485 | #define ST st->u.ifmatch |
| 5486 | |
| 5487 | case SUSPEND: /* (?>A) */ |
| 5488 | ST.wanted = 1; |
| 5489 | PL_reginput = locinput; |
| 5490 | goto do_ifmatch; |
| 5491 | |
| 5492 | case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */ |
| 5493 | ST.wanted = 0; |
| 5494 | goto ifmatch_trivial_fail_test; |
| 5495 | |
| 5496 | case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */ |
| 5497 | ST.wanted = 1; |
| 5498 | ifmatch_trivial_fail_test: |
| 5499 | if (scan->flags) { |
| 5500 | char * const s = HOPBACKc(locinput, scan->flags); |
| 5501 | if (!s) { |
| 5502 | /* trivial fail */ |
| 5503 | if (logical) { |
| 5504 | logical = 0; |
| 5505 | sw = 1 - cBOOL(ST.wanted); |
| 5506 | } |
| 5507 | else if (ST.wanted) |
| 5508 | sayNO; |
| 5509 | next = scan + ARG(scan); |
| 5510 | if (next == scan) |
| 5511 | next = NULL; |
| 5512 | break; |
| 5513 | } |
| 5514 | PL_reginput = s; |
| 5515 | } |
| 5516 | else |
| 5517 | PL_reginput = locinput; |
| 5518 | |
| 5519 | do_ifmatch: |
| 5520 | ST.me = scan; |
| 5521 | ST.logical = logical; |
| 5522 | logical = 0; /* XXX: reset state of logical once it has been saved into ST */ |
| 5523 | |
| 5524 | /* execute body of (?...A) */ |
| 5525 | PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan))); |
| 5526 | /* NOTREACHED */ |
| 5527 | |
| 5528 | case IFMATCH_A_fail: /* body of (?...A) failed */ |
| 5529 | ST.wanted = !ST.wanted; |
| 5530 | /* FALL THROUGH */ |
| 5531 | |
| 5532 | case IFMATCH_A: /* body of (?...A) succeeded */ |
| 5533 | if (ST.logical) { |
| 5534 | sw = cBOOL(ST.wanted); |
| 5535 | } |
| 5536 | else if (!ST.wanted) |
| 5537 | sayNO; |
| 5538 | |
| 5539 | if (OP(ST.me) == SUSPEND) |
| 5540 | locinput = PL_reginput; |
| 5541 | else { |
| 5542 | locinput = PL_reginput = st->locinput; |
| 5543 | nextchr = UCHARAT(locinput); |
| 5544 | } |
| 5545 | scan = ST.me + ARG(ST.me); |
| 5546 | if (scan == ST.me) |
| 5547 | scan = NULL; |
| 5548 | continue; /* execute B */ |
| 5549 | |
| 5550 | #undef ST |
| 5551 | |
| 5552 | case LONGJMP: |
| 5553 | next = scan + ARG(scan); |
| 5554 | if (next == scan) |
| 5555 | next = NULL; |
| 5556 | break; |
| 5557 | case COMMIT: |
| 5558 | reginfo->cutpoint = PL_regeol; |
| 5559 | /* FALLTHROUGH */ |
| 5560 | case PRUNE: |
| 5561 | PL_reginput = locinput; |
| 5562 | if (!scan->flags) |
| 5563 | sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 5564 | PUSH_STATE_GOTO(COMMIT_next,next); |
| 5565 | /* NOTREACHED */ |
| 5566 | case COMMIT_next_fail: |
| 5567 | no_final = 1; |
| 5568 | /* FALLTHROUGH */ |
| 5569 | case OPFAIL: |
| 5570 | sayNO; |
| 5571 | /* NOTREACHED */ |
| 5572 | |
| 5573 | #define ST st->u.mark |
| 5574 | case MARKPOINT: |
| 5575 | ST.prev_mark = mark_state; |
| 5576 | ST.mark_name = sv_commit = sv_yes_mark |
| 5577 | = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 5578 | mark_state = st; |
| 5579 | ST.mark_loc = PL_reginput = locinput; |
| 5580 | PUSH_YES_STATE_GOTO(MARKPOINT_next,next); |
| 5581 | /* NOTREACHED */ |
| 5582 | case MARKPOINT_next: |
| 5583 | mark_state = ST.prev_mark; |
| 5584 | sayYES; |
| 5585 | /* NOTREACHED */ |
| 5586 | case MARKPOINT_next_fail: |
| 5587 | if (popmark && sv_eq(ST.mark_name,popmark)) |
| 5588 | { |
| 5589 | if (ST.mark_loc > startpoint) |
| 5590 | reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1); |
| 5591 | popmark = NULL; /* we found our mark */ |
| 5592 | sv_commit = ST.mark_name; |
| 5593 | |
| 5594 | DEBUG_EXECUTE_r({ |
| 5595 | PerlIO_printf(Perl_debug_log, |
| 5596 | "%*s %ssetting cutpoint to mark:%"SVf"...%s\n", |
| 5597 | REPORT_CODE_OFF+depth*2, "", |
| 5598 | PL_colors[4], SVfARG(sv_commit), PL_colors[5]); |
| 5599 | }); |
| 5600 | } |
| 5601 | mark_state = ST.prev_mark; |
| 5602 | sv_yes_mark = mark_state ? |
| 5603 | mark_state->u.mark.mark_name : NULL; |
| 5604 | sayNO; |
| 5605 | /* NOTREACHED */ |
| 5606 | case SKIP: |
| 5607 | PL_reginput = locinput; |
| 5608 | if (scan->flags) { |
| 5609 | /* (*SKIP) : if we fail we cut here*/ |
| 5610 | ST.mark_name = NULL; |
| 5611 | ST.mark_loc = locinput; |
| 5612 | PUSH_STATE_GOTO(SKIP_next,next); |
| 5613 | } else { |
| 5614 | /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was, |
| 5615 | otherwise do nothing. Meaning we need to scan |
| 5616 | */ |
| 5617 | regmatch_state *cur = mark_state; |
| 5618 | SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 5619 | |
| 5620 | while (cur) { |
| 5621 | if ( sv_eq( cur->u.mark.mark_name, |
| 5622 | find ) ) |
| 5623 | { |
| 5624 | ST.mark_name = find; |
| 5625 | PUSH_STATE_GOTO( SKIP_next, next ); |
| 5626 | } |
| 5627 | cur = cur->u.mark.prev_mark; |
| 5628 | } |
| 5629 | } |
| 5630 | /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */ |
| 5631 | break; |
| 5632 | case SKIP_next_fail: |
| 5633 | if (ST.mark_name) { |
| 5634 | /* (*CUT:NAME) - Set up to search for the name as we |
| 5635 | collapse the stack*/ |
| 5636 | popmark = ST.mark_name; |
| 5637 | } else { |
| 5638 | /* (*CUT) - No name, we cut here.*/ |
| 5639 | if (ST.mark_loc > startpoint) |
| 5640 | reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1); |
| 5641 | /* but we set sv_commit to latest mark_name if there |
| 5642 | is one so they can test to see how things lead to this |
| 5643 | cut */ |
| 5644 | if (mark_state) |
| 5645 | sv_commit=mark_state->u.mark.mark_name; |
| 5646 | } |
| 5647 | no_final = 1; |
| 5648 | sayNO; |
| 5649 | /* NOTREACHED */ |
| 5650 | #undef ST |
| 5651 | case FOLDCHAR: |
| 5652 | n = ARG(scan); |
| 5653 | if ( n == (U32)what_len_TRICKYFOLD(locinput,utf8_target,ln) ) { |
| 5654 | locinput += ln; |
| 5655 | } else if ( LATIN_SMALL_LETTER_SHARP_S == n && !utf8_target && !UTF_PATTERN ) { |
| 5656 | sayNO; |
| 5657 | } else { |
| 5658 | U8 folded[UTF8_MAXBYTES_CASE+1]; |
| 5659 | STRLEN foldlen; |
| 5660 | const char * const l = locinput; |
| 5661 | char *e = PL_regeol; |
| 5662 | to_uni_fold(n, folded, &foldlen); |
| 5663 | |
| 5664 | if (! foldEQ_utf8((const char*) folded, 0, foldlen, 1, |
| 5665 | l, &e, 0, utf8_target)) { |
| 5666 | sayNO; |
| 5667 | } |
| 5668 | locinput = e; |
| 5669 | } |
| 5670 | nextchr = UCHARAT(locinput); |
| 5671 | break; |
| 5672 | case LNBREAK: |
| 5673 | if ((n=is_LNBREAK(locinput,utf8_target))) { |
| 5674 | locinput += n; |
| 5675 | nextchr = UCHARAT(locinput); |
| 5676 | } else |
| 5677 | sayNO; |
| 5678 | break; |
| 5679 | |
| 5680 | #define CASE_CLASS(nAmE) \ |
| 5681 | case nAmE: \ |
| 5682 | if ((n=is_##nAmE(locinput,utf8_target))) { \ |
| 5683 | locinput += n; \ |
| 5684 | nextchr = UCHARAT(locinput); \ |
| 5685 | } else \ |
| 5686 | sayNO; \ |
| 5687 | break; \ |
| 5688 | case N##nAmE: \ |
| 5689 | if ((n=is_##nAmE(locinput,utf8_target))) { \ |
| 5690 | sayNO; \ |
| 5691 | } else { \ |
| 5692 | locinput += UTF8SKIP(locinput); \ |
| 5693 | nextchr = UCHARAT(locinput); \ |
| 5694 | } \ |
| 5695 | break |
| 5696 | |
| 5697 | CASE_CLASS(VERTWS); |
| 5698 | CASE_CLASS(HORIZWS); |
| 5699 | #undef CASE_CLASS |
| 5700 | |
| 5701 | default: |
| 5702 | PerlIO_printf(Perl_error_log, "%"UVxf" %d\n", |
| 5703 | PTR2UV(scan), OP(scan)); |
| 5704 | Perl_croak(aTHX_ "regexp memory corruption"); |
| 5705 | |
| 5706 | } /* end switch */ |
| 5707 | |
| 5708 | /* switch break jumps here */ |
| 5709 | scan = next; /* prepare to execute the next op and ... */ |
| 5710 | continue; /* ... jump back to the top, reusing st */ |
| 5711 | /* NOTREACHED */ |
| 5712 | |
| 5713 | push_yes_state: |
| 5714 | /* push a state that backtracks on success */ |
| 5715 | st->u.yes.prev_yes_state = yes_state; |
| 5716 | yes_state = st; |
| 5717 | /* FALL THROUGH */ |
| 5718 | push_state: |
| 5719 | /* push a new regex state, then continue at scan */ |
| 5720 | { |
| 5721 | regmatch_state *newst; |
| 5722 | |
| 5723 | DEBUG_STACK_r({ |
| 5724 | regmatch_state *cur = st; |
| 5725 | regmatch_state *curyes = yes_state; |
| 5726 | int curd = depth; |
| 5727 | regmatch_slab *slab = PL_regmatch_slab; |
| 5728 | for (;curd > -1;cur--,curd--) { |
| 5729 | if (cur < SLAB_FIRST(slab)) { |
| 5730 | slab = slab->prev; |
| 5731 | cur = SLAB_LAST(slab); |
| 5732 | } |
| 5733 | PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n", |
| 5734 | REPORT_CODE_OFF + 2 + depth * 2,"", |
| 5735 | curd, PL_reg_name[cur->resume_state], |
| 5736 | (curyes == cur) ? "yes" : "" |
| 5737 | ); |
| 5738 | if (curyes == cur) |
| 5739 | curyes = cur->u.yes.prev_yes_state; |
| 5740 | } |
| 5741 | } else |
| 5742 | DEBUG_STATE_pp("push") |
| 5743 | ); |
| 5744 | depth++; |
| 5745 | st->locinput = locinput; |
| 5746 | newst = st+1; |
| 5747 | if (newst > SLAB_LAST(PL_regmatch_slab)) |
| 5748 | newst = S_push_slab(aTHX); |
| 5749 | PL_regmatch_state = newst; |
| 5750 | |
| 5751 | locinput = PL_reginput; |
| 5752 | nextchr = UCHARAT(locinput); |
| 5753 | st = newst; |
| 5754 | continue; |
| 5755 | /* NOTREACHED */ |
| 5756 | } |
| 5757 | } |
| 5758 | |
| 5759 | /* |
| 5760 | * We get here only if there's trouble -- normally "case END" is |
| 5761 | * the terminating point. |
| 5762 | */ |
| 5763 | Perl_croak(aTHX_ "corrupted regexp pointers"); |
| 5764 | /*NOTREACHED*/ |
| 5765 | sayNO; |
| 5766 | |
| 5767 | yes: |
| 5768 | if (yes_state) { |
| 5769 | /* we have successfully completed a subexpression, but we must now |
| 5770 | * pop to the state marked by yes_state and continue from there */ |
| 5771 | assert(st != yes_state); |
| 5772 | #ifdef DEBUGGING |
| 5773 | while (st != yes_state) { |
| 5774 | st--; |
| 5775 | if (st < SLAB_FIRST(PL_regmatch_slab)) { |
| 5776 | PL_regmatch_slab = PL_regmatch_slab->prev; |
| 5777 | st = SLAB_LAST(PL_regmatch_slab); |
| 5778 | } |
| 5779 | DEBUG_STATE_r({ |
| 5780 | if (no_final) { |
| 5781 | DEBUG_STATE_pp("pop (no final)"); |
| 5782 | } else { |
| 5783 | DEBUG_STATE_pp("pop (yes)"); |
| 5784 | } |
| 5785 | }); |
| 5786 | depth--; |
| 5787 | } |
| 5788 | #else |
| 5789 | while (yes_state < SLAB_FIRST(PL_regmatch_slab) |
| 5790 | || yes_state > SLAB_LAST(PL_regmatch_slab)) |
| 5791 | { |
| 5792 | /* not in this slab, pop slab */ |
| 5793 | depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1); |
| 5794 | PL_regmatch_slab = PL_regmatch_slab->prev; |
| 5795 | st = SLAB_LAST(PL_regmatch_slab); |
| 5796 | } |
| 5797 | depth -= (st - yes_state); |
| 5798 | #endif |
| 5799 | st = yes_state; |
| 5800 | yes_state = st->u.yes.prev_yes_state; |
| 5801 | PL_regmatch_state = st; |
| 5802 | |
| 5803 | if (no_final) { |
| 5804 | locinput= st->locinput; |
| 5805 | nextchr = UCHARAT(locinput); |
| 5806 | } |
| 5807 | state_num = st->resume_state + no_final; |
| 5808 | goto reenter_switch; |
| 5809 | } |
| 5810 | |
| 5811 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n", |
| 5812 | PL_colors[4], PL_colors[5])); |
| 5813 | |
| 5814 | if (PL_reg_eval_set) { |
| 5815 | /* each successfully executed (?{...}) block does the equivalent of |
| 5816 | * local $^R = do {...} |
| 5817 | * When popping the save stack, all these locals would be undone; |
| 5818 | * bypass this by setting the outermost saved $^R to the latest |
| 5819 | * value */ |
| 5820 | if (oreplsv != GvSV(PL_replgv)) |
| 5821 | sv_setsv(oreplsv, GvSV(PL_replgv)); |
| 5822 | } |
| 5823 | result = 1; |
| 5824 | goto final_exit; |
| 5825 | |
| 5826 | no: |
| 5827 | DEBUG_EXECUTE_r( |
| 5828 | PerlIO_printf(Perl_debug_log, |
| 5829 | "%*s %sfailed...%s\n", |
| 5830 | REPORT_CODE_OFF+depth*2, "", |
| 5831 | PL_colors[4], PL_colors[5]) |
| 5832 | ); |
| 5833 | |
| 5834 | no_silent: |
| 5835 | if (no_final) { |
| 5836 | if (yes_state) { |
| 5837 | goto yes; |
| 5838 | } else { |
| 5839 | goto final_exit; |
| 5840 | } |
| 5841 | } |
| 5842 | if (depth) { |
| 5843 | /* there's a previous state to backtrack to */ |
| 5844 | st--; |
| 5845 | if (st < SLAB_FIRST(PL_regmatch_slab)) { |
| 5846 | PL_regmatch_slab = PL_regmatch_slab->prev; |
| 5847 | st = SLAB_LAST(PL_regmatch_slab); |
| 5848 | } |
| 5849 | PL_regmatch_state = st; |
| 5850 | locinput= st->locinput; |
| 5851 | nextchr = UCHARAT(locinput); |
| 5852 | |
| 5853 | DEBUG_STATE_pp("pop"); |
| 5854 | depth--; |
| 5855 | if (yes_state == st) |
| 5856 | yes_state = st->u.yes.prev_yes_state; |
| 5857 | |
| 5858 | state_num = st->resume_state + 1; /* failure = success + 1 */ |
| 5859 | goto reenter_switch; |
| 5860 | } |
| 5861 | result = 0; |
| 5862 | |
| 5863 | final_exit: |
| 5864 | if (rex->intflags & PREGf_VERBARG_SEEN) { |
| 5865 | SV *sv_err = get_sv("REGERROR", 1); |
| 5866 | SV *sv_mrk = get_sv("REGMARK", 1); |
| 5867 | if (result) { |
| 5868 | sv_commit = &PL_sv_no; |
| 5869 | if (!sv_yes_mark) |
| 5870 | sv_yes_mark = &PL_sv_yes; |
| 5871 | } else { |
| 5872 | if (!sv_commit) |
| 5873 | sv_commit = &PL_sv_yes; |
| 5874 | sv_yes_mark = &PL_sv_no; |
| 5875 | } |
| 5876 | sv_setsv(sv_err, sv_commit); |
| 5877 | sv_setsv(sv_mrk, sv_yes_mark); |
| 5878 | } |
| 5879 | |
| 5880 | /* clean up; in particular, free all slabs above current one */ |
| 5881 | LEAVE_SCOPE(oldsave); |
| 5882 | |
| 5883 | return result; |
| 5884 | } |
| 5885 | |
| 5886 | /* |
| 5887 | - regrepeat - repeatedly match something simple, report how many |
| 5888 | */ |
| 5889 | /* |
| 5890 | * [This routine now assumes that it will only match on things of length 1. |
| 5891 | * That was true before, but now we assume scan - reginput is the count, |
| 5892 | * rather than incrementing count on every character. [Er, except utf8.]] |
| 5893 | */ |
| 5894 | STATIC I32 |
| 5895 | S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth) |
| 5896 | { |
| 5897 | dVAR; |
| 5898 | register char *scan; |
| 5899 | register I32 c; |
| 5900 | register char *loceol = PL_regeol; |
| 5901 | register I32 hardcount = 0; |
| 5902 | register bool utf8_target = PL_reg_match_utf8; |
| 5903 | UV utf8_flags; |
| 5904 | #ifndef DEBUGGING |
| 5905 | PERL_UNUSED_ARG(depth); |
| 5906 | #endif |
| 5907 | |
| 5908 | PERL_ARGS_ASSERT_REGREPEAT; |
| 5909 | |
| 5910 | scan = PL_reginput; |
| 5911 | if (max == REG_INFTY) |
| 5912 | max = I32_MAX; |
| 5913 | else if (max < loceol - scan) |
| 5914 | loceol = scan + max; |
| 5915 | switch (OP(p)) { |
| 5916 | case REG_ANY: |
| 5917 | if (utf8_target) { |
| 5918 | loceol = PL_regeol; |
| 5919 | while (scan < loceol && hardcount < max && *scan != '\n') { |
| 5920 | scan += UTF8SKIP(scan); |
| 5921 | hardcount++; |
| 5922 | } |
| 5923 | } else { |
| 5924 | while (scan < loceol && *scan != '\n') |
| 5925 | scan++; |
| 5926 | } |
| 5927 | break; |
| 5928 | case SANY: |
| 5929 | if (utf8_target) { |
| 5930 | loceol = PL_regeol; |
| 5931 | while (scan < loceol && hardcount < max) { |
| 5932 | scan += UTF8SKIP(scan); |
| 5933 | hardcount++; |
| 5934 | } |
| 5935 | } |
| 5936 | else |
| 5937 | scan = loceol; |
| 5938 | break; |
| 5939 | case CANY: |
| 5940 | scan = loceol; |
| 5941 | break; |
| 5942 | case EXACT: |
| 5943 | /* To get here, EXACTish nodes must have *byte* length == 1. That |
| 5944 | * means they match only characters in the string that can be expressed |
| 5945 | * as a single byte. For non-utf8 strings, that means a simple match. |
| 5946 | * For utf8 strings, the character matched must be an invariant, or |
| 5947 | * downgradable to a single byte. The pattern's utf8ness is |
| 5948 | * irrelevant, as since it's a single byte, it either isn't utf8, or if |
| 5949 | * it is, it's an invariant */ |
| 5950 | |
| 5951 | c = (U8)*STRING(p); |
| 5952 | assert(! UTF_PATTERN || UNI_IS_INVARIANT(c)); |
| 5953 | |
| 5954 | if (! utf8_target || UNI_IS_INVARIANT(c)) { |
| 5955 | while (scan < loceol && UCHARAT(scan) == c) { |
| 5956 | scan++; |
| 5957 | } |
| 5958 | } |
| 5959 | else { |
| 5960 | |
| 5961 | /* Here, the string is utf8, and the pattern char is different |
| 5962 | * in utf8 than not, so can't compare them directly. Outside the |
| 5963 | * loop, find find the two utf8 bytes that represent c, and then |
| 5964 | * look for those in sequence in the utf8 string */ |
| 5965 | U8 high = UTF8_TWO_BYTE_HI(c); |
| 5966 | U8 low = UTF8_TWO_BYTE_LO(c); |
| 5967 | loceol = PL_regeol; |
| 5968 | |
| 5969 | while (hardcount < max |
| 5970 | && scan + 1 < loceol |
| 5971 | && UCHARAT(scan) == high |
| 5972 | && UCHARAT(scan + 1) == low) |
| 5973 | { |
| 5974 | scan += 2; |
| 5975 | hardcount++; |
| 5976 | } |
| 5977 | } |
| 5978 | break; |
| 5979 | case EXACTFA: |
| 5980 | utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 5981 | goto do_exactf; |
| 5982 | |
| 5983 | case EXACTFL: |
| 5984 | PL_reg_flags |= RF_tainted; |
| 5985 | utf8_flags = FOLDEQ_UTF8_LOCALE; |
| 5986 | goto do_exactf; |
| 5987 | |
| 5988 | case EXACTF: |
| 5989 | case EXACTFU: |
| 5990 | utf8_flags = 0; |
| 5991 | |
| 5992 | /* The comments for the EXACT case above apply as well to these fold |
| 5993 | * ones */ |
| 5994 | |
| 5995 | do_exactf: |
| 5996 | c = (U8)*STRING(p); |
| 5997 | assert(! UTF_PATTERN || UNI_IS_INVARIANT(c)); |
| 5998 | |
| 5999 | if (utf8_target) { /* Use full Unicode fold matching */ |
| 6000 | char *tmpeol = loceol; |
| 6001 | while (hardcount < max |
| 6002 | && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target, |
| 6003 | STRING(p), NULL, 1, cBOOL(UTF_PATTERN), utf8_flags)) |
| 6004 | { |
| 6005 | scan = tmpeol; |
| 6006 | tmpeol = loceol; |
| 6007 | hardcount++; |
| 6008 | } |
| 6009 | |
| 6010 | /* XXX Note that the above handles properly the German sharp s in |
| 6011 | * the pattern matching ss in the string. But it doesn't handle |
| 6012 | * properly cases where the string contains say 'LIGATURE ff' and |
| 6013 | * the pattern is 'f+'. This would require, say, a new function or |
| 6014 | * revised interface to foldEQ_utf8(), in which the maximum number |
| 6015 | * of characters to match could be passed and it would return how |
| 6016 | * many actually did. This is just one of many cases where |
| 6017 | * multi-char folds don't work properly, and so the fix is being |
| 6018 | * deferred */ |
| 6019 | } |
| 6020 | else { |
| 6021 | U8 folded; |
| 6022 | |
| 6023 | /* Here, the string isn't utf8 and c is a single byte; and either |
| 6024 | * the pattern isn't utf8 or c is an invariant, so its utf8ness |
| 6025 | * doesn't affect c. Can just do simple comparisons for exact or |
| 6026 | * fold matching. */ |
| 6027 | switch (OP(p)) { |
| 6028 | case EXACTF: folded = PL_fold[c]; break; |
| 6029 | case EXACTFA: |
| 6030 | case EXACTFU: folded = PL_fold_latin1[c]; break; |
| 6031 | case EXACTFL: folded = PL_fold_locale[c]; break; |
| 6032 | default: Perl_croak(aTHX_ "panic: Unexpected op %u", OP(p)); |
| 6033 | } |
| 6034 | while (scan < loceol && |
| 6035 | (UCHARAT(scan) == c || UCHARAT(scan) == folded)) |
| 6036 | { |
| 6037 | scan++; |
| 6038 | } |
| 6039 | } |
| 6040 | break; |
| 6041 | case ANYOFV: |
| 6042 | case ANYOF: |
| 6043 | if (utf8_target || OP(p) == ANYOFV) { |
| 6044 | STRLEN inclasslen; |
| 6045 | loceol = PL_regeol; |
| 6046 | inclasslen = loceol - scan; |
| 6047 | while (hardcount < max |
| 6048 | && ((inclasslen = loceol - scan) > 0) |
| 6049 | && reginclass(prog, p, (U8*)scan, &inclasslen, utf8_target)) |
| 6050 | { |
| 6051 | scan += inclasslen; |
| 6052 | hardcount++; |
| 6053 | } |
| 6054 | } else { |
| 6055 | while (scan < loceol && REGINCLASS(prog, p, (U8*)scan)) |
| 6056 | scan++; |
| 6057 | } |
| 6058 | break; |
| 6059 | case ALNUMU: |
| 6060 | if (utf8_target) { |
| 6061 | utf8_wordchar: |
| 6062 | loceol = PL_regeol; |
| 6063 | LOAD_UTF8_CHARCLASS_ALNUM(); |
| 6064 | while (hardcount < max && scan < loceol && |
| 6065 | swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target)) |
| 6066 | { |
| 6067 | scan += UTF8SKIP(scan); |
| 6068 | hardcount++; |
| 6069 | } |
| 6070 | } else { |
| 6071 | while (scan < loceol && isWORDCHAR_L1((U8) *scan)) { |
| 6072 | scan++; |
| 6073 | } |
| 6074 | } |
| 6075 | break; |
| 6076 | case ALNUM: |
| 6077 | if (utf8_target) |
| 6078 | goto utf8_wordchar; |
| 6079 | while (scan < loceol && isALNUM((U8) *scan)) { |
| 6080 | scan++; |
| 6081 | } |
| 6082 | break; |
| 6083 | case ALNUMA: |
| 6084 | while (scan < loceol && isWORDCHAR_A((U8) *scan)) { |
| 6085 | scan++; |
| 6086 | } |
| 6087 | break; |
| 6088 | case ALNUML: |
| 6089 | PL_reg_flags |= RF_tainted; |
| 6090 | if (utf8_target) { |
| 6091 | loceol = PL_regeol; |
| 6092 | while (hardcount < max && scan < loceol && |
| 6093 | isALNUM_LC_utf8((U8*)scan)) { |
| 6094 | scan += UTF8SKIP(scan); |
| 6095 | hardcount++; |
| 6096 | } |
| 6097 | } else { |
| 6098 | while (scan < loceol && isALNUM_LC(*scan)) |
| 6099 | scan++; |
| 6100 | } |
| 6101 | break; |
| 6102 | case NALNUMU: |
| 6103 | if (utf8_target) { |
| 6104 | |
| 6105 | utf8_Nwordchar: |
| 6106 | |
| 6107 | loceol = PL_regeol; |
| 6108 | LOAD_UTF8_CHARCLASS_ALNUM(); |
| 6109 | while (hardcount < max && scan < loceol && |
| 6110 | ! swash_fetch(PL_utf8_alnum, (U8*)scan, utf8_target)) |
| 6111 | { |
| 6112 | scan += UTF8SKIP(scan); |
| 6113 | hardcount++; |
| 6114 | } |
| 6115 | } else { |
| 6116 | while (scan < loceol && ! isWORDCHAR_L1((U8) *scan)) { |
| 6117 | scan++; |
| 6118 | } |
| 6119 | } |
| 6120 | break; |
| 6121 | case NALNUM: |
| 6122 | if (utf8_target) |
| 6123 | goto utf8_Nwordchar; |
| 6124 | while (scan < loceol && ! isALNUM((U8) *scan)) { |
| 6125 | scan++; |
| 6126 | } |
| 6127 | break; |
| 6128 | case NALNUMA: |
| 6129 | if (utf8_target) { |
| 6130 | while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) { |
| 6131 | scan += UTF8SKIP(scan); |
| 6132 | } |
| 6133 | } |
| 6134 | else { |
| 6135 | while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) { |
| 6136 | scan++; |
| 6137 | } |
| 6138 | } |
| 6139 | break; |
| 6140 | case NALNUML: |
| 6141 | PL_reg_flags |= RF_tainted; |
| 6142 | if (utf8_target) { |
| 6143 | loceol = PL_regeol; |
| 6144 | while (hardcount < max && scan < loceol && |
| 6145 | !isALNUM_LC_utf8((U8*)scan)) { |
| 6146 | scan += UTF8SKIP(scan); |
| 6147 | hardcount++; |
| 6148 | } |
| 6149 | } else { |
| 6150 | while (scan < loceol && !isALNUM_LC(*scan)) |
| 6151 | scan++; |
| 6152 | } |
| 6153 | break; |
| 6154 | case SPACEU: |
| 6155 | if (utf8_target) { |
| 6156 | |
| 6157 | utf8_space: |
| 6158 | |
| 6159 | loceol = PL_regeol; |
| 6160 | LOAD_UTF8_CHARCLASS_SPACE(); |
| 6161 | while (hardcount < max && scan < loceol && |
| 6162 | (*scan == ' ' || |
| 6163 | swash_fetch(PL_utf8_space,(U8*)scan, utf8_target))) |
| 6164 | { |
| 6165 | scan += UTF8SKIP(scan); |
| 6166 | hardcount++; |
| 6167 | } |
| 6168 | break; |
| 6169 | } |
| 6170 | else { |
| 6171 | while (scan < loceol && isSPACE_L1((U8) *scan)) { |
| 6172 | scan++; |
| 6173 | } |
| 6174 | break; |
| 6175 | } |
| 6176 | case SPACE: |
| 6177 | if (utf8_target) |
| 6178 | goto utf8_space; |
| 6179 | |
| 6180 | while (scan < loceol && isSPACE((U8) *scan)) { |
| 6181 | scan++; |
| 6182 | } |
| 6183 | break; |
| 6184 | case SPACEA: |
| 6185 | while (scan < loceol && isSPACE_A((U8) *scan)) { |
| 6186 | scan++; |
| 6187 | } |
| 6188 | break; |
| 6189 | case SPACEL: |
| 6190 | PL_reg_flags |= RF_tainted; |
| 6191 | if (utf8_target) { |
| 6192 | loceol = PL_regeol; |
| 6193 | while (hardcount < max && scan < loceol && |
| 6194 | isSPACE_LC_utf8((U8*)scan)) { |
| 6195 | scan += UTF8SKIP(scan); |
| 6196 | hardcount++; |
| 6197 | } |
| 6198 | } else { |
| 6199 | while (scan < loceol && isSPACE_LC(*scan)) |
| 6200 | scan++; |
| 6201 | } |
| 6202 | break; |
| 6203 | case NSPACEU: |
| 6204 | if (utf8_target) { |
| 6205 | |
| 6206 | utf8_Nspace: |
| 6207 | |
| 6208 | loceol = PL_regeol; |
| 6209 | LOAD_UTF8_CHARCLASS_SPACE(); |
| 6210 | while (hardcount < max && scan < loceol && |
| 6211 | ! (*scan == ' ' || |
| 6212 | swash_fetch(PL_utf8_space,(U8*)scan, utf8_target))) |
| 6213 | { |
| 6214 | scan += UTF8SKIP(scan); |
| 6215 | hardcount++; |
| 6216 | } |
| 6217 | break; |
| 6218 | } |
| 6219 | else { |
| 6220 | while (scan < loceol && ! isSPACE_L1((U8) *scan)) { |
| 6221 | scan++; |
| 6222 | } |
| 6223 | } |
| 6224 | break; |
| 6225 | case NSPACE: |
| 6226 | if (utf8_target) |
| 6227 | goto utf8_Nspace; |
| 6228 | |
| 6229 | while (scan < loceol && ! isSPACE((U8) *scan)) { |
| 6230 | scan++; |
| 6231 | } |
| 6232 | break; |
| 6233 | case NSPACEA: |
| 6234 | if (utf8_target) { |
| 6235 | while (scan < loceol && ! isSPACE_A((U8) *scan)) { |
| 6236 | scan += UTF8SKIP(scan); |
| 6237 | } |
| 6238 | } |
| 6239 | else { |
| 6240 | while (scan < loceol && ! isSPACE_A((U8) *scan)) { |
| 6241 | scan++; |
| 6242 | } |
| 6243 | } |
| 6244 | break; |
| 6245 | case NSPACEL: |
| 6246 | PL_reg_flags |= RF_tainted; |
| 6247 | if (utf8_target) { |
| 6248 | loceol = PL_regeol; |
| 6249 | while (hardcount < max && scan < loceol && |
| 6250 | !isSPACE_LC_utf8((U8*)scan)) { |
| 6251 | scan += UTF8SKIP(scan); |
| 6252 | hardcount++; |
| 6253 | } |
| 6254 | } else { |
| 6255 | while (scan < loceol && !isSPACE_LC(*scan)) |
| 6256 | scan++; |
| 6257 | } |
| 6258 | break; |
| 6259 | case DIGIT: |
| 6260 | if (utf8_target) { |
| 6261 | loceol = PL_regeol; |
| 6262 | LOAD_UTF8_CHARCLASS_DIGIT(); |
| 6263 | while (hardcount < max && scan < loceol && |
| 6264 | swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) { |
| 6265 | scan += UTF8SKIP(scan); |
| 6266 | hardcount++; |
| 6267 | } |
| 6268 | } else { |
| 6269 | while (scan < loceol && isDIGIT(*scan)) |
| 6270 | scan++; |
| 6271 | } |
| 6272 | break; |
| 6273 | case DIGITA: |
| 6274 | while (scan < loceol && isDIGIT_A((U8) *scan)) { |
| 6275 | scan++; |
| 6276 | } |
| 6277 | break; |
| 6278 | case DIGITL: |
| 6279 | PL_reg_flags |= RF_tainted; |
| 6280 | if (utf8_target) { |
| 6281 | loceol = PL_regeol; |
| 6282 | while (hardcount < max && scan < loceol && |
| 6283 | isDIGIT_LC_utf8((U8*)scan)) { |
| 6284 | scan += UTF8SKIP(scan); |
| 6285 | hardcount++; |
| 6286 | } |
| 6287 | } else { |
| 6288 | while (scan < loceol && isDIGIT_LC(*scan)) |
| 6289 | scan++; |
| 6290 | } |
| 6291 | break; |
| 6292 | case NDIGIT: |
| 6293 | if (utf8_target) { |
| 6294 | loceol = PL_regeol; |
| 6295 | LOAD_UTF8_CHARCLASS_DIGIT(); |
| 6296 | while (hardcount < max && scan < loceol && |
| 6297 | !swash_fetch(PL_utf8_digit, (U8*)scan, utf8_target)) { |
| 6298 | scan += UTF8SKIP(scan); |
| 6299 | hardcount++; |
| 6300 | } |
| 6301 | } else { |
| 6302 | while (scan < loceol && !isDIGIT(*scan)) |
| 6303 | scan++; |
| 6304 | } |
| 6305 | break; |
| 6306 | case NDIGITA: |
| 6307 | if (utf8_target) { |
| 6308 | while (scan < loceol && ! isDIGIT_A((U8) *scan)) { |
| 6309 | scan += UTF8SKIP(scan); |
| 6310 | } |
| 6311 | } |
| 6312 | else { |
| 6313 | while (scan < loceol && ! isDIGIT_A((U8) *scan)) { |
| 6314 | scan++; |
| 6315 | } |
| 6316 | } |
| 6317 | break; |
| 6318 | case NDIGITL: |
| 6319 | PL_reg_flags |= RF_tainted; |
| 6320 | if (utf8_target) { |
| 6321 | loceol = PL_regeol; |
| 6322 | while (hardcount < max && scan < loceol && |
| 6323 | !isDIGIT_LC_utf8((U8*)scan)) { |
| 6324 | scan += UTF8SKIP(scan); |
| 6325 | hardcount++; |
| 6326 | } |
| 6327 | } else { |
| 6328 | while (scan < loceol && !isDIGIT_LC(*scan)) |
| 6329 | scan++; |
| 6330 | } |
| 6331 | break; |
| 6332 | case LNBREAK: |
| 6333 | if (utf8_target) { |
| 6334 | loceol = PL_regeol; |
| 6335 | while (hardcount < max && scan < loceol && (c=is_LNBREAK_utf8(scan))) { |
| 6336 | scan += c; |
| 6337 | hardcount++; |
| 6338 | } |
| 6339 | } else { |
| 6340 | /* |
| 6341 | LNBREAK can match two latin chars, which is ok, |
| 6342 | because we have a null terminated string, but we |
| 6343 | have to use hardcount in this situation |
| 6344 | */ |
| 6345 | while (scan < loceol && (c=is_LNBREAK_latin1(scan))) { |
| 6346 | scan+=c; |
| 6347 | hardcount++; |
| 6348 | } |
| 6349 | } |
| 6350 | break; |
| 6351 | case HORIZWS: |
| 6352 | if (utf8_target) { |
| 6353 | loceol = PL_regeol; |
| 6354 | while (hardcount < max && scan < loceol && (c=is_HORIZWS_utf8(scan))) { |
| 6355 | scan += c; |
| 6356 | hardcount++; |
| 6357 | } |
| 6358 | } else { |
| 6359 | while (scan < loceol && is_HORIZWS_latin1(scan)) |
| 6360 | scan++; |
| 6361 | } |
| 6362 | break; |
| 6363 | case NHORIZWS: |
| 6364 | if (utf8_target) { |
| 6365 | loceol = PL_regeol; |
| 6366 | while (hardcount < max && scan < loceol && !is_HORIZWS_utf8(scan)) { |
| 6367 | scan += UTF8SKIP(scan); |
| 6368 | hardcount++; |
| 6369 | } |
| 6370 | } else { |
| 6371 | while (scan < loceol && !is_HORIZWS_latin1(scan)) |
| 6372 | scan++; |
| 6373 | |
| 6374 | } |
| 6375 | break; |
| 6376 | case VERTWS: |
| 6377 | if (utf8_target) { |
| 6378 | loceol = PL_regeol; |
| 6379 | while (hardcount < max && scan < loceol && (c=is_VERTWS_utf8(scan))) { |
| 6380 | scan += c; |
| 6381 | hardcount++; |
| 6382 | } |
| 6383 | } else { |
| 6384 | while (scan < loceol && is_VERTWS_latin1(scan)) |
| 6385 | scan++; |
| 6386 | |
| 6387 | } |
| 6388 | break; |
| 6389 | case NVERTWS: |
| 6390 | if (utf8_target) { |
| 6391 | loceol = PL_regeol; |
| 6392 | while (hardcount < max && scan < loceol && !is_VERTWS_utf8(scan)) { |
| 6393 | scan += UTF8SKIP(scan); |
| 6394 | hardcount++; |
| 6395 | } |
| 6396 | } else { |
| 6397 | while (scan < loceol && !is_VERTWS_latin1(scan)) |
| 6398 | scan++; |
| 6399 | |
| 6400 | } |
| 6401 | break; |
| 6402 | |
| 6403 | default: /* Called on something of 0 width. */ |
| 6404 | break; /* So match right here or not at all. */ |
| 6405 | } |
| 6406 | |
| 6407 | if (hardcount) |
| 6408 | c = hardcount; |
| 6409 | else |
| 6410 | c = scan - PL_reginput; |
| 6411 | PL_reginput = scan; |
| 6412 | |
| 6413 | DEBUG_r({ |
| 6414 | GET_RE_DEBUG_FLAGS_DECL; |
| 6415 | DEBUG_EXECUTE_r({ |
| 6416 | SV * const prop = sv_newmortal(); |
| 6417 | regprop(prog, prop, p); |
| 6418 | PerlIO_printf(Perl_debug_log, |
| 6419 | "%*s %s can match %"IVdf" times out of %"IVdf"...\n", |
| 6420 | REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max); |
| 6421 | }); |
| 6422 | }); |
| 6423 | |
| 6424 | return(c); |
| 6425 | } |
| 6426 | |
| 6427 | |
| 6428 | #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) |
| 6429 | /* |
| 6430 | - regclass_swash - prepare the utf8 swash |
| 6431 | */ |
| 6432 | |
| 6433 | SV * |
| 6434 | Perl_regclass_swash(pTHX_ const regexp *prog, register const regnode* node, bool doinit, SV** listsvp, SV **altsvp) |
| 6435 | { |
| 6436 | dVAR; |
| 6437 | SV *sw = NULL; |
| 6438 | SV *si = NULL; |
| 6439 | SV *alt = NULL; |
| 6440 | RXi_GET_DECL(prog,progi); |
| 6441 | const struct reg_data * const data = prog ? progi->data : NULL; |
| 6442 | |
| 6443 | PERL_ARGS_ASSERT_REGCLASS_SWASH; |
| 6444 | |
| 6445 | assert(ANYOF_NONBITMAP(node)); |
| 6446 | |
| 6447 | if (data && data->count) { |
| 6448 | const U32 n = ARG(node); |
| 6449 | |
| 6450 | if (data->what[n] == 's') { |
| 6451 | SV * const rv = MUTABLE_SV(data->data[n]); |
| 6452 | AV * const av = MUTABLE_AV(SvRV(rv)); |
| 6453 | SV **const ary = AvARRAY(av); |
| 6454 | SV **a, **b; |
| 6455 | |
| 6456 | /* See the end of regcomp.c:S_regclass() for |
| 6457 | * documentation of these array elements. */ |
| 6458 | |
| 6459 | si = *ary; |
| 6460 | a = SvROK(ary[1]) ? &ary[1] : NULL; |
| 6461 | b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL; |
| 6462 | |
| 6463 | if (a) |
| 6464 | sw = *a; |
| 6465 | else if (si && doinit) { |
| 6466 | sw = swash_init("utf8", "", si, 1, 0); |
| 6467 | (void)av_store(av, 1, sw); |
| 6468 | } |
| 6469 | if (b) |
| 6470 | alt = *b; |
| 6471 | } |
| 6472 | } |
| 6473 | |
| 6474 | if (listsvp) |
| 6475 | *listsvp = si; |
| 6476 | if (altsvp) |
| 6477 | *altsvp = alt; |
| 6478 | |
| 6479 | return sw; |
| 6480 | } |
| 6481 | #endif |
| 6482 | |
| 6483 | /* |
| 6484 | - reginclass - determine if a character falls into a character class |
| 6485 | |
| 6486 | n is the ANYOF regnode |
| 6487 | p is the target string |
| 6488 | lenp is pointer to the maximum number of bytes of how far to go in p |
| 6489 | (This is assumed wthout checking to always be at least the current |
| 6490 | character's size) |
| 6491 | utf8_target tells whether p is in UTF-8. |
| 6492 | |
| 6493 | Returns true if matched; false otherwise. If lenp is not NULL, on return |
| 6494 | from a successful match, the value it points to will be updated to how many |
| 6495 | bytes in p were matched. If there was no match, the value is undefined, |
| 6496 | possibly changed from the input. |
| 6497 | |
| 6498 | Note that this can be a synthetic start class, a combination of various |
| 6499 | nodes, so things you think might be mutually exclusive, such as locale, |
| 6500 | aren't. It can match both locale and non-locale |
| 6501 | |
| 6502 | */ |
| 6503 | |
| 6504 | STATIC bool |
| 6505 | S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, register const U8* const p, STRLEN* lenp, register const bool utf8_target) |
| 6506 | { |
| 6507 | dVAR; |
| 6508 | const char flags = ANYOF_FLAGS(n); |
| 6509 | bool match = FALSE; |
| 6510 | UV c = *p; |
| 6511 | STRLEN c_len = 0; |
| 6512 | STRLEN maxlen; |
| 6513 | |
| 6514 | PERL_ARGS_ASSERT_REGINCLASS; |
| 6515 | |
| 6516 | /* If c is not already the code point, get it */ |
| 6517 | if (utf8_target && !UTF8_IS_INVARIANT(c)) { |
| 6518 | c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len, |
| 6519 | (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) |
| 6520 | | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY); |
| 6521 | /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for |
| 6522 | * UTF8_ALLOW_FFFF */ |
| 6523 | if (c_len == (STRLEN)-1) |
| 6524 | Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)"); |
| 6525 | } |
| 6526 | else { |
| 6527 | c_len = 1; |
| 6528 | } |
| 6529 | |
| 6530 | /* Use passed in max length, or one character if none passed in or less |
| 6531 | * than one character. And assume will match just one character. This is |
| 6532 | * overwritten later if matched more. */ |
| 6533 | if (lenp) { |
| 6534 | maxlen = (*lenp > c_len) ? *lenp : c_len; |
| 6535 | *lenp = c_len; |
| 6536 | |
| 6537 | } |
| 6538 | else { |
| 6539 | maxlen = c_len; |
| 6540 | } |
| 6541 | |
| 6542 | /* If this character is potentially in the bitmap, check it */ |
| 6543 | if (c < 256) { |
| 6544 | if (ANYOF_BITMAP_TEST(n, c)) |
| 6545 | match = TRUE; |
| 6546 | else if (flags & ANYOF_NON_UTF8_LATIN1_ALL |
| 6547 | && ! utf8_target |
| 6548 | && ! isASCII(c)) |
| 6549 | { |
| 6550 | match = TRUE; |
| 6551 | } |
| 6552 | |
| 6553 | else if (flags & ANYOF_LOCALE) { |
| 6554 | PL_reg_flags |= RF_tainted; |
| 6555 | |
| 6556 | if ((flags & ANYOF_LOC_NONBITMAP_FOLD) |
| 6557 | && ANYOF_BITMAP_TEST(n, PL_fold_locale[c])) |
| 6558 | { |
| 6559 | match = TRUE; |
| 6560 | } |
| 6561 | else if (ANYOF_CLASS_TEST_ANY_SET(n) && |
| 6562 | ((ANYOF_CLASS_TEST(n, ANYOF_ALNUM) && isALNUM_LC(c)) || |
| 6563 | (ANYOF_CLASS_TEST(n, ANYOF_NALNUM) && !isALNUM_LC(c)) || |
| 6564 | (ANYOF_CLASS_TEST(n, ANYOF_SPACE) && isSPACE_LC(c)) || |
| 6565 | (ANYOF_CLASS_TEST(n, ANYOF_NSPACE) && !isSPACE_LC(c)) || |
| 6566 | (ANYOF_CLASS_TEST(n, ANYOF_DIGIT) && isDIGIT_LC(c)) || |
| 6567 | (ANYOF_CLASS_TEST(n, ANYOF_NDIGIT) && !isDIGIT_LC(c)) || |
| 6568 | (ANYOF_CLASS_TEST(n, ANYOF_ALNUMC) && isALNUMC_LC(c)) || |
| 6569 | (ANYOF_CLASS_TEST(n, ANYOF_NALNUMC) && !isALNUMC_LC(c)) || |
| 6570 | (ANYOF_CLASS_TEST(n, ANYOF_ALPHA) && isALPHA_LC(c)) || |
| 6571 | (ANYOF_CLASS_TEST(n, ANYOF_NALPHA) && !isALPHA_LC(c)) || |
| 6572 | (ANYOF_CLASS_TEST(n, ANYOF_ASCII) && isASCII(c)) || |
| 6573 | (ANYOF_CLASS_TEST(n, ANYOF_NASCII) && !isASCII(c)) || |
| 6574 | (ANYOF_CLASS_TEST(n, ANYOF_CNTRL) && isCNTRL_LC(c)) || |
| 6575 | (ANYOF_CLASS_TEST(n, ANYOF_NCNTRL) && !isCNTRL_LC(c)) || |
| 6576 | (ANYOF_CLASS_TEST(n, ANYOF_GRAPH) && isGRAPH_LC(c)) || |
| 6577 | (ANYOF_CLASS_TEST(n, ANYOF_NGRAPH) && !isGRAPH_LC(c)) || |
| 6578 | (ANYOF_CLASS_TEST(n, ANYOF_LOWER) && isLOWER_LC(c)) || |
| 6579 | (ANYOF_CLASS_TEST(n, ANYOF_NLOWER) && !isLOWER_LC(c)) || |
| 6580 | (ANYOF_CLASS_TEST(n, ANYOF_PRINT) && isPRINT_LC(c)) || |
| 6581 | (ANYOF_CLASS_TEST(n, ANYOF_NPRINT) && !isPRINT_LC(c)) || |
| 6582 | (ANYOF_CLASS_TEST(n, ANYOF_PUNCT) && isPUNCT_LC(c)) || |
| 6583 | (ANYOF_CLASS_TEST(n, ANYOF_NPUNCT) && !isPUNCT_LC(c)) || |
| 6584 | (ANYOF_CLASS_TEST(n, ANYOF_UPPER) && isUPPER_LC(c)) || |
| 6585 | (ANYOF_CLASS_TEST(n, ANYOF_NUPPER) && !isUPPER_LC(c)) || |
| 6586 | (ANYOF_CLASS_TEST(n, ANYOF_XDIGIT) && isXDIGIT(c)) || |
| 6587 | (ANYOF_CLASS_TEST(n, ANYOF_NXDIGIT) && !isXDIGIT(c)) || |
| 6588 | (ANYOF_CLASS_TEST(n, ANYOF_PSXSPC) && isPSXSPC(c)) || |
| 6589 | (ANYOF_CLASS_TEST(n, ANYOF_NPSXSPC) && !isPSXSPC(c)) || |
| 6590 | (ANYOF_CLASS_TEST(n, ANYOF_BLANK) && isBLANK(c)) || |
| 6591 | (ANYOF_CLASS_TEST(n, ANYOF_NBLANK) && !isBLANK(c)) |
| 6592 | ) /* How's that for a conditional? */ |
| 6593 | ) { |
| 6594 | match = TRUE; |
| 6595 | } |
| 6596 | } |
| 6597 | } |
| 6598 | |
| 6599 | /* If the bitmap didn't (or couldn't) match, and something outside the |
| 6600 | * bitmap could match, try that. Locale nodes specifiy completely the |
| 6601 | * behavior of code points in the bit map (otherwise, a utf8 target would |
| 6602 | * cause them to be treated as Unicode and not locale), except in |
| 6603 | * the very unlikely event when this node is a synthetic start class, which |
| 6604 | * could be a combination of locale and non-locale nodes. So allow locale |
| 6605 | * to match for the synthetic start class, which will give a false |
| 6606 | * positive that will be resolved when the match is done again as not part |
| 6607 | * of the synthetic start class */ |
| 6608 | if (!match) { |
| 6609 | if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) { |
| 6610 | match = TRUE; /* Everything above 255 matches */ |
| 6611 | } |
| 6612 | else if (ANYOF_NONBITMAP(n) |
| 6613 | && ((flags & ANYOF_NONBITMAP_NON_UTF8) |
| 6614 | || (utf8_target |
| 6615 | && (c >=256 |
| 6616 | || (! (flags & ANYOF_LOCALE)) |
| 6617 | || (flags & ANYOF_IS_SYNTHETIC))))) |
| 6618 | { |
| 6619 | AV *av; |
| 6620 | SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av); |
| 6621 | |
| 6622 | if (sw) { |
| 6623 | U8 * utf8_p; |
| 6624 | if (utf8_target) { |
| 6625 | utf8_p = (U8 *) p; |
| 6626 | } else { |
| 6627 | |
| 6628 | /* Not utf8. Convert as much of the string as available up |
| 6629 | * to the limit of how far the (single) character in the |
| 6630 | * pattern can possibly match (no need to go further). If |
| 6631 | * the node is a straight ANYOF or not folding, it can't |
| 6632 | * match more than one. Otherwise, It can match up to how |
| 6633 | * far a single char can fold to. Since not utf8, each |
| 6634 | * character is a single byte, so the max it can be in |
| 6635 | * bytes is the same as the max it can be in characters */ |
| 6636 | STRLEN len = (OP(n) == ANYOF |
| 6637 | || ! (flags & ANYOF_LOC_NONBITMAP_FOLD)) |
| 6638 | ? 1 |
| 6639 | : (maxlen < UTF8_MAX_FOLD_CHAR_EXPAND) |
| 6640 | ? maxlen |
| 6641 | : UTF8_MAX_FOLD_CHAR_EXPAND; |
| 6642 | utf8_p = bytes_to_utf8(p, &len); |
| 6643 | } |
| 6644 | |
| 6645 | if (swash_fetch(sw, utf8_p, TRUE)) |
| 6646 | match = TRUE; |
| 6647 | else if (flags & ANYOF_LOC_NONBITMAP_FOLD) { |
| 6648 | |
| 6649 | /* Here, we need to test if the fold of the target string |
| 6650 | * matches. The non-multi char folds have all been moved to |
| 6651 | * the compilation phase, and the multi-char folds have |
| 6652 | * been stored by regcomp into 'av'; we linearly check to |
| 6653 | * see if any match the target string (folded). We know |
| 6654 | * that the originals were each one character, but we don't |
| 6655 | * currently know how many characters/bytes each folded to, |
| 6656 | * except we do know that there are small limits imposed by |
| 6657 | * Unicode. XXX A performance enhancement would be to have |
| 6658 | * regcomp.c store the max number of chars/bytes that are |
| 6659 | * in an av entry, as, say the 0th element. Even better |
| 6660 | * would be to have a hash of the few characters that can |
| 6661 | * start a multi-char fold to the max number of chars of |
| 6662 | * those folds. |
| 6663 | * |
| 6664 | * If there is a match, we will need to advance (if lenp is |
| 6665 | * specified) the match pointer in the target string. But |
| 6666 | * what we are comparing here isn't that string directly, |
| 6667 | * but its fold, whose length may differ from the original. |
| 6668 | * As we go along in constructing the fold, therefore, we |
| 6669 | * create a map so that we know how many bytes in the |
| 6670 | * source to advance given that we have matched a certain |
| 6671 | * number of bytes in the fold. This map is stored in |
| 6672 | * 'map_fold_len_back'. Let n mean the number of bytes in |
| 6673 | * the fold of the first character that we are folding. |
| 6674 | * Then map_fold_len_back[n] is set to the number of bytes |
| 6675 | * in that first character. Similarly let m be the |
| 6676 | * corresponding number for the second character to be |
| 6677 | * folded. Then map_fold_len_back[n+m] is set to the |
| 6678 | * number of bytes occupied by the first two source |
| 6679 | * characters. ... */ |
| 6680 | U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 }; |
| 6681 | U8 folded[UTF8_MAXBYTES_CASE+1]; |
| 6682 | STRLEN foldlen = 0; /* num bytes in fold of 1st char */ |
| 6683 | STRLEN total_foldlen = 0; /* num bytes in fold of all |
| 6684 | chars */ |
| 6685 | |
| 6686 | if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) { |
| 6687 | |
| 6688 | /* Here, only need to fold the first char of the target |
| 6689 | * string. It the source wasn't utf8, is 1 byte long */ |
| 6690 | to_utf8_fold(utf8_p, folded, &foldlen); |
| 6691 | total_foldlen = foldlen; |
| 6692 | map_fold_len_back[foldlen] = (utf8_target) |
| 6693 | ? UTF8SKIP(utf8_p) |
| 6694 | : 1; |
| 6695 | } |
| 6696 | else { |
| 6697 | |
| 6698 | /* Here, need to fold more than the first char. Do so |
| 6699 | * up to the limits */ |
| 6700 | U8* source_ptr = utf8_p; /* The source for the fold |
| 6701 | is the regex target |
| 6702 | string */ |
| 6703 | U8* folded_ptr = folded; |
| 6704 | U8* e = utf8_p + maxlen; /* Can't go beyond last |
| 6705 | available byte in the |
| 6706 | target string */ |
| 6707 | U8 i; |
| 6708 | for (i = 0; |
| 6709 | i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e; |
| 6710 | i++) |
| 6711 | { |
| 6712 | |
| 6713 | /* Fold the next character */ |
| 6714 | U8 this_char_folded[UTF8_MAXBYTES_CASE+1]; |
| 6715 | STRLEN this_char_foldlen; |
| 6716 | to_utf8_fold(source_ptr, |
| 6717 | this_char_folded, |
| 6718 | &this_char_foldlen); |
| 6719 | |
| 6720 | /* Bail if it would exceed the byte limit for |
| 6721 | * folding a single char. */ |
| 6722 | if (this_char_foldlen + folded_ptr - folded > |
| 6723 | UTF8_MAXBYTES_CASE) |
| 6724 | { |
| 6725 | break; |
| 6726 | } |
| 6727 | |
| 6728 | /* Add the fold of this character */ |
| 6729 | Copy(this_char_folded, |
| 6730 | folded_ptr, |
| 6731 | this_char_foldlen, |
| 6732 | U8); |
| 6733 | source_ptr += UTF8SKIP(source_ptr); |
| 6734 | folded_ptr += this_char_foldlen; |
| 6735 | total_foldlen = folded_ptr - folded; |
| 6736 | |
| 6737 | /* Create map from the number of bytes in the fold |
| 6738 | * back to the number of bytes in the source. If |
| 6739 | * the source isn't utf8, the byte count is just |
| 6740 | * the number of characters so far */ |
| 6741 | map_fold_len_back[total_foldlen] |
| 6742 | = (utf8_target) |
| 6743 | ? source_ptr - utf8_p |
| 6744 | : i + 1; |
| 6745 | } |
| 6746 | *folded_ptr = '\0'; |
| 6747 | } |
| 6748 | |
| 6749 | |
| 6750 | /* Do the linear search to see if the fold is in the list |
| 6751 | * of multi-char folds. */ |
| 6752 | if (av) { |
| 6753 | I32 i; |
| 6754 | for (i = 0; i <= av_len(av); i++) { |
| 6755 | SV* const sv = *av_fetch(av, i, FALSE); |
| 6756 | STRLEN len; |
| 6757 | const char * const s = SvPV_const(sv, len); |
| 6758 | |
| 6759 | if (len <= total_foldlen |
| 6760 | && memEQ(s, (char*)folded, len) |
| 6761 | |
| 6762 | /* If 0, means matched a partial char. See |
| 6763 | * [perl #90536] */ |
| 6764 | && map_fold_len_back[len]) |
| 6765 | { |
| 6766 | |
| 6767 | /* Advance the target string ptr to account for |
| 6768 | * this fold, but have to translate from the |
| 6769 | * folded length to the corresponding source |
| 6770 | * length. */ |
| 6771 | if (lenp) { |
| 6772 | *lenp = map_fold_len_back[len]; |
| 6773 | } |
| 6774 | match = TRUE; |
| 6775 | break; |
| 6776 | } |
| 6777 | } |
| 6778 | } |
| 6779 | } |
| 6780 | |
| 6781 | /* If we allocated a string above, free it */ |
| 6782 | if (! utf8_target) Safefree(utf8_p); |
| 6783 | } |
| 6784 | } |
| 6785 | } |
| 6786 | |
| 6787 | return (flags & ANYOF_INVERT) ? !match : match; |
| 6788 | } |
| 6789 | |
| 6790 | STATIC U8 * |
| 6791 | S_reghop3(U8 *s, I32 off, const U8* lim) |
| 6792 | { |
| 6793 | dVAR; |
| 6794 | |
| 6795 | PERL_ARGS_ASSERT_REGHOP3; |
| 6796 | |
| 6797 | if (off >= 0) { |
| 6798 | while (off-- && s < lim) { |
| 6799 | /* XXX could check well-formedness here */ |
| 6800 | s += UTF8SKIP(s); |
| 6801 | } |
| 6802 | } |
| 6803 | else { |
| 6804 | while (off++ && s > lim) { |
| 6805 | s--; |
| 6806 | if (UTF8_IS_CONTINUED(*s)) { |
| 6807 | while (s > lim && UTF8_IS_CONTINUATION(*s)) |
| 6808 | s--; |
| 6809 | } |
| 6810 | /* XXX could check well-formedness here */ |
| 6811 | } |
| 6812 | } |
| 6813 | return s; |
| 6814 | } |
| 6815 | |
| 6816 | #ifdef XXX_dmq |
| 6817 | /* there are a bunch of places where we use two reghop3's that should |
| 6818 | be replaced with this routine. but since thats not done yet |
| 6819 | we ifdef it out - dmq |
| 6820 | */ |
| 6821 | STATIC U8 * |
| 6822 | S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim) |
| 6823 | { |
| 6824 | dVAR; |
| 6825 | |
| 6826 | PERL_ARGS_ASSERT_REGHOP4; |
| 6827 | |
| 6828 | if (off >= 0) { |
| 6829 | while (off-- && s < rlim) { |
| 6830 | /* XXX could check well-formedness here */ |
| 6831 | s += UTF8SKIP(s); |
| 6832 | } |
| 6833 | } |
| 6834 | else { |
| 6835 | while (off++ && s > llim) { |
| 6836 | s--; |
| 6837 | if (UTF8_IS_CONTINUED(*s)) { |
| 6838 | while (s > llim && UTF8_IS_CONTINUATION(*s)) |
| 6839 | s--; |
| 6840 | } |
| 6841 | /* XXX could check well-formedness here */ |
| 6842 | } |
| 6843 | } |
| 6844 | return s; |
| 6845 | } |
| 6846 | #endif |
| 6847 | |
| 6848 | STATIC U8 * |
| 6849 | S_reghopmaybe3(U8* s, I32 off, const U8* lim) |
| 6850 | { |
| 6851 | dVAR; |
| 6852 | |
| 6853 | PERL_ARGS_ASSERT_REGHOPMAYBE3; |
| 6854 | |
| 6855 | if (off >= 0) { |
| 6856 | while (off-- && s < lim) { |
| 6857 | /* XXX could check well-formedness here */ |
| 6858 | s += UTF8SKIP(s); |
| 6859 | } |
| 6860 | if (off >= 0) |
| 6861 | return NULL; |
| 6862 | } |
| 6863 | else { |
| 6864 | while (off++ && s > lim) { |
| 6865 | s--; |
| 6866 | if (UTF8_IS_CONTINUED(*s)) { |
| 6867 | while (s > lim && UTF8_IS_CONTINUATION(*s)) |
| 6868 | s--; |
| 6869 | } |
| 6870 | /* XXX could check well-formedness here */ |
| 6871 | } |
| 6872 | if (off <= 0) |
| 6873 | return NULL; |
| 6874 | } |
| 6875 | return s; |
| 6876 | } |
| 6877 | |
| 6878 | static void |
| 6879 | restore_pos(pTHX_ void *arg) |
| 6880 | { |
| 6881 | dVAR; |
| 6882 | regexp * const rex = (regexp *)arg; |
| 6883 | if (PL_reg_eval_set) { |
| 6884 | if (PL_reg_oldsaved) { |
| 6885 | rex->subbeg = PL_reg_oldsaved; |
| 6886 | rex->sublen = PL_reg_oldsavedlen; |
| 6887 | #ifdef PERL_OLD_COPY_ON_WRITE |
| 6888 | rex->saved_copy = PL_nrs; |
| 6889 | #endif |
| 6890 | RXp_MATCH_COPIED_on(rex); |
| 6891 | } |
| 6892 | PL_reg_magic->mg_len = PL_reg_oldpos; |
| 6893 | PL_reg_eval_set = 0; |
| 6894 | PL_curpm = PL_reg_oldcurpm; |
| 6895 | } |
| 6896 | } |
| 6897 | |
| 6898 | STATIC void |
| 6899 | S_to_utf8_substr(pTHX_ register regexp *prog) |
| 6900 | { |
| 6901 | int i = 1; |
| 6902 | |
| 6903 | PERL_ARGS_ASSERT_TO_UTF8_SUBSTR; |
| 6904 | |
| 6905 | do { |
| 6906 | if (prog->substrs->data[i].substr |
| 6907 | && !prog->substrs->data[i].utf8_substr) { |
| 6908 | SV* const sv = newSVsv(prog->substrs->data[i].substr); |
| 6909 | prog->substrs->data[i].utf8_substr = sv; |
| 6910 | sv_utf8_upgrade(sv); |
| 6911 | if (SvVALID(prog->substrs->data[i].substr)) { |
| 6912 | if (SvTAIL(prog->substrs->data[i].substr)) { |
| 6913 | /* Trim the trailing \n that fbm_compile added last |
| 6914 | time. */ |
| 6915 | SvCUR_set(sv, SvCUR(sv) - 1); |
| 6916 | /* Whilst this makes the SV technically "invalid" (as its |
| 6917 | buffer is no longer followed by "\0") when fbm_compile() |
| 6918 | adds the "\n" back, a "\0" is restored. */ |
| 6919 | fbm_compile(sv, FBMcf_TAIL); |
| 6920 | } else |
| 6921 | fbm_compile(sv, 0); |
| 6922 | } |
| 6923 | if (prog->substrs->data[i].substr == prog->check_substr) |
| 6924 | prog->check_utf8 = sv; |
| 6925 | } |
| 6926 | } while (i--); |
| 6927 | } |
| 6928 | |
| 6929 | STATIC void |
| 6930 | S_to_byte_substr(pTHX_ register regexp *prog) |
| 6931 | { |
| 6932 | dVAR; |
| 6933 | int i = 1; |
| 6934 | |
| 6935 | PERL_ARGS_ASSERT_TO_BYTE_SUBSTR; |
| 6936 | |
| 6937 | do { |
| 6938 | if (prog->substrs->data[i].utf8_substr |
| 6939 | && !prog->substrs->data[i].substr) { |
| 6940 | SV* sv = newSVsv(prog->substrs->data[i].utf8_substr); |
| 6941 | if (sv_utf8_downgrade(sv, TRUE)) { |
| 6942 | if (SvVALID(prog->substrs->data[i].utf8_substr)) { |
| 6943 | if (SvTAIL(prog->substrs->data[i].utf8_substr)) { |
| 6944 | /* Trim the trailing \n that fbm_compile added last |
| 6945 | time. */ |
| 6946 | SvCUR_set(sv, SvCUR(sv) - 1); |
| 6947 | fbm_compile(sv, FBMcf_TAIL); |
| 6948 | } else |
| 6949 | fbm_compile(sv, 0); |
| 6950 | } |
| 6951 | } else { |
| 6952 | SvREFCNT_dec(sv); |
| 6953 | sv = &PL_sv_undef; |
| 6954 | } |
| 6955 | prog->substrs->data[i].substr = sv; |
| 6956 | if (prog->substrs->data[i].utf8_substr == prog->check_utf8) |
| 6957 | prog->check_substr = sv; |
| 6958 | } |
| 6959 | } while (i--); |
| 6960 | } |
| 6961 | |
| 6962 | /* |
| 6963 | * Local variables: |
| 6964 | * c-indentation-style: bsd |
| 6965 | * c-basic-offset: 4 |
| 6966 | * indent-tabs-mode: t |
| 6967 | * End: |
| 6968 | * |
| 6969 | * ex: set ts=8 sts=4 sw=4 noet: |
| 6970 | */ |