| 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 | /* At least one required character in the target string is expressible only in |
| 41 | * UTF-8. */ |
| 42 | static const char* const non_utf8_target_but_utf8_required |
| 43 | = "Can't match, because target string needs to be in UTF-8\n"; |
| 44 | |
| 45 | #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \ |
| 46 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\ |
| 47 | goto target; \ |
| 48 | } STMT_END |
| 49 | |
| 50 | /* |
| 51 | * pregcomp and pregexec -- regsub and regerror are not used in perl |
| 52 | * |
| 53 | * Copyright (c) 1986 by University of Toronto. |
| 54 | * Written by Henry Spencer. Not derived from licensed software. |
| 55 | * |
| 56 | * Permission is granted to anyone to use this software for any |
| 57 | * purpose on any computer system, and to redistribute it freely, |
| 58 | * subject to the following restrictions: |
| 59 | * |
| 60 | * 1. The author is not responsible for the consequences of use of |
| 61 | * this software, no matter how awful, even if they arise |
| 62 | * from defects in it. |
| 63 | * |
| 64 | * 2. The origin of this software must not be misrepresented, either |
| 65 | * by explicit claim or by omission. |
| 66 | * |
| 67 | * 3. Altered versions must be plainly marked as such, and must not |
| 68 | * be misrepresented as being the original software. |
| 69 | * |
| 70 | **** Alterations to Henry's code are... |
| 71 | **** |
| 72 | **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, |
| 73 | **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 |
| 74 | **** by Larry Wall and others |
| 75 | **** |
| 76 | **** You may distribute under the terms of either the GNU General Public |
| 77 | **** License or the Artistic License, as specified in the README file. |
| 78 | * |
| 79 | * Beware that some of this code is subtly aware of the way operator |
| 80 | * precedence is structured in regular expressions. Serious changes in |
| 81 | * regular-expression syntax might require a total rethink. |
| 82 | */ |
| 83 | #include "EXTERN.h" |
| 84 | #define PERL_IN_REGEXEC_C |
| 85 | #include "perl.h" |
| 86 | |
| 87 | #ifdef PERL_IN_XSUB_RE |
| 88 | # include "re_comp.h" |
| 89 | #else |
| 90 | # include "regcomp.h" |
| 91 | #endif |
| 92 | |
| 93 | #include "inline_invlist.c" |
| 94 | #include "unicode_constants.h" |
| 95 | |
| 96 | #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i) |
| 97 | |
| 98 | #ifndef STATIC |
| 99 | #define STATIC static |
| 100 | #endif |
| 101 | |
| 102 | /* Valid for non-utf8 strings: avoids the reginclass |
| 103 | * call if there are no complications: i.e., if everything matchable is |
| 104 | * straight forward in the bitmap */ |
| 105 | #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0) \ |
| 106 | : ANYOF_BITMAP_TEST(p,*(c))) |
| 107 | |
| 108 | /* |
| 109 | * Forwards. |
| 110 | */ |
| 111 | |
| 112 | #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv)) |
| 113 | #define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b) |
| 114 | |
| 115 | #define HOPc(pos,off) \ |
| 116 | (char *)(PL_reg_match_utf8 \ |
| 117 | ? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \ |
| 118 | : (U8*)(pos + off)) |
| 119 | #define HOPBACKc(pos, off) \ |
| 120 | (char*)(PL_reg_match_utf8\ |
| 121 | ? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \ |
| 122 | : (pos - off >= PL_bostr) \ |
| 123 | ? (U8*)pos - off \ |
| 124 | : NULL) |
| 125 | |
| 126 | #define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off)) |
| 127 | #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim)) |
| 128 | |
| 129 | |
| 130 | #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */ |
| 131 | #define NEXTCHR_IS_EOS (nextchr < 0) |
| 132 | |
| 133 | #define SET_nextchr \ |
| 134 | nextchr = ((locinput < PL_regeol) ? UCHARAT(locinput) : NEXTCHR_EOS) |
| 135 | |
| 136 | #define SET_locinput(p) \ |
| 137 | locinput = (p); \ |
| 138 | SET_nextchr |
| 139 | |
| 140 | |
| 141 | #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name) STMT_START { \ |
| 142 | if (!swash_ptr) { \ |
| 143 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \ |
| 144 | swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \ |
| 145 | 1, 0, NULL, &flags); \ |
| 146 | assert(swash_ptr); \ |
| 147 | } \ |
| 148 | } STMT_END |
| 149 | |
| 150 | /* If in debug mode, we test that a known character properly matches */ |
| 151 | #ifdef DEBUGGING |
| 152 | # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \ |
| 153 | property_name, \ |
| 154 | utf8_char_in_property) \ |
| 155 | LOAD_UTF8_CHARCLASS(swash_ptr, property_name); \ |
| 156 | assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE)); |
| 157 | #else |
| 158 | # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \ |
| 159 | property_name, \ |
| 160 | utf8_char_in_property) \ |
| 161 | LOAD_UTF8_CHARCLASS(swash_ptr, property_name) |
| 162 | #endif |
| 163 | |
| 164 | #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \ |
| 165 | PL_utf8_swash_ptrs[_CC_WORDCHAR], \ |
| 166 | swash_property_names[_CC_WORDCHAR], \ |
| 167 | GREEK_SMALL_LETTER_IOTA_UTF8) |
| 168 | |
| 169 | #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \ |
| 170 | STMT_START { \ |
| 171 | LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \ |
| 172 | "_X_regular_begin", \ |
| 173 | GREEK_SMALL_LETTER_IOTA_UTF8); \ |
| 174 | LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \ |
| 175 | "_X_extend", \ |
| 176 | COMBINING_GRAVE_ACCENT_UTF8); \ |
| 177 | } STMT_END |
| 178 | |
| 179 | #define PLACEHOLDER /* Something for the preprocessor to grab onto */ |
| 180 | /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */ |
| 181 | |
| 182 | /* for use after a quantifier and before an EXACT-like node -- japhy */ |
| 183 | /* it would be nice to rework regcomp.sym to generate this stuff. sigh |
| 184 | * |
| 185 | * NOTE that *nothing* that affects backtracking should be in here, specifically |
| 186 | * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a |
| 187 | * node that is in between two EXACT like nodes when ascertaining what the required |
| 188 | * "follow" character is. This should probably be moved to regex compile time |
| 189 | * although it may be done at run time beause of the REF possibility - more |
| 190 | * investigation required. -- demerphq |
| 191 | */ |
| 192 | #define JUMPABLE(rn) ( \ |
| 193 | OP(rn) == OPEN || \ |
| 194 | (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \ |
| 195 | OP(rn) == EVAL || \ |
| 196 | OP(rn) == SUSPEND || OP(rn) == IFMATCH || \ |
| 197 | OP(rn) == PLUS || OP(rn) == MINMOD || \ |
| 198 | OP(rn) == KEEPS || \ |
| 199 | (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \ |
| 200 | ) |
| 201 | #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT) |
| 202 | |
| 203 | #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF ) |
| 204 | |
| 205 | #if 0 |
| 206 | /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so |
| 207 | we don't need this definition. */ |
| 208 | #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF ) |
| 209 | #define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn)==EXACTFA || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF ) |
| 210 | #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL ) |
| 211 | |
| 212 | #else |
| 213 | /* ... so we use this as its faster. */ |
| 214 | #define IS_TEXT(rn) ( OP(rn)==EXACT ) |
| 215 | #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA) |
| 216 | #define IS_TEXTF(rn) ( OP(rn)==EXACTF ) |
| 217 | #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL ) |
| 218 | |
| 219 | #endif |
| 220 | |
| 221 | /* |
| 222 | Search for mandatory following text node; for lookahead, the text must |
| 223 | follow but for lookbehind (rn->flags != 0) we skip to the next step. |
| 224 | */ |
| 225 | #define FIND_NEXT_IMPT(rn) STMT_START { \ |
| 226 | while (JUMPABLE(rn)) { \ |
| 227 | const OPCODE type = OP(rn); \ |
| 228 | if (type == SUSPEND || PL_regkind[type] == CURLY) \ |
| 229 | rn = NEXTOPER(NEXTOPER(rn)); \ |
| 230 | else if (type == PLUS) \ |
| 231 | rn = NEXTOPER(rn); \ |
| 232 | else if (type == IFMATCH) \ |
| 233 | rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \ |
| 234 | else rn += NEXT_OFF(rn); \ |
| 235 | } \ |
| 236 | } STMT_END |
| 237 | |
| 238 | /* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode. |
| 239 | * These are for the pre-composed Hangul syllables, which are all in a |
| 240 | * contiguous block and arranged there in such a way so as to facilitate |
| 241 | * alorithmic determination of their characteristics. As such, they don't need |
| 242 | * a swash, but can be determined by simple arithmetic. Almost all are |
| 243 | * GCB=LVT, but every 28th one is a GCB=LV */ |
| 244 | #define SBASE 0xAC00 /* Start of block */ |
| 245 | #define SCount 11172 /* Length of block */ |
| 246 | #define TCount 28 |
| 247 | |
| 248 | static void restore_pos(pTHX_ void *arg); |
| 249 | |
| 250 | #define REGCP_PAREN_ELEMS 3 |
| 251 | #define REGCP_OTHER_ELEMS 3 |
| 252 | #define REGCP_FRAME_ELEMS 1 |
| 253 | /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and |
| 254 | * are needed for the regexp context stack bookkeeping. */ |
| 255 | |
| 256 | STATIC CHECKPOINT |
| 257 | S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen) |
| 258 | { |
| 259 | dVAR; |
| 260 | const int retval = PL_savestack_ix; |
| 261 | const int paren_elems_to_push = |
| 262 | (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS; |
| 263 | const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS; |
| 264 | const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT; |
| 265 | I32 p; |
| 266 | GET_RE_DEBUG_FLAGS_DECL; |
| 267 | |
| 268 | PERL_ARGS_ASSERT_REGCPPUSH; |
| 269 | |
| 270 | if (paren_elems_to_push < 0) |
| 271 | Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0", |
| 272 | paren_elems_to_push); |
| 273 | |
| 274 | if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems) |
| 275 | Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf |
| 276 | " out of range (%lu-%ld)", |
| 277 | total_elems, |
| 278 | (unsigned long)maxopenparen, |
| 279 | (long)parenfloor); |
| 280 | |
| 281 | SSGROW(total_elems + REGCP_FRAME_ELEMS); |
| 282 | |
| 283 | DEBUG_BUFFERS_r( |
| 284 | if ((int)maxopenparen > (int)parenfloor) |
| 285 | PerlIO_printf(Perl_debug_log, |
| 286 | "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n", |
| 287 | PTR2UV(rex), |
| 288 | PTR2UV(rex->offs) |
| 289 | ); |
| 290 | ); |
| 291 | for (p = parenfloor+1; p <= (I32)maxopenparen; p++) { |
| 292 | /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */ |
| 293 | SSPUSHINT(rex->offs[p].end); |
| 294 | SSPUSHINT(rex->offs[p].start); |
| 295 | SSPUSHINT(rex->offs[p].start_tmp); |
| 296 | DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, |
| 297 | " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n", |
| 298 | (UV)p, |
| 299 | (IV)rex->offs[p].start, |
| 300 | (IV)rex->offs[p].start_tmp, |
| 301 | (IV)rex->offs[p].end |
| 302 | )); |
| 303 | } |
| 304 | /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */ |
| 305 | SSPUSHINT(maxopenparen); |
| 306 | SSPUSHINT(rex->lastparen); |
| 307 | SSPUSHINT(rex->lastcloseparen); |
| 308 | SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */ |
| 309 | |
| 310 | return retval; |
| 311 | } |
| 312 | |
| 313 | /* These are needed since we do not localize EVAL nodes: */ |
| 314 | #define REGCP_SET(cp) \ |
| 315 | DEBUG_STATE_r( \ |
| 316 | PerlIO_printf(Perl_debug_log, \ |
| 317 | " Setting an EVAL scope, savestack=%"IVdf"\n", \ |
| 318 | (IV)PL_savestack_ix)); \ |
| 319 | cp = PL_savestack_ix |
| 320 | |
| 321 | #define REGCP_UNWIND(cp) \ |
| 322 | DEBUG_STATE_r( \ |
| 323 | if (cp != PL_savestack_ix) \ |
| 324 | PerlIO_printf(Perl_debug_log, \ |
| 325 | " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \ |
| 326 | (IV)(cp), (IV)PL_savestack_ix)); \ |
| 327 | regcpblow(cp) |
| 328 | |
| 329 | #define UNWIND_PAREN(lp, lcp) \ |
| 330 | for (n = rex->lastparen; n > lp; n--) \ |
| 331 | rex->offs[n].end = -1; \ |
| 332 | rex->lastparen = n; \ |
| 333 | rex->lastcloseparen = lcp; |
| 334 | |
| 335 | |
| 336 | STATIC void |
| 337 | S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p) |
| 338 | { |
| 339 | dVAR; |
| 340 | UV i; |
| 341 | U32 paren; |
| 342 | GET_RE_DEBUG_FLAGS_DECL; |
| 343 | |
| 344 | PERL_ARGS_ASSERT_REGCPPOP; |
| 345 | |
| 346 | /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */ |
| 347 | i = SSPOPUV; |
| 348 | assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */ |
| 349 | i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */ |
| 350 | rex->lastcloseparen = SSPOPINT; |
| 351 | rex->lastparen = SSPOPINT; |
| 352 | *maxopenparen_p = SSPOPINT; |
| 353 | |
| 354 | i -= REGCP_OTHER_ELEMS; |
| 355 | /* Now restore the parentheses context. */ |
| 356 | DEBUG_BUFFERS_r( |
| 357 | if (i || rex->lastparen + 1 <= rex->nparens) |
| 358 | PerlIO_printf(Perl_debug_log, |
| 359 | "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n", |
| 360 | PTR2UV(rex), |
| 361 | PTR2UV(rex->offs) |
| 362 | ); |
| 363 | ); |
| 364 | paren = *maxopenparen_p; |
| 365 | for ( ; i > 0; i -= REGCP_PAREN_ELEMS) { |
| 366 | I32 tmps; |
| 367 | rex->offs[paren].start_tmp = SSPOPINT; |
| 368 | rex->offs[paren].start = SSPOPINT; |
| 369 | tmps = SSPOPINT; |
| 370 | if (paren <= rex->lastparen) |
| 371 | rex->offs[paren].end = tmps; |
| 372 | DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log, |
| 373 | " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n", |
| 374 | (UV)paren, |
| 375 | (IV)rex->offs[paren].start, |
| 376 | (IV)rex->offs[paren].start_tmp, |
| 377 | (IV)rex->offs[paren].end, |
| 378 | (paren > rex->lastparen ? "(skipped)" : "")); |
| 379 | ); |
| 380 | paren--; |
| 381 | } |
| 382 | #if 1 |
| 383 | /* It would seem that the similar code in regtry() |
| 384 | * already takes care of this, and in fact it is in |
| 385 | * a better location to since this code can #if 0-ed out |
| 386 | * but the code in regtry() is needed or otherwise tests |
| 387 | * requiring null fields (pat.t#187 and split.t#{13,14} |
| 388 | * (as of patchlevel 7877) will fail. Then again, |
| 389 | * this code seems to be necessary or otherwise |
| 390 | * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/ |
| 391 | * --jhi updated by dapm */ |
| 392 | for (i = rex->lastparen + 1; i <= rex->nparens; i++) { |
| 393 | if (i > *maxopenparen_p) |
| 394 | rex->offs[i].start = -1; |
| 395 | rex->offs[i].end = -1; |
| 396 | DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log, |
| 397 | " \\%"UVuf": %s ..-1 undeffing\n", |
| 398 | (UV)i, |
| 399 | (i > *maxopenparen_p) ? "-1" : " " |
| 400 | )); |
| 401 | } |
| 402 | #endif |
| 403 | } |
| 404 | |
| 405 | /* restore the parens and associated vars at savestack position ix, |
| 406 | * but without popping the stack */ |
| 407 | |
| 408 | STATIC void |
| 409 | S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p) |
| 410 | { |
| 411 | I32 tmpix = PL_savestack_ix; |
| 412 | PL_savestack_ix = ix; |
| 413 | regcppop(rex, maxopenparen_p); |
| 414 | PL_savestack_ix = tmpix; |
| 415 | } |
| 416 | |
| 417 | #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */ |
| 418 | |
| 419 | STATIC bool |
| 420 | S_isFOO_lc(pTHX_ const U8 classnum, const U8 character) |
| 421 | { |
| 422 | /* Returns a boolean as to whether or not 'character' is a member of the |
| 423 | * Posix character class given by 'classnum' that should be equivalent to a |
| 424 | * value in the typedef '_char_class_number'. |
| 425 | * |
| 426 | * Ideally this could be replaced by a just an array of function pointers |
| 427 | * to the C library functions that implement the macros this calls. |
| 428 | * However, to compile, the precise function signatures are required, and |
| 429 | * these may vary from platform to to platform. To avoid having to figure |
| 430 | * out what those all are on each platform, I (khw) am using this method, |
| 431 | * which adds an extra layer of function call overhead (unless the C |
| 432 | * optimizer strips it away). But we don't particularly care about |
| 433 | * performance with locales anyway. */ |
| 434 | |
| 435 | switch ((_char_class_number) classnum) { |
| 436 | case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character); |
| 437 | case _CC_ENUM_ALPHA: return isALPHA_LC(character); |
| 438 | case _CC_ENUM_ASCII: return isASCII_LC(character); |
| 439 | case _CC_ENUM_BLANK: return isBLANK_LC(character); |
| 440 | case _CC_ENUM_CASED: return isLOWER_LC(character) |
| 441 | || isUPPER_LC(character); |
| 442 | case _CC_ENUM_CNTRL: return isCNTRL_LC(character); |
| 443 | case _CC_ENUM_DIGIT: return isDIGIT_LC(character); |
| 444 | case _CC_ENUM_GRAPH: return isGRAPH_LC(character); |
| 445 | case _CC_ENUM_LOWER: return isLOWER_LC(character); |
| 446 | case _CC_ENUM_PRINT: return isPRINT_LC(character); |
| 447 | case _CC_ENUM_PSXSPC: return isPSXSPC_LC(character); |
| 448 | case _CC_ENUM_PUNCT: return isPUNCT_LC(character); |
| 449 | case _CC_ENUM_SPACE: return isSPACE_LC(character); |
| 450 | case _CC_ENUM_UPPER: return isUPPER_LC(character); |
| 451 | case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character); |
| 452 | case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character); |
| 453 | default: /* VERTSPACE should never occur in locales */ |
| 454 | Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum); |
| 455 | } |
| 456 | |
| 457 | assert(0); /* NOTREACHED */ |
| 458 | return FALSE; |
| 459 | } |
| 460 | |
| 461 | STATIC bool |
| 462 | S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character) |
| 463 | { |
| 464 | /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded |
| 465 | * 'character' is a member of the Posix character class given by 'classnum' |
| 466 | * that should be equivalent to a value in the typedef |
| 467 | * '_char_class_number'. |
| 468 | * |
| 469 | * This just calls isFOO_lc on the code point for the character if it is in |
| 470 | * the range 0-255. Outside that range, all characters avoid Unicode |
| 471 | * rules, ignoring any locale. So use the Unicode function if this class |
| 472 | * requires a swash, and use the Unicode macro otherwise. */ |
| 473 | |
| 474 | PERL_ARGS_ASSERT_ISFOO_UTF8_LC; |
| 475 | |
| 476 | if (UTF8_IS_INVARIANT(*character)) { |
| 477 | return isFOO_lc(classnum, *character); |
| 478 | } |
| 479 | else if (UTF8_IS_DOWNGRADEABLE_START(*character)) { |
| 480 | return isFOO_lc(classnum, |
| 481 | TWO_BYTE_UTF8_TO_UNI(*character, *(character + 1))); |
| 482 | } |
| 483 | |
| 484 | if (classnum < _FIRST_NON_SWASH_CC) { |
| 485 | |
| 486 | /* Initialize the swash unless done already */ |
| 487 | if (! PL_utf8_swash_ptrs[classnum]) { |
| 488 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; |
| 489 | PL_utf8_swash_ptrs[classnum] = _core_swash_init("utf8", |
| 490 | swash_property_names[classnum], &PL_sv_undef, 1, 0, NULL, &flags); |
| 491 | } |
| 492 | |
| 493 | return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *) |
| 494 | character, |
| 495 | TRUE /* is UTF */ )); |
| 496 | } |
| 497 | |
| 498 | switch ((_char_class_number) classnum) { |
| 499 | case _CC_ENUM_SPACE: |
| 500 | case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character); |
| 501 | |
| 502 | case _CC_ENUM_BLANK: return is_HORIZWS_high(character); |
| 503 | case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character); |
| 504 | case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character); |
| 505 | default: return 0; /* Things like CNTRL are always |
| 506 | below 256 */ |
| 507 | } |
| 508 | |
| 509 | assert(0); /* NOTREACHED */ |
| 510 | return FALSE; |
| 511 | } |
| 512 | |
| 513 | /* |
| 514 | * pregexec and friends |
| 515 | */ |
| 516 | |
| 517 | #ifndef PERL_IN_XSUB_RE |
| 518 | /* |
| 519 | - pregexec - match a regexp against a string |
| 520 | */ |
| 521 | I32 |
| 522 | Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend, |
| 523 | char *strbeg, I32 minend, SV *screamer, U32 nosave) |
| 524 | /* stringarg: the point in the string at which to begin matching */ |
| 525 | /* strend: pointer to null at end of string */ |
| 526 | /* strbeg: real beginning of string */ |
| 527 | /* minend: end of match must be >= minend bytes after stringarg. */ |
| 528 | /* screamer: SV being matched: only used for utf8 flag, pos() etc; string |
| 529 | * itself is accessed via the pointers above */ |
| 530 | /* nosave: For optimizations. */ |
| 531 | { |
| 532 | PERL_ARGS_ASSERT_PREGEXEC; |
| 533 | |
| 534 | return |
| 535 | regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL, |
| 536 | nosave ? 0 : REXEC_COPY_STR); |
| 537 | } |
| 538 | #endif |
| 539 | |
| 540 | /* |
| 541 | * Need to implement the following flags for reg_anch: |
| 542 | * |
| 543 | * USE_INTUIT_NOML - Useful to call re_intuit_start() first |
| 544 | * USE_INTUIT_ML |
| 545 | * INTUIT_AUTORITATIVE_NOML - Can trust a positive answer |
| 546 | * INTUIT_AUTORITATIVE_ML |
| 547 | * INTUIT_ONCE_NOML - Intuit can match in one location only. |
| 548 | * INTUIT_ONCE_ML |
| 549 | * |
| 550 | * Another flag for this function: SECOND_TIME (so that float substrs |
| 551 | * with giant delta may be not rechecked). |
| 552 | */ |
| 553 | |
| 554 | /* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */ |
| 555 | |
| 556 | /* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend. |
| 557 | Otherwise, only SvCUR(sv) is used to get strbeg. */ |
| 558 | |
| 559 | /* XXXX We assume that strpos is strbeg unless sv. */ |
| 560 | |
| 561 | /* XXXX Some places assume that there is a fixed substring. |
| 562 | An update may be needed if optimizer marks as "INTUITable" |
| 563 | RExen without fixed substrings. Similarly, it is assumed that |
| 564 | lengths of all the strings are no more than minlen, thus they |
| 565 | cannot come from lookahead. |
| 566 | (Or minlen should take into account lookahead.) |
| 567 | NOTE: Some of this comment is not correct. minlen does now take account |
| 568 | of lookahead/behind. Further research is required. -- demerphq |
| 569 | |
| 570 | */ |
| 571 | |
| 572 | /* A failure to find a constant substring means that there is no need to make |
| 573 | an expensive call to REx engine, thus we celebrate a failure. Similarly, |
| 574 | finding a substring too deep into the string means that fewer calls to |
| 575 | regtry() should be needed. |
| 576 | |
| 577 | REx compiler's optimizer found 4 possible hints: |
| 578 | a) Anchored substring; |
| 579 | b) Fixed substring; |
| 580 | c) Whether we are anchored (beginning-of-line or \G); |
| 581 | d) First node (of those at offset 0) which may distinguish positions; |
| 582 | We use a)b)d) and multiline-part of c), and try to find a position in the |
| 583 | string which does not contradict any of them. |
| 584 | */ |
| 585 | |
| 586 | /* Most of decisions we do here should have been done at compile time. |
| 587 | The nodes of the REx which we used for the search should have been |
| 588 | deleted from the finite automaton. */ |
| 589 | |
| 590 | char * |
| 591 | Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos, |
| 592 | char *strend, const U32 flags, re_scream_pos_data *data) |
| 593 | { |
| 594 | dVAR; |
| 595 | struct regexp *const prog = ReANY(rx); |
| 596 | I32 start_shift = 0; |
| 597 | /* Should be nonnegative! */ |
| 598 | I32 end_shift = 0; |
| 599 | char *s; |
| 600 | SV *check; |
| 601 | char *strbeg; |
| 602 | char *t; |
| 603 | const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */ |
| 604 | I32 ml_anch; |
| 605 | char *other_last = NULL; /* other substr checked before this */ |
| 606 | char *check_at = NULL; /* check substr found at this pos */ |
| 607 | char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/ |
| 608 | const I32 multiline = prog->extflags & RXf_PMf_MULTILINE; |
| 609 | RXi_GET_DECL(prog,progi); |
| 610 | bool is_utf8_pat; |
| 611 | #ifdef DEBUGGING |
| 612 | const char * const i_strpos = strpos; |
| 613 | #endif |
| 614 | GET_RE_DEBUG_FLAGS_DECL; |
| 615 | |
| 616 | PERL_ARGS_ASSERT_RE_INTUIT_START; |
| 617 | PERL_UNUSED_ARG(flags); |
| 618 | PERL_UNUSED_ARG(data); |
| 619 | |
| 620 | RX_MATCH_UTF8_set(rx,utf8_target); |
| 621 | |
| 622 | is_utf8_pat = cBOOL(RX_UTF8(rx)); |
| 623 | |
| 624 | DEBUG_EXECUTE_r( |
| 625 | debug_start_match(rx, utf8_target, strpos, strend, |
| 626 | sv ? "Guessing start of match in sv for" |
| 627 | : "Guessing start of match in string for"); |
| 628 | ); |
| 629 | |
| 630 | /* CHR_DIST() would be more correct here but it makes things slow. */ |
| 631 | if (prog->minlen > strend - strpos) { |
| 632 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 633 | "String too short... [re_intuit_start]\n")); |
| 634 | goto fail; |
| 635 | } |
| 636 | |
| 637 | /* XXX we need to pass strbeg as a separate arg: the following is |
| 638 | * guesswork and can be wrong... */ |
| 639 | if (sv && SvPOK(sv)) { |
| 640 | char * p = SvPVX(sv); |
| 641 | STRLEN cur = SvCUR(sv); |
| 642 | if (p <= strpos && strpos < p + cur) { |
| 643 | strbeg = p; |
| 644 | assert(p <= strend && strend <= p + cur); |
| 645 | } |
| 646 | else |
| 647 | strbeg = strend - cur; |
| 648 | } |
| 649 | else |
| 650 | strbeg = strpos; |
| 651 | |
| 652 | PL_regeol = strend; |
| 653 | if (utf8_target) { |
| 654 | if (!prog->check_utf8 && prog->check_substr) |
| 655 | to_utf8_substr(prog); |
| 656 | check = prog->check_utf8; |
| 657 | } else { |
| 658 | if (!prog->check_substr && prog->check_utf8) { |
| 659 | if (! to_byte_substr(prog)) { |
| 660 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail); |
| 661 | } |
| 662 | } |
| 663 | check = prog->check_substr; |
| 664 | } |
| 665 | if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */ |
| 666 | ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE) |
| 667 | || ( (prog->extflags & RXf_ANCH_BOL) |
| 668 | && !multiline ) ); /* Check after \n? */ |
| 669 | |
| 670 | if (!ml_anch) { |
| 671 | if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */ |
| 672 | && !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */ |
| 673 | /* SvCUR is not set on references: SvRV and SvPVX_const overlap */ |
| 674 | && sv && !SvROK(sv) |
| 675 | && (strpos != strbeg)) { |
| 676 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n")); |
| 677 | goto fail; |
| 678 | } |
| 679 | if (prog->check_offset_min == prog->check_offset_max |
| 680 | && !(prog->extflags & RXf_CANY_SEEN) |
| 681 | && ! multiline) /* /m can cause \n's to match that aren't |
| 682 | accounted for in the string max length. |
| 683 | See [perl #115242] */ |
| 684 | { |
| 685 | /* Substring at constant offset from beg-of-str... */ |
| 686 | I32 slen; |
| 687 | |
| 688 | s = HOP3c(strpos, prog->check_offset_min, strend); |
| 689 | |
| 690 | if (SvTAIL(check)) { |
| 691 | slen = SvCUR(check); /* >= 1 */ |
| 692 | |
| 693 | if ( strend - s > slen || strend - s < slen - 1 |
| 694 | || (strend - s == slen && strend[-1] != '\n')) { |
| 695 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n")); |
| 696 | goto fail_finish; |
| 697 | } |
| 698 | /* Now should match s[0..slen-2] */ |
| 699 | slen--; |
| 700 | if (slen && (*SvPVX_const(check) != *s |
| 701 | || (slen > 1 |
| 702 | && memNE(SvPVX_const(check), s, slen)))) { |
| 703 | report_neq: |
| 704 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n")); |
| 705 | goto fail_finish; |
| 706 | } |
| 707 | } |
| 708 | else if (*SvPVX_const(check) != *s |
| 709 | || ((slen = SvCUR(check)) > 1 |
| 710 | && memNE(SvPVX_const(check), s, slen))) |
| 711 | goto report_neq; |
| 712 | check_at = s; |
| 713 | goto success_at_start; |
| 714 | } |
| 715 | } |
| 716 | /* Match is anchored, but substr is not anchored wrt beg-of-str. */ |
| 717 | s = strpos; |
| 718 | start_shift = prog->check_offset_min; /* okay to underestimate on CC */ |
| 719 | end_shift = prog->check_end_shift; |
| 720 | |
| 721 | if (!ml_anch) { |
| 722 | const I32 end = prog->check_offset_max + CHR_SVLEN(check) |
| 723 | - (SvTAIL(check) != 0); |
| 724 | const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end; |
| 725 | |
| 726 | if (end_shift < eshift) |
| 727 | end_shift = eshift; |
| 728 | } |
| 729 | } |
| 730 | else { /* Can match at random position */ |
| 731 | ml_anch = 0; |
| 732 | s = strpos; |
| 733 | start_shift = prog->check_offset_min; /* okay to underestimate on CC */ |
| 734 | end_shift = prog->check_end_shift; |
| 735 | |
| 736 | /* end shift should be non negative here */ |
| 737 | } |
| 738 | |
| 739 | #ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */ |
| 740 | if (end_shift < 0) |
| 741 | Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ", |
| 742 | (IV)end_shift, RX_PRECOMP(prog)); |
| 743 | #endif |
| 744 | |
| 745 | restart: |
| 746 | /* Find a possible match in the region s..strend by looking for |
| 747 | the "check" substring in the region corrected by start/end_shift. */ |
| 748 | |
| 749 | { |
| 750 | I32 srch_start_shift = start_shift; |
| 751 | I32 srch_end_shift = end_shift; |
| 752 | U8* start_point; |
| 753 | U8* end_point; |
| 754 | if (srch_start_shift < 0 && strbeg - s > srch_start_shift) { |
| 755 | srch_end_shift -= ((strbeg - s) - srch_start_shift); |
| 756 | srch_start_shift = strbeg - s; |
| 757 | } |
| 758 | DEBUG_OPTIMISE_MORE_r({ |
| 759 | PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n", |
| 760 | (IV)prog->check_offset_min, |
| 761 | (IV)srch_start_shift, |
| 762 | (IV)srch_end_shift, |
| 763 | (IV)prog->check_end_shift); |
| 764 | }); |
| 765 | |
| 766 | if (prog->extflags & RXf_CANY_SEEN) { |
| 767 | start_point= (U8*)(s + srch_start_shift); |
| 768 | end_point= (U8*)(strend - srch_end_shift); |
| 769 | } else { |
| 770 | start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend); |
| 771 | end_point= HOP3(strend, -srch_end_shift, strbeg); |
| 772 | } |
| 773 | DEBUG_OPTIMISE_MORE_r({ |
| 774 | PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n", |
| 775 | (int)(end_point - start_point), |
| 776 | (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point), |
| 777 | start_point); |
| 778 | }); |
| 779 | |
| 780 | s = fbm_instr( start_point, end_point, |
| 781 | check, multiline ? FBMrf_MULTILINE : 0); |
| 782 | } |
| 783 | /* Update the count-of-usability, remove useless subpatterns, |
| 784 | unshift s. */ |
| 785 | |
| 786 | DEBUG_EXECUTE_r({ |
| 787 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), |
| 788 | SvPVX_const(check), RE_SV_DUMPLEN(check), 30); |
| 789 | PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s", |
| 790 | (s ? "Found" : "Did not find"), |
| 791 | (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) |
| 792 | ? "anchored" : "floating"), |
| 793 | quoted, |
| 794 | RE_SV_TAIL(check), |
| 795 | (s ? " at offset " : "...\n") ); |
| 796 | }); |
| 797 | |
| 798 | if (!s) |
| 799 | goto fail_finish; |
| 800 | /* Finish the diagnostic message */ |
| 801 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) ); |
| 802 | |
| 803 | /* XXX dmq: first branch is for positive lookbehind... |
| 804 | Our check string is offset from the beginning of the pattern. |
| 805 | So we need to do any stclass tests offset forward from that |
| 806 | point. I think. :-( |
| 807 | */ |
| 808 | |
| 809 | |
| 810 | |
| 811 | check_at=s; |
| 812 | |
| 813 | |
| 814 | /* Got a candidate. Check MBOL anchoring, and the *other* substr. |
| 815 | Start with the other substr. |
| 816 | XXXX no SCREAM optimization yet - and a very coarse implementation |
| 817 | XXXX /ttx+/ results in anchored="ttx", floating="x". floating will |
| 818 | *always* match. Probably should be marked during compile... |
| 819 | Probably it is right to do no SCREAM here... |
| 820 | */ |
| 821 | |
| 822 | if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8) |
| 823 | : (prog->float_substr && prog->anchored_substr)) |
| 824 | { |
| 825 | /* Take into account the "other" substring. */ |
| 826 | /* XXXX May be hopelessly wrong for UTF... */ |
| 827 | if (!other_last) |
| 828 | other_last = strpos; |
| 829 | if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) { |
| 830 | do_other_anchored: |
| 831 | { |
| 832 | char * const last = HOP3c(s, -start_shift, strbeg); |
| 833 | char *last1, *last2; |
| 834 | char * const saved_s = s; |
| 835 | SV* must; |
| 836 | |
| 837 | t = s - prog->check_offset_max; |
| 838 | if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */ |
| 839 | && (!utf8_target |
| 840 | || ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos)) |
| 841 | && t > strpos))) |
| 842 | NOOP; |
| 843 | else |
| 844 | t = strpos; |
| 845 | t = HOP3c(t, prog->anchored_offset, strend); |
| 846 | if (t < other_last) /* These positions already checked */ |
| 847 | t = other_last; |
| 848 | last2 = last1 = HOP3c(strend, -prog->minlen, strbeg); |
| 849 | if (last < last1) |
| 850 | last1 = last; |
| 851 | /* XXXX It is not documented what units *_offsets are in. |
| 852 | We assume bytes, but this is clearly wrong. |
| 853 | Meaning this code needs to be carefully reviewed for errors. |
| 854 | dmq. |
| 855 | */ |
| 856 | |
| 857 | /* On end-of-str: see comment below. */ |
| 858 | must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr; |
| 859 | if (must == &PL_sv_undef) { |
| 860 | s = (char*)NULL; |
| 861 | DEBUG_r(must = prog->anchored_utf8); /* for debug */ |
| 862 | } |
| 863 | else |
| 864 | s = fbm_instr( |
| 865 | (unsigned char*)t, |
| 866 | HOP3(HOP3(last1, prog->anchored_offset, strend) |
| 867 | + SvCUR(must), -(SvTAIL(must)!=0), strbeg), |
| 868 | must, |
| 869 | multiline ? FBMrf_MULTILINE : 0 |
| 870 | ); |
| 871 | DEBUG_EXECUTE_r({ |
| 872 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), |
| 873 | SvPVX_const(must), RE_SV_DUMPLEN(must), 30); |
| 874 | PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s", |
| 875 | (s ? "Found" : "Contradicts"), |
| 876 | quoted, RE_SV_TAIL(must)); |
| 877 | }); |
| 878 | |
| 879 | |
| 880 | if (!s) { |
| 881 | if (last1 >= last2) { |
| 882 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 883 | ", giving up...\n")); |
| 884 | goto fail_finish; |
| 885 | } |
| 886 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 887 | ", trying floating at offset %ld...\n", |
| 888 | (long)(HOP3c(saved_s, 1, strend) - i_strpos))); |
| 889 | other_last = HOP3c(last1, prog->anchored_offset+1, strend); |
| 890 | s = HOP3c(last, 1, strend); |
| 891 | goto restart; |
| 892 | } |
| 893 | else { |
| 894 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n", |
| 895 | (long)(s - i_strpos))); |
| 896 | t = HOP3c(s, -prog->anchored_offset, strbeg); |
| 897 | other_last = HOP3c(s, 1, strend); |
| 898 | s = saved_s; |
| 899 | if (t == strpos) |
| 900 | goto try_at_start; |
| 901 | goto try_at_offset; |
| 902 | } |
| 903 | } |
| 904 | } |
| 905 | else { /* Take into account the floating substring. */ |
| 906 | char *last, *last1; |
| 907 | char * const saved_s = s; |
| 908 | SV* must; |
| 909 | |
| 910 | t = HOP3c(s, -start_shift, strbeg); |
| 911 | last1 = last = |
| 912 | HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg); |
| 913 | if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset) |
| 914 | last = HOP3c(t, prog->float_max_offset, strend); |
| 915 | s = HOP3c(t, prog->float_min_offset, strend); |
| 916 | if (s < other_last) |
| 917 | s = other_last; |
| 918 | /* XXXX It is not documented what units *_offsets are in. Assume bytes. */ |
| 919 | must = utf8_target ? prog->float_utf8 : prog->float_substr; |
| 920 | /* fbm_instr() takes into account exact value of end-of-str |
| 921 | if the check is SvTAIL(ed). Since false positives are OK, |
| 922 | and end-of-str is not later than strend we are OK. */ |
| 923 | if (must == &PL_sv_undef) { |
| 924 | s = (char*)NULL; |
| 925 | DEBUG_r(must = prog->float_utf8); /* for debug message */ |
| 926 | } |
| 927 | else |
| 928 | s = fbm_instr((unsigned char*)s, |
| 929 | (unsigned char*)last + SvCUR(must) |
| 930 | - (SvTAIL(must)!=0), |
| 931 | must, multiline ? FBMrf_MULTILINE : 0); |
| 932 | DEBUG_EXECUTE_r({ |
| 933 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), |
| 934 | SvPVX_const(must), RE_SV_DUMPLEN(must), 30); |
| 935 | PerlIO_printf(Perl_debug_log, "%s floating substr %s%s", |
| 936 | (s ? "Found" : "Contradicts"), |
| 937 | quoted, RE_SV_TAIL(must)); |
| 938 | }); |
| 939 | if (!s) { |
| 940 | if (last1 == last) { |
| 941 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 942 | ", giving up...\n")); |
| 943 | goto fail_finish; |
| 944 | } |
| 945 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 946 | ", trying anchored starting at offset %ld...\n", |
| 947 | (long)(saved_s + 1 - i_strpos))); |
| 948 | other_last = last; |
| 949 | s = HOP3c(t, 1, strend); |
| 950 | goto restart; |
| 951 | } |
| 952 | else { |
| 953 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n", |
| 954 | (long)(s - i_strpos))); |
| 955 | other_last = s; /* Fix this later. --Hugo */ |
| 956 | s = saved_s; |
| 957 | if (t == strpos) |
| 958 | goto try_at_start; |
| 959 | goto try_at_offset; |
| 960 | } |
| 961 | } |
| 962 | } |
| 963 | |
| 964 | |
| 965 | t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos); |
| 966 | |
| 967 | DEBUG_OPTIMISE_MORE_r( |
| 968 | PerlIO_printf(Perl_debug_log, |
| 969 | "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n", |
| 970 | (IV)prog->check_offset_min, |
| 971 | (IV)prog->check_offset_max, |
| 972 | (IV)(s-strpos), |
| 973 | (IV)(t-strpos), |
| 974 | (IV)(t-s), |
| 975 | (IV)(strend-strpos) |
| 976 | ) |
| 977 | ); |
| 978 | |
| 979 | if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */ |
| 980 | && (!utf8_target |
| 981 | || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos))) |
| 982 | && t > strpos))) |
| 983 | { |
| 984 | /* Fixed substring is found far enough so that the match |
| 985 | cannot start at strpos. */ |
| 986 | try_at_offset: |
| 987 | if (ml_anch && t[-1] != '\n') { |
| 988 | /* Eventually fbm_*() should handle this, but often |
| 989 | anchored_offset is not 0, so this check will not be wasted. */ |
| 990 | /* XXXX In the code below we prefer to look for "^" even in |
| 991 | presence of anchored substrings. And we search even |
| 992 | beyond the found float position. These pessimizations |
| 993 | are historical artefacts only. */ |
| 994 | find_anchor: |
| 995 | while (t < strend - prog->minlen) { |
| 996 | if (*t == '\n') { |
| 997 | if (t < check_at - prog->check_offset_min) { |
| 998 | if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) { |
| 999 | /* Since we moved from the found position, |
| 1000 | we definitely contradict the found anchored |
| 1001 | substr. Due to the above check we do not |
| 1002 | contradict "check" substr. |
| 1003 | Thus we can arrive here only if check substr |
| 1004 | is float. Redo checking for "other"=="fixed". |
| 1005 | */ |
| 1006 | strpos = t + 1; |
| 1007 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n", |
| 1008 | PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset))); |
| 1009 | goto do_other_anchored; |
| 1010 | } |
| 1011 | /* We don't contradict the found floating substring. */ |
| 1012 | /* XXXX Why not check for STCLASS? */ |
| 1013 | s = t + 1; |
| 1014 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n", |
| 1015 | PL_colors[0], PL_colors[1], (long)(s - i_strpos))); |
| 1016 | goto set_useful; |
| 1017 | } |
| 1018 | /* Position contradicts check-string */ |
| 1019 | /* XXXX probably better to look for check-string |
| 1020 | than for "\n", so one should lower the limit for t? */ |
| 1021 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n", |
| 1022 | PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos))); |
| 1023 | other_last = strpos = s = t + 1; |
| 1024 | goto restart; |
| 1025 | } |
| 1026 | t++; |
| 1027 | } |
| 1028 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n", |
| 1029 | PL_colors[0], PL_colors[1])); |
| 1030 | goto fail_finish; |
| 1031 | } |
| 1032 | else { |
| 1033 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n", |
| 1034 | PL_colors[0], PL_colors[1])); |
| 1035 | } |
| 1036 | s = t; |
| 1037 | set_useful: |
| 1038 | ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */ |
| 1039 | } |
| 1040 | else { |
| 1041 | /* The found string does not prohibit matching at strpos, |
| 1042 | - no optimization of calling REx engine can be performed, |
| 1043 | unless it was an MBOL and we are not after MBOL, |
| 1044 | or a future STCLASS check will fail this. */ |
| 1045 | try_at_start: |
| 1046 | /* Even in this situation we may use MBOL flag if strpos is offset |
| 1047 | wrt the start of the string. */ |
| 1048 | if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */ |
| 1049 | && (strpos != strbeg) && strpos[-1] != '\n' |
| 1050 | /* May be due to an implicit anchor of m{.*foo} */ |
| 1051 | && !(prog->intflags & PREGf_IMPLICIT)) |
| 1052 | { |
| 1053 | t = strpos; |
| 1054 | goto find_anchor; |
| 1055 | } |
| 1056 | DEBUG_EXECUTE_r( if (ml_anch) |
| 1057 | PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n", |
| 1058 | (long)(strpos - i_strpos), PL_colors[0], PL_colors[1]); |
| 1059 | ); |
| 1060 | success_at_start: |
| 1061 | if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */ |
| 1062 | && (utf8_target ? ( |
| 1063 | prog->check_utf8 /* Could be deleted already */ |
| 1064 | && --BmUSEFUL(prog->check_utf8) < 0 |
| 1065 | && (prog->check_utf8 == prog->float_utf8) |
| 1066 | ) : ( |
| 1067 | prog->check_substr /* Could be deleted already */ |
| 1068 | && --BmUSEFUL(prog->check_substr) < 0 |
| 1069 | && (prog->check_substr == prog->float_substr) |
| 1070 | ))) |
| 1071 | { |
| 1072 | /* If flags & SOMETHING - do not do it many times on the same match */ |
| 1073 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n")); |
| 1074 | /* XXX Does the destruction order has to change with utf8_target? */ |
| 1075 | SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr); |
| 1076 | SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8); |
| 1077 | prog->check_substr = prog->check_utf8 = NULL; /* disable */ |
| 1078 | prog->float_substr = prog->float_utf8 = NULL; /* clear */ |
| 1079 | check = NULL; /* abort */ |
| 1080 | s = strpos; |
| 1081 | /* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag |
| 1082 | see http://bugs.activestate.com/show_bug.cgi?id=87173 */ |
| 1083 | if (prog->intflags & PREGf_IMPLICIT) |
| 1084 | prog->extflags &= ~RXf_ANCH_MBOL; |
| 1085 | /* XXXX This is a remnant of the old implementation. It |
| 1086 | looks wasteful, since now INTUIT can use many |
| 1087 | other heuristics. */ |
| 1088 | prog->extflags &= ~RXf_USE_INTUIT; |
| 1089 | /* XXXX What other flags might need to be cleared in this branch? */ |
| 1090 | } |
| 1091 | else |
| 1092 | s = strpos; |
| 1093 | } |
| 1094 | |
| 1095 | /* Last resort... */ |
| 1096 | /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */ |
| 1097 | /* trie stclasses are too expensive to use here, we are better off to |
| 1098 | leave it to regmatch itself */ |
| 1099 | if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) { |
| 1100 | /* minlen == 0 is possible if regstclass is \b or \B, |
| 1101 | and the fixed substr is ''$. |
| 1102 | Since minlen is already taken into account, s+1 is before strend; |
| 1103 | accidentally, minlen >= 1 guaranties no false positives at s + 1 |
| 1104 | even for \b or \B. But (minlen? 1 : 0) below assumes that |
| 1105 | regstclass does not come from lookahead... */ |
| 1106 | /* If regstclass takes bytelength more than 1: If charlength==1, OK. |
| 1107 | This leaves EXACTF-ish only, which are dealt with in find_byclass(). */ |
| 1108 | const U8* const str = (U8*)STRING(progi->regstclass); |
| 1109 | const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT |
| 1110 | ? CHR_DIST(str+STR_LEN(progi->regstclass), str) |
| 1111 | : 1); |
| 1112 | char * endpos; |
| 1113 | if (prog->anchored_substr || prog->anchored_utf8 || ml_anch) |
| 1114 | endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend); |
| 1115 | else if (prog->float_substr || prog->float_utf8) |
| 1116 | endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend); |
| 1117 | else |
| 1118 | endpos= strend; |
| 1119 | |
| 1120 | if (checked_upto < s) |
| 1121 | checked_upto = s; |
| 1122 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n", |
| 1123 | (IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg))); |
| 1124 | |
| 1125 | t = s; |
| 1126 | s = find_byclass(prog, progi->regstclass, checked_upto, endpos, |
| 1127 | NULL, is_utf8_pat); |
| 1128 | if (s) { |
| 1129 | checked_upto = s; |
| 1130 | } else { |
| 1131 | #ifdef DEBUGGING |
| 1132 | const char *what = NULL; |
| 1133 | #endif |
| 1134 | if (endpos == strend) { |
| 1135 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 1136 | "Could not match STCLASS...\n") ); |
| 1137 | goto fail; |
| 1138 | } |
| 1139 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 1140 | "This position contradicts STCLASS...\n") ); |
| 1141 | if ((prog->extflags & RXf_ANCH) && !ml_anch) |
| 1142 | goto fail; |
| 1143 | checked_upto = HOPBACKc(endpos, start_shift); |
| 1144 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n", |
| 1145 | (IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg))); |
| 1146 | /* Contradict one of substrings */ |
| 1147 | if (prog->anchored_substr || prog->anchored_utf8) { |
| 1148 | if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) { |
| 1149 | DEBUG_EXECUTE_r( what = "anchored" ); |
| 1150 | hop_and_restart: |
| 1151 | s = HOP3c(t, 1, strend); |
| 1152 | if (s + start_shift + end_shift > strend) { |
| 1153 | /* XXXX Should be taken into account earlier? */ |
| 1154 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 1155 | "Could not match STCLASS...\n") ); |
| 1156 | goto fail; |
| 1157 | } |
| 1158 | if (!check) |
| 1159 | goto giveup; |
| 1160 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 1161 | "Looking for %s substr starting at offset %ld...\n", |
| 1162 | what, (long)(s + start_shift - i_strpos)) ); |
| 1163 | goto restart; |
| 1164 | } |
| 1165 | /* Have both, check_string is floating */ |
| 1166 | if (t + start_shift >= check_at) /* Contradicts floating=check */ |
| 1167 | goto retry_floating_check; |
| 1168 | /* Recheck anchored substring, but not floating... */ |
| 1169 | s = check_at; |
| 1170 | if (!check) |
| 1171 | goto giveup; |
| 1172 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 1173 | "Looking for anchored substr starting at offset %ld...\n", |
| 1174 | (long)(other_last - i_strpos)) ); |
| 1175 | goto do_other_anchored; |
| 1176 | } |
| 1177 | /* Another way we could have checked stclass at the |
| 1178 | current position only: */ |
| 1179 | if (ml_anch) { |
| 1180 | s = t = t + 1; |
| 1181 | if (!check) |
| 1182 | goto giveup; |
| 1183 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 1184 | "Looking for /%s^%s/m starting at offset %ld...\n", |
| 1185 | PL_colors[0], PL_colors[1], (long)(t - i_strpos)) ); |
| 1186 | goto try_at_offset; |
| 1187 | } |
| 1188 | if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */ |
| 1189 | goto fail; |
| 1190 | /* Check is floating substring. */ |
| 1191 | retry_floating_check: |
| 1192 | t = check_at - start_shift; |
| 1193 | DEBUG_EXECUTE_r( what = "floating" ); |
| 1194 | goto hop_and_restart; |
| 1195 | } |
| 1196 | if (t != s) { |
| 1197 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 1198 | "By STCLASS: moving %ld --> %ld\n", |
| 1199 | (long)(t - i_strpos), (long)(s - i_strpos)) |
| 1200 | ); |
| 1201 | } |
| 1202 | else { |
| 1203 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 1204 | "Does not contradict STCLASS...\n"); |
| 1205 | ); |
| 1206 | } |
| 1207 | } |
| 1208 | giveup: |
| 1209 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n", |
| 1210 | PL_colors[4], (check ? "Guessed" : "Giving up"), |
| 1211 | PL_colors[5], (long)(s - i_strpos)) ); |
| 1212 | return s; |
| 1213 | |
| 1214 | fail_finish: /* Substring not found */ |
| 1215 | if (prog->check_substr || prog->check_utf8) /* could be removed already */ |
| 1216 | BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */ |
| 1217 | fail: |
| 1218 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n", |
| 1219 | PL_colors[4], PL_colors[5])); |
| 1220 | return NULL; |
| 1221 | } |
| 1222 | |
| 1223 | #define DECL_TRIE_TYPE(scan) \ |
| 1224 | const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \ |
| 1225 | trie_type = ((scan->flags == EXACT) \ |
| 1226 | ? (utf8_target ? trie_utf8 : trie_plain) \ |
| 1227 | : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold)) |
| 1228 | |
| 1229 | #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \ |
| 1230 | STMT_START { \ |
| 1231 | STRLEN skiplen; \ |
| 1232 | switch (trie_type) { \ |
| 1233 | case trie_utf8_fold: \ |
| 1234 | if ( foldlen>0 ) { \ |
| 1235 | uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \ |
| 1236 | foldlen -= len; \ |
| 1237 | uscan += len; \ |
| 1238 | len=0; \ |
| 1239 | } else { \ |
| 1240 | uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \ |
| 1241 | len = UTF8SKIP(uc); \ |
| 1242 | skiplen = UNISKIP( uvc ); \ |
| 1243 | foldlen -= skiplen; \ |
| 1244 | uscan = foldbuf + skiplen; \ |
| 1245 | } \ |
| 1246 | break; \ |
| 1247 | case trie_latin_utf8_fold: \ |
| 1248 | if ( foldlen>0 ) { \ |
| 1249 | uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \ |
| 1250 | foldlen -= len; \ |
| 1251 | uscan += len; \ |
| 1252 | len=0; \ |
| 1253 | } else { \ |
| 1254 | len = 1; \ |
| 1255 | uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \ |
| 1256 | skiplen = UNISKIP( uvc ); \ |
| 1257 | foldlen -= skiplen; \ |
| 1258 | uscan = foldbuf + skiplen; \ |
| 1259 | } \ |
| 1260 | break; \ |
| 1261 | case trie_utf8: \ |
| 1262 | uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \ |
| 1263 | break; \ |
| 1264 | case trie_plain: \ |
| 1265 | uvc = (UV)*uc; \ |
| 1266 | len = 1; \ |
| 1267 | } \ |
| 1268 | if (uvc < 256) { \ |
| 1269 | charid = trie->charmap[ uvc ]; \ |
| 1270 | } \ |
| 1271 | else { \ |
| 1272 | charid = 0; \ |
| 1273 | if (widecharmap) { \ |
| 1274 | SV** const svpp = hv_fetch(widecharmap, \ |
| 1275 | (char*)&uvc, sizeof(UV), 0); \ |
| 1276 | if (svpp) \ |
| 1277 | charid = (U16)SvIV(*svpp); \ |
| 1278 | } \ |
| 1279 | } \ |
| 1280 | } STMT_END |
| 1281 | |
| 1282 | #define REXEC_FBC_EXACTISH_SCAN(CoNd) \ |
| 1283 | STMT_START { \ |
| 1284 | while (s <= e) { \ |
| 1285 | if ( (CoNd) \ |
| 1286 | && (ln == 1 || folder(s, pat_string, ln)) \ |
| 1287 | && (!reginfo || regtry(reginfo, &s)) ) \ |
| 1288 | goto got_it; \ |
| 1289 | s++; \ |
| 1290 | } \ |
| 1291 | } STMT_END |
| 1292 | |
| 1293 | #define REXEC_FBC_UTF8_SCAN(CoDe) \ |
| 1294 | STMT_START { \ |
| 1295 | while (s < strend) { \ |
| 1296 | CoDe \ |
| 1297 | s += UTF8SKIP(s); \ |
| 1298 | } \ |
| 1299 | } STMT_END |
| 1300 | |
| 1301 | #define REXEC_FBC_SCAN(CoDe) \ |
| 1302 | STMT_START { \ |
| 1303 | while (s < strend) { \ |
| 1304 | CoDe \ |
| 1305 | s++; \ |
| 1306 | } \ |
| 1307 | } STMT_END |
| 1308 | |
| 1309 | #define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \ |
| 1310 | REXEC_FBC_UTF8_SCAN( \ |
| 1311 | if (CoNd) { \ |
| 1312 | if (tmp && (!reginfo || regtry(reginfo, &s))) \ |
| 1313 | goto got_it; \ |
| 1314 | else \ |
| 1315 | tmp = doevery; \ |
| 1316 | } \ |
| 1317 | else \ |
| 1318 | tmp = 1; \ |
| 1319 | ) |
| 1320 | |
| 1321 | #define REXEC_FBC_CLASS_SCAN(CoNd) \ |
| 1322 | REXEC_FBC_SCAN( \ |
| 1323 | if (CoNd) { \ |
| 1324 | if (tmp && (!reginfo || regtry(reginfo, &s))) \ |
| 1325 | goto got_it; \ |
| 1326 | else \ |
| 1327 | tmp = doevery; \ |
| 1328 | } \ |
| 1329 | else \ |
| 1330 | tmp = 1; \ |
| 1331 | ) |
| 1332 | |
| 1333 | #define REXEC_FBC_TRYIT \ |
| 1334 | if ((!reginfo || regtry(reginfo, &s))) \ |
| 1335 | goto got_it |
| 1336 | |
| 1337 | #define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \ |
| 1338 | if (utf8_target) { \ |
| 1339 | REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \ |
| 1340 | } \ |
| 1341 | else { \ |
| 1342 | REXEC_FBC_CLASS_SCAN(CoNd); \ |
| 1343 | } |
| 1344 | |
| 1345 | #define DUMP_EXEC_POS(li,s,doutf8) \ |
| 1346 | dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8) |
| 1347 | |
| 1348 | |
| 1349 | #define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \ |
| 1350 | tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \ |
| 1351 | tmp = TEST_NON_UTF8(tmp); \ |
| 1352 | REXEC_FBC_UTF8_SCAN( \ |
| 1353 | if (tmp == ! TEST_NON_UTF8((U8) *s)) { \ |
| 1354 | tmp = !tmp; \ |
| 1355 | IF_SUCCESS; \ |
| 1356 | } \ |
| 1357 | else { \ |
| 1358 | IF_FAIL; \ |
| 1359 | } \ |
| 1360 | ); \ |
| 1361 | |
| 1362 | #define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \ |
| 1363 | if (s == PL_bostr) { \ |
| 1364 | tmp = '\n'; \ |
| 1365 | } \ |
| 1366 | else { \ |
| 1367 | U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \ |
| 1368 | tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \ |
| 1369 | } \ |
| 1370 | tmp = TeSt1_UtF8; \ |
| 1371 | LOAD_UTF8_CHARCLASS_ALNUM(); \ |
| 1372 | REXEC_FBC_UTF8_SCAN( \ |
| 1373 | if (tmp == ! (TeSt2_UtF8)) { \ |
| 1374 | tmp = !tmp; \ |
| 1375 | IF_SUCCESS; \ |
| 1376 | } \ |
| 1377 | else { \ |
| 1378 | IF_FAIL; \ |
| 1379 | } \ |
| 1380 | ); \ |
| 1381 | |
| 1382 | /* The only difference between the BOUND and NBOUND cases is that |
| 1383 | * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in |
| 1384 | * NBOUND. This is accomplished by passing it in either the if or else clause, |
| 1385 | * with the other one being empty */ |
| 1386 | #define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \ |
| 1387 | FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER) |
| 1388 | |
| 1389 | #define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \ |
| 1390 | FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER) |
| 1391 | |
| 1392 | #define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \ |
| 1393 | FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT) |
| 1394 | |
| 1395 | #define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \ |
| 1396 | FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT) |
| 1397 | |
| 1398 | |
| 1399 | /* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to |
| 1400 | * be passed in completely with the variable name being tested, which isn't |
| 1401 | * such a clean interface, but this is easier to read than it was before. We |
| 1402 | * are looking for the boundary (or non-boundary between a word and non-word |
| 1403 | * character. The utf8 and non-utf8 cases have the same logic, but the details |
| 1404 | * must be different. Find the "wordness" of the character just prior to this |
| 1405 | * one, and compare it with the wordness of this one. If they differ, we have |
| 1406 | * a boundary. At the beginning of the string, pretend that the previous |
| 1407 | * character was a new-line */ |
| 1408 | #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \ |
| 1409 | if (utf8_target) { \ |
| 1410 | UTF8_CODE \ |
| 1411 | } \ |
| 1412 | else { /* Not utf8 */ \ |
| 1413 | tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \ |
| 1414 | tmp = TEST_NON_UTF8(tmp); \ |
| 1415 | REXEC_FBC_SCAN( \ |
| 1416 | if (tmp == ! TEST_NON_UTF8((U8) *s)) { \ |
| 1417 | tmp = !tmp; \ |
| 1418 | IF_SUCCESS; \ |
| 1419 | } \ |
| 1420 | else { \ |
| 1421 | IF_FAIL; \ |
| 1422 | } \ |
| 1423 | ); \ |
| 1424 | } \ |
| 1425 | if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \ |
| 1426 | goto got_it; |
| 1427 | |
| 1428 | /* We know what class REx starts with. Try to find this position... */ |
| 1429 | /* if reginfo is NULL, its a dryrun */ |
| 1430 | /* annoyingly all the vars in this routine have different names from their counterparts |
| 1431 | in regmatch. /grrr */ |
| 1432 | |
| 1433 | STATIC char * |
| 1434 | S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, |
| 1435 | const char *strend, regmatch_info *reginfo, bool is_utf8_pat) |
| 1436 | { |
| 1437 | dVAR; |
| 1438 | const I32 doevery = (prog->intflags & PREGf_SKIP) == 0; |
| 1439 | char *pat_string; /* The pattern's exactish string */ |
| 1440 | char *pat_end; /* ptr to end char of pat_string */ |
| 1441 | re_fold_t folder; /* Function for computing non-utf8 folds */ |
| 1442 | const U8 *fold_array; /* array for folding ords < 256 */ |
| 1443 | STRLEN ln; |
| 1444 | STRLEN lnc; |
| 1445 | U8 c1; |
| 1446 | U8 c2; |
| 1447 | char *e; |
| 1448 | I32 tmp = 1; /* Scratch variable? */ |
| 1449 | const bool utf8_target = PL_reg_match_utf8; |
| 1450 | UV utf8_fold_flags = 0; |
| 1451 | bool to_complement = FALSE; /* Invert the result? Taking the xor of this |
| 1452 | with a result inverts that result, as 0^1 = |
| 1453 | 1 and 1^1 = 0 */ |
| 1454 | _char_class_number classnum; |
| 1455 | |
| 1456 | RXi_GET_DECL(prog,progi); |
| 1457 | |
| 1458 | PERL_ARGS_ASSERT_FIND_BYCLASS; |
| 1459 | |
| 1460 | /* We know what class it must start with. */ |
| 1461 | switch (OP(c)) { |
| 1462 | case ANYOF: |
| 1463 | case ANYOF_SYNTHETIC: |
| 1464 | case ANYOF_WARN_SUPER: |
| 1465 | if (utf8_target) { |
| 1466 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 1467 | reginclass(prog, c, (U8*)s, utf8_target)); |
| 1468 | } |
| 1469 | else { |
| 1470 | REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s)); |
| 1471 | } |
| 1472 | break; |
| 1473 | case CANY: |
| 1474 | REXEC_FBC_SCAN( |
| 1475 | if (tmp && (!reginfo || regtry(reginfo, &s))) |
| 1476 | goto got_it; |
| 1477 | else |
| 1478 | tmp = doevery; |
| 1479 | ); |
| 1480 | break; |
| 1481 | |
| 1482 | case EXACTFA: |
| 1483 | if (is_utf8_pat || utf8_target) { |
| 1484 | utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 1485 | goto do_exactf_utf8; |
| 1486 | } |
| 1487 | fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */ |
| 1488 | folder = foldEQ_latin1; /* /a, except the sharp s one which */ |
| 1489 | goto do_exactf_non_utf8; /* isn't dealt with by these */ |
| 1490 | |
| 1491 | case EXACTF: |
| 1492 | if (utf8_target) { |
| 1493 | |
| 1494 | /* regcomp.c already folded this if pattern is in UTF-8 */ |
| 1495 | utf8_fold_flags = 0; |
| 1496 | goto do_exactf_utf8; |
| 1497 | } |
| 1498 | fold_array = PL_fold; |
| 1499 | folder = foldEQ; |
| 1500 | goto do_exactf_non_utf8; |
| 1501 | |
| 1502 | case EXACTFL: |
| 1503 | if (is_utf8_pat || utf8_target) { |
| 1504 | utf8_fold_flags = FOLDEQ_UTF8_LOCALE; |
| 1505 | goto do_exactf_utf8; |
| 1506 | } |
| 1507 | fold_array = PL_fold_locale; |
| 1508 | folder = foldEQ_locale; |
| 1509 | goto do_exactf_non_utf8; |
| 1510 | |
| 1511 | case EXACTFU_SS: |
| 1512 | if (is_utf8_pat) { |
| 1513 | utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED; |
| 1514 | } |
| 1515 | goto do_exactf_utf8; |
| 1516 | |
| 1517 | case EXACTFU_TRICKYFOLD: |
| 1518 | case EXACTFU: |
| 1519 | if (is_utf8_pat || utf8_target) { |
| 1520 | utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0; |
| 1521 | goto do_exactf_utf8; |
| 1522 | } |
| 1523 | |
| 1524 | /* Any 'ss' in the pattern should have been replaced by regcomp, |
| 1525 | * so we don't have to worry here about this single special case |
| 1526 | * in the Latin1 range */ |
| 1527 | fold_array = PL_fold_latin1; |
| 1528 | folder = foldEQ_latin1; |
| 1529 | |
| 1530 | /* FALL THROUGH */ |
| 1531 | |
| 1532 | do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there |
| 1533 | are no glitches with fold-length differences |
| 1534 | between the target string and pattern */ |
| 1535 | |
| 1536 | /* The idea in the non-utf8 EXACTF* cases is to first find the |
| 1537 | * first character of the EXACTF* node and then, if necessary, |
| 1538 | * case-insensitively compare the full text of the node. c1 is the |
| 1539 | * first character. c2 is its fold. This logic will not work for |
| 1540 | * Unicode semantics and the german sharp ss, which hence should |
| 1541 | * not be compiled into a node that gets here. */ |
| 1542 | pat_string = STRING(c); |
| 1543 | ln = STR_LEN(c); /* length to match in octets/bytes */ |
| 1544 | |
| 1545 | /* We know that we have to match at least 'ln' bytes (which is the |
| 1546 | * same as characters, since not utf8). If we have to match 3 |
| 1547 | * characters, and there are only 2 availabe, we know without |
| 1548 | * trying that it will fail; so don't start a match past the |
| 1549 | * required minimum number from the far end */ |
| 1550 | e = HOP3c(strend, -((I32)ln), s); |
| 1551 | |
| 1552 | if (!reginfo && e < s) { |
| 1553 | e = s; /* Due to minlen logic of intuit() */ |
| 1554 | } |
| 1555 | |
| 1556 | c1 = *pat_string; |
| 1557 | c2 = fold_array[c1]; |
| 1558 | if (c1 == c2) { /* If char and fold are the same */ |
| 1559 | REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1); |
| 1560 | } |
| 1561 | else { |
| 1562 | REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2); |
| 1563 | } |
| 1564 | break; |
| 1565 | |
| 1566 | do_exactf_utf8: |
| 1567 | { |
| 1568 | unsigned expansion; |
| 1569 | |
| 1570 | /* If one of the operands is in utf8, we can't use the simpler folding |
| 1571 | * above, due to the fact that many different characters can have the |
| 1572 | * same fold, or portion of a fold, or different- length fold */ |
| 1573 | pat_string = STRING(c); |
| 1574 | ln = STR_LEN(c); /* length to match in octets/bytes */ |
| 1575 | pat_end = pat_string + ln; |
| 1576 | lnc = is_utf8_pat /* length to match in characters */ |
| 1577 | ? utf8_length((U8 *) pat_string, (U8 *) pat_end) |
| 1578 | : ln; |
| 1579 | |
| 1580 | /* We have 'lnc' characters to match in the pattern, but because of |
| 1581 | * multi-character folding, each character in the target can match |
| 1582 | * up to 3 characters (Unicode guarantees it will never exceed |
| 1583 | * this) if it is utf8-encoded; and up to 2 if not (based on the |
| 1584 | * fact that the Latin 1 folds are already determined, and the |
| 1585 | * only multi-char fold in that range is the sharp-s folding to |
| 1586 | * 'ss'. Thus, a pattern character can match as little as 1/3 of a |
| 1587 | * string character. Adjust lnc accordingly, rounding up, so that |
| 1588 | * if we need to match at least 4+1/3 chars, that really is 5. */ |
| 1589 | expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2; |
| 1590 | lnc = (lnc + expansion - 1) / expansion; |
| 1591 | |
| 1592 | /* As in the non-UTF8 case, if we have to match 3 characters, and |
| 1593 | * only 2 are left, it's guaranteed to fail, so don't start a |
| 1594 | * match that would require us to go beyond the end of the string |
| 1595 | */ |
| 1596 | e = HOP3c(strend, -((I32)lnc), s); |
| 1597 | |
| 1598 | if (!reginfo && e < s) { |
| 1599 | e = s; /* Due to minlen logic of intuit() */ |
| 1600 | } |
| 1601 | |
| 1602 | /* XXX Note that we could recalculate e to stop the loop earlier, |
| 1603 | * as the worst case expansion above will rarely be met, and as we |
| 1604 | * go along we would usually find that e moves further to the left. |
| 1605 | * This would happen only after we reached the point in the loop |
| 1606 | * where if there were no expansion we should fail. Unclear if |
| 1607 | * worth the expense */ |
| 1608 | |
| 1609 | while (s <= e) { |
| 1610 | char *my_strend= (char *)strend; |
| 1611 | if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target, |
| 1612 | pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags) |
| 1613 | && (!reginfo || regtry(reginfo, &s)) ) |
| 1614 | { |
| 1615 | goto got_it; |
| 1616 | } |
| 1617 | s += (utf8_target) ? UTF8SKIP(s) : 1; |
| 1618 | } |
| 1619 | break; |
| 1620 | } |
| 1621 | case BOUNDL: |
| 1622 | RXp_MATCH_TAINTED_on(prog); |
| 1623 | FBC_BOUND(isWORDCHAR_LC, |
| 1624 | isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(tmp)), |
| 1625 | isWORDCHAR_LC_utf8((U8*)s)); |
| 1626 | break; |
| 1627 | case NBOUNDL: |
| 1628 | RXp_MATCH_TAINTED_on(prog); |
| 1629 | FBC_NBOUND(isWORDCHAR_LC, |
| 1630 | isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(tmp)), |
| 1631 | isWORDCHAR_LC_utf8((U8*)s)); |
| 1632 | break; |
| 1633 | case BOUND: |
| 1634 | FBC_BOUND(isWORDCHAR, |
| 1635 | isWORDCHAR_uni(tmp), |
| 1636 | cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target))); |
| 1637 | break; |
| 1638 | case BOUNDA: |
| 1639 | FBC_BOUND_NOLOAD(isWORDCHAR_A, |
| 1640 | isWORDCHAR_A(tmp), |
| 1641 | isWORDCHAR_A((U8*)s)); |
| 1642 | break; |
| 1643 | case NBOUND: |
| 1644 | FBC_NBOUND(isWORDCHAR, |
| 1645 | isWORDCHAR_uni(tmp), |
| 1646 | cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target))); |
| 1647 | break; |
| 1648 | case NBOUNDA: |
| 1649 | FBC_NBOUND_NOLOAD(isWORDCHAR_A, |
| 1650 | isWORDCHAR_A(tmp), |
| 1651 | isWORDCHAR_A((U8*)s)); |
| 1652 | break; |
| 1653 | case BOUNDU: |
| 1654 | FBC_BOUND(isWORDCHAR_L1, |
| 1655 | isWORDCHAR_uni(tmp), |
| 1656 | cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target))); |
| 1657 | break; |
| 1658 | case NBOUNDU: |
| 1659 | FBC_NBOUND(isWORDCHAR_L1, |
| 1660 | isWORDCHAR_uni(tmp), |
| 1661 | cBOOL(swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)s, utf8_target))); |
| 1662 | break; |
| 1663 | case LNBREAK: |
| 1664 | REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend), |
| 1665 | is_LNBREAK_latin1_safe(s, strend) |
| 1666 | ); |
| 1667 | break; |
| 1668 | |
| 1669 | /* The argument to all the POSIX node types is the class number to pass to |
| 1670 | * _generic_isCC() to build a mask for searching in PL_charclass[] */ |
| 1671 | |
| 1672 | case NPOSIXL: |
| 1673 | to_complement = 1; |
| 1674 | /* FALLTHROUGH */ |
| 1675 | |
| 1676 | case POSIXL: |
| 1677 | RXp_MATCH_TAINTED_on(prog); |
| 1678 | REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)), |
| 1679 | to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s))); |
| 1680 | break; |
| 1681 | |
| 1682 | case NPOSIXD: |
| 1683 | to_complement = 1; |
| 1684 | /* FALLTHROUGH */ |
| 1685 | |
| 1686 | case POSIXD: |
| 1687 | if (utf8_target) { |
| 1688 | goto posix_utf8; |
| 1689 | } |
| 1690 | goto posixa; |
| 1691 | |
| 1692 | case NPOSIXA: |
| 1693 | if (utf8_target) { |
| 1694 | /* The complement of something that matches only ASCII matches all |
| 1695 | * UTF-8 variant code points, plus everything in ASCII that isn't |
| 1696 | * in the class */ |
| 1697 | REXEC_FBC_UTF8_CLASS_SCAN(! UTF8_IS_INVARIANT(*s) |
| 1698 | || ! _generic_isCC_A(*s, FLAGS(c))); |
| 1699 | break; |
| 1700 | } |
| 1701 | |
| 1702 | to_complement = 1; |
| 1703 | /* FALLTHROUGH */ |
| 1704 | |
| 1705 | case POSIXA: |
| 1706 | posixa: |
| 1707 | /* Don't need to worry about utf8, as it can match only a single |
| 1708 | * byte invariant character. */ |
| 1709 | REXEC_FBC_CLASS_SCAN( |
| 1710 | to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c)))); |
| 1711 | break; |
| 1712 | |
| 1713 | case NPOSIXU: |
| 1714 | to_complement = 1; |
| 1715 | /* FALLTHROUGH */ |
| 1716 | |
| 1717 | case POSIXU: |
| 1718 | if (! utf8_target) { |
| 1719 | REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s, |
| 1720 | FLAGS(c)))); |
| 1721 | } |
| 1722 | else { |
| 1723 | |
| 1724 | posix_utf8: |
| 1725 | classnum = (_char_class_number) FLAGS(c); |
| 1726 | if (classnum < _FIRST_NON_SWASH_CC) { |
| 1727 | while (s < strend) { |
| 1728 | |
| 1729 | /* We avoid loading in the swash as long as possible, but |
| 1730 | * should we have to, we jump to a separate loop. This |
| 1731 | * extra 'if' statement is what keeps this code from being |
| 1732 | * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */ |
| 1733 | if (UTF8_IS_ABOVE_LATIN1(*s)) { |
| 1734 | goto found_above_latin1; |
| 1735 | } |
| 1736 | if ((UTF8_IS_INVARIANT(*s) |
| 1737 | && to_complement ^ cBOOL(_generic_isCC((U8) *s, |
| 1738 | classnum))) |
| 1739 | || (UTF8_IS_DOWNGRADEABLE_START(*s) |
| 1740 | && to_complement ^ cBOOL( |
| 1741 | _generic_isCC(TWO_BYTE_UTF8_TO_UNI(*s, *(s + 1)), |
| 1742 | classnum)))) |
| 1743 | { |
| 1744 | if (tmp && (!reginfo || regtry(reginfo, &s))) |
| 1745 | goto got_it; |
| 1746 | else { |
| 1747 | tmp = doevery; |
| 1748 | } |
| 1749 | } |
| 1750 | else { |
| 1751 | tmp = 1; |
| 1752 | } |
| 1753 | s += UTF8SKIP(s); |
| 1754 | } |
| 1755 | } |
| 1756 | else switch (classnum) { /* These classes are implemented as |
| 1757 | macros */ |
| 1758 | case _CC_ENUM_SPACE: /* XXX would require separate code if we |
| 1759 | revert the change of \v matching this */ |
| 1760 | /* FALL THROUGH */ |
| 1761 | |
| 1762 | case _CC_ENUM_PSXSPC: |
| 1763 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 1764 | to_complement ^ cBOOL(isSPACE_utf8(s))); |
| 1765 | break; |
| 1766 | |
| 1767 | case _CC_ENUM_BLANK: |
| 1768 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 1769 | to_complement ^ cBOOL(isBLANK_utf8(s))); |
| 1770 | break; |
| 1771 | |
| 1772 | case _CC_ENUM_XDIGIT: |
| 1773 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 1774 | to_complement ^ cBOOL(isXDIGIT_utf8(s))); |
| 1775 | break; |
| 1776 | |
| 1777 | case _CC_ENUM_VERTSPACE: |
| 1778 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 1779 | to_complement ^ cBOOL(isVERTWS_utf8(s))); |
| 1780 | break; |
| 1781 | |
| 1782 | case _CC_ENUM_CNTRL: |
| 1783 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 1784 | to_complement ^ cBOOL(isCNTRL_utf8(s))); |
| 1785 | break; |
| 1786 | |
| 1787 | default: |
| 1788 | Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum); |
| 1789 | assert(0); /* NOTREACHED */ |
| 1790 | } |
| 1791 | } |
| 1792 | break; |
| 1793 | |
| 1794 | found_above_latin1: /* Here we have to load a swash to get the result |
| 1795 | for the current code point */ |
| 1796 | if (! PL_utf8_swash_ptrs[classnum]) { |
| 1797 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; |
| 1798 | PL_utf8_swash_ptrs[classnum] = |
| 1799 | _core_swash_init("utf8", swash_property_names[classnum], |
| 1800 | &PL_sv_undef, 1, 0, NULL, &flags); |
| 1801 | } |
| 1802 | |
| 1803 | /* This is a copy of the loop above for swash classes, though using the |
| 1804 | * FBC macro instead of being expanded out. Since we've loaded the |
| 1805 | * swash, we don't have to check for that each time through the loop */ |
| 1806 | REXEC_FBC_UTF8_CLASS_SCAN( |
| 1807 | to_complement ^ cBOOL(_generic_utf8( |
| 1808 | classnum, |
| 1809 | s, |
| 1810 | swash_fetch(PL_utf8_swash_ptrs[classnum], |
| 1811 | (U8 *) s, TRUE)))); |
| 1812 | break; |
| 1813 | |
| 1814 | case AHOCORASICKC: |
| 1815 | case AHOCORASICK: |
| 1816 | { |
| 1817 | DECL_TRIE_TYPE(c); |
| 1818 | /* what trie are we using right now */ |
| 1819 | reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ]; |
| 1820 | reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ]; |
| 1821 | HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]); |
| 1822 | |
| 1823 | const char *last_start = strend - trie->minlen; |
| 1824 | #ifdef DEBUGGING |
| 1825 | const char *real_start = s; |
| 1826 | #endif |
| 1827 | STRLEN maxlen = trie->maxlen; |
| 1828 | SV *sv_points; |
| 1829 | U8 **points; /* map of where we were in the input string |
| 1830 | when reading a given char. For ASCII this |
| 1831 | is unnecessary overhead as the relationship |
| 1832 | is always 1:1, but for Unicode, especially |
| 1833 | case folded Unicode this is not true. */ |
| 1834 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
| 1835 | U8 *bitmap=NULL; |
| 1836 | |
| 1837 | |
| 1838 | GET_RE_DEBUG_FLAGS_DECL; |
| 1839 | |
| 1840 | /* We can't just allocate points here. We need to wrap it in |
| 1841 | * an SV so it gets freed properly if there is a croak while |
| 1842 | * running the match */ |
| 1843 | ENTER; |
| 1844 | SAVETMPS; |
| 1845 | sv_points=newSV(maxlen * sizeof(U8 *)); |
| 1846 | SvCUR_set(sv_points, |
| 1847 | maxlen * sizeof(U8 *)); |
| 1848 | SvPOK_on(sv_points); |
| 1849 | sv_2mortal(sv_points); |
| 1850 | points=(U8**)SvPV_nolen(sv_points ); |
| 1851 | if ( trie_type != trie_utf8_fold |
| 1852 | && (trie->bitmap || OP(c)==AHOCORASICKC) ) |
| 1853 | { |
| 1854 | if (trie->bitmap) |
| 1855 | bitmap=(U8*)trie->bitmap; |
| 1856 | else |
| 1857 | bitmap=(U8*)ANYOF_BITMAP(c); |
| 1858 | } |
| 1859 | /* this is the Aho-Corasick algorithm modified a touch |
| 1860 | to include special handling for long "unknown char" sequences. |
| 1861 | The basic idea being that we use AC as long as we are dealing |
| 1862 | with a possible matching char, when we encounter an unknown char |
| 1863 | (and we have not encountered an accepting state) we scan forward |
| 1864 | until we find a legal starting char. |
| 1865 | AC matching is basically that of trie matching, except that when |
| 1866 | we encounter a failing transition, we fall back to the current |
| 1867 | states "fail state", and try the current char again, a process |
| 1868 | we repeat until we reach the root state, state 1, or a legal |
| 1869 | transition. If we fail on the root state then we can either |
| 1870 | terminate if we have reached an accepting state previously, or |
| 1871 | restart the entire process from the beginning if we have not. |
| 1872 | |
| 1873 | */ |
| 1874 | while (s <= last_start) { |
| 1875 | const U32 uniflags = UTF8_ALLOW_DEFAULT; |
| 1876 | U8 *uc = (U8*)s; |
| 1877 | U16 charid = 0; |
| 1878 | U32 base = 1; |
| 1879 | U32 state = 1; |
| 1880 | UV uvc = 0; |
| 1881 | STRLEN len = 0; |
| 1882 | STRLEN foldlen = 0; |
| 1883 | U8 *uscan = (U8*)NULL; |
| 1884 | U8 *leftmost = NULL; |
| 1885 | #ifdef DEBUGGING |
| 1886 | U32 accepted_word= 0; |
| 1887 | #endif |
| 1888 | U32 pointpos = 0; |
| 1889 | |
| 1890 | while ( state && uc <= (U8*)strend ) { |
| 1891 | int failed=0; |
| 1892 | U32 word = aho->states[ state ].wordnum; |
| 1893 | |
| 1894 | if( state==1 ) { |
| 1895 | if ( bitmap ) { |
| 1896 | DEBUG_TRIE_EXECUTE_r( |
| 1897 | if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { |
| 1898 | dump_exec_pos( (char *)uc, c, strend, real_start, |
| 1899 | (char *)uc, utf8_target ); |
| 1900 | PerlIO_printf( Perl_debug_log, |
| 1901 | " Scanning for legal start char...\n"); |
| 1902 | } |
| 1903 | ); |
| 1904 | if (utf8_target) { |
| 1905 | while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { |
| 1906 | uc += UTF8SKIP(uc); |
| 1907 | } |
| 1908 | } else { |
| 1909 | while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { |
| 1910 | uc++; |
| 1911 | } |
| 1912 | } |
| 1913 | s= (char *)uc; |
| 1914 | } |
| 1915 | if (uc >(U8*)last_start) break; |
| 1916 | } |
| 1917 | |
| 1918 | if ( word ) { |
| 1919 | U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ]; |
| 1920 | if (!leftmost || lpos < leftmost) { |
| 1921 | DEBUG_r(accepted_word=word); |
| 1922 | leftmost= lpos; |
| 1923 | } |
| 1924 | if (base==0) break; |
| 1925 | |
| 1926 | } |
| 1927 | points[pointpos++ % maxlen]= uc; |
| 1928 | if (foldlen || uc < (U8*)strend) { |
| 1929 | REXEC_TRIE_READ_CHAR(trie_type, trie, |
| 1930 | widecharmap, uc, |
| 1931 | uscan, len, uvc, charid, foldlen, |
| 1932 | foldbuf, uniflags); |
| 1933 | DEBUG_TRIE_EXECUTE_r({ |
| 1934 | dump_exec_pos( (char *)uc, c, strend, |
| 1935 | real_start, s, utf8_target); |
| 1936 | PerlIO_printf(Perl_debug_log, |
| 1937 | " Charid:%3u CP:%4"UVxf" ", |
| 1938 | charid, uvc); |
| 1939 | }); |
| 1940 | } |
| 1941 | else { |
| 1942 | len = 0; |
| 1943 | charid = 0; |
| 1944 | } |
| 1945 | |
| 1946 | |
| 1947 | do { |
| 1948 | #ifdef DEBUGGING |
| 1949 | word = aho->states[ state ].wordnum; |
| 1950 | #endif |
| 1951 | base = aho->states[ state ].trans.base; |
| 1952 | |
| 1953 | DEBUG_TRIE_EXECUTE_r({ |
| 1954 | if (failed) |
| 1955 | dump_exec_pos( (char *)uc, c, strend, real_start, |
| 1956 | s, utf8_target ); |
| 1957 | PerlIO_printf( Perl_debug_log, |
| 1958 | "%sState: %4"UVxf", word=%"UVxf, |
| 1959 | failed ? " Fail transition to " : "", |
| 1960 | (UV)state, (UV)word); |
| 1961 | }); |
| 1962 | if ( base ) { |
| 1963 | U32 tmp; |
| 1964 | I32 offset; |
| 1965 | if (charid && |
| 1966 | ( ((offset = base + charid |
| 1967 | - 1 - trie->uniquecharcount)) >= 0) |
| 1968 | && ((U32)offset < trie->lasttrans) |
| 1969 | && trie->trans[offset].check == state |
| 1970 | && (tmp=trie->trans[offset].next)) |
| 1971 | { |
| 1972 | DEBUG_TRIE_EXECUTE_r( |
| 1973 | PerlIO_printf( Perl_debug_log," - legal\n")); |
| 1974 | state = tmp; |
| 1975 | break; |
| 1976 | } |
| 1977 | else { |
| 1978 | DEBUG_TRIE_EXECUTE_r( |
| 1979 | PerlIO_printf( Perl_debug_log," - fail\n")); |
| 1980 | failed = 1; |
| 1981 | state = aho->fail[state]; |
| 1982 | } |
| 1983 | } |
| 1984 | else { |
| 1985 | /* we must be accepting here */ |
| 1986 | DEBUG_TRIE_EXECUTE_r( |
| 1987 | PerlIO_printf( Perl_debug_log," - accepting\n")); |
| 1988 | failed = 1; |
| 1989 | break; |
| 1990 | } |
| 1991 | } while(state); |
| 1992 | uc += len; |
| 1993 | if (failed) { |
| 1994 | if (leftmost) |
| 1995 | break; |
| 1996 | if (!state) state = 1; |
| 1997 | } |
| 1998 | } |
| 1999 | if ( aho->states[ state ].wordnum ) { |
| 2000 | U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ]; |
| 2001 | if (!leftmost || lpos < leftmost) { |
| 2002 | DEBUG_r(accepted_word=aho->states[ state ].wordnum); |
| 2003 | leftmost = lpos; |
| 2004 | } |
| 2005 | } |
| 2006 | if (leftmost) { |
| 2007 | s = (char*)leftmost; |
| 2008 | DEBUG_TRIE_EXECUTE_r({ |
| 2009 | PerlIO_printf( |
| 2010 | Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n", |
| 2011 | (UV)accepted_word, (IV)(s - real_start) |
| 2012 | ); |
| 2013 | }); |
| 2014 | if (!reginfo || regtry(reginfo, &s)) { |
| 2015 | FREETMPS; |
| 2016 | LEAVE; |
| 2017 | goto got_it; |
| 2018 | } |
| 2019 | s = HOPc(s,1); |
| 2020 | DEBUG_TRIE_EXECUTE_r({ |
| 2021 | PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n"); |
| 2022 | }); |
| 2023 | } else { |
| 2024 | DEBUG_TRIE_EXECUTE_r( |
| 2025 | PerlIO_printf( Perl_debug_log,"No match.\n")); |
| 2026 | break; |
| 2027 | } |
| 2028 | } |
| 2029 | FREETMPS; |
| 2030 | LEAVE; |
| 2031 | } |
| 2032 | break; |
| 2033 | default: |
| 2034 | Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c)); |
| 2035 | break; |
| 2036 | } |
| 2037 | return 0; |
| 2038 | got_it: |
| 2039 | return s; |
| 2040 | } |
| 2041 | |
| 2042 | |
| 2043 | /* |
| 2044 | - regexec_flags - match a regexp against a string |
| 2045 | */ |
| 2046 | I32 |
| 2047 | Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend, |
| 2048 | char *strbeg, I32 minend, SV *sv, void *data, U32 flags) |
| 2049 | /* stringarg: the point in the string at which to begin matching */ |
| 2050 | /* strend: pointer to null at end of string */ |
| 2051 | /* strbeg: real beginning of string */ |
| 2052 | /* minend: end of match must be >= minend bytes after stringarg. */ |
| 2053 | /* sv: SV being matched: only used for utf8 flag, pos() etc; string |
| 2054 | * itself is accessed via the pointers above */ |
| 2055 | /* data: May be used for some additional optimizations. |
| 2056 | Currently its only used, with a U32 cast, for transmitting |
| 2057 | the ganch offset when doing a /g match. This will change */ |
| 2058 | /* nosave: For optimizations. */ |
| 2059 | |
| 2060 | { |
| 2061 | dVAR; |
| 2062 | struct regexp *const prog = ReANY(rx); |
| 2063 | char *s; |
| 2064 | regnode *c; |
| 2065 | char *startpos = stringarg; |
| 2066 | I32 minlen; /* must match at least this many chars */ |
| 2067 | I32 dontbother = 0; /* how many characters not to try at end */ |
| 2068 | I32 end_shift = 0; /* Same for the end. */ /* CC */ |
| 2069 | I32 scream_pos = -1; /* Internal iterator of scream. */ |
| 2070 | char *scream_olds = NULL; |
| 2071 | const bool utf8_target = cBOOL(DO_UTF8(sv)); |
| 2072 | I32 multiline; |
| 2073 | RXi_GET_DECL(prog,progi); |
| 2074 | regmatch_info reginfo; /* create some info to pass to regtry etc */ |
| 2075 | regexp_paren_pair *swap = NULL; |
| 2076 | GET_RE_DEBUG_FLAGS_DECL; |
| 2077 | |
| 2078 | PERL_ARGS_ASSERT_REGEXEC_FLAGS; |
| 2079 | PERL_UNUSED_ARG(data); |
| 2080 | |
| 2081 | /* Be paranoid... */ |
| 2082 | if (prog == NULL || startpos == NULL) { |
| 2083 | Perl_croak(aTHX_ "NULL regexp parameter"); |
| 2084 | return 0; |
| 2085 | } |
| 2086 | |
| 2087 | multiline = prog->extflags & RXf_PMf_MULTILINE; |
| 2088 | reginfo.prog = rx; /* Yes, sorry that this is confusing. */ |
| 2089 | |
| 2090 | RX_MATCH_UTF8_set(rx, utf8_target); |
| 2091 | DEBUG_EXECUTE_r( |
| 2092 | debug_start_match(rx, utf8_target, startpos, strend, |
| 2093 | "Matching"); |
| 2094 | ); |
| 2095 | |
| 2096 | minlen = prog->minlen; |
| 2097 | |
| 2098 | if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) { |
| 2099 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 2100 | "String too short [regexec_flags]...\n")); |
| 2101 | goto phooey; |
| 2102 | } |
| 2103 | |
| 2104 | |
| 2105 | /* Check validity of program. */ |
| 2106 | if (UCHARAT(progi->program) != REG_MAGIC) { |
| 2107 | Perl_croak(aTHX_ "corrupted regexp program"); |
| 2108 | } |
| 2109 | |
| 2110 | RX_MATCH_TAINTED_off(rx); |
| 2111 | PL_reg_state.re_state_eval_setup_done = FALSE; |
| 2112 | PL_reg_maxiter = 0; |
| 2113 | |
| 2114 | reginfo.is_utf8_pat = cBOOL(RX_UTF8(rx)); |
| 2115 | reginfo.warned = FALSE; |
| 2116 | /* Mark beginning of line for ^ and lookbehind. */ |
| 2117 | reginfo.bol = startpos; /* XXX not used ??? */ |
| 2118 | PL_bostr = strbeg; |
| 2119 | reginfo.sv = sv; |
| 2120 | |
| 2121 | /* Mark end of line for $ (and such) */ |
| 2122 | PL_regeol = strend; |
| 2123 | |
| 2124 | /* see how far we have to get to not match where we matched before */ |
| 2125 | reginfo.till = startpos+minend; |
| 2126 | |
| 2127 | /* If there is a "must appear" string, look for it. */ |
| 2128 | s = startpos; |
| 2129 | |
| 2130 | if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */ |
| 2131 | MAGIC *mg; |
| 2132 | if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */ |
| 2133 | reginfo.ganch = startpos + prog->gofs; |
| 2134 | DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log, |
| 2135 | "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)prog->gofs)); |
| 2136 | } else if (sv && SvTYPE(sv) >= SVt_PVMG |
| 2137 | && SvMAGIC(sv) |
| 2138 | && (mg = mg_find(sv, PERL_MAGIC_regex_global)) |
| 2139 | && mg->mg_len >= 0) { |
| 2140 | reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */ |
| 2141 | DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log, |
| 2142 | "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",(IV)mg->mg_len)); |
| 2143 | |
| 2144 | if (prog->extflags & RXf_ANCH_GPOS) { |
| 2145 | if (s > reginfo.ganch) |
| 2146 | goto phooey; |
| 2147 | s = reginfo.ganch - prog->gofs; |
| 2148 | DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log, |
| 2149 | "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",(UV)prog->gofs)); |
| 2150 | if (s < strbeg) |
| 2151 | goto phooey; |
| 2152 | } |
| 2153 | } |
| 2154 | else if (data) { |
| 2155 | reginfo.ganch = strbeg + PTR2UV(data); |
| 2156 | DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log, |
| 2157 | "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data))); |
| 2158 | |
| 2159 | } else { /* pos() not defined */ |
| 2160 | reginfo.ganch = strbeg; |
| 2161 | DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log, |
| 2162 | "GPOS: reginfo.ganch = strbeg\n")); |
| 2163 | } |
| 2164 | } |
| 2165 | if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) { |
| 2166 | /* We have to be careful. If the previous successful match |
| 2167 | was from this regex we don't want a subsequent partially |
| 2168 | successful match to clobber the old results. |
| 2169 | So when we detect this possibility we add a swap buffer |
| 2170 | to the re, and switch the buffer each match. If we fail, |
| 2171 | we switch it back; otherwise we leave it swapped. |
| 2172 | */ |
| 2173 | swap = prog->offs; |
| 2174 | /* do we need a save destructor here for eval dies? */ |
| 2175 | Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair); |
| 2176 | DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, |
| 2177 | "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n", |
| 2178 | PTR2UV(prog), |
| 2179 | PTR2UV(swap), |
| 2180 | PTR2UV(prog->offs) |
| 2181 | )); |
| 2182 | } |
| 2183 | if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) { |
| 2184 | re_scream_pos_data d; |
| 2185 | |
| 2186 | d.scream_olds = &scream_olds; |
| 2187 | d.scream_pos = &scream_pos; |
| 2188 | s = re_intuit_start(rx, sv, s, strend, flags, &d); |
| 2189 | if (!s) { |
| 2190 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n")); |
| 2191 | goto phooey; /* not present */ |
| 2192 | } |
| 2193 | } |
| 2194 | |
| 2195 | |
| 2196 | |
| 2197 | /* Simplest case: anchored match need be tried only once. */ |
| 2198 | /* [unless only anchor is BOL and multiline is set] */ |
| 2199 | if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) { |
| 2200 | if (s == startpos && regtry(®info, &startpos)) |
| 2201 | goto got_it; |
| 2202 | else if (multiline || (prog->intflags & PREGf_IMPLICIT) |
| 2203 | || (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */ |
| 2204 | { |
| 2205 | char *end; |
| 2206 | |
| 2207 | if (minlen) |
| 2208 | dontbother = minlen - 1; |
| 2209 | end = HOP3c(strend, -dontbother, strbeg) - 1; |
| 2210 | /* for multiline we only have to try after newlines */ |
| 2211 | if (prog->check_substr || prog->check_utf8) { |
| 2212 | /* because of the goto we can not easily reuse the macros for bifurcating the |
| 2213 | unicode/non-unicode match modes here like we do elsewhere - demerphq */ |
| 2214 | if (utf8_target) { |
| 2215 | if (s == startpos) |
| 2216 | goto after_try_utf8; |
| 2217 | while (1) { |
| 2218 | if (regtry(®info, &s)) { |
| 2219 | goto got_it; |
| 2220 | } |
| 2221 | after_try_utf8: |
| 2222 | if (s > end) { |
| 2223 | goto phooey; |
| 2224 | } |
| 2225 | if (prog->extflags & RXf_USE_INTUIT) { |
| 2226 | s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL); |
| 2227 | if (!s) { |
| 2228 | goto phooey; |
| 2229 | } |
| 2230 | } |
| 2231 | else { |
| 2232 | s += UTF8SKIP(s); |
| 2233 | } |
| 2234 | } |
| 2235 | } /* end search for check string in unicode */ |
| 2236 | else { |
| 2237 | if (s == startpos) { |
| 2238 | goto after_try_latin; |
| 2239 | } |
| 2240 | while (1) { |
| 2241 | if (regtry(®info, &s)) { |
| 2242 | goto got_it; |
| 2243 | } |
| 2244 | after_try_latin: |
| 2245 | if (s > end) { |
| 2246 | goto phooey; |
| 2247 | } |
| 2248 | if (prog->extflags & RXf_USE_INTUIT) { |
| 2249 | s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL); |
| 2250 | if (!s) { |
| 2251 | goto phooey; |
| 2252 | } |
| 2253 | } |
| 2254 | else { |
| 2255 | s++; |
| 2256 | } |
| 2257 | } |
| 2258 | } /* end search for check string in latin*/ |
| 2259 | } /* end search for check string */ |
| 2260 | else { /* search for newline */ |
| 2261 | if (s > startpos) { |
| 2262 | /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/ |
| 2263 | s--; |
| 2264 | } |
| 2265 | /* We can use a more efficient search as newlines are the same in unicode as they are in latin */ |
| 2266 | while (s <= end) { /* note it could be possible to match at the end of the string */ |
| 2267 | if (*s++ == '\n') { /* don't need PL_utf8skip here */ |
| 2268 | if (regtry(®info, &s)) |
| 2269 | goto got_it; |
| 2270 | } |
| 2271 | } |
| 2272 | } /* end search for newline */ |
| 2273 | } /* end anchored/multiline check string search */ |
| 2274 | goto phooey; |
| 2275 | } else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK)) |
| 2276 | { |
| 2277 | /* the warning about reginfo.ganch being used without initialization |
| 2278 | is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN |
| 2279 | and we only enter this block when the same bit is set. */ |
| 2280 | char *tmp_s = reginfo.ganch - prog->gofs; |
| 2281 | |
| 2282 | if (tmp_s >= strbeg && regtry(®info, &tmp_s)) |
| 2283 | goto got_it; |
| 2284 | goto phooey; |
| 2285 | } |
| 2286 | |
| 2287 | /* Messy cases: unanchored match. */ |
| 2288 | if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) { |
| 2289 | /* we have /x+whatever/ */ |
| 2290 | /* it must be a one character string (XXXX Except is_utf8_pat?) */ |
| 2291 | char ch; |
| 2292 | #ifdef DEBUGGING |
| 2293 | int did_match = 0; |
| 2294 | #endif |
| 2295 | if (utf8_target) { |
| 2296 | if (! prog->anchored_utf8) { |
| 2297 | to_utf8_substr(prog); |
| 2298 | } |
| 2299 | ch = SvPVX_const(prog->anchored_utf8)[0]; |
| 2300 | REXEC_FBC_SCAN( |
| 2301 | if (*s == ch) { |
| 2302 | DEBUG_EXECUTE_r( did_match = 1 ); |
| 2303 | if (regtry(®info, &s)) goto got_it; |
| 2304 | s += UTF8SKIP(s); |
| 2305 | while (s < strend && *s == ch) |
| 2306 | s += UTF8SKIP(s); |
| 2307 | } |
| 2308 | ); |
| 2309 | |
| 2310 | } |
| 2311 | else { |
| 2312 | if (! prog->anchored_substr) { |
| 2313 | if (! to_byte_substr(prog)) { |
| 2314 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); |
| 2315 | } |
| 2316 | } |
| 2317 | ch = SvPVX_const(prog->anchored_substr)[0]; |
| 2318 | REXEC_FBC_SCAN( |
| 2319 | if (*s == ch) { |
| 2320 | DEBUG_EXECUTE_r( did_match = 1 ); |
| 2321 | if (regtry(®info, &s)) goto got_it; |
| 2322 | s++; |
| 2323 | while (s < strend && *s == ch) |
| 2324 | s++; |
| 2325 | } |
| 2326 | ); |
| 2327 | } |
| 2328 | DEBUG_EXECUTE_r(if (!did_match) |
| 2329 | PerlIO_printf(Perl_debug_log, |
| 2330 | "Did not find anchored character...\n") |
| 2331 | ); |
| 2332 | } |
| 2333 | else if (prog->anchored_substr != NULL |
| 2334 | || prog->anchored_utf8 != NULL |
| 2335 | || ((prog->float_substr != NULL || prog->float_utf8 != NULL) |
| 2336 | && prog->float_max_offset < strend - s)) { |
| 2337 | SV *must; |
| 2338 | I32 back_max; |
| 2339 | I32 back_min; |
| 2340 | char *last; |
| 2341 | char *last1; /* Last position checked before */ |
| 2342 | #ifdef DEBUGGING |
| 2343 | int did_match = 0; |
| 2344 | #endif |
| 2345 | if (prog->anchored_substr || prog->anchored_utf8) { |
| 2346 | if (utf8_target) { |
| 2347 | if (! prog->anchored_utf8) { |
| 2348 | to_utf8_substr(prog); |
| 2349 | } |
| 2350 | must = prog->anchored_utf8; |
| 2351 | } |
| 2352 | else { |
| 2353 | if (! prog->anchored_substr) { |
| 2354 | if (! to_byte_substr(prog)) { |
| 2355 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); |
| 2356 | } |
| 2357 | } |
| 2358 | must = prog->anchored_substr; |
| 2359 | } |
| 2360 | back_max = back_min = prog->anchored_offset; |
| 2361 | } else { |
| 2362 | if (utf8_target) { |
| 2363 | if (! prog->float_utf8) { |
| 2364 | to_utf8_substr(prog); |
| 2365 | } |
| 2366 | must = prog->float_utf8; |
| 2367 | } |
| 2368 | else { |
| 2369 | if (! prog->float_substr) { |
| 2370 | if (! to_byte_substr(prog)) { |
| 2371 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); |
| 2372 | } |
| 2373 | } |
| 2374 | must = prog->float_substr; |
| 2375 | } |
| 2376 | back_max = prog->float_max_offset; |
| 2377 | back_min = prog->float_min_offset; |
| 2378 | } |
| 2379 | |
| 2380 | if (back_min<0) { |
| 2381 | last = strend; |
| 2382 | } else { |
| 2383 | last = HOP3c(strend, /* Cannot start after this */ |
| 2384 | -(I32)(CHR_SVLEN(must) |
| 2385 | - (SvTAIL(must) != 0) + back_min), strbeg); |
| 2386 | } |
| 2387 | if (s > PL_bostr) |
| 2388 | last1 = HOPc(s, -1); |
| 2389 | else |
| 2390 | last1 = s - 1; /* bogus */ |
| 2391 | |
| 2392 | /* XXXX check_substr already used to find "s", can optimize if |
| 2393 | check_substr==must. */ |
| 2394 | scream_pos = -1; |
| 2395 | dontbother = end_shift; |
| 2396 | strend = HOPc(strend, -dontbother); |
| 2397 | while ( (s <= last) && |
| 2398 | (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)), |
| 2399 | (unsigned char*)strend, must, |
| 2400 | multiline ? FBMrf_MULTILINE : 0)) ) { |
| 2401 | DEBUG_EXECUTE_r( did_match = 1 ); |
| 2402 | if (HOPc(s, -back_max) > last1) { |
| 2403 | last1 = HOPc(s, -back_min); |
| 2404 | s = HOPc(s, -back_max); |
| 2405 | } |
| 2406 | else { |
| 2407 | char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1; |
| 2408 | |
| 2409 | last1 = HOPc(s, -back_min); |
| 2410 | s = t; |
| 2411 | } |
| 2412 | if (utf8_target) { |
| 2413 | while (s <= last1) { |
| 2414 | if (regtry(®info, &s)) |
| 2415 | goto got_it; |
| 2416 | if (s >= last1) { |
| 2417 | s++; /* to break out of outer loop */ |
| 2418 | break; |
| 2419 | } |
| 2420 | s += UTF8SKIP(s); |
| 2421 | } |
| 2422 | } |
| 2423 | else { |
| 2424 | while (s <= last1) { |
| 2425 | if (regtry(®info, &s)) |
| 2426 | goto got_it; |
| 2427 | s++; |
| 2428 | } |
| 2429 | } |
| 2430 | } |
| 2431 | DEBUG_EXECUTE_r(if (!did_match) { |
| 2432 | RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0), |
| 2433 | SvPVX_const(must), RE_SV_DUMPLEN(must), 30); |
| 2434 | PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n", |
| 2435 | ((must == prog->anchored_substr || must == prog->anchored_utf8) |
| 2436 | ? "anchored" : "floating"), |
| 2437 | quoted, RE_SV_TAIL(must)); |
| 2438 | }); |
| 2439 | goto phooey; |
| 2440 | } |
| 2441 | else if ( (c = progi->regstclass) ) { |
| 2442 | if (minlen) { |
| 2443 | const OPCODE op = OP(progi->regstclass); |
| 2444 | /* don't bother with what can't match */ |
| 2445 | if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE) |
| 2446 | strend = HOPc(strend, -(minlen - 1)); |
| 2447 | } |
| 2448 | DEBUG_EXECUTE_r({ |
| 2449 | SV * const prop = sv_newmortal(); |
| 2450 | regprop(prog, prop, c); |
| 2451 | { |
| 2452 | RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1), |
| 2453 | s,strend-s,60); |
| 2454 | PerlIO_printf(Perl_debug_log, |
| 2455 | "Matching stclass %.*s against %s (%d bytes)\n", |
| 2456 | (int)SvCUR(prop), SvPVX_const(prop), |
| 2457 | quoted, (int)(strend - s)); |
| 2458 | } |
| 2459 | }); |
| 2460 | if (find_byclass(prog, c, s, strend, ®info, reginfo.is_utf8_pat)) |
| 2461 | goto got_it; |
| 2462 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n")); |
| 2463 | } |
| 2464 | else { |
| 2465 | dontbother = 0; |
| 2466 | if (prog->float_substr != NULL || prog->float_utf8 != NULL) { |
| 2467 | /* Trim the end. */ |
| 2468 | char *last= NULL; |
| 2469 | SV* float_real; |
| 2470 | STRLEN len; |
| 2471 | const char *little; |
| 2472 | |
| 2473 | if (utf8_target) { |
| 2474 | if (! prog->float_utf8) { |
| 2475 | to_utf8_substr(prog); |
| 2476 | } |
| 2477 | float_real = prog->float_utf8; |
| 2478 | } |
| 2479 | else { |
| 2480 | if (! prog->float_substr) { |
| 2481 | if (! to_byte_substr(prog)) { |
| 2482 | NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey); |
| 2483 | } |
| 2484 | } |
| 2485 | float_real = prog->float_substr; |
| 2486 | } |
| 2487 | |
| 2488 | little = SvPV_const(float_real, len); |
| 2489 | if (SvTAIL(float_real)) { |
| 2490 | /* This means that float_real contains an artificial \n on |
| 2491 | * the end due to the presence of something like this: |
| 2492 | * /foo$/ where we can match both "foo" and "foo\n" at the |
| 2493 | * end of the string. So we have to compare the end of the |
| 2494 | * string first against the float_real without the \n and |
| 2495 | * then against the full float_real with the string. We |
| 2496 | * have to watch out for cases where the string might be |
| 2497 | * smaller than the float_real or the float_real without |
| 2498 | * the \n. */ |
| 2499 | char *checkpos= strend - len; |
| 2500 | DEBUG_OPTIMISE_r( |
| 2501 | PerlIO_printf(Perl_debug_log, |
| 2502 | "%sChecking for float_real.%s\n", |
| 2503 | PL_colors[4], PL_colors[5])); |
| 2504 | if (checkpos + 1 < strbeg) { |
| 2505 | /* can't match, even if we remove the trailing \n |
| 2506 | * string is too short to match */ |
| 2507 | DEBUG_EXECUTE_r( |
| 2508 | PerlIO_printf(Perl_debug_log, |
| 2509 | "%sString shorter than required trailing substring, cannot match.%s\n", |
| 2510 | PL_colors[4], PL_colors[5])); |
| 2511 | goto phooey; |
| 2512 | } else if (memEQ(checkpos + 1, little, len - 1)) { |
| 2513 | /* can match, the end of the string matches without the |
| 2514 | * "\n" */ |
| 2515 | last = checkpos + 1; |
| 2516 | } else if (checkpos < strbeg) { |
| 2517 | /* cant match, string is too short when the "\n" is |
| 2518 | * included */ |
| 2519 | DEBUG_EXECUTE_r( |
| 2520 | PerlIO_printf(Perl_debug_log, |
| 2521 | "%sString does not contain required trailing substring, cannot match.%s\n", |
| 2522 | PL_colors[4], PL_colors[5])); |
| 2523 | goto phooey; |
| 2524 | } else if (!multiline) { |
| 2525 | /* non multiline match, so compare with the "\n" at the |
| 2526 | * end of the string */ |
| 2527 | if (memEQ(checkpos, little, len)) { |
| 2528 | last= checkpos; |
| 2529 | } else { |
| 2530 | DEBUG_EXECUTE_r( |
| 2531 | PerlIO_printf(Perl_debug_log, |
| 2532 | "%sString does not contain required trailing substring, cannot match.%s\n", |
| 2533 | PL_colors[4], PL_colors[5])); |
| 2534 | goto phooey; |
| 2535 | } |
| 2536 | } else { |
| 2537 | /* multiline match, so we have to search for a place |
| 2538 | * where the full string is located */ |
| 2539 | goto find_last; |
| 2540 | } |
| 2541 | } else { |
| 2542 | find_last: |
| 2543 | if (len) |
| 2544 | last = rninstr(s, strend, little, little + len); |
| 2545 | else |
| 2546 | last = strend; /* matching "$" */ |
| 2547 | } |
| 2548 | if (!last) { |
| 2549 | /* at one point this block contained a comment which was |
| 2550 | * probably incorrect, which said that this was a "should not |
| 2551 | * happen" case. Even if it was true when it was written I am |
| 2552 | * pretty sure it is not anymore, so I have removed the comment |
| 2553 | * and replaced it with this one. Yves */ |
| 2554 | DEBUG_EXECUTE_r( |
| 2555 | PerlIO_printf(Perl_debug_log, |
| 2556 | "String does not contain required substring, cannot match.\n" |
| 2557 | )); |
| 2558 | goto phooey; |
| 2559 | } |
| 2560 | dontbother = strend - last + prog->float_min_offset; |
| 2561 | } |
| 2562 | if (minlen && (dontbother < minlen)) |
| 2563 | dontbother = minlen - 1; |
| 2564 | strend -= dontbother; /* this one's always in bytes! */ |
| 2565 | /* We don't know much -- general case. */ |
| 2566 | if (utf8_target) { |
| 2567 | for (;;) { |
| 2568 | if (regtry(®info, &s)) |
| 2569 | goto got_it; |
| 2570 | if (s >= strend) |
| 2571 | break; |
| 2572 | s += UTF8SKIP(s); |
| 2573 | }; |
| 2574 | } |
| 2575 | else { |
| 2576 | do { |
| 2577 | if (regtry(®info, &s)) |
| 2578 | goto got_it; |
| 2579 | } while (s++ < strend); |
| 2580 | } |
| 2581 | } |
| 2582 | |
| 2583 | /* Failure. */ |
| 2584 | goto phooey; |
| 2585 | |
| 2586 | got_it: |
| 2587 | DEBUG_BUFFERS_r( |
| 2588 | if (swap) |
| 2589 | PerlIO_printf(Perl_debug_log, |
| 2590 | "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n", |
| 2591 | PTR2UV(prog), |
| 2592 | PTR2UV(swap) |
| 2593 | ); |
| 2594 | ); |
| 2595 | Safefree(swap); |
| 2596 | |
| 2597 | if (PL_reg_state.re_state_eval_setup_done) |
| 2598 | restore_pos(aTHX_ prog); |
| 2599 | if (RXp_PAREN_NAMES(prog)) |
| 2600 | (void)hv_iterinit(RXp_PAREN_NAMES(prog)); |
| 2601 | |
| 2602 | /* make sure $`, $&, $', and $digit will work later */ |
| 2603 | if ( !(flags & REXEC_NOT_FIRST) ) { |
| 2604 | if (flags & REXEC_COPY_STR) { |
| 2605 | #ifdef PERL_ANY_COW |
| 2606 | if (SvCANCOW(sv)) { |
| 2607 | if (DEBUG_C_TEST) { |
| 2608 | PerlIO_printf(Perl_debug_log, |
| 2609 | "Copy on write: regexp capture, type %d\n", |
| 2610 | (int) SvTYPE(sv)); |
| 2611 | } |
| 2612 | RX_MATCH_COPY_FREE(rx); |
| 2613 | prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv); |
| 2614 | prog->subbeg = (char *)SvPVX_const(prog->saved_copy); |
| 2615 | assert (SvPOKp(prog->saved_copy)); |
| 2616 | prog->sublen = PL_regeol - strbeg; |
| 2617 | prog->suboffset = 0; |
| 2618 | prog->subcoffset = 0; |
| 2619 | } else |
| 2620 | #endif |
| 2621 | { |
| 2622 | I32 min = 0; |
| 2623 | I32 max = PL_regeol - strbeg; |
| 2624 | I32 sublen; |
| 2625 | |
| 2626 | if ( (flags & REXEC_COPY_SKIP_POST) |
| 2627 | && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */ |
| 2628 | && !(PL_sawampersand & SAWAMPERSAND_RIGHT) |
| 2629 | ) { /* don't copy $' part of string */ |
| 2630 | U32 n = 0; |
| 2631 | max = -1; |
| 2632 | /* calculate the right-most part of the string covered |
| 2633 | * by a capture. Due to look-ahead, this may be to |
| 2634 | * the right of $&, so we have to scan all captures */ |
| 2635 | while (n <= prog->lastparen) { |
| 2636 | if (prog->offs[n].end > max) |
| 2637 | max = prog->offs[n].end; |
| 2638 | n++; |
| 2639 | } |
| 2640 | if (max == -1) |
| 2641 | max = (PL_sawampersand & SAWAMPERSAND_LEFT) |
| 2642 | ? prog->offs[0].start |
| 2643 | : 0; |
| 2644 | assert(max >= 0 && max <= PL_regeol - strbeg); |
| 2645 | } |
| 2646 | |
| 2647 | if ( (flags & REXEC_COPY_SKIP_PRE) |
| 2648 | && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */ |
| 2649 | && !(PL_sawampersand & SAWAMPERSAND_LEFT) |
| 2650 | ) { /* don't copy $` part of string */ |
| 2651 | U32 n = 0; |
| 2652 | min = max; |
| 2653 | /* calculate the left-most part of the string covered |
| 2654 | * by a capture. Due to look-behind, this may be to |
| 2655 | * the left of $&, so we have to scan all captures */ |
| 2656 | while (min && n <= prog->lastparen) { |
| 2657 | if ( prog->offs[n].start != -1 |
| 2658 | && prog->offs[n].start < min) |
| 2659 | { |
| 2660 | min = prog->offs[n].start; |
| 2661 | } |
| 2662 | n++; |
| 2663 | } |
| 2664 | if ((PL_sawampersand & SAWAMPERSAND_RIGHT) |
| 2665 | && min > prog->offs[0].end |
| 2666 | ) |
| 2667 | min = prog->offs[0].end; |
| 2668 | |
| 2669 | } |
| 2670 | |
| 2671 | assert(min >= 0 && min <= max && min <= PL_regeol - strbeg); |
| 2672 | sublen = max - min; |
| 2673 | |
| 2674 | if (RX_MATCH_COPIED(rx)) { |
| 2675 | if (sublen > prog->sublen) |
| 2676 | prog->subbeg = |
| 2677 | (char*)saferealloc(prog->subbeg, sublen+1); |
| 2678 | } |
| 2679 | else |
| 2680 | prog->subbeg = (char*)safemalloc(sublen+1); |
| 2681 | Copy(strbeg + min, prog->subbeg, sublen, char); |
| 2682 | prog->subbeg[sublen] = '\0'; |
| 2683 | prog->suboffset = min; |
| 2684 | prog->sublen = sublen; |
| 2685 | RX_MATCH_COPIED_on(rx); |
| 2686 | } |
| 2687 | prog->subcoffset = prog->suboffset; |
| 2688 | if (prog->suboffset && utf8_target) { |
| 2689 | /* Convert byte offset to chars. |
| 2690 | * XXX ideally should only compute this if @-/@+ |
| 2691 | * has been seen, a la PL_sawampersand ??? */ |
| 2692 | |
| 2693 | /* If there's a direct correspondence between the |
| 2694 | * string which we're matching and the original SV, |
| 2695 | * then we can use the utf8 len cache associated with |
| 2696 | * the SV. In particular, it means that under //g, |
| 2697 | * sv_pos_b2u() will use the previously cached |
| 2698 | * position to speed up working out the new length of |
| 2699 | * subcoffset, rather than counting from the start of |
| 2700 | * the string each time. This stops |
| 2701 | * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g; |
| 2702 | * from going quadratic */ |
| 2703 | if (SvPOKp(sv) && SvPVX(sv) == strbeg) |
| 2704 | sv_pos_b2u(sv, &(prog->subcoffset)); |
| 2705 | else |
| 2706 | prog->subcoffset = utf8_length((U8*)strbeg, |
| 2707 | (U8*)(strbeg+prog->suboffset)); |
| 2708 | } |
| 2709 | } |
| 2710 | else { |
| 2711 | RX_MATCH_COPY_FREE(rx); |
| 2712 | prog->subbeg = strbeg; |
| 2713 | prog->suboffset = 0; |
| 2714 | prog->subcoffset = 0; |
| 2715 | prog->sublen = PL_regeol - strbeg; /* strend may have been modified */ |
| 2716 | } |
| 2717 | } |
| 2718 | |
| 2719 | return 1; |
| 2720 | |
| 2721 | phooey: |
| 2722 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n", |
| 2723 | PL_colors[4], PL_colors[5])); |
| 2724 | if (PL_reg_state.re_state_eval_setup_done) |
| 2725 | restore_pos(aTHX_ prog); |
| 2726 | if (swap) { |
| 2727 | /* we failed :-( roll it back */ |
| 2728 | DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, |
| 2729 | "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n", |
| 2730 | PTR2UV(prog), |
| 2731 | PTR2UV(prog->offs), |
| 2732 | PTR2UV(swap) |
| 2733 | )); |
| 2734 | Safefree(prog->offs); |
| 2735 | prog->offs = swap; |
| 2736 | } |
| 2737 | return 0; |
| 2738 | } |
| 2739 | |
| 2740 | |
| 2741 | /* Set which rex is pointed to by PL_reg_state, handling ref counting. |
| 2742 | * Do inc before dec, in case old and new rex are the same */ |
| 2743 | #define SET_reg_curpm(Re2) \ |
| 2744 | if (PL_reg_state.re_state_eval_setup_done) { \ |
| 2745 | (void)ReREFCNT_inc(Re2); \ |
| 2746 | ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \ |
| 2747 | PM_SETRE((PL_reg_curpm), (Re2)); \ |
| 2748 | } |
| 2749 | |
| 2750 | |
| 2751 | /* |
| 2752 | - regtry - try match at specific point |
| 2753 | */ |
| 2754 | STATIC I32 /* 0 failure, 1 success */ |
| 2755 | S_regtry(pTHX_ regmatch_info *reginfo, char **startposp) |
| 2756 | { |
| 2757 | dVAR; |
| 2758 | CHECKPOINT lastcp; |
| 2759 | REGEXP *const rx = reginfo->prog; |
| 2760 | regexp *const prog = ReANY(rx); |
| 2761 | I32 result; |
| 2762 | RXi_GET_DECL(prog,progi); |
| 2763 | GET_RE_DEBUG_FLAGS_DECL; |
| 2764 | |
| 2765 | PERL_ARGS_ASSERT_REGTRY; |
| 2766 | |
| 2767 | reginfo->cutpoint=NULL; |
| 2768 | |
| 2769 | if ((prog->extflags & RXf_EVAL_SEEN) |
| 2770 | && !PL_reg_state.re_state_eval_setup_done) |
| 2771 | { |
| 2772 | MAGIC *mg; |
| 2773 | |
| 2774 | PL_reg_state.re_state_eval_setup_done = TRUE; |
| 2775 | if (reginfo->sv) { |
| 2776 | /* Make $_ available to executed code. */ |
| 2777 | if (reginfo->sv != DEFSV) { |
| 2778 | SAVE_DEFSV; |
| 2779 | DEFSV_set(reginfo->sv); |
| 2780 | } |
| 2781 | |
| 2782 | if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv) |
| 2783 | && (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) { |
| 2784 | /* prepare for quick setting of pos */ |
| 2785 | #ifdef PERL_OLD_COPY_ON_WRITE |
| 2786 | if (SvIsCOW(reginfo->sv)) |
| 2787 | sv_force_normal_flags(reginfo->sv, 0); |
| 2788 | #endif |
| 2789 | mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global, |
| 2790 | &PL_vtbl_mglob, NULL, 0); |
| 2791 | mg->mg_len = -1; |
| 2792 | } |
| 2793 | PL_reg_magic = mg; |
| 2794 | PL_reg_oldpos = mg->mg_len; |
| 2795 | SAVEDESTRUCTOR_X(restore_pos, prog); |
| 2796 | } |
| 2797 | if (!PL_reg_curpm) { |
| 2798 | Newxz(PL_reg_curpm, 1, PMOP); |
| 2799 | #ifdef USE_ITHREADS |
| 2800 | { |
| 2801 | SV* const repointer = &PL_sv_undef; |
| 2802 | /* this regexp is also owned by the new PL_reg_curpm, which |
| 2803 | will try to free it. */ |
| 2804 | av_push(PL_regex_padav, repointer); |
| 2805 | PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav); |
| 2806 | PL_regex_pad = AvARRAY(PL_regex_padav); |
| 2807 | } |
| 2808 | #endif |
| 2809 | } |
| 2810 | SET_reg_curpm(rx); |
| 2811 | PL_reg_oldcurpm = PL_curpm; |
| 2812 | PL_curpm = PL_reg_curpm; |
| 2813 | if (RXp_MATCH_COPIED(prog)) { |
| 2814 | /* Here is a serious problem: we cannot rewrite subbeg, |
| 2815 | since it may be needed if this match fails. Thus |
| 2816 | $` inside (?{}) could fail... */ |
| 2817 | PL_reg_oldsaved = prog->subbeg; |
| 2818 | PL_reg_oldsavedlen = prog->sublen; |
| 2819 | PL_reg_oldsavedoffset = prog->suboffset; |
| 2820 | PL_reg_oldsavedcoffset = prog->suboffset; |
| 2821 | #ifdef PERL_ANY_COW |
| 2822 | PL_nrs = prog->saved_copy; |
| 2823 | #endif |
| 2824 | RXp_MATCH_COPIED_off(prog); |
| 2825 | } |
| 2826 | else |
| 2827 | PL_reg_oldsaved = NULL; |
| 2828 | prog->subbeg = PL_bostr; |
| 2829 | prog->suboffset = 0; |
| 2830 | prog->subcoffset = 0; |
| 2831 | prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */ |
| 2832 | } |
| 2833 | #ifdef DEBUGGING |
| 2834 | PL_reg_starttry = *startposp; |
| 2835 | #endif |
| 2836 | prog->offs[0].start = *startposp - PL_bostr; |
| 2837 | prog->lastparen = 0; |
| 2838 | prog->lastcloseparen = 0; |
| 2839 | |
| 2840 | /* XXXX What this code is doing here?!!! There should be no need |
| 2841 | to do this again and again, prog->lastparen should take care of |
| 2842 | this! --ilya*/ |
| 2843 | |
| 2844 | /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code. |
| 2845 | * Actually, the code in regcppop() (which Ilya may be meaning by |
| 2846 | * prog->lastparen), is not needed at all by the test suite |
| 2847 | * (op/regexp, op/pat, op/split), but that code is needed otherwise |
| 2848 | * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/ |
| 2849 | * Meanwhile, this code *is* needed for the |
| 2850 | * above-mentioned test suite tests to succeed. The common theme |
| 2851 | * on those tests seems to be returning null fields from matches. |
| 2852 | * --jhi updated by dapm */ |
| 2853 | #if 1 |
| 2854 | if (prog->nparens) { |
| 2855 | regexp_paren_pair *pp = prog->offs; |
| 2856 | I32 i; |
| 2857 | for (i = prog->nparens; i > (I32)prog->lastparen; i--) { |
| 2858 | ++pp; |
| 2859 | pp->start = -1; |
| 2860 | pp->end = -1; |
| 2861 | } |
| 2862 | } |
| 2863 | #endif |
| 2864 | REGCP_SET(lastcp); |
| 2865 | result = regmatch(reginfo, *startposp, progi->program + 1); |
| 2866 | if (result != -1) { |
| 2867 | prog->offs[0].end = result; |
| 2868 | return 1; |
| 2869 | } |
| 2870 | if (reginfo->cutpoint) |
| 2871 | *startposp= reginfo->cutpoint; |
| 2872 | REGCP_UNWIND(lastcp); |
| 2873 | return 0; |
| 2874 | } |
| 2875 | |
| 2876 | |
| 2877 | #define sayYES goto yes |
| 2878 | #define sayNO goto no |
| 2879 | #define sayNO_SILENT goto no_silent |
| 2880 | |
| 2881 | /* we dont use STMT_START/END here because it leads to |
| 2882 | "unreachable code" warnings, which are bogus, but distracting. */ |
| 2883 | #define CACHEsayNO \ |
| 2884 | if (ST.cache_mask) \ |
| 2885 | PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \ |
| 2886 | sayNO |
| 2887 | |
| 2888 | /* this is used to determine how far from the left messages like |
| 2889 | 'failed...' are printed. It should be set such that messages |
| 2890 | are inline with the regop output that created them. |
| 2891 | */ |
| 2892 | #define REPORT_CODE_OFF 32 |
| 2893 | |
| 2894 | |
| 2895 | #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */ |
| 2896 | #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */ |
| 2897 | #define CHRTEST_NOT_A_CP_1 -999 |
| 2898 | #define CHRTEST_NOT_A_CP_2 -998 |
| 2899 | |
| 2900 | #define SLAB_FIRST(s) (&(s)->states[0]) |
| 2901 | #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1]) |
| 2902 | |
| 2903 | /* grab a new slab and return the first slot in it */ |
| 2904 | |
| 2905 | STATIC regmatch_state * |
| 2906 | S_push_slab(pTHX) |
| 2907 | { |
| 2908 | #if PERL_VERSION < 9 && !defined(PERL_CORE) |
| 2909 | dMY_CXT; |
| 2910 | #endif |
| 2911 | regmatch_slab *s = PL_regmatch_slab->next; |
| 2912 | if (!s) { |
| 2913 | Newx(s, 1, regmatch_slab); |
| 2914 | s->prev = PL_regmatch_slab; |
| 2915 | s->next = NULL; |
| 2916 | PL_regmatch_slab->next = s; |
| 2917 | } |
| 2918 | PL_regmatch_slab = s; |
| 2919 | return SLAB_FIRST(s); |
| 2920 | } |
| 2921 | |
| 2922 | |
| 2923 | /* push a new state then goto it */ |
| 2924 | |
| 2925 | #define PUSH_STATE_GOTO(state, node, input) \ |
| 2926 | pushinput = input; \ |
| 2927 | scan = node; \ |
| 2928 | st->resume_state = state; \ |
| 2929 | goto push_state; |
| 2930 | |
| 2931 | /* push a new state with success backtracking, then goto it */ |
| 2932 | |
| 2933 | #define PUSH_YES_STATE_GOTO(state, node, input) \ |
| 2934 | pushinput = input; \ |
| 2935 | scan = node; \ |
| 2936 | st->resume_state = state; \ |
| 2937 | goto push_yes_state; |
| 2938 | |
| 2939 | |
| 2940 | |
| 2941 | |
| 2942 | /* |
| 2943 | |
| 2944 | regmatch() - main matching routine |
| 2945 | |
| 2946 | This is basically one big switch statement in a loop. We execute an op, |
| 2947 | set 'next' to point the next op, and continue. If we come to a point which |
| 2948 | we may need to backtrack to on failure such as (A|B|C), we push a |
| 2949 | backtrack state onto the backtrack stack. On failure, we pop the top |
| 2950 | state, and re-enter the loop at the state indicated. If there are no more |
| 2951 | states to pop, we return failure. |
| 2952 | |
| 2953 | Sometimes we also need to backtrack on success; for example /A+/, where |
| 2954 | after successfully matching one A, we need to go back and try to |
| 2955 | match another one; similarly for lookahead assertions: if the assertion |
| 2956 | completes successfully, we backtrack to the state just before the assertion |
| 2957 | and then carry on. In these cases, the pushed state is marked as |
| 2958 | 'backtrack on success too'. This marking is in fact done by a chain of |
| 2959 | pointers, each pointing to the previous 'yes' state. On success, we pop to |
| 2960 | the nearest yes state, discarding any intermediate failure-only states. |
| 2961 | Sometimes a yes state is pushed just to force some cleanup code to be |
| 2962 | called at the end of a successful match or submatch; e.g. (??{$re}) uses |
| 2963 | it to free the inner regex. |
| 2964 | |
| 2965 | Note that failure backtracking rewinds the cursor position, while |
| 2966 | success backtracking leaves it alone. |
| 2967 | |
| 2968 | A pattern is complete when the END op is executed, while a subpattern |
| 2969 | such as (?=foo) is complete when the SUCCESS op is executed. Both of these |
| 2970 | ops trigger the "pop to last yes state if any, otherwise return true" |
| 2971 | behaviour. |
| 2972 | |
| 2973 | A common convention in this function is to use A and B to refer to the two |
| 2974 | subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is |
| 2975 | the subpattern to be matched possibly multiple times, while B is the entire |
| 2976 | rest of the pattern. Variable and state names reflect this convention. |
| 2977 | |
| 2978 | The states in the main switch are the union of ops and failure/success of |
| 2979 | substates associated with with that op. For example, IFMATCH is the op |
| 2980 | that does lookahead assertions /(?=A)B/ and so the IFMATCH state means |
| 2981 | 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just |
| 2982 | successfully matched A and IFMATCH_A_fail is a state saying that we have |
| 2983 | just failed to match A. Resume states always come in pairs. The backtrack |
| 2984 | state we push is marked as 'IFMATCH_A', but when that is popped, we resume |
| 2985 | at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking |
| 2986 | on success or failure. |
| 2987 | |
| 2988 | The struct that holds a backtracking state is actually a big union, with |
| 2989 | one variant for each major type of op. The variable st points to the |
| 2990 | top-most backtrack struct. To make the code clearer, within each |
| 2991 | block of code we #define ST to alias the relevant union. |
| 2992 | |
| 2993 | Here's a concrete example of a (vastly oversimplified) IFMATCH |
| 2994 | implementation: |
| 2995 | |
| 2996 | switch (state) { |
| 2997 | .... |
| 2998 | |
| 2999 | #define ST st->u.ifmatch |
| 3000 | |
| 3001 | case IFMATCH: // we are executing the IFMATCH op, (?=A)B |
| 3002 | ST.foo = ...; // some state we wish to save |
| 3003 | ... |
| 3004 | // push a yes backtrack state with a resume value of |
| 3005 | // IFMATCH_A/IFMATCH_A_fail, then continue execution at the |
| 3006 | // first node of A: |
| 3007 | PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput); |
| 3008 | // NOTREACHED |
| 3009 | |
| 3010 | case IFMATCH_A: // we have successfully executed A; now continue with B |
| 3011 | next = B; |
| 3012 | bar = ST.foo; // do something with the preserved value |
| 3013 | break; |
| 3014 | |
| 3015 | case IFMATCH_A_fail: // A failed, so the assertion failed |
| 3016 | ...; // do some housekeeping, then ... |
| 3017 | sayNO; // propagate the failure |
| 3018 | |
| 3019 | #undef ST |
| 3020 | |
| 3021 | ... |
| 3022 | } |
| 3023 | |
| 3024 | For any old-timers reading this who are familiar with the old recursive |
| 3025 | approach, the code above is equivalent to: |
| 3026 | |
| 3027 | case IFMATCH: // we are executing the IFMATCH op, (?=A)B |
| 3028 | { |
| 3029 | int foo = ... |
| 3030 | ... |
| 3031 | if (regmatch(A)) { |
| 3032 | next = B; |
| 3033 | bar = foo; |
| 3034 | break; |
| 3035 | } |
| 3036 | ...; // do some housekeeping, then ... |
| 3037 | sayNO; // propagate the failure |
| 3038 | } |
| 3039 | |
| 3040 | The topmost backtrack state, pointed to by st, is usually free. If you |
| 3041 | want to claim it, populate any ST.foo fields in it with values you wish to |
| 3042 | save, then do one of |
| 3043 | |
| 3044 | PUSH_STATE_GOTO(resume_state, node, newinput); |
| 3045 | PUSH_YES_STATE_GOTO(resume_state, node, newinput); |
| 3046 | |
| 3047 | which sets that backtrack state's resume value to 'resume_state', pushes a |
| 3048 | new free entry to the top of the backtrack stack, then goes to 'node'. |
| 3049 | On backtracking, the free slot is popped, and the saved state becomes the |
| 3050 | new free state. An ST.foo field in this new top state can be temporarily |
| 3051 | accessed to retrieve values, but once the main loop is re-entered, it |
| 3052 | becomes available for reuse. |
| 3053 | |
| 3054 | Note that the depth of the backtrack stack constantly increases during the |
| 3055 | left-to-right execution of the pattern, rather than going up and down with |
| 3056 | the pattern nesting. For example the stack is at its maximum at Z at the |
| 3057 | end of the pattern, rather than at X in the following: |
| 3058 | |
| 3059 | /(((X)+)+)+....(Y)+....Z/ |
| 3060 | |
| 3061 | The only exceptions to this are lookahead/behind assertions and the cut, |
| 3062 | (?>A), which pop all the backtrack states associated with A before |
| 3063 | continuing. |
| 3064 | |
| 3065 | Backtrack state structs are allocated in slabs of about 4K in size. |
| 3066 | PL_regmatch_state and st always point to the currently active state, |
| 3067 | and PL_regmatch_slab points to the slab currently containing |
| 3068 | PL_regmatch_state. The first time regmatch() is called, the first slab is |
| 3069 | allocated, and is never freed until interpreter destruction. When the slab |
| 3070 | is full, a new one is allocated and chained to the end. At exit from |
| 3071 | regmatch(), slabs allocated since entry are freed. |
| 3072 | |
| 3073 | */ |
| 3074 | |
| 3075 | |
| 3076 | #define DEBUG_STATE_pp(pp) \ |
| 3077 | DEBUG_STATE_r({ \ |
| 3078 | DUMP_EXEC_POS(locinput, scan, utf8_target); \ |
| 3079 | PerlIO_printf(Perl_debug_log, \ |
| 3080 | " %*s"pp" %s%s%s%s%s\n", \ |
| 3081 | depth*2, "", \ |
| 3082 | PL_reg_name[st->resume_state], \ |
| 3083 | ((st==yes_state||st==mark_state) ? "[" : ""), \ |
| 3084 | ((st==yes_state) ? "Y" : ""), \ |
| 3085 | ((st==mark_state) ? "M" : ""), \ |
| 3086 | ((st==yes_state||st==mark_state) ? "]" : "") \ |
| 3087 | ); \ |
| 3088 | }); |
| 3089 | |
| 3090 | |
| 3091 | #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1) |
| 3092 | |
| 3093 | #ifdef DEBUGGING |
| 3094 | |
| 3095 | STATIC void |
| 3096 | S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target, |
| 3097 | const char *start, const char *end, const char *blurb) |
| 3098 | { |
| 3099 | const bool utf8_pat = RX_UTF8(prog) ? 1 : 0; |
| 3100 | |
| 3101 | PERL_ARGS_ASSERT_DEBUG_START_MATCH; |
| 3102 | |
| 3103 | if (!PL_colorset) |
| 3104 | reginitcolors(); |
| 3105 | { |
| 3106 | RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0), |
| 3107 | RX_PRECOMP_const(prog), RX_PRELEN(prog), 60); |
| 3108 | |
| 3109 | RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1), |
| 3110 | start, end - start, 60); |
| 3111 | |
| 3112 | PerlIO_printf(Perl_debug_log, |
| 3113 | "%s%s REx%s %s against %s\n", |
| 3114 | PL_colors[4], blurb, PL_colors[5], s0, s1); |
| 3115 | |
| 3116 | if (utf8_target||utf8_pat) |
| 3117 | PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n", |
| 3118 | utf8_pat ? "pattern" : "", |
| 3119 | utf8_pat && utf8_target ? " and " : "", |
| 3120 | utf8_target ? "string" : "" |
| 3121 | ); |
| 3122 | } |
| 3123 | } |
| 3124 | |
| 3125 | STATIC void |
| 3126 | S_dump_exec_pos(pTHX_ const char *locinput, |
| 3127 | const regnode *scan, |
| 3128 | const char *loc_regeol, |
| 3129 | const char *loc_bostr, |
| 3130 | const char *loc_reg_starttry, |
| 3131 | const bool utf8_target) |
| 3132 | { |
| 3133 | const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4]; |
| 3134 | const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */ |
| 3135 | int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput); |
| 3136 | /* The part of the string before starttry has one color |
| 3137 | (pref0_len chars), between starttry and current |
| 3138 | position another one (pref_len - pref0_len chars), |
| 3139 | after the current position the third one. |
| 3140 | We assume that pref0_len <= pref_len, otherwise we |
| 3141 | decrease pref0_len. */ |
| 3142 | int pref_len = (locinput - loc_bostr) > (5 + taill) - l |
| 3143 | ? (5 + taill) - l : locinput - loc_bostr; |
| 3144 | int pref0_len; |
| 3145 | |
| 3146 | PERL_ARGS_ASSERT_DUMP_EXEC_POS; |
| 3147 | |
| 3148 | while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len))) |
| 3149 | pref_len++; |
| 3150 | pref0_len = pref_len - (locinput - loc_reg_starttry); |
| 3151 | if (l + pref_len < (5 + taill) && l < loc_regeol - locinput) |
| 3152 | l = ( loc_regeol - locinput > (5 + taill) - pref_len |
| 3153 | ? (5 + taill) - pref_len : loc_regeol - locinput); |
| 3154 | while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l))) |
| 3155 | l--; |
| 3156 | if (pref0_len < 0) |
| 3157 | pref0_len = 0; |
| 3158 | if (pref0_len > pref_len) |
| 3159 | pref0_len = pref_len; |
| 3160 | { |
| 3161 | const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0; |
| 3162 | |
| 3163 | RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0), |
| 3164 | (locinput - pref_len),pref0_len, 60, 4, 5); |
| 3165 | |
| 3166 | RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1), |
| 3167 | (locinput - pref_len + pref0_len), |
| 3168 | pref_len - pref0_len, 60, 2, 3); |
| 3169 | |
| 3170 | RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2), |
| 3171 | locinput, loc_regeol - locinput, 10, 0, 1); |
| 3172 | |
| 3173 | const STRLEN tlen=len0+len1+len2; |
| 3174 | PerlIO_printf(Perl_debug_log, |
| 3175 | "%4"IVdf" <%.*s%.*s%s%.*s>%*s|", |
| 3176 | (IV)(locinput - loc_bostr), |
| 3177 | len0, s0, |
| 3178 | len1, s1, |
| 3179 | (docolor ? "" : "> <"), |
| 3180 | len2, s2, |
| 3181 | (int)(tlen > 19 ? 0 : 19 - tlen), |
| 3182 | ""); |
| 3183 | } |
| 3184 | } |
| 3185 | |
| 3186 | #endif |
| 3187 | |
| 3188 | /* reg_check_named_buff_matched() |
| 3189 | * Checks to see if a named buffer has matched. The data array of |
| 3190 | * buffer numbers corresponding to the buffer is expected to reside |
| 3191 | * in the regexp->data->data array in the slot stored in the ARG() of |
| 3192 | * node involved. Note that this routine doesn't actually care about the |
| 3193 | * name, that information is not preserved from compilation to execution. |
| 3194 | * Returns the index of the leftmost defined buffer with the given name |
| 3195 | * or 0 if non of the buffers matched. |
| 3196 | */ |
| 3197 | STATIC I32 |
| 3198 | S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) |
| 3199 | { |
| 3200 | I32 n; |
| 3201 | RXi_GET_DECL(rex,rexi); |
| 3202 | SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 3203 | I32 *nums=(I32*)SvPVX(sv_dat); |
| 3204 | |
| 3205 | PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED; |
| 3206 | |
| 3207 | for ( n=0; n<SvIVX(sv_dat); n++ ) { |
| 3208 | if ((I32)rex->lastparen >= nums[n] && |
| 3209 | rex->offs[nums[n]].end != -1) |
| 3210 | { |
| 3211 | return nums[n]; |
| 3212 | } |
| 3213 | } |
| 3214 | return 0; |
| 3215 | } |
| 3216 | |
| 3217 | |
| 3218 | /* free all slabs above current one - called during LEAVE_SCOPE */ |
| 3219 | |
| 3220 | STATIC void |
| 3221 | S_clear_backtrack_stack(pTHX_ void *p) |
| 3222 | { |
| 3223 | regmatch_slab *s = PL_regmatch_slab->next; |
| 3224 | PERL_UNUSED_ARG(p); |
| 3225 | |
| 3226 | if (!s) |
| 3227 | return; |
| 3228 | PL_regmatch_slab->next = NULL; |
| 3229 | while (s) { |
| 3230 | regmatch_slab * const osl = s; |
| 3231 | s = s->next; |
| 3232 | Safefree(osl); |
| 3233 | } |
| 3234 | } |
| 3235 | static bool |
| 3236 | S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p, |
| 3237 | U8* c1_utf8, int *c2p, U8* c2_utf8, bool is_utf8_pat) |
| 3238 | { |
| 3239 | /* This function determines if there are one or two characters that match |
| 3240 | * the first character of the passed-in EXACTish node <text_node>, and if |
| 3241 | * so, returns them in the passed-in pointers. |
| 3242 | * |
| 3243 | * If it determines that no possible character in the target string can |
| 3244 | * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if |
| 3245 | * the first character in <text_node> requires UTF-8 to represent, and the |
| 3246 | * target string isn't in UTF-8.) |
| 3247 | * |
| 3248 | * If there are more than two characters that could match the beginning of |
| 3249 | * <text_node>, or if more context is required to determine a match or not, |
| 3250 | * it sets both *<c1p> and *<c2p> to CHRTEST_VOID. |
| 3251 | * |
| 3252 | * The motiviation behind this function is to allow the caller to set up |
| 3253 | * tight loops for matching. If <text_node> is of type EXACT, there is |
| 3254 | * only one possible character that can match its first character, and so |
| 3255 | * the situation is quite simple. But things get much more complicated if |
| 3256 | * folding is involved. It may be that the first character of an EXACTFish |
| 3257 | * node doesn't participate in any possible fold, e.g., punctuation, so it |
| 3258 | * can be matched only by itself. The vast majority of characters that are |
| 3259 | * in folds match just two things, their lower and upper-case equivalents. |
| 3260 | * But not all are like that; some have multiple possible matches, or match |
| 3261 | * sequences of more than one character. This function sorts all that out. |
| 3262 | * |
| 3263 | * Consider the patterns A*B or A*?B where A and B are arbitrary. In a |
| 3264 | * loop of trying to match A*, we know we can't exit where the thing |
| 3265 | * following it isn't a B. And something can't be a B unless it is the |
| 3266 | * beginning of B. By putting a quick test for that beginning in a tight |
| 3267 | * loop, we can rule out things that can't possibly be B without having to |
| 3268 | * break out of the loop, thus avoiding work. Similarly, if A is a single |
| 3269 | * character, we can make a tight loop matching A*, using the outputs of |
| 3270 | * this function. |
| 3271 | * |
| 3272 | * If the target string to match isn't in UTF-8, and there aren't |
| 3273 | * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to |
| 3274 | * the one or two possible octets (which are characters in this situation) |
| 3275 | * that can match. In all cases, if there is only one character that can |
| 3276 | * match, *<c1p> and *<c2p> will be identical. |
| 3277 | * |
| 3278 | * If the target string is in UTF-8, the buffers pointed to by <c1_utf8> |
| 3279 | * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that |
| 3280 | * can match the beginning of <text_node>. They should be declared with at |
| 3281 | * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is |
| 3282 | * undefined what these contain.) If one or both of the buffers are |
| 3283 | * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the |
| 3284 | * corresponding invariant. If variant, the corresponding *<c1p> and/or |
| 3285 | * *<c2p> will be set to a negative number(s) that shouldn't match any code |
| 3286 | * point (unless inappropriately coerced to unsigned). *<c1p> will equal |
| 3287 | * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */ |
| 3288 | |
| 3289 | const bool utf8_target = PL_reg_match_utf8; |
| 3290 | |
| 3291 | UV c1 = CHRTEST_NOT_A_CP_1; |
| 3292 | UV c2 = CHRTEST_NOT_A_CP_2; |
| 3293 | bool use_chrtest_void = FALSE; |
| 3294 | |
| 3295 | /* Used when we have both utf8 input and utf8 output, to avoid converting |
| 3296 | * to/from code points */ |
| 3297 | bool utf8_has_been_setup = FALSE; |
| 3298 | |
| 3299 | dVAR; |
| 3300 | |
| 3301 | U8 *pat = (U8*)STRING(text_node); |
| 3302 | |
| 3303 | if (OP(text_node) == EXACT) { |
| 3304 | |
| 3305 | /* In an exact node, only one thing can be matched, that first |
| 3306 | * character. If both the pat and the target are UTF-8, we can just |
| 3307 | * copy the input to the output, avoiding finding the code point of |
| 3308 | * that character */ |
| 3309 | if (!is_utf8_pat) { |
| 3310 | c2 = c1 = *pat; |
| 3311 | } |
| 3312 | else if (utf8_target) { |
| 3313 | Copy(pat, c1_utf8, UTF8SKIP(pat), U8); |
| 3314 | Copy(pat, c2_utf8, UTF8SKIP(pat), U8); |
| 3315 | utf8_has_been_setup = TRUE; |
| 3316 | } |
| 3317 | else { |
| 3318 | c2 = c1 = valid_utf8_to_uvchr(pat, NULL); |
| 3319 | } |
| 3320 | } |
| 3321 | else /* an EXACTFish node */ |
| 3322 | if ((is_utf8_pat |
| 3323 | && is_MULTI_CHAR_FOLD_utf8_safe(pat, |
| 3324 | pat + STR_LEN(text_node))) |
| 3325 | || (!is_utf8_pat |
| 3326 | && is_MULTI_CHAR_FOLD_latin1_safe(pat, |
| 3327 | pat + STR_LEN(text_node)))) |
| 3328 | { |
| 3329 | /* Multi-character folds require more context to sort out. Also |
| 3330 | * PL_utf8_foldclosures used below doesn't handle them, so have to be |
| 3331 | * handled outside this routine */ |
| 3332 | use_chrtest_void = TRUE; |
| 3333 | } |
| 3334 | else { /* an EXACTFish node which doesn't begin with a multi-char fold */ |
| 3335 | c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat; |
| 3336 | if (c1 > 256) { |
| 3337 | /* Load the folds hash, if not already done */ |
| 3338 | SV** listp; |
| 3339 | if (! PL_utf8_foldclosures) { |
| 3340 | if (! PL_utf8_tofold) { |
| 3341 | U8 dummy[UTF8_MAXBYTES+1]; |
| 3342 | |
| 3343 | /* Force loading this by folding an above-Latin1 char */ |
| 3344 | to_utf8_fold((U8*) HYPHEN_UTF8, dummy, NULL); |
| 3345 | assert(PL_utf8_tofold); /* Verify that worked */ |
| 3346 | } |
| 3347 | PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold); |
| 3348 | } |
| 3349 | |
| 3350 | /* The fold closures data structure is a hash with the keys being |
| 3351 | * the UTF-8 of every character that is folded to, like 'k', and |
| 3352 | * the values each an array of all code points that fold to its |
| 3353 | * key. e.g. [ 'k', 'K', KELVIN_SIGN ]. Multi-character folds are |
| 3354 | * not included */ |
| 3355 | if ((! (listp = hv_fetch(PL_utf8_foldclosures, |
| 3356 | (char *) pat, |
| 3357 | UTF8SKIP(pat), |
| 3358 | FALSE)))) |
| 3359 | { |
| 3360 | /* Not found in the hash, therefore there are no folds |
| 3361 | * containing it, so there is only a single character that |
| 3362 | * could match */ |
| 3363 | c2 = c1; |
| 3364 | } |
| 3365 | else { /* Does participate in folds */ |
| 3366 | AV* list = (AV*) *listp; |
| 3367 | if (av_len(list) != 1) { |
| 3368 | |
| 3369 | /* If there aren't exactly two folds to this, it is outside |
| 3370 | * the scope of this function */ |
| 3371 | use_chrtest_void = TRUE; |
| 3372 | } |
| 3373 | else { /* There are two. Get them */ |
| 3374 | SV** c_p = av_fetch(list, 0, FALSE); |
| 3375 | if (c_p == NULL) { |
| 3376 | Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); |
| 3377 | } |
| 3378 | c1 = SvUV(*c_p); |
| 3379 | |
| 3380 | c_p = av_fetch(list, 1, FALSE); |
| 3381 | if (c_p == NULL) { |
| 3382 | Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); |
| 3383 | } |
| 3384 | c2 = SvUV(*c_p); |
| 3385 | |
| 3386 | /* Folds that cross the 255/256 boundary are forbidden if |
| 3387 | * EXACTFL, or EXACTFA and one is ASCIII. Since the |
| 3388 | * pattern character is above 256, and its only other match |
| 3389 | * is below 256, the only legal match will be to itself. |
| 3390 | * We have thrown away the original, so have to compute |
| 3391 | * which is the one above 255 */ |
| 3392 | if ((c1 < 256) != (c2 < 256)) { |
| 3393 | if (OP(text_node) == EXACTFL |
| 3394 | || (OP(text_node) == EXACTFA |
| 3395 | && (isASCII(c1) || isASCII(c2)))) |
| 3396 | { |
| 3397 | if (c1 < 256) { |
| 3398 | c1 = c2; |
| 3399 | } |
| 3400 | else { |
| 3401 | c2 = c1; |
| 3402 | } |
| 3403 | } |
| 3404 | } |
| 3405 | } |
| 3406 | } |
| 3407 | } |
| 3408 | else /* Here, c1 is < 255 */ |
| 3409 | if (utf8_target |
| 3410 | && HAS_NONLATIN1_FOLD_CLOSURE(c1) |
| 3411 | && OP(text_node) != EXACTFL |
| 3412 | && (OP(text_node) != EXACTFA || ! isASCII(c1))) |
| 3413 | { |
| 3414 | /* Here, there could be something above Latin1 in the target which |
| 3415 | * folds to this character in the pattern. All such cases except |
| 3416 | * LATIN SMALL LETTER Y WITH DIAERESIS have more than two characters |
| 3417 | * involved in their folds, so are outside the scope of this |
| 3418 | * function */ |
| 3419 | if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) { |
| 3420 | c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS; |
| 3421 | } |
| 3422 | else { |
| 3423 | use_chrtest_void = TRUE; |
| 3424 | } |
| 3425 | } |
| 3426 | else { /* Here nothing above Latin1 can fold to the pattern character */ |
| 3427 | switch (OP(text_node)) { |
| 3428 | |
| 3429 | case EXACTFL: /* /l rules */ |
| 3430 | c2 = PL_fold_locale[c1]; |
| 3431 | break; |
| 3432 | |
| 3433 | case EXACTF: |
| 3434 | if (! utf8_target) { /* /d rules */ |
| 3435 | c2 = PL_fold[c1]; |
| 3436 | break; |
| 3437 | } |
| 3438 | /* FALLTHROUGH */ |
| 3439 | /* /u rules for all these. This happens to work for |
| 3440 | * EXACTFA as nothing in Latin1 folds to ASCII */ |
| 3441 | case EXACTFA: |
| 3442 | case EXACTFU_TRICKYFOLD: |
| 3443 | case EXACTFU_SS: |
| 3444 | case EXACTFU: |
| 3445 | c2 = PL_fold_latin1[c1]; |
| 3446 | break; |
| 3447 | |
| 3448 | default: |
| 3449 | Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node)); |
| 3450 | assert(0); /* NOTREACHED */ |
| 3451 | } |
| 3452 | } |
| 3453 | } |
| 3454 | |
| 3455 | /* Here have figured things out. Set up the returns */ |
| 3456 | if (use_chrtest_void) { |
| 3457 | *c2p = *c1p = CHRTEST_VOID; |
| 3458 | } |
| 3459 | else if (utf8_target) { |
| 3460 | if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */ |
| 3461 | uvchr_to_utf8(c1_utf8, c1); |
| 3462 | uvchr_to_utf8(c2_utf8, c2); |
| 3463 | } |
| 3464 | |
| 3465 | /* Invariants are stored in both the utf8 and byte outputs; Use |
| 3466 | * negative numbers otherwise for the byte ones. Make sure that the |
| 3467 | * byte ones are the same iff the utf8 ones are the same */ |
| 3468 | *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1; |
| 3469 | *c2p = (UTF8_IS_INVARIANT(*c2_utf8)) |
| 3470 | ? *c2_utf8 |
| 3471 | : (c1 == c2) |
| 3472 | ? CHRTEST_NOT_A_CP_1 |
| 3473 | : CHRTEST_NOT_A_CP_2; |
| 3474 | } |
| 3475 | else if (c1 > 255) { |
| 3476 | if (c2 > 255) { /* both possibilities are above what a non-utf8 string |
| 3477 | can represent */ |
| 3478 | return FALSE; |
| 3479 | } |
| 3480 | |
| 3481 | *c1p = *c2p = c2; /* c2 is the only representable value */ |
| 3482 | } |
| 3483 | else { /* c1 is representable; see about c2 */ |
| 3484 | *c1p = c1; |
| 3485 | *c2p = (c2 < 256) ? c2 : c1; |
| 3486 | } |
| 3487 | |
| 3488 | return TRUE; |
| 3489 | } |
| 3490 | |
| 3491 | /* returns -1 on failure, $+[0] on success */ |
| 3492 | STATIC I32 |
| 3493 | S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog) |
| 3494 | { |
| 3495 | #if PERL_VERSION < 9 && !defined(PERL_CORE) |
| 3496 | dMY_CXT; |
| 3497 | #endif |
| 3498 | dVAR; |
| 3499 | const bool utf8_target = PL_reg_match_utf8; |
| 3500 | const U32 uniflags = UTF8_ALLOW_DEFAULT; |
| 3501 | REGEXP *rex_sv = reginfo->prog; |
| 3502 | regexp *rex = ReANY(rex_sv); |
| 3503 | RXi_GET_DECL(rex,rexi); |
| 3504 | I32 oldsave; |
| 3505 | /* the current state. This is a cached copy of PL_regmatch_state */ |
| 3506 | regmatch_state *st; |
| 3507 | /* cache heavy used fields of st in registers */ |
| 3508 | regnode *scan; |
| 3509 | regnode *next; |
| 3510 | U32 n = 0; /* general value; init to avoid compiler warning */ |
| 3511 | I32 ln = 0; /* len or last; init to avoid compiler warning */ |
| 3512 | char *locinput = startpos; |
| 3513 | char *pushinput; /* where to continue after a PUSH */ |
| 3514 | I32 nextchr; /* is always set to UCHARAT(locinput) */ |
| 3515 | |
| 3516 | bool result = 0; /* return value of S_regmatch */ |
| 3517 | int depth = 0; /* depth of backtrack stack */ |
| 3518 | U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */ |
| 3519 | const U32 max_nochange_depth = |
| 3520 | (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ? |
| 3521 | 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH; |
| 3522 | regmatch_state *yes_state = NULL; /* state to pop to on success of |
| 3523 | subpattern */ |
| 3524 | /* mark_state piggy backs on the yes_state logic so that when we unwind |
| 3525 | the stack on success we can update the mark_state as we go */ |
| 3526 | regmatch_state *mark_state = NULL; /* last mark state we have seen */ |
| 3527 | regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */ |
| 3528 | struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */ |
| 3529 | U32 state_num; |
| 3530 | bool no_final = 0; /* prevent failure from backtracking? */ |
| 3531 | bool do_cutgroup = 0; /* no_final only until next branch/trie entry */ |
| 3532 | char *startpoint = locinput; |
| 3533 | SV *popmark = NULL; /* are we looking for a mark? */ |
| 3534 | SV *sv_commit = NULL; /* last mark name seen in failure */ |
| 3535 | SV *sv_yes_mark = NULL; /* last mark name we have seen |
| 3536 | during a successful match */ |
| 3537 | U32 lastopen = 0; /* last open we saw */ |
| 3538 | bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0; |
| 3539 | SV* const oreplsv = GvSV(PL_replgv); |
| 3540 | /* these three flags are set by various ops to signal information to |
| 3541 | * the very next op. They have a useful lifetime of exactly one loop |
| 3542 | * iteration, and are not preserved or restored by state pushes/pops |
| 3543 | */ |
| 3544 | bool sw = 0; /* the condition value in (?(cond)a|b) */ |
| 3545 | bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */ |
| 3546 | int logical = 0; /* the following EVAL is: |
| 3547 | 0: (?{...}) |
| 3548 | 1: (?(?{...})X|Y) |
| 3549 | 2: (??{...}) |
| 3550 | or the following IFMATCH/UNLESSM is: |
| 3551 | false: plain (?=foo) |
| 3552 | true: used as a condition: (?(?=foo)) |
| 3553 | */ |
| 3554 | PAD* last_pad = NULL; |
| 3555 | dMULTICALL; |
| 3556 | I32 gimme = G_SCALAR; |
| 3557 | CV *caller_cv = NULL; /* who called us */ |
| 3558 | CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */ |
| 3559 | CHECKPOINT runops_cp; /* savestack position before executing EVAL */ |
| 3560 | U32 maxopenparen = 0; /* max '(' index seen so far */ |
| 3561 | int to_complement; /* Invert the result? */ |
| 3562 | _char_class_number classnum; |
| 3563 | bool is_utf8_pat = reginfo->is_utf8_pat; |
| 3564 | |
| 3565 | #ifdef DEBUGGING |
| 3566 | GET_RE_DEBUG_FLAGS_DECL; |
| 3567 | #endif |
| 3568 | |
| 3569 | /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */ |
| 3570 | multicall_oldcatch = 0; |
| 3571 | multicall_cv = NULL; |
| 3572 | cx = NULL; |
| 3573 | PERL_UNUSED_VAR(multicall_cop); |
| 3574 | PERL_UNUSED_VAR(newsp); |
| 3575 | |
| 3576 | |
| 3577 | PERL_ARGS_ASSERT_REGMATCH; |
| 3578 | |
| 3579 | DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({ |
| 3580 | PerlIO_printf(Perl_debug_log,"regmatch start\n"); |
| 3581 | })); |
| 3582 | /* on first ever call to regmatch, allocate first slab */ |
| 3583 | if (!PL_regmatch_slab) { |
| 3584 | Newx(PL_regmatch_slab, 1, regmatch_slab); |
| 3585 | PL_regmatch_slab->prev = NULL; |
| 3586 | PL_regmatch_slab->next = NULL; |
| 3587 | PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab); |
| 3588 | } |
| 3589 | |
| 3590 | oldsave = PL_savestack_ix; |
| 3591 | SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL); |
| 3592 | SAVEVPTR(PL_regmatch_slab); |
| 3593 | SAVEVPTR(PL_regmatch_state); |
| 3594 | |
| 3595 | /* grab next free state slot */ |
| 3596 | st = ++PL_regmatch_state; |
| 3597 | if (st > SLAB_LAST(PL_regmatch_slab)) |
| 3598 | st = PL_regmatch_state = S_push_slab(aTHX); |
| 3599 | |
| 3600 | /* Note that nextchr is a byte even in UTF */ |
| 3601 | SET_nextchr; |
| 3602 | scan = prog; |
| 3603 | while (scan != NULL) { |
| 3604 | |
| 3605 | DEBUG_EXECUTE_r( { |
| 3606 | SV * const prop = sv_newmortal(); |
| 3607 | regnode *rnext=regnext(scan); |
| 3608 | DUMP_EXEC_POS( locinput, scan, utf8_target ); |
| 3609 | regprop(rex, prop, scan); |
| 3610 | |
| 3611 | PerlIO_printf(Perl_debug_log, |
| 3612 | "%3"IVdf":%*s%s(%"IVdf")\n", |
| 3613 | (IV)(scan - rexi->program), depth*2, "", |
| 3614 | SvPVX_const(prop), |
| 3615 | (PL_regkind[OP(scan)] == END || !rnext) ? |
| 3616 | 0 : (IV)(rnext - rexi->program)); |
| 3617 | }); |
| 3618 | |
| 3619 | next = scan + NEXT_OFF(scan); |
| 3620 | if (next == scan) |
| 3621 | next = NULL; |
| 3622 | state_num = OP(scan); |
| 3623 | |
| 3624 | reenter_switch: |
| 3625 | to_complement = 0; |
| 3626 | |
| 3627 | SET_nextchr; |
| 3628 | assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS)); |
| 3629 | |
| 3630 | switch (state_num) { |
| 3631 | case BOL: /* /^../ */ |
| 3632 | if (locinput == PL_bostr) |
| 3633 | { |
| 3634 | /* reginfo->till = reginfo->bol; */ |
| 3635 | break; |
| 3636 | } |
| 3637 | sayNO; |
| 3638 | |
| 3639 | case MBOL: /* /^../m */ |
| 3640 | if (locinput == PL_bostr || |
| 3641 | (!NEXTCHR_IS_EOS && locinput[-1] == '\n')) |
| 3642 | { |
| 3643 | break; |
| 3644 | } |
| 3645 | sayNO; |
| 3646 | |
| 3647 | case SBOL: /* /^../s */ |
| 3648 | if (locinput == PL_bostr) |
| 3649 | break; |
| 3650 | sayNO; |
| 3651 | |
| 3652 | case GPOS: /* \G */ |
| 3653 | if (locinput == reginfo->ganch) |
| 3654 | break; |
| 3655 | sayNO; |
| 3656 | |
| 3657 | case KEEPS: /* \K */ |
| 3658 | /* update the startpoint */ |
| 3659 | st->u.keeper.val = rex->offs[0].start; |
| 3660 | rex->offs[0].start = locinput - PL_bostr; |
| 3661 | PUSH_STATE_GOTO(KEEPS_next, next, locinput); |
| 3662 | assert(0); /*NOTREACHED*/ |
| 3663 | case KEEPS_next_fail: |
| 3664 | /* rollback the start point change */ |
| 3665 | rex->offs[0].start = st->u.keeper.val; |
| 3666 | sayNO_SILENT; |
| 3667 | assert(0); /*NOTREACHED*/ |
| 3668 | |
| 3669 | case EOL: /* /..$/ */ |
| 3670 | goto seol; |
| 3671 | |
| 3672 | case MEOL: /* /..$/m */ |
| 3673 | if (!NEXTCHR_IS_EOS && nextchr != '\n') |
| 3674 | sayNO; |
| 3675 | break; |
| 3676 | |
| 3677 | case SEOL: /* /..$/s */ |
| 3678 | seol: |
| 3679 | if (!NEXTCHR_IS_EOS && nextchr != '\n') |
| 3680 | sayNO; |
| 3681 | if (PL_regeol - locinput > 1) |
| 3682 | sayNO; |
| 3683 | break; |
| 3684 | |
| 3685 | case EOS: /* \z */ |
| 3686 | if (!NEXTCHR_IS_EOS) |
| 3687 | sayNO; |
| 3688 | break; |
| 3689 | |
| 3690 | case SANY: /* /./s */ |
| 3691 | if (NEXTCHR_IS_EOS) |
| 3692 | sayNO; |
| 3693 | goto increment_locinput; |
| 3694 | |
| 3695 | case CANY: /* \C */ |
| 3696 | if (NEXTCHR_IS_EOS) |
| 3697 | sayNO; |
| 3698 | locinput++; |
| 3699 | break; |
| 3700 | |
| 3701 | case REG_ANY: /* /./ */ |
| 3702 | if ((NEXTCHR_IS_EOS) || nextchr == '\n') |
| 3703 | sayNO; |
| 3704 | goto increment_locinput; |
| 3705 | |
| 3706 | |
| 3707 | #undef ST |
| 3708 | #define ST st->u.trie |
| 3709 | case TRIEC: /* (ab|cd) with known charclass */ |
| 3710 | /* In this case the charclass data is available inline so |
| 3711 | we can fail fast without a lot of extra overhead. |
| 3712 | */ |
| 3713 | if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) { |
| 3714 | DEBUG_EXECUTE_r( |
| 3715 | PerlIO_printf(Perl_debug_log, |
| 3716 | "%*s %sfailed to match trie start class...%s\n", |
| 3717 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) |
| 3718 | ); |
| 3719 | sayNO_SILENT; |
| 3720 | assert(0); /* NOTREACHED */ |
| 3721 | } |
| 3722 | /* FALL THROUGH */ |
| 3723 | case TRIE: /* (ab|cd) */ |
| 3724 | /* the basic plan of execution of the trie is: |
| 3725 | * At the beginning, run though all the states, and |
| 3726 | * find the longest-matching word. Also remember the position |
| 3727 | * of the shortest matching word. For example, this pattern: |
| 3728 | * 1 2 3 4 5 |
| 3729 | * ab|a|x|abcd|abc |
| 3730 | * when matched against the string "abcde", will generate |
| 3731 | * accept states for all words except 3, with the longest |
| 3732 | * matching word being 4, and the shortest being 2 (with |
| 3733 | * the position being after char 1 of the string). |
| 3734 | * |
| 3735 | * Then for each matching word, in word order (i.e. 1,2,4,5), |
| 3736 | * we run the remainder of the pattern; on each try setting |
| 3737 | * the current position to the character following the word, |
| 3738 | * returning to try the next word on failure. |
| 3739 | * |
| 3740 | * We avoid having to build a list of words at runtime by |
| 3741 | * using a compile-time structure, wordinfo[].prev, which |
| 3742 | * gives, for each word, the previous accepting word (if any). |
| 3743 | * In the case above it would contain the mappings 1->2, 2->0, |
| 3744 | * 3->0, 4->5, 5->1. We can use this table to generate, from |
| 3745 | * the longest word (4 above), a list of all words, by |
| 3746 | * following the list of prev pointers; this gives us the |
| 3747 | * unordered list 4,5,1,2. Then given the current word we have |
| 3748 | * just tried, we can go through the list and find the |
| 3749 | * next-biggest word to try (so if we just failed on word 2, |
| 3750 | * the next in the list is 4). |
| 3751 | * |
| 3752 | * Since at runtime we don't record the matching position in |
| 3753 | * the string for each word, we have to work that out for |
| 3754 | * each word we're about to process. The wordinfo table holds |
| 3755 | * the character length of each word; given that we recorded |
| 3756 | * at the start: the position of the shortest word and its |
| 3757 | * length in chars, we just need to move the pointer the |
| 3758 | * difference between the two char lengths. Depending on |
| 3759 | * Unicode status and folding, that's cheap or expensive. |
| 3760 | * |
| 3761 | * This algorithm is optimised for the case where are only a |
| 3762 | * small number of accept states, i.e. 0,1, or maybe 2. |
| 3763 | * With lots of accepts states, and having to try all of them, |
| 3764 | * it becomes quadratic on number of accept states to find all |
| 3765 | * the next words. |
| 3766 | */ |
| 3767 | |
| 3768 | { |
| 3769 | /* what type of TRIE am I? (utf8 makes this contextual) */ |
| 3770 | DECL_TRIE_TYPE(scan); |
| 3771 | |
| 3772 | /* what trie are we using right now */ |
| 3773 | reg_trie_data * const trie |
| 3774 | = (reg_trie_data*)rexi->data->data[ ARG( scan ) ]; |
| 3775 | HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]); |
| 3776 | U32 state = trie->startstate; |
| 3777 | |
| 3778 | if ( trie->bitmap |
| 3779 | && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr))) |
| 3780 | { |
| 3781 | if (trie->states[ state ].wordnum) { |
| 3782 | DEBUG_EXECUTE_r( |
| 3783 | PerlIO_printf(Perl_debug_log, |
| 3784 | "%*s %smatched empty string...%s\n", |
| 3785 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) |
| 3786 | ); |
| 3787 | if (!trie->jump) |
| 3788 | break; |
| 3789 | } else { |
| 3790 | DEBUG_EXECUTE_r( |
| 3791 | PerlIO_printf(Perl_debug_log, |
| 3792 | "%*s %sfailed to match trie start class...%s\n", |
| 3793 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) |
| 3794 | ); |
| 3795 | sayNO_SILENT; |
| 3796 | } |
| 3797 | } |
| 3798 | |
| 3799 | { |
| 3800 | U8 *uc = ( U8* )locinput; |
| 3801 | |
| 3802 | STRLEN len = 0; |
| 3803 | STRLEN foldlen = 0; |
| 3804 | U8 *uscan = (U8*)NULL; |
| 3805 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
| 3806 | U32 charcount = 0; /* how many input chars we have matched */ |
| 3807 | U32 accepted = 0; /* have we seen any accepting states? */ |
| 3808 | |
| 3809 | ST.jump = trie->jump; |
| 3810 | ST.me = scan; |
| 3811 | ST.firstpos = NULL; |
| 3812 | ST.longfold = FALSE; /* char longer if folded => it's harder */ |
| 3813 | ST.nextword = 0; |
| 3814 | |
| 3815 | /* fully traverse the TRIE; note the position of the |
| 3816 | shortest accept state and the wordnum of the longest |
| 3817 | accept state */ |
| 3818 | |
| 3819 | while ( state && uc <= (U8*)PL_regeol ) { |
| 3820 | U32 base = trie->states[ state ].trans.base; |
| 3821 | UV uvc = 0; |
| 3822 | U16 charid = 0; |
| 3823 | U16 wordnum; |
| 3824 | wordnum = trie->states[ state ].wordnum; |
| 3825 | |
| 3826 | if (wordnum) { /* it's an accept state */ |
| 3827 | if (!accepted) { |
| 3828 | accepted = 1; |
| 3829 | /* record first match position */ |
| 3830 | if (ST.longfold) { |
| 3831 | ST.firstpos = (U8*)locinput; |
| 3832 | ST.firstchars = 0; |
| 3833 | } |
| 3834 | else { |
| 3835 | ST.firstpos = uc; |
| 3836 | ST.firstchars = charcount; |
| 3837 | } |
| 3838 | } |
| 3839 | if (!ST.nextword || wordnum < ST.nextword) |
| 3840 | ST.nextword = wordnum; |
| 3841 | ST.topword = wordnum; |
| 3842 | } |
| 3843 | |
| 3844 | DEBUG_TRIE_EXECUTE_r({ |
| 3845 | DUMP_EXEC_POS( (char *)uc, scan, utf8_target ); |
| 3846 | PerlIO_printf( Perl_debug_log, |
| 3847 | "%*s %sState: %4"UVxf" Accepted: %c ", |
| 3848 | 2+depth * 2, "", PL_colors[4], |
| 3849 | (UV)state, (accepted ? 'Y' : 'N')); |
| 3850 | }); |
| 3851 | |
| 3852 | /* read a char and goto next state */ |
| 3853 | if ( base && (foldlen || uc < (U8*)PL_regeol)) { |
| 3854 | I32 offset; |
| 3855 | REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, |
| 3856 | uscan, len, uvc, charid, foldlen, |
| 3857 | foldbuf, uniflags); |
| 3858 | charcount++; |
| 3859 | if (foldlen>0) |
| 3860 | ST.longfold = TRUE; |
| 3861 | if (charid && |
| 3862 | ( ((offset = |
| 3863 | base + charid - 1 - trie->uniquecharcount)) >= 0) |
| 3864 | |
| 3865 | && ((U32)offset < trie->lasttrans) |
| 3866 | && trie->trans[offset].check == state) |
| 3867 | { |
| 3868 | state = trie->trans[offset].next; |
| 3869 | } |
| 3870 | else { |
| 3871 | state = 0; |
| 3872 | } |
| 3873 | uc += len; |
| 3874 | |
| 3875 | } |
| 3876 | else { |
| 3877 | state = 0; |
| 3878 | } |
| 3879 | DEBUG_TRIE_EXECUTE_r( |
| 3880 | PerlIO_printf( Perl_debug_log, |
| 3881 | "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n", |
| 3882 | charid, uvc, (UV)state, PL_colors[5] ); |
| 3883 | ); |
| 3884 | } |
| 3885 | if (!accepted) |
| 3886 | sayNO; |
| 3887 | |
| 3888 | /* calculate total number of accept states */ |
| 3889 | { |
| 3890 | U16 w = ST.topword; |
| 3891 | accepted = 0; |
| 3892 | while (w) { |
| 3893 | w = trie->wordinfo[w].prev; |
| 3894 | accepted++; |
| 3895 | } |
| 3896 | ST.accepted = accepted; |
| 3897 | } |
| 3898 | |
| 3899 | DEBUG_EXECUTE_r( |
| 3900 | PerlIO_printf( Perl_debug_log, |
| 3901 | "%*s %sgot %"IVdf" possible matches%s\n", |
| 3902 | REPORT_CODE_OFF + depth * 2, "", |
| 3903 | PL_colors[4], (IV)ST.accepted, PL_colors[5] ); |
| 3904 | ); |
| 3905 | goto trie_first_try; /* jump into the fail handler */ |
| 3906 | }} |
| 3907 | assert(0); /* NOTREACHED */ |
| 3908 | |
| 3909 | case TRIE_next_fail: /* we failed - try next alternative */ |
| 3910 | { |
| 3911 | U8 *uc; |
| 3912 | if ( ST.jump) { |
| 3913 | REGCP_UNWIND(ST.cp); |
| 3914 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
| 3915 | } |
| 3916 | if (!--ST.accepted) { |
| 3917 | DEBUG_EXECUTE_r({ |
| 3918 | PerlIO_printf( Perl_debug_log, |
| 3919 | "%*s %sTRIE failed...%s\n", |
| 3920 | REPORT_CODE_OFF+depth*2, "", |
| 3921 | PL_colors[4], |
| 3922 | PL_colors[5] ); |
| 3923 | }); |
| 3924 | sayNO_SILENT; |
| 3925 | } |
| 3926 | { |
| 3927 | /* Find next-highest word to process. Note that this code |
| 3928 | * is O(N^2) per trie run (O(N) per branch), so keep tight */ |
| 3929 | U16 min = 0; |
| 3930 | U16 word; |
| 3931 | U16 const nextword = ST.nextword; |
| 3932 | reg_trie_wordinfo * const wordinfo |
| 3933 | = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo; |
| 3934 | for (word=ST.topword; word; word=wordinfo[word].prev) { |
| 3935 | if (word > nextword && (!min || word < min)) |
| 3936 | min = word; |
| 3937 | } |
| 3938 | ST.nextword = min; |
| 3939 | } |
| 3940 | |
| 3941 | trie_first_try: |
| 3942 | if (do_cutgroup) { |
| 3943 | do_cutgroup = 0; |
| 3944 | no_final = 0; |
| 3945 | } |
| 3946 | |
| 3947 | if ( ST.jump) { |
| 3948 | ST.lastparen = rex->lastparen; |
| 3949 | ST.lastcloseparen = rex->lastcloseparen; |
| 3950 | REGCP_SET(ST.cp); |
| 3951 | } |
| 3952 | |
| 3953 | /* find start char of end of current word */ |
| 3954 | { |
| 3955 | U32 chars; /* how many chars to skip */ |
| 3956 | reg_trie_data * const trie |
| 3957 | = (reg_trie_data*)rexi->data->data[ARG(ST.me)]; |
| 3958 | |
| 3959 | assert((trie->wordinfo[ST.nextword].len - trie->prefixlen) |
| 3960 | >= ST.firstchars); |
| 3961 | chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen) |
| 3962 | - ST.firstchars; |
| 3963 | uc = ST.firstpos; |
| 3964 | |
| 3965 | if (ST.longfold) { |
| 3966 | /* the hard option - fold each char in turn and find |
| 3967 | * its folded length (which may be different */ |
| 3968 | U8 foldbuf[UTF8_MAXBYTES_CASE + 1]; |
| 3969 | STRLEN foldlen; |
| 3970 | STRLEN len; |
| 3971 | UV uvc; |
| 3972 | U8 *uscan; |
| 3973 | |
| 3974 | while (chars) { |
| 3975 | if (utf8_target) { |
| 3976 | uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len, |
| 3977 | uniflags); |
| 3978 | uc += len; |
| 3979 | } |
| 3980 | else { |
| 3981 | uvc = *uc; |
| 3982 | uc++; |
| 3983 | } |
| 3984 | uvc = to_uni_fold(uvc, foldbuf, &foldlen); |
| 3985 | uscan = foldbuf; |
| 3986 | while (foldlen) { |
| 3987 | if (!--chars) |
| 3988 | break; |
| 3989 | uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len, |
| 3990 | uniflags); |
| 3991 | uscan += len; |
| 3992 | foldlen -= len; |
| 3993 | } |
| 3994 | } |
| 3995 | } |
| 3996 | else { |
| 3997 | if (utf8_target) |
| 3998 | while (chars--) |
| 3999 | uc += UTF8SKIP(uc); |
| 4000 | else |
| 4001 | uc += chars; |
| 4002 | } |
| 4003 | } |
| 4004 | |
| 4005 | scan = ST.me + ((ST.jump && ST.jump[ST.nextword]) |
| 4006 | ? ST.jump[ST.nextword] |
| 4007 | : NEXT_OFF(ST.me)); |
| 4008 | |
| 4009 | DEBUG_EXECUTE_r({ |
| 4010 | PerlIO_printf( Perl_debug_log, |
| 4011 | "%*s %sTRIE matched word #%d, continuing%s\n", |
| 4012 | REPORT_CODE_OFF+depth*2, "", |
| 4013 | PL_colors[4], |
| 4014 | ST.nextword, |
| 4015 | PL_colors[5] |
| 4016 | ); |
| 4017 | }); |
| 4018 | |
| 4019 | if (ST.accepted > 1 || has_cutgroup) { |
| 4020 | PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc); |
| 4021 | assert(0); /* NOTREACHED */ |
| 4022 | } |
| 4023 | /* only one choice left - just continue */ |
| 4024 | DEBUG_EXECUTE_r({ |
| 4025 | AV *const trie_words |
| 4026 | = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]); |
| 4027 | SV ** const tmp = av_fetch( trie_words, |
| 4028 | ST.nextword-1, 0 ); |
| 4029 | SV *sv= tmp ? sv_newmortal() : NULL; |
| 4030 | |
| 4031 | PerlIO_printf( Perl_debug_log, |
| 4032 | "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n", |
| 4033 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], |
| 4034 | ST.nextword, |
| 4035 | tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0, |
| 4036 | PL_colors[0], PL_colors[1], |
| 4037 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII |
| 4038 | ) |
| 4039 | : "not compiled under -Dr", |
| 4040 | PL_colors[5] ); |
| 4041 | }); |
| 4042 | |
| 4043 | locinput = (char*)uc; |
| 4044 | continue; /* execute rest of RE */ |
| 4045 | assert(0); /* NOTREACHED */ |
| 4046 | } |
| 4047 | #undef ST |
| 4048 | |
| 4049 | case EXACT: { /* /abc/ */ |
| 4050 | char *s = STRING(scan); |
| 4051 | ln = STR_LEN(scan); |
| 4052 | if (utf8_target != is_utf8_pat) { |
| 4053 | /* The target and the pattern have differing utf8ness. */ |
| 4054 | char *l = locinput; |
| 4055 | const char * const e = s + ln; |
| 4056 | |
| 4057 | if (utf8_target) { |
| 4058 | /* The target is utf8, the pattern is not utf8. |
| 4059 | * Above-Latin1 code points can't match the pattern; |
| 4060 | * invariants match exactly, and the other Latin1 ones need |
| 4061 | * to be downgraded to a single byte in order to do the |
| 4062 | * comparison. (If we could be confident that the target |
| 4063 | * is not malformed, this could be refactored to have fewer |
| 4064 | * tests by just assuming that if the first bytes match, it |
| 4065 | * is an invariant, but there are tests in the test suite |
| 4066 | * dealing with (??{...}) which violate this) */ |
| 4067 | while (s < e) { |
| 4068 | if (l >= PL_regeol || UTF8_IS_ABOVE_LATIN1(* (U8*) l)) { |
| 4069 | sayNO; |
| 4070 | } |
| 4071 | if (UTF8_IS_INVARIANT(*(U8*)l)) { |
| 4072 | if (*l != *s) { |
| 4073 | sayNO; |
| 4074 | } |
| 4075 | l++; |
| 4076 | } |
| 4077 | else { |
| 4078 | if (TWO_BYTE_UTF8_TO_UNI(*l, *(l+1)) != * (U8*) s) { |
| 4079 | sayNO; |
| 4080 | } |
| 4081 | l += 2; |
| 4082 | } |
| 4083 | s++; |
| 4084 | } |
| 4085 | } |
| 4086 | else { |
| 4087 | /* The target is not utf8, the pattern is utf8. */ |
| 4088 | while (s < e) { |
| 4089 | if (l >= PL_regeol || UTF8_IS_ABOVE_LATIN1(* (U8*) s)) |
| 4090 | { |
| 4091 | sayNO; |
| 4092 | } |
| 4093 | if (UTF8_IS_INVARIANT(*(U8*)s)) { |
| 4094 | if (*s != *l) { |
| 4095 | sayNO; |
| 4096 | } |
| 4097 | s++; |
| 4098 | } |
| 4099 | else { |
| 4100 | if (TWO_BYTE_UTF8_TO_UNI(*s, *(s+1)) != * (U8*) l) { |
| 4101 | sayNO; |
| 4102 | } |
| 4103 | s += 2; |
| 4104 | } |
| 4105 | l++; |
| 4106 | } |
| 4107 | } |
| 4108 | locinput = l; |
| 4109 | } |
| 4110 | else { |
| 4111 | /* The target and the pattern have the same utf8ness. */ |
| 4112 | /* Inline the first character, for speed. */ |
| 4113 | if (PL_regeol - locinput < ln |
| 4114 | || UCHARAT(s) != nextchr |
| 4115 | || (ln > 1 && memNE(s, locinput, ln))) |
| 4116 | { |
| 4117 | sayNO; |
| 4118 | } |
| 4119 | locinput += ln; |
| 4120 | } |
| 4121 | break; |
| 4122 | } |
| 4123 | |
| 4124 | case EXACTFL: { /* /abc/il */ |
| 4125 | re_fold_t folder; |
| 4126 | const U8 * fold_array; |
| 4127 | const char * s; |
| 4128 | U32 fold_utf8_flags; |
| 4129 | |
| 4130 | RX_MATCH_TAINTED_on(reginfo->prog); |
| 4131 | folder = foldEQ_locale; |
| 4132 | fold_array = PL_fold_locale; |
| 4133 | fold_utf8_flags = FOLDEQ_UTF8_LOCALE; |
| 4134 | goto do_exactf; |
| 4135 | |
| 4136 | case EXACTFU_SS: /* /\x{df}/iu */ |
| 4137 | case EXACTFU_TRICKYFOLD: /* /\x{390}/iu */ |
| 4138 | case EXACTFU: /* /abc/iu */ |
| 4139 | folder = foldEQ_latin1; |
| 4140 | fold_array = PL_fold_latin1; |
| 4141 | fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0; |
| 4142 | goto do_exactf; |
| 4143 | |
| 4144 | case EXACTFA: /* /abc/iaa */ |
| 4145 | folder = foldEQ_latin1; |
| 4146 | fold_array = PL_fold_latin1; |
| 4147 | fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 4148 | goto do_exactf; |
| 4149 | |
| 4150 | case EXACTF: /* /abc/i */ |
| 4151 | folder = foldEQ; |
| 4152 | fold_array = PL_fold; |
| 4153 | fold_utf8_flags = 0; |
| 4154 | |
| 4155 | do_exactf: |
| 4156 | s = STRING(scan); |
| 4157 | ln = STR_LEN(scan); |
| 4158 | |
| 4159 | if (utf8_target || is_utf8_pat || state_num == EXACTFU_SS) { |
| 4160 | /* Either target or the pattern are utf8, or has the issue where |
| 4161 | * the fold lengths may differ. */ |
| 4162 | const char * const l = locinput; |
| 4163 | char *e = PL_regeol; |
| 4164 | |
| 4165 | if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat, |
| 4166 | l, &e, 0, utf8_target, fold_utf8_flags)) |
| 4167 | { |
| 4168 | sayNO; |
| 4169 | } |
| 4170 | locinput = e; |
| 4171 | break; |
| 4172 | } |
| 4173 | |
| 4174 | /* Neither the target nor the pattern are utf8 */ |
| 4175 | if (UCHARAT(s) != nextchr |
| 4176 | && !NEXTCHR_IS_EOS |
| 4177 | && UCHARAT(s) != fold_array[nextchr]) |
| 4178 | { |
| 4179 | sayNO; |
| 4180 | } |
| 4181 | if (PL_regeol - locinput < ln) |
| 4182 | sayNO; |
| 4183 | if (ln > 1 && ! folder(s, locinput, ln)) |
| 4184 | sayNO; |
| 4185 | locinput += ln; |
| 4186 | break; |
| 4187 | } |
| 4188 | |
| 4189 | /* XXX Could improve efficiency by separating these all out using a |
| 4190 | * macro or in-line function. At that point regcomp.c would no longer |
| 4191 | * have to set the FLAGS fields of these */ |
| 4192 | case BOUNDL: /* /\b/l */ |
| 4193 | case NBOUNDL: /* /\B/l */ |
| 4194 | RX_MATCH_TAINTED_on(reginfo->prog); |
| 4195 | /* FALL THROUGH */ |
| 4196 | case BOUND: /* /\b/ */ |
| 4197 | case BOUNDU: /* /\b/u */ |
| 4198 | case BOUNDA: /* /\b/a */ |
| 4199 | case NBOUND: /* /\B/ */ |
| 4200 | case NBOUNDU: /* /\B/u */ |
| 4201 | case NBOUNDA: /* /\B/a */ |
| 4202 | /* was last char in word? */ |
| 4203 | if (utf8_target |
| 4204 | && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET |
| 4205 | && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET) |
| 4206 | { |
| 4207 | if (locinput == PL_bostr) |
| 4208 | ln = '\n'; |
| 4209 | else { |
| 4210 | const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr); |
| 4211 | |
| 4212 | ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags); |
| 4213 | } |
| 4214 | if (FLAGS(scan) != REGEX_LOCALE_CHARSET) { |
| 4215 | ln = isWORDCHAR_uni(ln); |
| 4216 | if (NEXTCHR_IS_EOS) |
| 4217 | n = 0; |
| 4218 | else { |
| 4219 | LOAD_UTF8_CHARCLASS_ALNUM(); |
| 4220 | n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput, |
| 4221 | utf8_target); |
| 4222 | } |
| 4223 | } |
| 4224 | else { |
| 4225 | ln = isWORDCHAR_LC_uvchr(UNI_TO_NATIVE(ln)); |
| 4226 | n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput); |
| 4227 | } |
| 4228 | } |
| 4229 | else { |
| 4230 | |
| 4231 | /* Here the string isn't utf8, or is utf8 and only ascii |
| 4232 | * characters are to match \w. In the latter case looking at |
| 4233 | * the byte just prior to the current one may be just the final |
| 4234 | * byte of a multi-byte character. This is ok. There are two |
| 4235 | * cases: |
| 4236 | * 1) it is a single byte character, and then the test is doing |
| 4237 | * just what it's supposed to. |
| 4238 | * 2) it is a multi-byte character, in which case the final |
| 4239 | * byte is never mistakable for ASCII, and so the test |
| 4240 | * will say it is not a word character, which is the |
| 4241 | * correct answer. */ |
| 4242 | ln = (locinput != PL_bostr) ? |
| 4243 | UCHARAT(locinput - 1) : '\n'; |
| 4244 | switch (FLAGS(scan)) { |
| 4245 | case REGEX_UNICODE_CHARSET: |
| 4246 | ln = isWORDCHAR_L1(ln); |
| 4247 | n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr); |
| 4248 | break; |
| 4249 | case REGEX_LOCALE_CHARSET: |
| 4250 | ln = isWORDCHAR_LC(ln); |
| 4251 | n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr); |
| 4252 | break; |
| 4253 | case REGEX_DEPENDS_CHARSET: |
| 4254 | ln = isWORDCHAR(ln); |
| 4255 | n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr); |
| 4256 | break; |
| 4257 | case REGEX_ASCII_RESTRICTED_CHARSET: |
| 4258 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
| 4259 | ln = isWORDCHAR_A(ln); |
| 4260 | n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr); |
| 4261 | break; |
| 4262 | default: |
| 4263 | Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan)); |
| 4264 | break; |
| 4265 | } |
| 4266 | } |
| 4267 | /* Note requires that all BOUNDs be lower than all NBOUNDs in |
| 4268 | * regcomp.sym */ |
| 4269 | if (((!ln) == (!n)) == (OP(scan) < NBOUND)) |
| 4270 | sayNO; |
| 4271 | break; |
| 4272 | |
| 4273 | case ANYOF: /* /[abc]/ */ |
| 4274 | case ANYOF_WARN_SUPER: |
| 4275 | if (NEXTCHR_IS_EOS) |
| 4276 | sayNO; |
| 4277 | if (utf8_target) { |
| 4278 | if (!reginclass(rex, scan, (U8*)locinput, utf8_target)) |
| 4279 | sayNO; |
| 4280 | locinput += UTF8SKIP(locinput); |
| 4281 | } |
| 4282 | else { |
| 4283 | if (!REGINCLASS(rex, scan, (U8*)locinput)) |
| 4284 | sayNO; |
| 4285 | locinput++; |
| 4286 | } |
| 4287 | break; |
| 4288 | |
| 4289 | /* The argument (FLAGS) to all the POSIX node types is the class number |
| 4290 | * */ |
| 4291 | |
| 4292 | case NPOSIXL: /* \W or [:^punct:] etc. under /l */ |
| 4293 | to_complement = 1; |
| 4294 | /* FALLTHROUGH */ |
| 4295 | |
| 4296 | case POSIXL: /* \w or [:punct:] etc. under /l */ |
| 4297 | if (NEXTCHR_IS_EOS) |
| 4298 | sayNO; |
| 4299 | |
| 4300 | /* The locale hasn't influenced the outcome before this, so defer |
| 4301 | * tainting until now */ |
| 4302 | RX_MATCH_TAINTED_on(reginfo->prog); |
| 4303 | |
| 4304 | /* Use isFOO_lc() for characters within Latin1. (Note that |
| 4305 | * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else |
| 4306 | * wouldn't be invariant) */ |
| 4307 | if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) { |
| 4308 | if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) { |
| 4309 | sayNO; |
| 4310 | } |
| 4311 | } |
| 4312 | else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) { |
| 4313 | if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), |
| 4314 | (U8) TWO_BYTE_UTF8_TO_UNI(nextchr, |
| 4315 | *(locinput + 1)))))) |
| 4316 | { |
| 4317 | sayNO; |
| 4318 | } |
| 4319 | } |
| 4320 | else { /* Here, must be an above Latin-1 code point */ |
| 4321 | goto utf8_posix_not_eos; |
| 4322 | } |
| 4323 | |
| 4324 | /* Here, must be utf8 */ |
| 4325 | locinput += UTF8SKIP(locinput); |
| 4326 | break; |
| 4327 | |
| 4328 | case NPOSIXD: /* \W or [:^punct:] etc. under /d */ |
| 4329 | to_complement = 1; |
| 4330 | /* FALLTHROUGH */ |
| 4331 | |
| 4332 | case POSIXD: /* \w or [:punct:] etc. under /d */ |
| 4333 | if (utf8_target) { |
| 4334 | goto utf8_posix; |
| 4335 | } |
| 4336 | goto posixa; |
| 4337 | |
| 4338 | case NPOSIXA: /* \W or [:^punct:] etc. under /a */ |
| 4339 | |
| 4340 | if (NEXTCHR_IS_EOS) { |
| 4341 | sayNO; |
| 4342 | } |
| 4343 | |
| 4344 | /* All UTF-8 variants match */ |
| 4345 | if (! UTF8_IS_INVARIANT(nextchr)) { |
| 4346 | goto increment_locinput; |
| 4347 | } |
| 4348 | |
| 4349 | to_complement = 1; |
| 4350 | /* FALLTHROUGH */ |
| 4351 | |
| 4352 | case POSIXA: /* \w or [:punct:] etc. under /a */ |
| 4353 | |
| 4354 | posixa: |
| 4355 | /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in |
| 4356 | * UTF-8, and also from NPOSIXA even in UTF-8 when the current |
| 4357 | * character is a single byte */ |
| 4358 | |
| 4359 | if (NEXTCHR_IS_EOS |
| 4360 | || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr, |
| 4361 | FLAGS(scan))))) |
| 4362 | { |
| 4363 | sayNO; |
| 4364 | } |
| 4365 | |
| 4366 | /* Here we are either not in utf8, or we matched a utf8-invariant, |
| 4367 | * so the next char is the next byte */ |
| 4368 | locinput++; |
| 4369 | break; |
| 4370 | |
| 4371 | case NPOSIXU: /* \W or [:^punct:] etc. under /u */ |
| 4372 | to_complement = 1; |
| 4373 | /* FALLTHROUGH */ |
| 4374 | |
| 4375 | case POSIXU: /* \w or [:punct:] etc. under /u */ |
| 4376 | utf8_posix: |
| 4377 | if (NEXTCHR_IS_EOS) { |
| 4378 | sayNO; |
| 4379 | } |
| 4380 | utf8_posix_not_eos: |
| 4381 | |
| 4382 | /* Use _generic_isCC() for characters within Latin1. (Note that |
| 4383 | * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else |
| 4384 | * wouldn't be invariant) */ |
| 4385 | if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) { |
| 4386 | if (! (to_complement ^ cBOOL(_generic_isCC(nextchr, |
| 4387 | FLAGS(scan))))) |
| 4388 | { |
| 4389 | sayNO; |
| 4390 | } |
| 4391 | locinput++; |
| 4392 | } |
| 4393 | else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) { |
| 4394 | if (! (to_complement |
| 4395 | ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_UNI(nextchr, |
| 4396 | *(locinput + 1)), |
| 4397 | FLAGS(scan))))) |
| 4398 | { |
| 4399 | sayNO; |
| 4400 | } |
| 4401 | locinput += 2; |
| 4402 | } |
| 4403 | else { /* Handle above Latin-1 code points */ |
| 4404 | classnum = (_char_class_number) FLAGS(scan); |
| 4405 | if (classnum < _FIRST_NON_SWASH_CC) { |
| 4406 | |
| 4407 | /* Here, uses a swash to find such code points. Load if if |
| 4408 | * not done already */ |
| 4409 | if (! PL_utf8_swash_ptrs[classnum]) { |
| 4410 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; |
| 4411 | PL_utf8_swash_ptrs[classnum] |
| 4412 | = _core_swash_init("utf8", |
| 4413 | swash_property_names[classnum], |
| 4414 | &PL_sv_undef, 1, 0, NULL, &flags); |
| 4415 | } |
| 4416 | if (! (to_complement |
| 4417 | ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], |
| 4418 | (U8 *) locinput, TRUE)))) |
| 4419 | { |
| 4420 | sayNO; |
| 4421 | } |
| 4422 | } |
| 4423 | else { /* Here, uses macros to find above Latin-1 code points */ |
| 4424 | switch (classnum) { |
| 4425 | case _CC_ENUM_SPACE: /* XXX would require separate |
| 4426 | code if we revert the change |
| 4427 | of \v matching this */ |
| 4428 | case _CC_ENUM_PSXSPC: |
| 4429 | if (! (to_complement |
| 4430 | ^ cBOOL(is_XPERLSPACE_high(locinput)))) |
| 4431 | { |
| 4432 | sayNO; |
| 4433 | } |
| 4434 | break; |
| 4435 | case _CC_ENUM_BLANK: |
| 4436 | if (! (to_complement |
| 4437 | ^ cBOOL(is_HORIZWS_high(locinput)))) |
| 4438 | { |
| 4439 | sayNO; |
| 4440 | } |
| 4441 | break; |
| 4442 | case _CC_ENUM_XDIGIT: |
| 4443 | if (! (to_complement |
| 4444 | ^ cBOOL(is_XDIGIT_high(locinput)))) |
| 4445 | { |
| 4446 | sayNO; |
| 4447 | } |
| 4448 | break; |
| 4449 | case _CC_ENUM_VERTSPACE: |
| 4450 | if (! (to_complement |
| 4451 | ^ cBOOL(is_VERTWS_high(locinput)))) |
| 4452 | { |
| 4453 | sayNO; |
| 4454 | } |
| 4455 | break; |
| 4456 | default: /* The rest, e.g. [:cntrl:], can't match |
| 4457 | above Latin1 */ |
| 4458 | if (! to_complement) { |
| 4459 | sayNO; |
| 4460 | } |
| 4461 | break; |
| 4462 | } |
| 4463 | } |
| 4464 | locinput += UTF8SKIP(locinput); |
| 4465 | } |
| 4466 | break; |
| 4467 | |
| 4468 | case CLUMP: /* Match \X: logical Unicode character. This is defined as |
| 4469 | a Unicode extended Grapheme Cluster */ |
| 4470 | /* From http://www.unicode.org/reports/tr29 (5.2 version). An |
| 4471 | extended Grapheme Cluster is: |
| 4472 | |
| 4473 | CR LF |
| 4474 | | Prepend* Begin Extend* |
| 4475 | | . |
| 4476 | |
| 4477 | Begin is: ( Special_Begin | ! Control ) |
| 4478 | Special_Begin is: ( Regional-Indicator+ | Hangul-syllable ) |
| 4479 | Extend is: ( Grapheme_Extend | Spacing_Mark ) |
| 4480 | Control is: [ GCB_Control | CR | LF ] |
| 4481 | Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) )) |
| 4482 | |
| 4483 | If we create a 'Regular_Begin' = Begin - Special_Begin, then |
| 4484 | we can rewrite |
| 4485 | |
| 4486 | Begin is ( Regular_Begin + Special Begin ) |
| 4487 | |
| 4488 | It turns out that 98.4% of all Unicode code points match |
| 4489 | Regular_Begin. Doing it this way eliminates a table match in |
| 4490 | the previous implementation for almost all Unicode code points. |
| 4491 | |
| 4492 | There is a subtlety with Prepend* which showed up in testing. |
| 4493 | Note that the Begin, and only the Begin is required in: |
| 4494 | | Prepend* Begin Extend* |
| 4495 | Also, Begin contains '! Control'. A Prepend must be a |
| 4496 | '! Control', which means it must also be a Begin. What it |
| 4497 | comes down to is that if we match Prepend* and then find no |
| 4498 | suitable Begin afterwards, that if we backtrack the last |
| 4499 | Prepend, that one will be a suitable Begin. |
| 4500 | */ |
| 4501 | |
| 4502 | if (NEXTCHR_IS_EOS) |
| 4503 | sayNO; |
| 4504 | if (! utf8_target) { |
| 4505 | |
| 4506 | /* Match either CR LF or '.', as all the other possibilities |
| 4507 | * require utf8 */ |
| 4508 | locinput++; /* Match the . or CR */ |
| 4509 | if (nextchr == '\r' /* And if it was CR, and the next is LF, |
| 4510 | match the LF */ |
| 4511 | && locinput < PL_regeol |
| 4512 | && UCHARAT(locinput) == '\n') |
| 4513 | { |
| 4514 | locinput++; |
| 4515 | } |
| 4516 | } |
| 4517 | else { |
| 4518 | |
| 4519 | /* Utf8: See if is ( CR LF ); already know that locinput < |
| 4520 | * PL_regeol, so locinput+1 is in bounds */ |
| 4521 | if ( nextchr == '\r' && locinput+1 < PL_regeol |
| 4522 | && UCHARAT(locinput + 1) == '\n') |
| 4523 | { |
| 4524 | locinput += 2; |
| 4525 | } |
| 4526 | else { |
| 4527 | STRLEN len; |
| 4528 | |
| 4529 | /* In case have to backtrack to beginning, then match '.' */ |
| 4530 | char *starting = locinput; |
| 4531 | |
| 4532 | /* In case have to backtrack the last prepend */ |
| 4533 | char *previous_prepend = NULL; |
| 4534 | |
| 4535 | LOAD_UTF8_CHARCLASS_GCB(); |
| 4536 | |
| 4537 | /* Match (prepend)* */ |
| 4538 | while (locinput < PL_regeol |
| 4539 | && (len = is_GCB_Prepend_utf8(locinput))) |
| 4540 | { |
| 4541 | previous_prepend = locinput; |
| 4542 | locinput += len; |
| 4543 | } |
| 4544 | |
| 4545 | /* As noted above, if we matched a prepend character, but |
| 4546 | * the next thing won't match, back off the last prepend we |
| 4547 | * matched, as it is guaranteed to match the begin */ |
| 4548 | if (previous_prepend |
| 4549 | && (locinput >= PL_regeol |
| 4550 | || (! swash_fetch(PL_utf8_X_regular_begin, |
| 4551 | (U8*)locinput, utf8_target) |
| 4552 | && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput))) |
| 4553 | ) |
| 4554 | { |
| 4555 | locinput = previous_prepend; |
| 4556 | } |
| 4557 | |
| 4558 | /* Note that here we know PL_regeol > locinput, as we |
| 4559 | * tested that upon input to this switch case, and if we |
| 4560 | * moved locinput forward, we tested the result just above |
| 4561 | * and it either passed, or we backed off so that it will |
| 4562 | * now pass */ |
| 4563 | if (swash_fetch(PL_utf8_X_regular_begin, |
| 4564 | (U8*)locinput, utf8_target)) { |
| 4565 | locinput += UTF8SKIP(locinput); |
| 4566 | } |
| 4567 | else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) { |
| 4568 | |
| 4569 | /* Here did not match the required 'Begin' in the |
| 4570 | * second term. So just match the very first |
| 4571 | * character, the '.' of the final term of the regex */ |
| 4572 | locinput = starting + UTF8SKIP(starting); |
| 4573 | goto exit_utf8; |
| 4574 | } else { |
| 4575 | |
| 4576 | /* Here is a special begin. It can be composed of |
| 4577 | * several individual characters. One possibility is |
| 4578 | * RI+ */ |
| 4579 | if ((len = is_GCB_RI_utf8(locinput))) { |
| 4580 | locinput += len; |
| 4581 | while (locinput < PL_regeol |
| 4582 | && (len = is_GCB_RI_utf8(locinput))) |
| 4583 | { |
| 4584 | locinput += len; |
| 4585 | } |
| 4586 | } else if ((len = is_GCB_T_utf8(locinput))) { |
| 4587 | /* Another possibility is T+ */ |
| 4588 | locinput += len; |
| 4589 | while (locinput < PL_regeol |
| 4590 | && (len = is_GCB_T_utf8(locinput))) |
| 4591 | { |
| 4592 | locinput += len; |
| 4593 | } |
| 4594 | } else { |
| 4595 | |
| 4596 | /* Here, neither RI+ nor T+; must be some other |
| 4597 | * Hangul. That means it is one of the others: L, |
| 4598 | * LV, LVT or V, and matches: |
| 4599 | * L* (L | LVT T* | V * V* T* | LV V* T*) */ |
| 4600 | |
| 4601 | /* Match L* */ |
| 4602 | while (locinput < PL_regeol |
| 4603 | && (len = is_GCB_L_utf8(locinput))) |
| 4604 | { |
| 4605 | locinput += len; |
| 4606 | } |
| 4607 | |
| 4608 | /* Here, have exhausted L*. If the next character |
| 4609 | * is not an LV, LVT nor V, it means we had to have |
| 4610 | * at least one L, so matches L+ in the original |
| 4611 | * equation, we have a complete hangul syllable. |
| 4612 | * Are done. */ |
| 4613 | |
| 4614 | if (locinput < PL_regeol |
| 4615 | && is_GCB_LV_LVT_V_utf8(locinput)) |
| 4616 | { |
| 4617 | /* Otherwise keep going. Must be LV, LVT or V. |
| 4618 | * See if LVT, by first ruling out V, then LV */ |
| 4619 | if (! is_GCB_V_utf8(locinput) |
| 4620 | /* All but every TCount one is LV */ |
| 4621 | && (valid_utf8_to_uvchr((U8 *) locinput, |
| 4622 | NULL) |
| 4623 | - SBASE) |
| 4624 | % TCount != 0) |
| 4625 | { |
| 4626 | locinput += UTF8SKIP(locinput); |
| 4627 | } else { |
| 4628 | |
| 4629 | /* Must be V or LV. Take it, then match |
| 4630 | * V* */ |
| 4631 | locinput += UTF8SKIP(locinput); |
| 4632 | while (locinput < PL_regeol |
| 4633 | && (len = is_GCB_V_utf8(locinput))) |
| 4634 | { |
| 4635 | locinput += len; |
| 4636 | } |
| 4637 | } |
| 4638 | |
| 4639 | /* And any of LV, LVT, or V can be followed |
| 4640 | * by T* */ |
| 4641 | while (locinput < PL_regeol |
| 4642 | && (len = is_GCB_T_utf8(locinput))) |
| 4643 | { |
| 4644 | locinput += len; |
| 4645 | } |
| 4646 | } |
| 4647 | } |
| 4648 | } |
| 4649 | |
| 4650 | /* Match any extender */ |
| 4651 | while (locinput < PL_regeol |
| 4652 | && swash_fetch(PL_utf8_X_extend, |
| 4653 | (U8*)locinput, utf8_target)) |
| 4654 | { |
| 4655 | locinput += UTF8SKIP(locinput); |
| 4656 | } |
| 4657 | } |
| 4658 | exit_utf8: |
| 4659 | if (locinput > PL_regeol) sayNO; |
| 4660 | } |
| 4661 | break; |
| 4662 | |
| 4663 | case NREFFL: /* /\g{name}/il */ |
| 4664 | { /* The capture buffer cases. The ones beginning with N for the |
| 4665 | named buffers just convert to the equivalent numbered and |
| 4666 | pretend they were called as the corresponding numbered buffer |
| 4667 | op. */ |
| 4668 | /* don't initialize these in the declaration, it makes C++ |
| 4669 | unhappy */ |
| 4670 | char *s; |
| 4671 | char type; |
| 4672 | re_fold_t folder; |
| 4673 | const U8 *fold_array; |
| 4674 | UV utf8_fold_flags; |
| 4675 | |
| 4676 | RX_MATCH_TAINTED_on(reginfo->prog); |
| 4677 | folder = foldEQ_locale; |
| 4678 | fold_array = PL_fold_locale; |
| 4679 | type = REFFL; |
| 4680 | utf8_fold_flags = FOLDEQ_UTF8_LOCALE; |
| 4681 | goto do_nref; |
| 4682 | |
| 4683 | case NREFFA: /* /\g{name}/iaa */ |
| 4684 | folder = foldEQ_latin1; |
| 4685 | fold_array = PL_fold_latin1; |
| 4686 | type = REFFA; |
| 4687 | utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 4688 | goto do_nref; |
| 4689 | |
| 4690 | case NREFFU: /* /\g{name}/iu */ |
| 4691 | folder = foldEQ_latin1; |
| 4692 | fold_array = PL_fold_latin1; |
| 4693 | type = REFFU; |
| 4694 | utf8_fold_flags = 0; |
| 4695 | goto do_nref; |
| 4696 | |
| 4697 | case NREFF: /* /\g{name}/i */ |
| 4698 | folder = foldEQ; |
| 4699 | fold_array = PL_fold; |
| 4700 | type = REFF; |
| 4701 | utf8_fold_flags = 0; |
| 4702 | goto do_nref; |
| 4703 | |
| 4704 | case NREF: /* /\g{name}/ */ |
| 4705 | type = REF; |
| 4706 | folder = NULL; |
| 4707 | fold_array = NULL; |
| 4708 | utf8_fold_flags = 0; |
| 4709 | do_nref: |
| 4710 | |
| 4711 | /* For the named back references, find the corresponding buffer |
| 4712 | * number */ |
| 4713 | n = reg_check_named_buff_matched(rex,scan); |
| 4714 | |
| 4715 | if ( ! n ) { |
| 4716 | sayNO; |
| 4717 | } |
| 4718 | goto do_nref_ref_common; |
| 4719 | |
| 4720 | case REFFL: /* /\1/il */ |
| 4721 | RX_MATCH_TAINTED_on(reginfo->prog); |
| 4722 | folder = foldEQ_locale; |
| 4723 | fold_array = PL_fold_locale; |
| 4724 | utf8_fold_flags = FOLDEQ_UTF8_LOCALE; |
| 4725 | goto do_ref; |
| 4726 | |
| 4727 | case REFFA: /* /\1/iaa */ |
| 4728 | folder = foldEQ_latin1; |
| 4729 | fold_array = PL_fold_latin1; |
| 4730 | utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 4731 | goto do_ref; |
| 4732 | |
| 4733 | case REFFU: /* /\1/iu */ |
| 4734 | folder = foldEQ_latin1; |
| 4735 | fold_array = PL_fold_latin1; |
| 4736 | utf8_fold_flags = 0; |
| 4737 | goto do_ref; |
| 4738 | |
| 4739 | case REFF: /* /\1/i */ |
| 4740 | folder = foldEQ; |
| 4741 | fold_array = PL_fold; |
| 4742 | utf8_fold_flags = 0; |
| 4743 | goto do_ref; |
| 4744 | |
| 4745 | case REF: /* /\1/ */ |
| 4746 | folder = NULL; |
| 4747 | fold_array = NULL; |
| 4748 | utf8_fold_flags = 0; |
| 4749 | |
| 4750 | do_ref: |
| 4751 | type = OP(scan); |
| 4752 | n = ARG(scan); /* which paren pair */ |
| 4753 | |
| 4754 | do_nref_ref_common: |
| 4755 | ln = rex->offs[n].start; |
| 4756 | PL_reg_leftiter = PL_reg_maxiter; /* Void cache */ |
| 4757 | if (rex->lastparen < n || ln == -1) |
| 4758 | sayNO; /* Do not match unless seen CLOSEn. */ |
| 4759 | if (ln == rex->offs[n].end) |
| 4760 | break; |
| 4761 | |
| 4762 | s = PL_bostr + ln; |
| 4763 | if (type != REF /* REF can do byte comparison */ |
| 4764 | && (utf8_target || type == REFFU)) |
| 4765 | { /* XXX handle REFFL better */ |
| 4766 | char * limit = PL_regeol; |
| 4767 | |
| 4768 | /* This call case insensitively compares the entire buffer |
| 4769 | * at s, with the current input starting at locinput, but |
| 4770 | * not going off the end given by PL_regeol, and returns in |
| 4771 | * <limit> upon success, how much of the current input was |
| 4772 | * matched */ |
| 4773 | if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target, |
| 4774 | locinput, &limit, 0, utf8_target, utf8_fold_flags)) |
| 4775 | { |
| 4776 | sayNO; |
| 4777 | } |
| 4778 | locinput = limit; |
| 4779 | break; |
| 4780 | } |
| 4781 | |
| 4782 | /* Not utf8: Inline the first character, for speed. */ |
| 4783 | if (!NEXTCHR_IS_EOS && |
| 4784 | UCHARAT(s) != nextchr && |
| 4785 | (type == REF || |
| 4786 | UCHARAT(s) != fold_array[nextchr])) |
| 4787 | sayNO; |
| 4788 | ln = rex->offs[n].end - ln; |
| 4789 | if (locinput + ln > PL_regeol) |
| 4790 | sayNO; |
| 4791 | if (ln > 1 && (type == REF |
| 4792 | ? memNE(s, locinput, ln) |
| 4793 | : ! folder(s, locinput, ln))) |
| 4794 | sayNO; |
| 4795 | locinput += ln; |
| 4796 | break; |
| 4797 | } |
| 4798 | |
| 4799 | case NOTHING: /* null op; e.g. the 'nothing' following |
| 4800 | * the '*' in m{(a+|b)*}' */ |
| 4801 | break; |
| 4802 | case TAIL: /* placeholder while compiling (A|B|C) */ |
| 4803 | break; |
| 4804 | |
| 4805 | case BACK: /* ??? doesn't appear to be used ??? */ |
| 4806 | break; |
| 4807 | |
| 4808 | #undef ST |
| 4809 | #define ST st->u.eval |
| 4810 | { |
| 4811 | SV *ret; |
| 4812 | REGEXP *re_sv; |
| 4813 | regexp *re; |
| 4814 | regexp_internal *rei; |
| 4815 | regnode *startpoint; |
| 4816 | |
| 4817 | case GOSTART: /* (?R) */ |
| 4818 | case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */ |
| 4819 | if (cur_eval && cur_eval->locinput==locinput) { |
| 4820 | if (cur_eval->u.eval.close_paren == (U32)ARG(scan)) |
| 4821 | Perl_croak(aTHX_ "Infinite recursion in regex"); |
| 4822 | if ( ++nochange_depth > max_nochange_depth ) |
| 4823 | Perl_croak(aTHX_ |
| 4824 | "Pattern subroutine nesting without pos change" |
| 4825 | " exceeded limit in regex"); |
| 4826 | } else { |
| 4827 | nochange_depth = 0; |
| 4828 | } |
| 4829 | re_sv = rex_sv; |
| 4830 | re = rex; |
| 4831 | rei = rexi; |
| 4832 | if (OP(scan)==GOSUB) { |
| 4833 | startpoint = scan + ARG2L(scan); |
| 4834 | ST.close_paren = ARG(scan); |
| 4835 | } else { |
| 4836 | startpoint = rei->program+1; |
| 4837 | ST.close_paren = 0; |
| 4838 | } |
| 4839 | goto eval_recurse_doit; |
| 4840 | assert(0); /* NOTREACHED */ |
| 4841 | |
| 4842 | case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */ |
| 4843 | if (cur_eval && cur_eval->locinput==locinput) { |
| 4844 | if ( ++nochange_depth > max_nochange_depth ) |
| 4845 | Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex"); |
| 4846 | } else { |
| 4847 | nochange_depth = 0; |
| 4848 | } |
| 4849 | { |
| 4850 | /* execute the code in the {...} */ |
| 4851 | |
| 4852 | dSP; |
| 4853 | IV before; |
| 4854 | OP * const oop = PL_op; |
| 4855 | COP * const ocurcop = PL_curcop; |
| 4856 | OP *nop; |
| 4857 | char *saved_regeol = PL_regeol; |
| 4858 | struct re_save_state saved_state; |
| 4859 | CV *newcv; |
| 4860 | |
| 4861 | /* save *all* paren positions */ |
| 4862 | regcppush(rex, 0, maxopenparen); |
| 4863 | REGCP_SET(runops_cp); |
| 4864 | |
| 4865 | /* To not corrupt the existing regex state while executing the |
| 4866 | * eval we would normally put it on the save stack, like with |
| 4867 | * save_re_context. However, re-evals have a weird scoping so we |
| 4868 | * can't just add ENTER/LEAVE here. With that, things like |
| 4869 | * |
| 4870 | * (?{$a=2})(a(?{local$a=$a+1}))*aak*c(?{$b=$a}) |
| 4871 | * |
| 4872 | * would break, as they expect the localisation to be unwound |
| 4873 | * only when the re-engine backtracks through the bit that |
| 4874 | * localised it. |
| 4875 | * |
| 4876 | * What we do instead is just saving the state in a local c |
| 4877 | * variable. |
| 4878 | */ |
| 4879 | Copy(&PL_reg_state, &saved_state, 1, struct re_save_state); |
| 4880 | |
| 4881 | if (!caller_cv) |
| 4882 | caller_cv = find_runcv(NULL); |
| 4883 | |
| 4884 | n = ARG(scan); |
| 4885 | |
| 4886 | if (rexi->data->what[n] == 'r') { /* code from an external qr */ |
| 4887 | newcv = (ReANY( |
| 4888 | (REGEXP*)(rexi->data->data[n]) |
| 4889 | ))->qr_anoncv |
| 4890 | ; |
| 4891 | nop = (OP*)rexi->data->data[n+1]; |
| 4892 | } |
| 4893 | else if (rexi->data->what[n] == 'l') { /* literal code */ |
| 4894 | newcv = caller_cv; |
| 4895 | nop = (OP*)rexi->data->data[n]; |
| 4896 | assert(CvDEPTH(newcv)); |
| 4897 | } |
| 4898 | else { |
| 4899 | /* literal with own CV */ |
| 4900 | assert(rexi->data->what[n] == 'L'); |
| 4901 | newcv = rex->qr_anoncv; |
| 4902 | nop = (OP*)rexi->data->data[n]; |
| 4903 | } |
| 4904 | |
| 4905 | /* normally if we're about to execute code from the same |
| 4906 | * CV that we used previously, we just use the existing |
| 4907 | * CX stack entry. However, its possible that in the |
| 4908 | * meantime we may have backtracked, popped from the save |
| 4909 | * stack, and undone the SAVECOMPPAD(s) associated with |
| 4910 | * PUSH_MULTICALL; in which case PL_comppad no longer |
| 4911 | * points to newcv's pad. */ |
| 4912 | if (newcv != last_pushed_cv || PL_comppad != last_pad) |
| 4913 | { |
| 4914 | U8 flags = (CXp_SUB_RE | |
| 4915 | ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0)); |
| 4916 | if (last_pushed_cv) { |
| 4917 | CHANGE_MULTICALL_FLAGS(newcv, flags); |
| 4918 | } |
| 4919 | else { |
| 4920 | PUSH_MULTICALL_FLAGS(newcv, flags); |
| 4921 | } |
| 4922 | last_pushed_cv = newcv; |
| 4923 | } |
| 4924 | else { |
| 4925 | /* these assignments are just to silence compiler |
| 4926 | * warnings */ |
| 4927 | multicall_cop = NULL; |
| 4928 | newsp = NULL; |
| 4929 | } |
| 4930 | last_pad = PL_comppad; |
| 4931 | |
| 4932 | /* the initial nextstate you would normally execute |
| 4933 | * at the start of an eval (which would cause error |
| 4934 | * messages to come from the eval), may be optimised |
| 4935 | * away from the execution path in the regex code blocks; |
| 4936 | * so manually set PL_curcop to it initially */ |
| 4937 | { |
| 4938 | OP *o = cUNOPx(nop)->op_first; |
| 4939 | assert(o->op_type == OP_NULL); |
| 4940 | if (o->op_targ == OP_SCOPE) { |
| 4941 | o = cUNOPo->op_first; |
| 4942 | } |
| 4943 | else { |
| 4944 | assert(o->op_targ == OP_LEAVE); |
| 4945 | o = cUNOPo->op_first; |
| 4946 | assert(o->op_type == OP_ENTER); |
| 4947 | o = o->op_sibling; |
| 4948 | } |
| 4949 | |
| 4950 | if (o->op_type != OP_STUB) { |
| 4951 | assert( o->op_type == OP_NEXTSTATE |
| 4952 | || o->op_type == OP_DBSTATE |
| 4953 | || (o->op_type == OP_NULL |
| 4954 | && ( o->op_targ == OP_NEXTSTATE |
| 4955 | || o->op_targ == OP_DBSTATE |
| 4956 | ) |
| 4957 | ) |
| 4958 | ); |
| 4959 | PL_curcop = (COP*)o; |
| 4960 | } |
| 4961 | } |
| 4962 | nop = nop->op_next; |
| 4963 | |
| 4964 | DEBUG_STATE_r( PerlIO_printf(Perl_debug_log, |
| 4965 | " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) ); |
| 4966 | |
| 4967 | rex->offs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr; |
| 4968 | |
| 4969 | if (sv_yes_mark) { |
| 4970 | SV *sv_mrk = get_sv("REGMARK", 1); |
| 4971 | sv_setsv(sv_mrk, sv_yes_mark); |
| 4972 | } |
| 4973 | |
| 4974 | /* we don't use MULTICALL here as we want to call the |
| 4975 | * first op of the block of interest, rather than the |
| 4976 | * first op of the sub */ |
| 4977 | before = (IV)(SP-PL_stack_base); |
| 4978 | PL_op = nop; |
| 4979 | CALLRUNOPS(aTHX); /* Scalar context. */ |
| 4980 | SPAGAIN; |
| 4981 | if ((IV)(SP-PL_stack_base) == before) |
| 4982 | ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */ |
| 4983 | else { |
| 4984 | ret = POPs; |
| 4985 | PUTBACK; |
| 4986 | } |
| 4987 | |
| 4988 | /* before restoring everything, evaluate the returned |
| 4989 | * value, so that 'uninit' warnings don't use the wrong |
| 4990 | * PL_op or pad. Also need to process any magic vars |
| 4991 | * (e.g. $1) *before* parentheses are restored */ |
| 4992 | |
| 4993 | PL_op = NULL; |
| 4994 | |
| 4995 | re_sv = NULL; |
| 4996 | if (logical == 0) /* (?{})/ */ |
| 4997 | sv_setsv(save_scalar(PL_replgv), ret); /* $^R */ |
| 4998 | else if (logical == 1) { /* /(?(?{...})X|Y)/ */ |
| 4999 | sw = cBOOL(SvTRUE(ret)); |
| 5000 | logical = 0; |
| 5001 | } |
| 5002 | else { /* /(??{}) */ |
| 5003 | /* if its overloaded, let the regex compiler handle |
| 5004 | * it; otherwise extract regex, or stringify */ |
| 5005 | if (!SvAMAGIC(ret)) { |
| 5006 | SV *sv = ret; |
| 5007 | if (SvROK(sv)) |
| 5008 | sv = SvRV(sv); |
| 5009 | if (SvTYPE(sv) == SVt_REGEXP) |
| 5010 | re_sv = (REGEXP*) sv; |
| 5011 | else if (SvSMAGICAL(sv)) { |
| 5012 | MAGIC *mg = mg_find(sv, PERL_MAGIC_qr); |
| 5013 | if (mg) |
| 5014 | re_sv = (REGEXP *) mg->mg_obj; |
| 5015 | } |
| 5016 | |
| 5017 | /* force any magic, undef warnings here */ |
| 5018 | if (!re_sv) { |
| 5019 | ret = sv_mortalcopy(ret); |
| 5020 | (void) SvPV_force_nolen(ret); |
| 5021 | } |
| 5022 | } |
| 5023 | |
| 5024 | } |
| 5025 | |
| 5026 | Copy(&saved_state, &PL_reg_state, 1, struct re_save_state); |
| 5027 | |
| 5028 | /* *** Note that at this point we don't restore |
| 5029 | * PL_comppad, (or pop the CxSUB) on the assumption it may |
| 5030 | * be used again soon. This is safe as long as nothing |
| 5031 | * in the regexp code uses the pad ! */ |
| 5032 | PL_op = oop; |
| 5033 | PL_curcop = ocurcop; |
| 5034 | PL_regeol = saved_regeol; |
| 5035 | S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen); |
| 5036 | |
| 5037 | if (logical != 2) |
| 5038 | break; |
| 5039 | } |
| 5040 | |
| 5041 | /* only /(??{})/ from now on */ |
| 5042 | logical = 0; |
| 5043 | { |
| 5044 | /* extract RE object from returned value; compiling if |
| 5045 | * necessary */ |
| 5046 | |
| 5047 | if (re_sv) { |
| 5048 | re_sv = reg_temp_copy(NULL, re_sv); |
| 5049 | } |
| 5050 | else { |
| 5051 | U32 pm_flags = 0; |
| 5052 | |
| 5053 | if (SvUTF8(ret) && IN_BYTES) { |
| 5054 | /* In use 'bytes': make a copy of the octet |
| 5055 | * sequence, but without the flag on */ |
| 5056 | STRLEN len; |
| 5057 | const char *const p = SvPV(ret, len); |
| 5058 | ret = newSVpvn_flags(p, len, SVs_TEMP); |
| 5059 | } |
| 5060 | if (rex->intflags & PREGf_USE_RE_EVAL) |
| 5061 | pm_flags |= PMf_USE_RE_EVAL; |
| 5062 | |
| 5063 | /* if we got here, it should be an engine which |
| 5064 | * supports compiling code blocks and stuff */ |
| 5065 | assert(rex->engine && rex->engine->op_comp); |
| 5066 | assert(!(scan->flags & ~RXf_PMf_COMPILETIME)); |
| 5067 | re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL, |
| 5068 | rex->engine, NULL, NULL, |
| 5069 | /* copy /msix etc to inner pattern */ |
| 5070 | scan->flags, |
| 5071 | pm_flags); |
| 5072 | |
| 5073 | if (!(SvFLAGS(ret) |
| 5074 | & (SVs_TEMP | SVs_PADTMP | SVf_READONLY |
| 5075 | | SVs_GMG))) { |
| 5076 | /* This isn't a first class regexp. Instead, it's |
| 5077 | caching a regexp onto an existing, Perl visible |
| 5078 | scalar. */ |
| 5079 | sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0); |
| 5080 | } |
| 5081 | /* safe to do now that any $1 etc has been |
| 5082 | * interpolated into the new pattern string and |
| 5083 | * compiled */ |
| 5084 | S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen); |
| 5085 | } |
| 5086 | SAVEFREESV(re_sv); |
| 5087 | re = ReANY(re_sv); |
| 5088 | } |
| 5089 | RXp_MATCH_COPIED_off(re); |
| 5090 | re->subbeg = rex->subbeg; |
| 5091 | re->sublen = rex->sublen; |
| 5092 | re->suboffset = rex->suboffset; |
| 5093 | re->subcoffset = rex->subcoffset; |
| 5094 | rei = RXi_GET(re); |
| 5095 | DEBUG_EXECUTE_r( |
| 5096 | debug_start_match(re_sv, utf8_target, locinput, PL_regeol, |
| 5097 | "Matching embedded"); |
| 5098 | ); |
| 5099 | startpoint = rei->program + 1; |
| 5100 | ST.close_paren = 0; /* only used for GOSUB */ |
| 5101 | |
| 5102 | eval_recurse_doit: /* Share code with GOSUB below this line */ |
| 5103 | /* run the pattern returned from (??{...}) */ |
| 5104 | |
| 5105 | /* Save *all* the positions. */ |
| 5106 | ST.cp = regcppush(rex, 0, maxopenparen); |
| 5107 | REGCP_SET(ST.lastcp); |
| 5108 | |
| 5109 | re->lastparen = 0; |
| 5110 | re->lastcloseparen = 0; |
| 5111 | |
| 5112 | maxopenparen = 0; |
| 5113 | |
| 5114 | /* XXXX This is too dramatic a measure... */ |
| 5115 | PL_reg_maxiter = 0; |
| 5116 | |
| 5117 | ST.saved_utf8_pat = is_utf8_pat; |
| 5118 | is_utf8_pat = cBOOL(RX_UTF8(re_sv)); |
| 5119 | |
| 5120 | ST.prev_rex = rex_sv; |
| 5121 | ST.prev_curlyx = cur_curlyx; |
| 5122 | rex_sv = re_sv; |
| 5123 | SET_reg_curpm(rex_sv); |
| 5124 | rex = re; |
| 5125 | rexi = rei; |
| 5126 | cur_curlyx = NULL; |
| 5127 | ST.B = next; |
| 5128 | ST.prev_eval = cur_eval; |
| 5129 | cur_eval = st; |
| 5130 | /* now continue from first node in postoned RE */ |
| 5131 | PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput); |
| 5132 | assert(0); /* NOTREACHED */ |
| 5133 | } |
| 5134 | |
| 5135 | case EVAL_AB: /* cleanup after a successful (??{A})B */ |
| 5136 | /* note: this is called twice; first after popping B, then A */ |
| 5137 | is_utf8_pat = ST.saved_utf8_pat; |
| 5138 | rex_sv = ST.prev_rex; |
| 5139 | SET_reg_curpm(rex_sv); |
| 5140 | rex = ReANY(rex_sv); |
| 5141 | rexi = RXi_GET(rex); |
| 5142 | regcpblow(ST.cp); |
| 5143 | cur_eval = ST.prev_eval; |
| 5144 | cur_curlyx = ST.prev_curlyx; |
| 5145 | |
| 5146 | /* XXXX This is too dramatic a measure... */ |
| 5147 | PL_reg_maxiter = 0; |
| 5148 | if ( nochange_depth ) |
| 5149 | nochange_depth--; |
| 5150 | sayYES; |
| 5151 | |
| 5152 | |
| 5153 | case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */ |
| 5154 | /* note: this is called twice; first after popping B, then A */ |
| 5155 | is_utf8_pat = ST.saved_utf8_pat; |
| 5156 | rex_sv = ST.prev_rex; |
| 5157 | SET_reg_curpm(rex_sv); |
| 5158 | rex = ReANY(rex_sv); |
| 5159 | rexi = RXi_GET(rex); |
| 5160 | |
| 5161 | REGCP_UNWIND(ST.lastcp); |
| 5162 | regcppop(rex, &maxopenparen); |
| 5163 | cur_eval = ST.prev_eval; |
| 5164 | cur_curlyx = ST.prev_curlyx; |
| 5165 | /* XXXX This is too dramatic a measure... */ |
| 5166 | PL_reg_maxiter = 0; |
| 5167 | if ( nochange_depth ) |
| 5168 | nochange_depth--; |
| 5169 | sayNO_SILENT; |
| 5170 | #undef ST |
| 5171 | |
| 5172 | case OPEN: /* ( */ |
| 5173 | n = ARG(scan); /* which paren pair */ |
| 5174 | rex->offs[n].start_tmp = locinput - PL_bostr; |
| 5175 | if (n > maxopenparen) |
| 5176 | maxopenparen = n; |
| 5177 | DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, |
| 5178 | "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n", |
| 5179 | PTR2UV(rex), |
| 5180 | PTR2UV(rex->offs), |
| 5181 | (UV)n, |
| 5182 | (IV)rex->offs[n].start_tmp, |
| 5183 | (UV)maxopenparen |
| 5184 | )); |
| 5185 | lastopen = n; |
| 5186 | break; |
| 5187 | |
| 5188 | /* XXX really need to log other places start/end are set too */ |
| 5189 | #define CLOSE_CAPTURE \ |
| 5190 | rex->offs[n].start = rex->offs[n].start_tmp; \ |
| 5191 | rex->offs[n].end = locinput - PL_bostr; \ |
| 5192 | DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \ |
| 5193 | "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \ |
| 5194 | PTR2UV(rex), \ |
| 5195 | PTR2UV(rex->offs), \ |
| 5196 | (UV)n, \ |
| 5197 | (IV)rex->offs[n].start, \ |
| 5198 | (IV)rex->offs[n].end \ |
| 5199 | )) |
| 5200 | |
| 5201 | case CLOSE: /* ) */ |
| 5202 | n = ARG(scan); /* which paren pair */ |
| 5203 | CLOSE_CAPTURE; |
| 5204 | if (n > rex->lastparen) |
| 5205 | rex->lastparen = n; |
| 5206 | rex->lastcloseparen = n; |
| 5207 | if (cur_eval && cur_eval->u.eval.close_paren == n) { |
| 5208 | goto fake_end; |
| 5209 | } |
| 5210 | break; |
| 5211 | |
| 5212 | case ACCEPT: /* (*ACCEPT) */ |
| 5213 | if (ARG(scan)){ |
| 5214 | regnode *cursor; |
| 5215 | for (cursor=scan; |
| 5216 | cursor && OP(cursor)!=END; |
| 5217 | cursor=regnext(cursor)) |
| 5218 | { |
| 5219 | if ( OP(cursor)==CLOSE ){ |
| 5220 | n = ARG(cursor); |
| 5221 | if ( n <= lastopen ) { |
| 5222 | CLOSE_CAPTURE; |
| 5223 | if (n > rex->lastparen) |
| 5224 | rex->lastparen = n; |
| 5225 | rex->lastcloseparen = n; |
| 5226 | if ( n == ARG(scan) || (cur_eval && |
| 5227 | cur_eval->u.eval.close_paren == n)) |
| 5228 | break; |
| 5229 | } |
| 5230 | } |
| 5231 | } |
| 5232 | } |
| 5233 | goto fake_end; |
| 5234 | /*NOTREACHED*/ |
| 5235 | |
| 5236 | case GROUPP: /* (?(1)) */ |
| 5237 | n = ARG(scan); /* which paren pair */ |
| 5238 | sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1); |
| 5239 | break; |
| 5240 | |
| 5241 | case NGROUPP: /* (?(<name>)) */ |
| 5242 | /* reg_check_named_buff_matched returns 0 for no match */ |
| 5243 | sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan)); |
| 5244 | break; |
| 5245 | |
| 5246 | case INSUBP: /* (?(R)) */ |
| 5247 | n = ARG(scan); |
| 5248 | sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n)); |
| 5249 | break; |
| 5250 | |
| 5251 | case DEFINEP: /* (?(DEFINE)) */ |
| 5252 | sw = 0; |
| 5253 | break; |
| 5254 | |
| 5255 | case IFTHEN: /* (?(cond)A|B) */ |
| 5256 | PL_reg_leftiter = PL_reg_maxiter; /* Void cache */ |
| 5257 | if (sw) |
| 5258 | next = NEXTOPER(NEXTOPER(scan)); |
| 5259 | else { |
| 5260 | next = scan + ARG(scan); |
| 5261 | if (OP(next) == IFTHEN) /* Fake one. */ |
| 5262 | next = NEXTOPER(NEXTOPER(next)); |
| 5263 | } |
| 5264 | break; |
| 5265 | |
| 5266 | case LOGICAL: /* modifier for EVAL and IFMATCH */ |
| 5267 | logical = scan->flags; |
| 5268 | break; |
| 5269 | |
| 5270 | /******************************************************************* |
| 5271 | |
| 5272 | The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/ |
| 5273 | pattern, where A and B are subpatterns. (For simple A, CURLYM or |
| 5274 | STAR/PLUS/CURLY/CURLYN are used instead.) |
| 5275 | |
| 5276 | A*B is compiled as <CURLYX><A><WHILEM><B> |
| 5277 | |
| 5278 | On entry to the subpattern, CURLYX is called. This pushes a CURLYX |
| 5279 | state, which contains the current count, initialised to -1. It also sets |
| 5280 | cur_curlyx to point to this state, with any previous value saved in the |
| 5281 | state block. |
| 5282 | |
| 5283 | CURLYX then jumps straight to the WHILEM op, rather than executing A, |
| 5284 | since the pattern may possibly match zero times (i.e. it's a while {} loop |
| 5285 | rather than a do {} while loop). |
| 5286 | |
| 5287 | Each entry to WHILEM represents a successful match of A. The count in the |
| 5288 | CURLYX block is incremented, another WHILEM state is pushed, and execution |
| 5289 | passes to A or B depending on greediness and the current count. |
| 5290 | |
| 5291 | For example, if matching against the string a1a2a3b (where the aN are |
| 5292 | substrings that match /A/), then the match progresses as follows: (the |
| 5293 | pushed states are interspersed with the bits of strings matched so far): |
| 5294 | |
| 5295 | <CURLYX cnt=-1> |
| 5296 | <CURLYX cnt=0><WHILEM> |
| 5297 | <CURLYX cnt=1><WHILEM> a1 <WHILEM> |
| 5298 | <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM> |
| 5299 | <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> |
| 5300 | <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b |
| 5301 | |
| 5302 | (Contrast this with something like CURLYM, which maintains only a single |
| 5303 | backtrack state: |
| 5304 | |
| 5305 | <CURLYM cnt=0> a1 |
| 5306 | a1 <CURLYM cnt=1> a2 |
| 5307 | a1 a2 <CURLYM cnt=2> a3 |
| 5308 | a1 a2 a3 <CURLYM cnt=3> b |
| 5309 | ) |
| 5310 | |
| 5311 | Each WHILEM state block marks a point to backtrack to upon partial failure |
| 5312 | of A or B, and also contains some minor state data related to that |
| 5313 | iteration. The CURLYX block, pointed to by cur_curlyx, contains the |
| 5314 | overall state, such as the count, and pointers to the A and B ops. |
| 5315 | |
| 5316 | This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx |
| 5317 | must always point to the *current* CURLYX block, the rules are: |
| 5318 | |
| 5319 | When executing CURLYX, save the old cur_curlyx in the CURLYX state block, |
| 5320 | and set cur_curlyx to point the new block. |
| 5321 | |
| 5322 | When popping the CURLYX block after a successful or unsuccessful match, |
| 5323 | restore the previous cur_curlyx. |
| 5324 | |
| 5325 | When WHILEM is about to execute B, save the current cur_curlyx, and set it |
| 5326 | to the outer one saved in the CURLYX block. |
| 5327 | |
| 5328 | When popping the WHILEM block after a successful or unsuccessful B match, |
| 5329 | restore the previous cur_curlyx. |
| 5330 | |
| 5331 | Here's an example for the pattern (AI* BI)*BO |
| 5332 | I and O refer to inner and outer, C and W refer to CURLYX and WHILEM: |
| 5333 | |
| 5334 | cur_ |
| 5335 | curlyx backtrack stack |
| 5336 | ------ --------------- |
| 5337 | NULL |
| 5338 | CO <CO prev=NULL> <WO> |
| 5339 | CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai |
| 5340 | CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi |
| 5341 | NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo |
| 5342 | |
| 5343 | At this point the pattern succeeds, and we work back down the stack to |
| 5344 | clean up, restoring as we go: |
| 5345 | |
| 5346 | CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi |
| 5347 | CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai |
| 5348 | CO <CO prev=NULL> <WO> |
| 5349 | NULL |
| 5350 | |
| 5351 | *******************************************************************/ |
| 5352 | |
| 5353 | #define ST st->u.curlyx |
| 5354 | |
| 5355 | case CURLYX: /* start of /A*B/ (for complex A) */ |
| 5356 | { |
| 5357 | /* No need to save/restore up to this paren */ |
| 5358 | I32 parenfloor = scan->flags; |
| 5359 | |
| 5360 | assert(next); /* keep Coverity happy */ |
| 5361 | if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */ |
| 5362 | next += ARG(next); |
| 5363 | |
| 5364 | /* XXXX Probably it is better to teach regpush to support |
| 5365 | parenfloor > maxopenparen ... */ |
| 5366 | if (parenfloor > (I32)rex->lastparen) |
| 5367 | parenfloor = rex->lastparen; /* Pessimization... */ |
| 5368 | |
| 5369 | ST.prev_curlyx= cur_curlyx; |
| 5370 | cur_curlyx = st; |
| 5371 | ST.cp = PL_savestack_ix; |
| 5372 | |
| 5373 | /* these fields contain the state of the current curly. |
| 5374 | * they are accessed by subsequent WHILEMs */ |
| 5375 | ST.parenfloor = parenfloor; |
| 5376 | ST.me = scan; |
| 5377 | ST.B = next; |
| 5378 | ST.minmod = minmod; |
| 5379 | minmod = 0; |
| 5380 | ST.count = -1; /* this will be updated by WHILEM */ |
| 5381 | ST.lastloc = NULL; /* this will be updated by WHILEM */ |
| 5382 | |
| 5383 | PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput); |
| 5384 | assert(0); /* NOTREACHED */ |
| 5385 | } |
| 5386 | |
| 5387 | case CURLYX_end: /* just finished matching all of A*B */ |
| 5388 | cur_curlyx = ST.prev_curlyx; |
| 5389 | sayYES; |
| 5390 | assert(0); /* NOTREACHED */ |
| 5391 | |
| 5392 | case CURLYX_end_fail: /* just failed to match all of A*B */ |
| 5393 | regcpblow(ST.cp); |
| 5394 | cur_curlyx = ST.prev_curlyx; |
| 5395 | sayNO; |
| 5396 | assert(0); /* NOTREACHED */ |
| 5397 | |
| 5398 | |
| 5399 | #undef ST |
| 5400 | #define ST st->u.whilem |
| 5401 | |
| 5402 | case WHILEM: /* just matched an A in /A*B/ (for complex A) */ |
| 5403 | { |
| 5404 | /* see the discussion above about CURLYX/WHILEM */ |
| 5405 | I32 n; |
| 5406 | int min = ARG1(cur_curlyx->u.curlyx.me); |
| 5407 | int max = ARG2(cur_curlyx->u.curlyx.me); |
| 5408 | regnode *A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS; |
| 5409 | |
| 5410 | assert(cur_curlyx); /* keep Coverity happy */ |
| 5411 | n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */ |
| 5412 | ST.save_lastloc = cur_curlyx->u.curlyx.lastloc; |
| 5413 | ST.cache_offset = 0; |
| 5414 | ST.cache_mask = 0; |
| 5415 | |
| 5416 | |
| 5417 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 5418 | "%*s whilem: matched %ld out of %d..%d\n", |
| 5419 | REPORT_CODE_OFF+depth*2, "", (long)n, min, max) |
| 5420 | ); |
| 5421 | |
| 5422 | /* First just match a string of min A's. */ |
| 5423 | |
| 5424 | if (n < min) { |
| 5425 | ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, |
| 5426 | maxopenparen); |
| 5427 | cur_curlyx->u.curlyx.lastloc = locinput; |
| 5428 | REGCP_SET(ST.lastcp); |
| 5429 | |
| 5430 | PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput); |
| 5431 | assert(0); /* NOTREACHED */ |
| 5432 | } |
| 5433 | |
| 5434 | /* If degenerate A matches "", assume A done. */ |
| 5435 | |
| 5436 | if (locinput == cur_curlyx->u.curlyx.lastloc) { |
| 5437 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 5438 | "%*s whilem: empty match detected, trying continuation...\n", |
| 5439 | REPORT_CODE_OFF+depth*2, "") |
| 5440 | ); |
| 5441 | goto do_whilem_B_max; |
| 5442 | } |
| 5443 | |
| 5444 | /* super-linear cache processing */ |
| 5445 | |
| 5446 | if (scan->flags) { |
| 5447 | |
| 5448 | if (!PL_reg_maxiter) { |
| 5449 | /* start the countdown: Postpone detection until we |
| 5450 | * know the match is not *that* much linear. */ |
| 5451 | PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4); |
| 5452 | /* possible overflow for long strings and many CURLYX's */ |
| 5453 | if (PL_reg_maxiter < 0) |
| 5454 | PL_reg_maxiter = I32_MAX; |
| 5455 | PL_reg_leftiter = PL_reg_maxiter; |
| 5456 | } |
| 5457 | |
| 5458 | if (PL_reg_leftiter-- == 0) { |
| 5459 | /* initialise cache */ |
| 5460 | const I32 size = (PL_reg_maxiter + 7)/8; |
| 5461 | if (PL_reg_poscache) { |
| 5462 | if ((I32)PL_reg_poscache_size < size) { |
| 5463 | Renew(PL_reg_poscache, size, char); |
| 5464 | PL_reg_poscache_size = size; |
| 5465 | } |
| 5466 | Zero(PL_reg_poscache, size, char); |
| 5467 | } |
| 5468 | else { |
| 5469 | PL_reg_poscache_size = size; |
| 5470 | Newxz(PL_reg_poscache, size, char); |
| 5471 | } |
| 5472 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 5473 | "%swhilem: Detected a super-linear match, switching on caching%s...\n", |
| 5474 | PL_colors[4], PL_colors[5]) |
| 5475 | ); |
| 5476 | } |
| 5477 | |
| 5478 | if (PL_reg_leftiter < 0) { |
| 5479 | /* have we already failed at this position? */ |
| 5480 | I32 offset, mask; |
| 5481 | offset = (scan->flags & 0xf) - 1 |
| 5482 | + (locinput - PL_bostr) * (scan->flags>>4); |
| 5483 | mask = 1 << (offset % 8); |
| 5484 | offset /= 8; |
| 5485 | if (PL_reg_poscache[offset] & mask) { |
| 5486 | DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, |
| 5487 | "%*s whilem: (cache) already tried at this position...\n", |
| 5488 | REPORT_CODE_OFF+depth*2, "") |
| 5489 | ); |
| 5490 | sayNO; /* cache records failure */ |
| 5491 | } |
| 5492 | ST.cache_offset = offset; |
| 5493 | ST.cache_mask = mask; |
| 5494 | } |
| 5495 | } |
| 5496 | |
| 5497 | /* Prefer B over A for minimal matching. */ |
| 5498 | |
| 5499 | if (cur_curlyx->u.curlyx.minmod) { |
| 5500 | ST.save_curlyx = cur_curlyx; |
| 5501 | cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; |
| 5502 | ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor, |
| 5503 | maxopenparen); |
| 5504 | REGCP_SET(ST.lastcp); |
| 5505 | PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B, |
| 5506 | locinput); |
| 5507 | assert(0); /* NOTREACHED */ |
| 5508 | } |
| 5509 | |
| 5510 | /* Prefer A over B for maximal matching. */ |
| 5511 | |
| 5512 | if (n < max) { /* More greed allowed? */ |
| 5513 | ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, |
| 5514 | maxopenparen); |
| 5515 | cur_curlyx->u.curlyx.lastloc = locinput; |
| 5516 | REGCP_SET(ST.lastcp); |
| 5517 | PUSH_STATE_GOTO(WHILEM_A_max, A, locinput); |
| 5518 | assert(0); /* NOTREACHED */ |
| 5519 | } |
| 5520 | goto do_whilem_B_max; |
| 5521 | } |
| 5522 | assert(0); /* NOTREACHED */ |
| 5523 | |
| 5524 | case WHILEM_B_min: /* just matched B in a minimal match */ |
| 5525 | case WHILEM_B_max: /* just matched B in a maximal match */ |
| 5526 | cur_curlyx = ST.save_curlyx; |
| 5527 | sayYES; |
| 5528 | assert(0); /* NOTREACHED */ |
| 5529 | |
| 5530 | case WHILEM_B_max_fail: /* just failed to match B in a maximal match */ |
| 5531 | cur_curlyx = ST.save_curlyx; |
| 5532 | cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; |
| 5533 | cur_curlyx->u.curlyx.count--; |
| 5534 | CACHEsayNO; |
| 5535 | assert(0); /* NOTREACHED */ |
| 5536 | |
| 5537 | case WHILEM_A_min_fail: /* just failed to match A in a minimal match */ |
| 5538 | /* FALL THROUGH */ |
| 5539 | case WHILEM_A_pre_fail: /* just failed to match even minimal A */ |
| 5540 | REGCP_UNWIND(ST.lastcp); |
| 5541 | regcppop(rex, &maxopenparen); |
| 5542 | cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; |
| 5543 | cur_curlyx->u.curlyx.count--; |
| 5544 | CACHEsayNO; |
| 5545 | assert(0); /* NOTREACHED */ |
| 5546 | |
| 5547 | case WHILEM_A_max_fail: /* just failed to match A in a maximal match */ |
| 5548 | REGCP_UNWIND(ST.lastcp); |
| 5549 | regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */ |
| 5550 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 5551 | "%*s whilem: failed, trying continuation...\n", |
| 5552 | REPORT_CODE_OFF+depth*2, "") |
| 5553 | ); |
| 5554 | do_whilem_B_max: |
| 5555 | if (cur_curlyx->u.curlyx.count >= REG_INFTY |
| 5556 | && ckWARN(WARN_REGEXP) |
| 5557 | && !reginfo->warned) |
| 5558 | { |
| 5559 | reginfo->warned = TRUE; |
| 5560 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), |
| 5561 | "Complex regular subexpression recursion limit (%d) " |
| 5562 | "exceeded", |
| 5563 | REG_INFTY - 1); |
| 5564 | } |
| 5565 | |
| 5566 | /* now try B */ |
| 5567 | ST.save_curlyx = cur_curlyx; |
| 5568 | cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; |
| 5569 | PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B, |
| 5570 | locinput); |
| 5571 | assert(0); /* NOTREACHED */ |
| 5572 | |
| 5573 | case WHILEM_B_min_fail: /* just failed to match B in a minimal match */ |
| 5574 | cur_curlyx = ST.save_curlyx; |
| 5575 | REGCP_UNWIND(ST.lastcp); |
| 5576 | regcppop(rex, &maxopenparen); |
| 5577 | |
| 5578 | if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) { |
| 5579 | /* Maximum greed exceeded */ |
| 5580 | if (cur_curlyx->u.curlyx.count >= REG_INFTY |
| 5581 | && ckWARN(WARN_REGEXP) |
| 5582 | && !reginfo->warned) |
| 5583 | { |
| 5584 | reginfo->warned = TRUE; |
| 5585 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), |
| 5586 | "Complex regular subexpression recursion " |
| 5587 | "limit (%d) exceeded", |
| 5588 | REG_INFTY - 1); |
| 5589 | } |
| 5590 | cur_curlyx->u.curlyx.count--; |
| 5591 | CACHEsayNO; |
| 5592 | } |
| 5593 | |
| 5594 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 5595 | "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "") |
| 5596 | ); |
| 5597 | /* Try grabbing another A and see if it helps. */ |
| 5598 | cur_curlyx->u.curlyx.lastloc = locinput; |
| 5599 | ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, |
| 5600 | maxopenparen); |
| 5601 | REGCP_SET(ST.lastcp); |
| 5602 | PUSH_STATE_GOTO(WHILEM_A_min, |
| 5603 | /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS, |
| 5604 | locinput); |
| 5605 | assert(0); /* NOTREACHED */ |
| 5606 | |
| 5607 | #undef ST |
| 5608 | #define ST st->u.branch |
| 5609 | |
| 5610 | case BRANCHJ: /* /(...|A|...)/ with long next pointer */ |
| 5611 | next = scan + ARG(scan); |
| 5612 | if (next == scan) |
| 5613 | next = NULL; |
| 5614 | scan = NEXTOPER(scan); |
| 5615 | /* FALL THROUGH */ |
| 5616 | |
| 5617 | case BRANCH: /* /(...|A|...)/ */ |
| 5618 | scan = NEXTOPER(scan); /* scan now points to inner node */ |
| 5619 | ST.lastparen = rex->lastparen; |
| 5620 | ST.lastcloseparen = rex->lastcloseparen; |
| 5621 | ST.next_branch = next; |
| 5622 | REGCP_SET(ST.cp); |
| 5623 | |
| 5624 | /* Now go into the branch */ |
| 5625 | if (has_cutgroup) { |
| 5626 | PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput); |
| 5627 | } else { |
| 5628 | PUSH_STATE_GOTO(BRANCH_next, scan, locinput); |
| 5629 | } |
| 5630 | assert(0); /* NOTREACHED */ |
| 5631 | |
| 5632 | case CUTGROUP: /* /(*THEN)/ */ |
| 5633 | sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL : |
| 5634 | MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 5635 | PUSH_STATE_GOTO(CUTGROUP_next, next, locinput); |
| 5636 | assert(0); /* NOTREACHED */ |
| 5637 | |
| 5638 | case CUTGROUP_next_fail: |
| 5639 | do_cutgroup = 1; |
| 5640 | no_final = 1; |
| 5641 | if (st->u.mark.mark_name) |
| 5642 | sv_commit = st->u.mark.mark_name; |
| 5643 | sayNO; |
| 5644 | assert(0); /* NOTREACHED */ |
| 5645 | |
| 5646 | case BRANCH_next: |
| 5647 | sayYES; |
| 5648 | assert(0); /* NOTREACHED */ |
| 5649 | |
| 5650 | case BRANCH_next_fail: /* that branch failed; try the next, if any */ |
| 5651 | if (do_cutgroup) { |
| 5652 | do_cutgroup = 0; |
| 5653 | no_final = 0; |
| 5654 | } |
| 5655 | REGCP_UNWIND(ST.cp); |
| 5656 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
| 5657 | scan = ST.next_branch; |
| 5658 | /* no more branches? */ |
| 5659 | if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) { |
| 5660 | DEBUG_EXECUTE_r({ |
| 5661 | PerlIO_printf( Perl_debug_log, |
| 5662 | "%*s %sBRANCH failed...%s\n", |
| 5663 | REPORT_CODE_OFF+depth*2, "", |
| 5664 | PL_colors[4], |
| 5665 | PL_colors[5] ); |
| 5666 | }); |
| 5667 | sayNO_SILENT; |
| 5668 | } |
| 5669 | continue; /* execute next BRANCH[J] op */ |
| 5670 | assert(0); /* NOTREACHED */ |
| 5671 | |
| 5672 | case MINMOD: /* next op will be non-greedy, e.g. A*? */ |
| 5673 | minmod = 1; |
| 5674 | break; |
| 5675 | |
| 5676 | #undef ST |
| 5677 | #define ST st->u.curlym |
| 5678 | |
| 5679 | case CURLYM: /* /A{m,n}B/ where A is fixed-length */ |
| 5680 | |
| 5681 | /* This is an optimisation of CURLYX that enables us to push |
| 5682 | * only a single backtracking state, no matter how many matches |
| 5683 | * there are in {m,n}. It relies on the pattern being constant |
| 5684 | * length, with no parens to influence future backrefs |
| 5685 | */ |
| 5686 | |
| 5687 | ST.me = scan; |
| 5688 | scan = NEXTOPER(scan) + NODE_STEP_REGNODE; |
| 5689 | |
| 5690 | ST.lastparen = rex->lastparen; |
| 5691 | ST.lastcloseparen = rex->lastcloseparen; |
| 5692 | |
| 5693 | /* if paren positive, emulate an OPEN/CLOSE around A */ |
| 5694 | if (ST.me->flags) { |
| 5695 | U32 paren = ST.me->flags; |
| 5696 | if (paren > maxopenparen) |
| 5697 | maxopenparen = paren; |
| 5698 | scan += NEXT_OFF(scan); /* Skip former OPEN. */ |
| 5699 | } |
| 5700 | ST.A = scan; |
| 5701 | ST.B = next; |
| 5702 | ST.alen = 0; |
| 5703 | ST.count = 0; |
| 5704 | ST.minmod = minmod; |
| 5705 | minmod = 0; |
| 5706 | ST.c1 = CHRTEST_UNINIT; |
| 5707 | REGCP_SET(ST.cp); |
| 5708 | |
| 5709 | if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */ |
| 5710 | goto curlym_do_B; |
| 5711 | |
| 5712 | curlym_do_A: /* execute the A in /A{m,n}B/ */ |
| 5713 | PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */ |
| 5714 | assert(0); /* NOTREACHED */ |
| 5715 | |
| 5716 | case CURLYM_A: /* we've just matched an A */ |
| 5717 | ST.count++; |
| 5718 | /* after first match, determine A's length: u.curlym.alen */ |
| 5719 | if (ST.count == 1) { |
| 5720 | if (PL_reg_match_utf8) { |
| 5721 | char *s = st->locinput; |
| 5722 | while (s < locinput) { |
| 5723 | ST.alen++; |
| 5724 | s += UTF8SKIP(s); |
| 5725 | } |
| 5726 | } |
| 5727 | else { |
| 5728 | ST.alen = locinput - st->locinput; |
| 5729 | } |
| 5730 | if (ST.alen == 0) |
| 5731 | ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me); |
| 5732 | } |
| 5733 | DEBUG_EXECUTE_r( |
| 5734 | PerlIO_printf(Perl_debug_log, |
| 5735 | "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n", |
| 5736 | (int)(REPORT_CODE_OFF+(depth*2)), "", |
| 5737 | (IV) ST.count, (IV)ST.alen) |
| 5738 | ); |
| 5739 | |
| 5740 | if (cur_eval && cur_eval->u.eval.close_paren && |
| 5741 | cur_eval->u.eval.close_paren == (U32)ST.me->flags) |
| 5742 | goto fake_end; |
| 5743 | |
| 5744 | { |
| 5745 | I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)); |
| 5746 | if ( max == REG_INFTY || ST.count < max ) |
| 5747 | goto curlym_do_A; /* try to match another A */ |
| 5748 | } |
| 5749 | goto curlym_do_B; /* try to match B */ |
| 5750 | |
| 5751 | case CURLYM_A_fail: /* just failed to match an A */ |
| 5752 | REGCP_UNWIND(ST.cp); |
| 5753 | |
| 5754 | if (ST.minmod || ST.count < ARG1(ST.me) /* min*/ |
| 5755 | || (cur_eval && cur_eval->u.eval.close_paren && |
| 5756 | cur_eval->u.eval.close_paren == (U32)ST.me->flags)) |
| 5757 | sayNO; |
| 5758 | |
| 5759 | curlym_do_B: /* execute the B in /A{m,n}B/ */ |
| 5760 | if (ST.c1 == CHRTEST_UNINIT) { |
| 5761 | /* calculate c1 and c2 for possible match of 1st char |
| 5762 | * following curly */ |
| 5763 | ST.c1 = ST.c2 = CHRTEST_VOID; |
| 5764 | if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) { |
| 5765 | regnode *text_node = ST.B; |
| 5766 | if (! HAS_TEXT(text_node)) |
| 5767 | FIND_NEXT_IMPT(text_node); |
| 5768 | /* this used to be |
| 5769 | |
| 5770 | (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT) |
| 5771 | |
| 5772 | But the former is redundant in light of the latter. |
| 5773 | |
| 5774 | if this changes back then the macro for |
| 5775 | IS_TEXT and friends need to change. |
| 5776 | */ |
| 5777 | if (PL_regkind[OP(text_node)] == EXACT) { |
| 5778 | if (! S_setup_EXACTISH_ST_c1_c2(aTHX_ |
| 5779 | text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8, |
| 5780 | is_utf8_pat)) |
| 5781 | { |
| 5782 | sayNO; |
| 5783 | } |
| 5784 | } |
| 5785 | } |
| 5786 | } |
| 5787 | |
| 5788 | DEBUG_EXECUTE_r( |
| 5789 | PerlIO_printf(Perl_debug_log, |
| 5790 | "%*s CURLYM trying tail with matches=%"IVdf"...\n", |
| 5791 | (int)(REPORT_CODE_OFF+(depth*2)), |
| 5792 | "", (IV)ST.count) |
| 5793 | ); |
| 5794 | if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) { |
| 5795 | if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) { |
| 5796 | if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)) |
| 5797 | && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput))) |
| 5798 | { |
| 5799 | /* simulate B failing */ |
| 5800 | DEBUG_OPTIMISE_r( |
| 5801 | PerlIO_printf(Perl_debug_log, |
| 5802 | "%*s CURLYM Fast bail next target=U+%"UVXf" c1=U+%"UVXf" c2=U+%"UVXf"\n", |
| 5803 | (int)(REPORT_CODE_OFF+(depth*2)),"", |
| 5804 | valid_utf8_to_uvchr((U8 *) locinput, NULL), |
| 5805 | valid_utf8_to_uvchr(ST.c1_utf8, NULL), |
| 5806 | valid_utf8_to_uvchr(ST.c2_utf8, NULL)) |
| 5807 | ); |
| 5808 | state_num = CURLYM_B_fail; |
| 5809 | goto reenter_switch; |
| 5810 | } |
| 5811 | } |
| 5812 | else if (nextchr != ST.c1 && nextchr != ST.c2) { |
| 5813 | /* simulate B failing */ |
| 5814 | DEBUG_OPTIMISE_r( |
| 5815 | PerlIO_printf(Perl_debug_log, |
| 5816 | "%*s CURLYM Fast bail next target=U+%X c1=U+%X c2=U+%X\n", |
| 5817 | (int)(REPORT_CODE_OFF+(depth*2)),"", |
| 5818 | (int) nextchr, ST.c1, ST.c2) |
| 5819 | ); |
| 5820 | state_num = CURLYM_B_fail; |
| 5821 | goto reenter_switch; |
| 5822 | } |
| 5823 | } |
| 5824 | |
| 5825 | if (ST.me->flags) { |
| 5826 | /* emulate CLOSE: mark current A as captured */ |
| 5827 | I32 paren = ST.me->flags; |
| 5828 | if (ST.count) { |
| 5829 | rex->offs[paren].start |
| 5830 | = HOPc(locinput, -ST.alen) - PL_bostr; |
| 5831 | rex->offs[paren].end = locinput - PL_bostr; |
| 5832 | if ((U32)paren > rex->lastparen) |
| 5833 | rex->lastparen = paren; |
| 5834 | rex->lastcloseparen = paren; |
| 5835 | } |
| 5836 | else |
| 5837 | rex->offs[paren].end = -1; |
| 5838 | if (cur_eval && cur_eval->u.eval.close_paren && |
| 5839 | cur_eval->u.eval.close_paren == (U32)ST.me->flags) |
| 5840 | { |
| 5841 | if (ST.count) |
| 5842 | goto fake_end; |
| 5843 | else |
| 5844 | sayNO; |
| 5845 | } |
| 5846 | } |
| 5847 | |
| 5848 | PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */ |
| 5849 | assert(0); /* NOTREACHED */ |
| 5850 | |
| 5851 | case CURLYM_B_fail: /* just failed to match a B */ |
| 5852 | REGCP_UNWIND(ST.cp); |
| 5853 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
| 5854 | if (ST.minmod) { |
| 5855 | I32 max = ARG2(ST.me); |
| 5856 | if (max != REG_INFTY && ST.count == max) |
| 5857 | sayNO; |
| 5858 | goto curlym_do_A; /* try to match a further A */ |
| 5859 | } |
| 5860 | /* backtrack one A */ |
| 5861 | if (ST.count == ARG1(ST.me) /* min */) |
| 5862 | sayNO; |
| 5863 | ST.count--; |
| 5864 | SET_locinput(HOPc(locinput, -ST.alen)); |
| 5865 | goto curlym_do_B; /* try to match B */ |
| 5866 | |
| 5867 | #undef ST |
| 5868 | #define ST st->u.curly |
| 5869 | |
| 5870 | #define CURLY_SETPAREN(paren, success) \ |
| 5871 | if (paren) { \ |
| 5872 | if (success) { \ |
| 5873 | rex->offs[paren].start = HOPc(locinput, -1) - PL_bostr; \ |
| 5874 | rex->offs[paren].end = locinput - PL_bostr; \ |
| 5875 | if (paren > rex->lastparen) \ |
| 5876 | rex->lastparen = paren; \ |
| 5877 | rex->lastcloseparen = paren; \ |
| 5878 | } \ |
| 5879 | else { \ |
| 5880 | rex->offs[paren].end = -1; \ |
| 5881 | rex->lastparen = ST.lastparen; \ |
| 5882 | rex->lastcloseparen = ST.lastcloseparen; \ |
| 5883 | } \ |
| 5884 | } |
| 5885 | |
| 5886 | case STAR: /* /A*B/ where A is width 1 char */ |
| 5887 | ST.paren = 0; |
| 5888 | ST.min = 0; |
| 5889 | ST.max = REG_INFTY; |
| 5890 | scan = NEXTOPER(scan); |
| 5891 | goto repeat; |
| 5892 | |
| 5893 | case PLUS: /* /A+B/ where A is width 1 char */ |
| 5894 | ST.paren = 0; |
| 5895 | ST.min = 1; |
| 5896 | ST.max = REG_INFTY; |
| 5897 | scan = NEXTOPER(scan); |
| 5898 | goto repeat; |
| 5899 | |
| 5900 | case CURLYN: /* /(A){m,n}B/ where A is width 1 char */ |
| 5901 | ST.paren = scan->flags; /* Which paren to set */ |
| 5902 | ST.lastparen = rex->lastparen; |
| 5903 | ST.lastcloseparen = rex->lastcloseparen; |
| 5904 | if (ST.paren > maxopenparen) |
| 5905 | maxopenparen = ST.paren; |
| 5906 | ST.min = ARG1(scan); /* min to match */ |
| 5907 | ST.max = ARG2(scan); /* max to match */ |
| 5908 | if (cur_eval && cur_eval->u.eval.close_paren && |
| 5909 | cur_eval->u.eval.close_paren == (U32)ST.paren) { |
| 5910 | ST.min=1; |
| 5911 | ST.max=1; |
| 5912 | } |
| 5913 | scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE); |
| 5914 | goto repeat; |
| 5915 | |
| 5916 | case CURLY: /* /A{m,n}B/ where A is width 1 char */ |
| 5917 | ST.paren = 0; |
| 5918 | ST.min = ARG1(scan); /* min to match */ |
| 5919 | ST.max = ARG2(scan); /* max to match */ |
| 5920 | scan = NEXTOPER(scan) + NODE_STEP_REGNODE; |
| 5921 | repeat: |
| 5922 | /* |
| 5923 | * Lookahead to avoid useless match attempts |
| 5924 | * when we know what character comes next. |
| 5925 | * |
| 5926 | * Used to only do .*x and .*?x, but now it allows |
| 5927 | * for )'s, ('s and (?{ ... })'s to be in the way |
| 5928 | * of the quantifier and the EXACT-like node. -- japhy |
| 5929 | */ |
| 5930 | |
| 5931 | assert(ST.min <= ST.max); |
| 5932 | if (! HAS_TEXT(next) && ! JUMPABLE(next)) { |
| 5933 | ST.c1 = ST.c2 = CHRTEST_VOID; |
| 5934 | } |
| 5935 | else { |
| 5936 | regnode *text_node = next; |
| 5937 | |
| 5938 | if (! HAS_TEXT(text_node)) |
| 5939 | FIND_NEXT_IMPT(text_node); |
| 5940 | |
| 5941 | if (! HAS_TEXT(text_node)) |
| 5942 | ST.c1 = ST.c2 = CHRTEST_VOID; |
| 5943 | else { |
| 5944 | if ( PL_regkind[OP(text_node)] != EXACT ) { |
| 5945 | ST.c1 = ST.c2 = CHRTEST_VOID; |
| 5946 | } |
| 5947 | else { |
| 5948 | |
| 5949 | /* Currently we only get here when |
| 5950 | |
| 5951 | PL_rekind[OP(text_node)] == EXACT |
| 5952 | |
| 5953 | if this changes back then the macro for IS_TEXT and |
| 5954 | friends need to change. */ |
| 5955 | if (! S_setup_EXACTISH_ST_c1_c2(aTHX_ |
| 5956 | text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8, |
| 5957 | is_utf8_pat)) |
| 5958 | { |
| 5959 | sayNO; |
| 5960 | } |
| 5961 | } |
| 5962 | } |
| 5963 | } |
| 5964 | |
| 5965 | ST.A = scan; |
| 5966 | ST.B = next; |
| 5967 | if (minmod) { |
| 5968 | char *li = locinput; |
| 5969 | minmod = 0; |
| 5970 | if (ST.min && |
| 5971 | regrepeat(rex, &li, ST.A, ST.min, depth, is_utf8_pat) |
| 5972 | < ST.min) |
| 5973 | sayNO; |
| 5974 | SET_locinput(li); |
| 5975 | ST.count = ST.min; |
| 5976 | REGCP_SET(ST.cp); |
| 5977 | if (ST.c1 == CHRTEST_VOID) |
| 5978 | goto curly_try_B_min; |
| 5979 | |
| 5980 | ST.oldloc = locinput; |
| 5981 | |
| 5982 | /* set ST.maxpos to the furthest point along the |
| 5983 | * string that could possibly match */ |
| 5984 | if (ST.max == REG_INFTY) { |
| 5985 | ST.maxpos = PL_regeol - 1; |
| 5986 | if (utf8_target) |
| 5987 | while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos)) |
| 5988 | ST.maxpos--; |
| 5989 | } |
| 5990 | else if (utf8_target) { |
| 5991 | int m = ST.max - ST.min; |
| 5992 | for (ST.maxpos = locinput; |
| 5993 | m >0 && ST.maxpos < PL_regeol; m--) |
| 5994 | ST.maxpos += UTF8SKIP(ST.maxpos); |
| 5995 | } |
| 5996 | else { |
| 5997 | ST.maxpos = locinput + ST.max - ST.min; |
| 5998 | if (ST.maxpos >= PL_regeol) |
| 5999 | ST.maxpos = PL_regeol - 1; |
| 6000 | } |
| 6001 | goto curly_try_B_min_known; |
| 6002 | |
| 6003 | } |
| 6004 | else { |
| 6005 | /* avoid taking address of locinput, so it can remain |
| 6006 | * a register var */ |
| 6007 | char *li = locinput; |
| 6008 | ST.count = regrepeat(rex, &li, ST.A, ST.max, depth, |
| 6009 | is_utf8_pat); |
| 6010 | if (ST.count < ST.min) |
| 6011 | sayNO; |
| 6012 | SET_locinput(li); |
| 6013 | if ((ST.count > ST.min) |
| 6014 | && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL)) |
| 6015 | { |
| 6016 | /* A{m,n} must come at the end of the string, there's |
| 6017 | * no point in backing off ... */ |
| 6018 | ST.min = ST.count; |
| 6019 | /* ...except that $ and \Z can match before *and* after |
| 6020 | newline at the end. Consider "\n\n" =~ /\n+\Z\n/. |
| 6021 | We may back off by one in this case. */ |
| 6022 | if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS) |
| 6023 | ST.min--; |
| 6024 | } |
| 6025 | REGCP_SET(ST.cp); |
| 6026 | goto curly_try_B_max; |
| 6027 | } |
| 6028 | assert(0); /* NOTREACHED */ |
| 6029 | |
| 6030 | |
| 6031 | case CURLY_B_min_known_fail: |
| 6032 | /* failed to find B in a non-greedy match where c1,c2 valid */ |
| 6033 | |
| 6034 | REGCP_UNWIND(ST.cp); |
| 6035 | if (ST.paren) { |
| 6036 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
| 6037 | } |
| 6038 | /* Couldn't or didn't -- move forward. */ |
| 6039 | ST.oldloc = locinput; |
| 6040 | if (utf8_target) |
| 6041 | locinput += UTF8SKIP(locinput); |
| 6042 | else |
| 6043 | locinput++; |
| 6044 | ST.count++; |
| 6045 | curly_try_B_min_known: |
| 6046 | /* find the next place where 'B' could work, then call B */ |
| 6047 | { |
| 6048 | int n; |
| 6049 | if (utf8_target) { |
| 6050 | n = (ST.oldloc == locinput) ? 0 : 1; |
| 6051 | if (ST.c1 == ST.c2) { |
| 6052 | /* set n to utf8_distance(oldloc, locinput) */ |
| 6053 | while (locinput <= ST.maxpos |
| 6054 | && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))) |
| 6055 | { |
| 6056 | locinput += UTF8SKIP(locinput); |
| 6057 | n++; |
| 6058 | } |
| 6059 | } |
| 6060 | else { |
| 6061 | /* set n to utf8_distance(oldloc, locinput) */ |
| 6062 | while (locinput <= ST.maxpos |
| 6063 | && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)) |
| 6064 | && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput))) |
| 6065 | { |
| 6066 | locinput += UTF8SKIP(locinput); |
| 6067 | n++; |
| 6068 | } |
| 6069 | } |
| 6070 | } |
| 6071 | else { /* Not utf8_target */ |
| 6072 | if (ST.c1 == ST.c2) { |
| 6073 | while (locinput <= ST.maxpos && |
| 6074 | UCHARAT(locinput) != ST.c1) |
| 6075 | locinput++; |
| 6076 | } |
| 6077 | else { |
| 6078 | while (locinput <= ST.maxpos |
| 6079 | && UCHARAT(locinput) != ST.c1 |
| 6080 | && UCHARAT(locinput) != ST.c2) |
| 6081 | locinput++; |
| 6082 | } |
| 6083 | n = locinput - ST.oldloc; |
| 6084 | } |
| 6085 | if (locinput > ST.maxpos) |
| 6086 | sayNO; |
| 6087 | if (n) { |
| 6088 | /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is |
| 6089 | * at b; check that everything between oldloc and |
| 6090 | * locinput matches */ |
| 6091 | char *li = ST.oldloc; |
| 6092 | ST.count += n; |
| 6093 | if (regrepeat(rex, &li, ST.A, n, depth, is_utf8_pat) < n) |
| 6094 | sayNO; |
| 6095 | assert(n == REG_INFTY || locinput == li); |
| 6096 | } |
| 6097 | CURLY_SETPAREN(ST.paren, ST.count); |
| 6098 | if (cur_eval && cur_eval->u.eval.close_paren && |
| 6099 | cur_eval->u.eval.close_paren == (U32)ST.paren) { |
| 6100 | goto fake_end; |
| 6101 | } |
| 6102 | PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput); |
| 6103 | } |
| 6104 | assert(0); /* NOTREACHED */ |
| 6105 | |
| 6106 | |
| 6107 | case CURLY_B_min_fail: |
| 6108 | /* failed to find B in a non-greedy match where c1,c2 invalid */ |
| 6109 | |
| 6110 | REGCP_UNWIND(ST.cp); |
| 6111 | if (ST.paren) { |
| 6112 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
| 6113 | } |
| 6114 | /* failed -- move forward one */ |
| 6115 | { |
| 6116 | char *li = locinput; |
| 6117 | if (!regrepeat(rex, &li, ST.A, 1, depth, is_utf8_pat)) { |
| 6118 | sayNO; |
| 6119 | } |
| 6120 | locinput = li; |
| 6121 | } |
| 6122 | { |
| 6123 | ST.count++; |
| 6124 | if (ST.count <= ST.max || (ST.max == REG_INFTY && |
| 6125 | ST.count > 0)) /* count overflow ? */ |
| 6126 | { |
| 6127 | curly_try_B_min: |
| 6128 | CURLY_SETPAREN(ST.paren, ST.count); |
| 6129 | if (cur_eval && cur_eval->u.eval.close_paren && |
| 6130 | cur_eval->u.eval.close_paren == (U32)ST.paren) { |
| 6131 | goto fake_end; |
| 6132 | } |
| 6133 | PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput); |
| 6134 | } |
| 6135 | } |
| 6136 | sayNO; |
| 6137 | assert(0); /* NOTREACHED */ |
| 6138 | |
| 6139 | |
| 6140 | curly_try_B_max: |
| 6141 | /* a successful greedy match: now try to match B */ |
| 6142 | if (cur_eval && cur_eval->u.eval.close_paren && |
| 6143 | cur_eval->u.eval.close_paren == (U32)ST.paren) { |
| 6144 | goto fake_end; |
| 6145 | } |
| 6146 | { |
| 6147 | bool could_match = locinput < PL_regeol; |
| 6148 | |
| 6149 | /* If it could work, try it. */ |
| 6150 | if (ST.c1 != CHRTEST_VOID && could_match) { |
| 6151 | if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target) |
| 6152 | { |
| 6153 | could_match = memEQ(locinput, |
| 6154 | ST.c1_utf8, |
| 6155 | UTF8SKIP(locinput)) |
| 6156 | || memEQ(locinput, |
| 6157 | ST.c2_utf8, |
| 6158 | UTF8SKIP(locinput)); |
| 6159 | } |
| 6160 | else { |
| 6161 | could_match = UCHARAT(locinput) == ST.c1 |
| 6162 | || UCHARAT(locinput) == ST.c2; |
| 6163 | } |
| 6164 | } |
| 6165 | if (ST.c1 == CHRTEST_VOID || could_match) { |
| 6166 | CURLY_SETPAREN(ST.paren, ST.count); |
| 6167 | PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput); |
| 6168 | assert(0); /* NOTREACHED */ |
| 6169 | } |
| 6170 | } |
| 6171 | /* FALL THROUGH */ |
| 6172 | |
| 6173 | case CURLY_B_max_fail: |
| 6174 | /* failed to find B in a greedy match */ |
| 6175 | |
| 6176 | REGCP_UNWIND(ST.cp); |
| 6177 | if (ST.paren) { |
| 6178 | UNWIND_PAREN(ST.lastparen, ST.lastcloseparen); |
| 6179 | } |
| 6180 | /* back up. */ |
| 6181 | if (--ST.count < ST.min) |
| 6182 | sayNO; |
| 6183 | locinput = HOPc(locinput, -1); |
| 6184 | goto curly_try_B_max; |
| 6185 | |
| 6186 | #undef ST |
| 6187 | |
| 6188 | case END: /* last op of main pattern */ |
| 6189 | fake_end: |
| 6190 | if (cur_eval) { |
| 6191 | /* we've just finished A in /(??{A})B/; now continue with B */ |
| 6192 | st->u.eval.saved_utf8_pat = is_utf8_pat; |
| 6193 | is_utf8_pat = cur_eval->u.eval.saved_utf8_pat; |
| 6194 | |
| 6195 | st->u.eval.prev_rex = rex_sv; /* inner */ |
| 6196 | |
| 6197 | /* Save *all* the positions. */ |
| 6198 | st->u.eval.cp = regcppush(rex, 0, maxopenparen); |
| 6199 | rex_sv = cur_eval->u.eval.prev_rex; |
| 6200 | SET_reg_curpm(rex_sv); |
| 6201 | rex = ReANY(rex_sv); |
| 6202 | rexi = RXi_GET(rex); |
| 6203 | cur_curlyx = cur_eval->u.eval.prev_curlyx; |
| 6204 | |
| 6205 | REGCP_SET(st->u.eval.lastcp); |
| 6206 | |
| 6207 | /* Restore parens of the outer rex without popping the |
| 6208 | * savestack */ |
| 6209 | S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp, |
| 6210 | &maxopenparen); |
| 6211 | |
| 6212 | st->u.eval.prev_eval = cur_eval; |
| 6213 | cur_eval = cur_eval->u.eval.prev_eval; |
| 6214 | DEBUG_EXECUTE_r( |
| 6215 | PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n", |
| 6216 | REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval));); |
| 6217 | if ( nochange_depth ) |
| 6218 | nochange_depth--; |
| 6219 | |
| 6220 | PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B, |
| 6221 | locinput); /* match B */ |
| 6222 | } |
| 6223 | |
| 6224 | if (locinput < reginfo->till) { |
| 6225 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, |
| 6226 | "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n", |
| 6227 | PL_colors[4], |
| 6228 | (long)(locinput - PL_reg_starttry), |
| 6229 | (long)(reginfo->till - PL_reg_starttry), |
| 6230 | PL_colors[5])); |
| 6231 | |
| 6232 | sayNO_SILENT; /* Cannot match: too short. */ |
| 6233 | } |
| 6234 | sayYES; /* Success! */ |
| 6235 | |
| 6236 | case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */ |
| 6237 | DEBUG_EXECUTE_r( |
| 6238 | PerlIO_printf(Perl_debug_log, |
| 6239 | "%*s %ssubpattern success...%s\n", |
| 6240 | REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])); |
| 6241 | sayYES; /* Success! */ |
| 6242 | |
| 6243 | #undef ST |
| 6244 | #define ST st->u.ifmatch |
| 6245 | |
| 6246 | { |
| 6247 | char *newstart; |
| 6248 | |
| 6249 | case SUSPEND: /* (?>A) */ |
| 6250 | ST.wanted = 1; |
| 6251 | newstart = locinput; |
| 6252 | goto do_ifmatch; |
| 6253 | |
| 6254 | case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */ |
| 6255 | ST.wanted = 0; |
| 6256 | goto ifmatch_trivial_fail_test; |
| 6257 | |
| 6258 | case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */ |
| 6259 | ST.wanted = 1; |
| 6260 | ifmatch_trivial_fail_test: |
| 6261 | if (scan->flags) { |
| 6262 | char * const s = HOPBACKc(locinput, scan->flags); |
| 6263 | if (!s) { |
| 6264 | /* trivial fail */ |
| 6265 | if (logical) { |
| 6266 | logical = 0; |
| 6267 | sw = 1 - cBOOL(ST.wanted); |
| 6268 | } |
| 6269 | else if (ST.wanted) |
| 6270 | sayNO; |
| 6271 | next = scan + ARG(scan); |
| 6272 | if (next == scan) |
| 6273 | next = NULL; |
| 6274 | break; |
| 6275 | } |
| 6276 | newstart = s; |
| 6277 | } |
| 6278 | else |
| 6279 | newstart = locinput; |
| 6280 | |
| 6281 | do_ifmatch: |
| 6282 | ST.me = scan; |
| 6283 | ST.logical = logical; |
| 6284 | logical = 0; /* XXX: reset state of logical once it has been saved into ST */ |
| 6285 | |
| 6286 | /* execute body of (?...A) */ |
| 6287 | PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart); |
| 6288 | assert(0); /* NOTREACHED */ |
| 6289 | } |
| 6290 | |
| 6291 | case IFMATCH_A_fail: /* body of (?...A) failed */ |
| 6292 | ST.wanted = !ST.wanted; |
| 6293 | /* FALL THROUGH */ |
| 6294 | |
| 6295 | case IFMATCH_A: /* body of (?...A) succeeded */ |
| 6296 | if (ST.logical) { |
| 6297 | sw = cBOOL(ST.wanted); |
| 6298 | } |
| 6299 | else if (!ST.wanted) |
| 6300 | sayNO; |
| 6301 | |
| 6302 | if (OP(ST.me) != SUSPEND) { |
| 6303 | /* restore old position except for (?>...) */ |
| 6304 | locinput = st->locinput; |
| 6305 | } |
| 6306 | scan = ST.me + ARG(ST.me); |
| 6307 | if (scan == ST.me) |
| 6308 | scan = NULL; |
| 6309 | continue; /* execute B */ |
| 6310 | |
| 6311 | #undef ST |
| 6312 | |
| 6313 | case LONGJMP: /* alternative with many branches compiles to |
| 6314 | * (BRANCHJ; EXACT ...; LONGJMP ) x N */ |
| 6315 | next = scan + ARG(scan); |
| 6316 | if (next == scan) |
| 6317 | next = NULL; |
| 6318 | break; |
| 6319 | |
| 6320 | case COMMIT: /* (*COMMIT) */ |
| 6321 | reginfo->cutpoint = PL_regeol; |
| 6322 | /* FALLTHROUGH */ |
| 6323 | |
| 6324 | case PRUNE: /* (*PRUNE) */ |
| 6325 | if (!scan->flags) |
| 6326 | sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 6327 | PUSH_STATE_GOTO(COMMIT_next, next, locinput); |
| 6328 | assert(0); /* NOTREACHED */ |
| 6329 | |
| 6330 | case COMMIT_next_fail: |
| 6331 | no_final = 1; |
| 6332 | /* FALLTHROUGH */ |
| 6333 | |
| 6334 | case OPFAIL: /* (*FAIL) */ |
| 6335 | sayNO; |
| 6336 | assert(0); /* NOTREACHED */ |
| 6337 | |
| 6338 | #define ST st->u.mark |
| 6339 | case MARKPOINT: /* (*MARK:foo) */ |
| 6340 | ST.prev_mark = mark_state; |
| 6341 | ST.mark_name = sv_commit = sv_yes_mark |
| 6342 | = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 6343 | mark_state = st; |
| 6344 | ST.mark_loc = locinput; |
| 6345 | PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput); |
| 6346 | assert(0); /* NOTREACHED */ |
| 6347 | |
| 6348 | case MARKPOINT_next: |
| 6349 | mark_state = ST.prev_mark; |
| 6350 | sayYES; |
| 6351 | assert(0); /* NOTREACHED */ |
| 6352 | |
| 6353 | case MARKPOINT_next_fail: |
| 6354 | if (popmark && sv_eq(ST.mark_name,popmark)) |
| 6355 | { |
| 6356 | if (ST.mark_loc > startpoint) |
| 6357 | reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1); |
| 6358 | popmark = NULL; /* we found our mark */ |
| 6359 | sv_commit = ST.mark_name; |
| 6360 | |
| 6361 | DEBUG_EXECUTE_r({ |
| 6362 | PerlIO_printf(Perl_debug_log, |
| 6363 | "%*s %ssetting cutpoint to mark:%"SVf"...%s\n", |
| 6364 | REPORT_CODE_OFF+depth*2, "", |
| 6365 | PL_colors[4], SVfARG(sv_commit), PL_colors[5]); |
| 6366 | }); |
| 6367 | } |
| 6368 | mark_state = ST.prev_mark; |
| 6369 | sv_yes_mark = mark_state ? |
| 6370 | mark_state->u.mark.mark_name : NULL; |
| 6371 | sayNO; |
| 6372 | assert(0); /* NOTREACHED */ |
| 6373 | |
| 6374 | case SKIP: /* (*SKIP) */ |
| 6375 | if (scan->flags) { |
| 6376 | /* (*SKIP) : if we fail we cut here*/ |
| 6377 | ST.mark_name = NULL; |
| 6378 | ST.mark_loc = locinput; |
| 6379 | PUSH_STATE_GOTO(SKIP_next,next, locinput); |
| 6380 | } else { |
| 6381 | /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was, |
| 6382 | otherwise do nothing. Meaning we need to scan |
| 6383 | */ |
| 6384 | regmatch_state *cur = mark_state; |
| 6385 | SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]); |
| 6386 | |
| 6387 | while (cur) { |
| 6388 | if ( sv_eq( cur->u.mark.mark_name, |
| 6389 | find ) ) |
| 6390 | { |
| 6391 | ST.mark_name = find; |
| 6392 | PUSH_STATE_GOTO( SKIP_next, next, locinput); |
| 6393 | } |
| 6394 | cur = cur->u.mark.prev_mark; |
| 6395 | } |
| 6396 | } |
| 6397 | /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */ |
| 6398 | break; |
| 6399 | |
| 6400 | case SKIP_next_fail: |
| 6401 | if (ST.mark_name) { |
| 6402 | /* (*CUT:NAME) - Set up to search for the name as we |
| 6403 | collapse the stack*/ |
| 6404 | popmark = ST.mark_name; |
| 6405 | } else { |
| 6406 | /* (*CUT) - No name, we cut here.*/ |
| 6407 | if (ST.mark_loc > startpoint) |
| 6408 | reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1); |
| 6409 | /* but we set sv_commit to latest mark_name if there |
| 6410 | is one so they can test to see how things lead to this |
| 6411 | cut */ |
| 6412 | if (mark_state) |
| 6413 | sv_commit=mark_state->u.mark.mark_name; |
| 6414 | } |
| 6415 | no_final = 1; |
| 6416 | sayNO; |
| 6417 | assert(0); /* NOTREACHED */ |
| 6418 | #undef ST |
| 6419 | |
| 6420 | case LNBREAK: /* \R */ |
| 6421 | if ((n=is_LNBREAK_safe(locinput, PL_regeol, utf8_target))) { |
| 6422 | locinput += n; |
| 6423 | } else |
| 6424 | sayNO; |
| 6425 | break; |
| 6426 | |
| 6427 | default: |
| 6428 | PerlIO_printf(Perl_error_log, "%"UVxf" %d\n", |
| 6429 | PTR2UV(scan), OP(scan)); |
| 6430 | Perl_croak(aTHX_ "regexp memory corruption"); |
| 6431 | |
| 6432 | /* this is a point to jump to in order to increment |
| 6433 | * locinput by one character */ |
| 6434 | increment_locinput: |
| 6435 | assert(!NEXTCHR_IS_EOS); |
| 6436 | if (utf8_target) { |
| 6437 | locinput += PL_utf8skip[nextchr]; |
| 6438 | /* locinput is allowed to go 1 char off the end, but not 2+ */ |
| 6439 | if (locinput > PL_regeol) |
| 6440 | sayNO; |
| 6441 | } |
| 6442 | else |
| 6443 | locinput++; |
| 6444 | break; |
| 6445 | |
| 6446 | } /* end switch */ |
| 6447 | |
| 6448 | /* switch break jumps here */ |
| 6449 | scan = next; /* prepare to execute the next op and ... */ |
| 6450 | continue; /* ... jump back to the top, reusing st */ |
| 6451 | assert(0); /* NOTREACHED */ |
| 6452 | |
| 6453 | push_yes_state: |
| 6454 | /* push a state that backtracks on success */ |
| 6455 | st->u.yes.prev_yes_state = yes_state; |
| 6456 | yes_state = st; |
| 6457 | /* FALL THROUGH */ |
| 6458 | push_state: |
| 6459 | /* push a new regex state, then continue at scan */ |
| 6460 | { |
| 6461 | regmatch_state *newst; |
| 6462 | |
| 6463 | DEBUG_STACK_r({ |
| 6464 | regmatch_state *cur = st; |
| 6465 | regmatch_state *curyes = yes_state; |
| 6466 | int curd = depth; |
| 6467 | regmatch_slab *slab = PL_regmatch_slab; |
| 6468 | for (;curd > -1;cur--,curd--) { |
| 6469 | if (cur < SLAB_FIRST(slab)) { |
| 6470 | slab = slab->prev; |
| 6471 | cur = SLAB_LAST(slab); |
| 6472 | } |
| 6473 | PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n", |
| 6474 | REPORT_CODE_OFF + 2 + depth * 2,"", |
| 6475 | curd, PL_reg_name[cur->resume_state], |
| 6476 | (curyes == cur) ? "yes" : "" |
| 6477 | ); |
| 6478 | if (curyes == cur) |
| 6479 | curyes = cur->u.yes.prev_yes_state; |
| 6480 | } |
| 6481 | } else |
| 6482 | DEBUG_STATE_pp("push") |
| 6483 | ); |
| 6484 | depth++; |
| 6485 | st->locinput = locinput; |
| 6486 | newst = st+1; |
| 6487 | if (newst > SLAB_LAST(PL_regmatch_slab)) |
| 6488 | newst = S_push_slab(aTHX); |
| 6489 | PL_regmatch_state = newst; |
| 6490 | |
| 6491 | locinput = pushinput; |
| 6492 | st = newst; |
| 6493 | continue; |
| 6494 | assert(0); /* NOTREACHED */ |
| 6495 | } |
| 6496 | } |
| 6497 | |
| 6498 | /* |
| 6499 | * We get here only if there's trouble -- normally "case END" is |
| 6500 | * the terminating point. |
| 6501 | */ |
| 6502 | Perl_croak(aTHX_ "corrupted regexp pointers"); |
| 6503 | /*NOTREACHED*/ |
| 6504 | sayNO; |
| 6505 | |
| 6506 | yes: |
| 6507 | if (yes_state) { |
| 6508 | /* we have successfully completed a subexpression, but we must now |
| 6509 | * pop to the state marked by yes_state and continue from there */ |
| 6510 | assert(st != yes_state); |
| 6511 | #ifdef DEBUGGING |
| 6512 | while (st != yes_state) { |
| 6513 | st--; |
| 6514 | if (st < SLAB_FIRST(PL_regmatch_slab)) { |
| 6515 | PL_regmatch_slab = PL_regmatch_slab->prev; |
| 6516 | st = SLAB_LAST(PL_regmatch_slab); |
| 6517 | } |
| 6518 | DEBUG_STATE_r({ |
| 6519 | if (no_final) { |
| 6520 | DEBUG_STATE_pp("pop (no final)"); |
| 6521 | } else { |
| 6522 | DEBUG_STATE_pp("pop (yes)"); |
| 6523 | } |
| 6524 | }); |
| 6525 | depth--; |
| 6526 | } |
| 6527 | #else |
| 6528 | while (yes_state < SLAB_FIRST(PL_regmatch_slab) |
| 6529 | || yes_state > SLAB_LAST(PL_regmatch_slab)) |
| 6530 | { |
| 6531 | /* not in this slab, pop slab */ |
| 6532 | depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1); |
| 6533 | PL_regmatch_slab = PL_regmatch_slab->prev; |
| 6534 | st = SLAB_LAST(PL_regmatch_slab); |
| 6535 | } |
| 6536 | depth -= (st - yes_state); |
| 6537 | #endif |
| 6538 | st = yes_state; |
| 6539 | yes_state = st->u.yes.prev_yes_state; |
| 6540 | PL_regmatch_state = st; |
| 6541 | |
| 6542 | if (no_final) |
| 6543 | locinput= st->locinput; |
| 6544 | state_num = st->resume_state + no_final; |
| 6545 | goto reenter_switch; |
| 6546 | } |
| 6547 | |
| 6548 | DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n", |
| 6549 | PL_colors[4], PL_colors[5])); |
| 6550 | |
| 6551 | if (PL_reg_state.re_state_eval_setup_done) { |
| 6552 | /* each successfully executed (?{...}) block does the equivalent of |
| 6553 | * local $^R = do {...} |
| 6554 | * When popping the save stack, all these locals would be undone; |
| 6555 | * bypass this by setting the outermost saved $^R to the latest |
| 6556 | * value */ |
| 6557 | if (oreplsv != GvSV(PL_replgv)) |
| 6558 | sv_setsv(oreplsv, GvSV(PL_replgv)); |
| 6559 | } |
| 6560 | result = 1; |
| 6561 | goto final_exit; |
| 6562 | |
| 6563 | no: |
| 6564 | DEBUG_EXECUTE_r( |
| 6565 | PerlIO_printf(Perl_debug_log, |
| 6566 | "%*s %sfailed...%s\n", |
| 6567 | REPORT_CODE_OFF+depth*2, "", |
| 6568 | PL_colors[4], PL_colors[5]) |
| 6569 | ); |
| 6570 | |
| 6571 | no_silent: |
| 6572 | if (no_final) { |
| 6573 | if (yes_state) { |
| 6574 | goto yes; |
| 6575 | } else { |
| 6576 | goto final_exit; |
| 6577 | } |
| 6578 | } |
| 6579 | if (depth) { |
| 6580 | /* there's a previous state to backtrack to */ |
| 6581 | st--; |
| 6582 | if (st < SLAB_FIRST(PL_regmatch_slab)) { |
| 6583 | PL_regmatch_slab = PL_regmatch_slab->prev; |
| 6584 | st = SLAB_LAST(PL_regmatch_slab); |
| 6585 | } |
| 6586 | PL_regmatch_state = st; |
| 6587 | locinput= st->locinput; |
| 6588 | |
| 6589 | DEBUG_STATE_pp("pop"); |
| 6590 | depth--; |
| 6591 | if (yes_state == st) |
| 6592 | yes_state = st->u.yes.prev_yes_state; |
| 6593 | |
| 6594 | state_num = st->resume_state + 1; /* failure = success + 1 */ |
| 6595 | goto reenter_switch; |
| 6596 | } |
| 6597 | result = 0; |
| 6598 | |
| 6599 | final_exit: |
| 6600 | if (rex->intflags & PREGf_VERBARG_SEEN) { |
| 6601 | SV *sv_err = get_sv("REGERROR", 1); |
| 6602 | SV *sv_mrk = get_sv("REGMARK", 1); |
| 6603 | if (result) { |
| 6604 | sv_commit = &PL_sv_no; |
| 6605 | if (!sv_yes_mark) |
| 6606 | sv_yes_mark = &PL_sv_yes; |
| 6607 | } else { |
| 6608 | if (!sv_commit) |
| 6609 | sv_commit = &PL_sv_yes; |
| 6610 | sv_yes_mark = &PL_sv_no; |
| 6611 | } |
| 6612 | sv_setsv(sv_err, sv_commit); |
| 6613 | sv_setsv(sv_mrk, sv_yes_mark); |
| 6614 | } |
| 6615 | |
| 6616 | |
| 6617 | if (last_pushed_cv) { |
| 6618 | dSP; |
| 6619 | POP_MULTICALL; |
| 6620 | PERL_UNUSED_VAR(SP); |
| 6621 | } |
| 6622 | |
| 6623 | /* clean up; in particular, free all slabs above current one */ |
| 6624 | LEAVE_SCOPE(oldsave); |
| 6625 | |
| 6626 | assert(!result || locinput - PL_bostr >= 0); |
| 6627 | return result ? locinput - PL_bostr : -1; |
| 6628 | } |
| 6629 | |
| 6630 | /* |
| 6631 | - regrepeat - repeatedly match something simple, report how many |
| 6632 | * |
| 6633 | * What 'simple' means is a node which can be the operand of a quantifier like |
| 6634 | * '+', or {1,3} |
| 6635 | * |
| 6636 | * startposp - pointer a pointer to the start position. This is updated |
| 6637 | * to point to the byte following the highest successful |
| 6638 | * match. |
| 6639 | * p - the regnode to be repeatedly matched against. |
| 6640 | * max - maximum number of things to match. |
| 6641 | * depth - (for debugging) backtracking depth. |
| 6642 | */ |
| 6643 | STATIC I32 |
| 6644 | S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p, |
| 6645 | I32 max, int depth, bool is_utf8_pat) |
| 6646 | { |
| 6647 | dVAR; |
| 6648 | char *scan; /* Pointer to current position in target string */ |
| 6649 | I32 c; |
| 6650 | char *loceol = PL_regeol; /* local version */ |
| 6651 | I32 hardcount = 0; /* How many matches so far */ |
| 6652 | bool utf8_target = PL_reg_match_utf8; |
| 6653 | int to_complement = 0; /* Invert the result? */ |
| 6654 | UV utf8_flags; |
| 6655 | _char_class_number classnum; |
| 6656 | #ifndef DEBUGGING |
| 6657 | PERL_UNUSED_ARG(depth); |
| 6658 | #endif |
| 6659 | |
| 6660 | PERL_ARGS_ASSERT_REGREPEAT; |
| 6661 | |
| 6662 | scan = *startposp; |
| 6663 | if (max == REG_INFTY) |
| 6664 | max = I32_MAX; |
| 6665 | else if (! utf8_target && scan + max < loceol) |
| 6666 | loceol = scan + max; |
| 6667 | |
| 6668 | /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down |
| 6669 | * to the maximum of how far we should go in it (leaving it set to the real |
| 6670 | * end, if the maximum permissible would take us beyond that). This allows |
| 6671 | * us to make the loop exit condition that we haven't gone past <loceol> to |
| 6672 | * also mean that we haven't exceeded the max permissible count, saving a |
| 6673 | * test each time through the loop. But it assumes that the OP matches a |
| 6674 | * single byte, which is true for most of the OPs below when applied to a |
| 6675 | * non-UTF-8 target. Those relatively few OPs that don't have this |
| 6676 | * characteristic will have to compensate. |
| 6677 | * |
| 6678 | * There is no adjustment for UTF-8 targets, as the number of bytes per |
| 6679 | * character varies. OPs will have to test both that the count is less |
| 6680 | * than the max permissible (using <hardcount> to keep track), and that we |
| 6681 | * are still within the bounds of the string (using <loceol>. A few OPs |
| 6682 | * match a single byte no matter what the encoding. They can omit the max |
| 6683 | * test if, for the UTF-8 case, they do the adjustment that was skipped |
| 6684 | * above. |
| 6685 | * |
| 6686 | * Thus, the code above sets things up for the common case; and exceptional |
| 6687 | * cases need extra work; the common case is to make sure <scan> doesn't |
| 6688 | * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the |
| 6689 | * count doesn't exceed the maximum permissible */ |
| 6690 | |
| 6691 | switch (OP(p)) { |
| 6692 | case REG_ANY: |
| 6693 | if (utf8_target) { |
| 6694 | while (scan < loceol && hardcount < max && *scan != '\n') { |
| 6695 | scan += UTF8SKIP(scan); |
| 6696 | hardcount++; |
| 6697 | } |
| 6698 | } else { |
| 6699 | while (scan < loceol && *scan != '\n') |
| 6700 | scan++; |
| 6701 | } |
| 6702 | break; |
| 6703 | case SANY: |
| 6704 | if (utf8_target) { |
| 6705 | while (scan < loceol && hardcount < max) { |
| 6706 | scan += UTF8SKIP(scan); |
| 6707 | hardcount++; |
| 6708 | } |
| 6709 | } |
| 6710 | else |
| 6711 | scan = loceol; |
| 6712 | break; |
| 6713 | case CANY: /* Move <scan> forward <max> bytes, unless goes off end */ |
| 6714 | if (utf8_target && scan + max < loceol) { |
| 6715 | |
| 6716 | /* <loceol> hadn't been adjusted in the UTF-8 case */ |
| 6717 | scan += max; |
| 6718 | } |
| 6719 | else { |
| 6720 | scan = loceol; |
| 6721 | } |
| 6722 | break; |
| 6723 | case EXACT: |
| 6724 | assert(STR_LEN(p) == is_utf8_pat ? UTF8SKIP(STRING(p)) : 1); |
| 6725 | |
| 6726 | c = (U8)*STRING(p); |
| 6727 | |
| 6728 | /* Can use a simple loop if the pattern char to match on is invariant |
| 6729 | * under UTF-8, or both target and pattern aren't UTF-8. Note that we |
| 6730 | * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's |
| 6731 | * true iff it doesn't matter if the argument is in UTF-8 or not */ |
| 6732 | if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! is_utf8_pat)) { |
| 6733 | if (utf8_target && scan + max < loceol) { |
| 6734 | /* We didn't adjust <loceol> because is UTF-8, but ok to do so, |
| 6735 | * since here, to match at all, 1 char == 1 byte */ |
| 6736 | loceol = scan + max; |
| 6737 | } |
| 6738 | while (scan < loceol && UCHARAT(scan) == c) { |
| 6739 | scan++; |
| 6740 | } |
| 6741 | } |
| 6742 | else if (is_utf8_pat) { |
| 6743 | if (utf8_target) { |
| 6744 | STRLEN scan_char_len; |
| 6745 | |
| 6746 | /* When both target and pattern are UTF-8, we have to do |
| 6747 | * string EQ */ |
| 6748 | while (hardcount < max |
| 6749 | && scan < loceol |
| 6750 | && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p) |
| 6751 | && memEQ(scan, STRING(p), scan_char_len)) |
| 6752 | { |
| 6753 | scan += scan_char_len; |
| 6754 | hardcount++; |
| 6755 | } |
| 6756 | } |
| 6757 | else if (! UTF8_IS_ABOVE_LATIN1(c)) { |
| 6758 | |
| 6759 | /* Target isn't utf8; convert the character in the UTF-8 |
| 6760 | * pattern to non-UTF8, and do a simple loop */ |
| 6761 | c = TWO_BYTE_UTF8_TO_UNI(c, *(STRING(p) + 1)); |
| 6762 | while (scan < loceol && UCHARAT(scan) == c) { |
| 6763 | scan++; |
| 6764 | } |
| 6765 | } /* else pattern char is above Latin1, can't possibly match the |
| 6766 | non-UTF-8 target */ |
| 6767 | } |
| 6768 | else { |
| 6769 | |
| 6770 | /* Here, the string must be utf8; pattern isn't, and <c> is |
| 6771 | * different in utf8 than not, so can't compare them directly. |
| 6772 | * Outside the loop, find the two utf8 bytes that represent c, and |
| 6773 | * then look for those in sequence in the utf8 string */ |
| 6774 | U8 high = UTF8_TWO_BYTE_HI(c); |
| 6775 | U8 low = UTF8_TWO_BYTE_LO(c); |
| 6776 | |
| 6777 | while (hardcount < max |
| 6778 | && scan + 1 < loceol |
| 6779 | && UCHARAT(scan) == high |
| 6780 | && UCHARAT(scan + 1) == low) |
| 6781 | { |
| 6782 | scan += 2; |
| 6783 | hardcount++; |
| 6784 | } |
| 6785 | } |
| 6786 | break; |
| 6787 | |
| 6788 | case EXACTFA: |
| 6789 | utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII; |
| 6790 | goto do_exactf; |
| 6791 | |
| 6792 | case EXACTFL: |
| 6793 | RXp_MATCH_TAINTED_on(prog); |
| 6794 | utf8_flags = FOLDEQ_UTF8_LOCALE; |
| 6795 | goto do_exactf; |
| 6796 | |
| 6797 | case EXACTF: |
| 6798 | utf8_flags = 0; |
| 6799 | goto do_exactf; |
| 6800 | |
| 6801 | case EXACTFU_SS: |
| 6802 | case EXACTFU_TRICKYFOLD: |
| 6803 | case EXACTFU: |
| 6804 | utf8_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0; |
| 6805 | |
| 6806 | do_exactf: { |
| 6807 | int c1, c2; |
| 6808 | U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1]; |
| 6809 | |
| 6810 | assert(STR_LEN(p) == is_utf8_pat ? UTF8SKIP(STRING(p)) : 1); |
| 6811 | |
| 6812 | if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8, |
| 6813 | is_utf8_pat)) |
| 6814 | { |
| 6815 | if (c1 == CHRTEST_VOID) { |
| 6816 | /* Use full Unicode fold matching */ |
| 6817 | char *tmpeol = PL_regeol; |
| 6818 | STRLEN pat_len = is_utf8_pat ? UTF8SKIP(STRING(p)) : 1; |
| 6819 | while (hardcount < max |
| 6820 | && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target, |
| 6821 | STRING(p), NULL, pat_len, |
| 6822 | is_utf8_pat, utf8_flags)) |
| 6823 | { |
| 6824 | scan = tmpeol; |
| 6825 | tmpeol = PL_regeol; |
| 6826 | hardcount++; |
| 6827 | } |
| 6828 | } |
| 6829 | else if (utf8_target) { |
| 6830 | if (c1 == c2) { |
| 6831 | while (scan < loceol |
| 6832 | && hardcount < max |
| 6833 | && memEQ(scan, c1_utf8, UTF8SKIP(scan))) |
| 6834 | { |
| 6835 | scan += UTF8SKIP(scan); |
| 6836 | hardcount++; |
| 6837 | } |
| 6838 | } |
| 6839 | else { |
| 6840 | while (scan < loceol |
| 6841 | && hardcount < max |
| 6842 | && (memEQ(scan, c1_utf8, UTF8SKIP(scan)) |
| 6843 | || memEQ(scan, c2_utf8, UTF8SKIP(scan)))) |
| 6844 | { |
| 6845 | scan += UTF8SKIP(scan); |
| 6846 | hardcount++; |
| 6847 | } |
| 6848 | } |
| 6849 | } |
| 6850 | else if (c1 == c2) { |
| 6851 | while (scan < loceol && UCHARAT(scan) == c1) { |
| 6852 | scan++; |
| 6853 | } |
| 6854 | } |
| 6855 | else { |
| 6856 | while (scan < loceol && |
| 6857 | (UCHARAT(scan) == c1 || UCHARAT(scan) == c2)) |
| 6858 | { |
| 6859 | scan++; |
| 6860 | } |
| 6861 | } |
| 6862 | } |
| 6863 | break; |
| 6864 | } |
| 6865 | case ANYOF: |
| 6866 | case ANYOF_WARN_SUPER: |
| 6867 | if (utf8_target) { |
| 6868 | while (hardcount < max |
| 6869 | && scan < loceol |
| 6870 | && reginclass(prog, p, (U8*)scan, utf8_target)) |
| 6871 | { |
| 6872 | scan += UTF8SKIP(scan); |
| 6873 | hardcount++; |
| 6874 | } |
| 6875 | } else { |
| 6876 | while (scan < loceol && REGINCLASS(prog, p, (U8*)scan)) |
| 6877 | scan++; |
| 6878 | } |
| 6879 | break; |
| 6880 | |
| 6881 | /* The argument (FLAGS) to all the POSIX node types is the class number */ |
| 6882 | |
| 6883 | case NPOSIXL: |
| 6884 | to_complement = 1; |
| 6885 | /* FALLTHROUGH */ |
| 6886 | |
| 6887 | case POSIXL: |
| 6888 | RXp_MATCH_TAINTED_on(prog); |
| 6889 | if (! utf8_target) { |
| 6890 | while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p), |
| 6891 | *scan))) |
| 6892 | { |
| 6893 | scan++; |
| 6894 | } |
| 6895 | } else { |
| 6896 | while (hardcount < max && scan < loceol |
| 6897 | && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p), |
| 6898 | (U8 *) scan))) |
| 6899 | { |
| 6900 | scan += UTF8SKIP(scan); |
| 6901 | hardcount++; |
| 6902 | } |
| 6903 | } |
| 6904 | break; |
| 6905 | |
| 6906 | case POSIXD: |
| 6907 | if (utf8_target) { |
| 6908 | goto utf8_posix; |
| 6909 | } |
| 6910 | /* FALLTHROUGH */ |
| 6911 | |
| 6912 | case POSIXA: |
| 6913 | if (utf8_target && scan + max < loceol) { |
| 6914 | |
| 6915 | /* We didn't adjust <loceol> at the beginning of this routine |
| 6916 | * because is UTF-8, but it is actually ok to do so, since here, to |
| 6917 | * match, 1 char == 1 byte. */ |
| 6918 | loceol = scan + max; |
| 6919 | } |
| 6920 | while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) { |
| 6921 | scan++; |
| 6922 | } |
| 6923 | break; |
| 6924 | |
| 6925 | case NPOSIXD: |
| 6926 | if (utf8_target) { |
| 6927 | to_complement = 1; |
| 6928 | goto utf8_posix; |
| 6929 | } |
| 6930 | /* FALL THROUGH */ |
| 6931 | |
| 6932 | case NPOSIXA: |
| 6933 | if (! utf8_target) { |
| 6934 | while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) { |
| 6935 | scan++; |
| 6936 | } |
| 6937 | } |
| 6938 | else { |
| 6939 | |
| 6940 | /* The complement of something that matches only ASCII matches all |
| 6941 | * UTF-8 variant code points, plus everything in ASCII that isn't |
| 6942 | * in the class. */ |
| 6943 | while (hardcount < max && scan < loceol |
| 6944 | && (! UTF8_IS_INVARIANT(*scan) |
| 6945 | || ! _generic_isCC_A((U8) *scan, FLAGS(p)))) |
| 6946 | { |
| 6947 | scan += UTF8SKIP(scan); |
| 6948 | hardcount++; |
| 6949 | } |
| 6950 | } |
| 6951 | break; |
| 6952 | |
| 6953 | case NPOSIXU: |
| 6954 | to_complement = 1; |
| 6955 | /* FALLTHROUGH */ |
| 6956 | |
| 6957 | case POSIXU: |
| 6958 | if (! utf8_target) { |
| 6959 | while (scan < loceol && to_complement |
| 6960 | ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p)))) |
| 6961 | { |
| 6962 | scan++; |
| 6963 | } |
| 6964 | } |
| 6965 | else { |
| 6966 | utf8_posix: |
| 6967 | classnum = (_char_class_number) FLAGS(p); |
| 6968 | if (classnum < _FIRST_NON_SWASH_CC) { |
| 6969 | |
| 6970 | /* Here, a swash is needed for above-Latin1 code points. |
| 6971 | * Process as many Latin1 code points using the built-in rules. |
| 6972 | * Go to another loop to finish processing upon encountering |
| 6973 | * the first Latin1 code point. We could do that in this loop |
| 6974 | * as well, but the other way saves having to test if the swash |
| 6975 | * has been loaded every time through the loop: extra space to |
| 6976 | * save a test. */ |
| 6977 | while (hardcount < max && scan < loceol) { |
| 6978 | if (UTF8_IS_INVARIANT(*scan)) { |
| 6979 | if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan, |
| 6980 | classnum)))) |
| 6981 | { |
| 6982 | break; |
| 6983 | } |
| 6984 | scan++; |
| 6985 | } |
| 6986 | else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) { |
| 6987 | if (! (to_complement |
| 6988 | ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_UNI(*scan, |
| 6989 | *(scan + 1)), |
| 6990 | classnum)))) |
| 6991 | { |
| 6992 | break; |
| 6993 | } |
| 6994 | scan += 2; |
| 6995 | } |
| 6996 | else { |
| 6997 | goto found_above_latin1; |
| 6998 | } |
| 6999 | |
| 7000 | hardcount++; |
| 7001 | } |
| 7002 | } |
| 7003 | else { |
| 7004 | /* For these character classes, the knowledge of how to handle |
| 7005 | * every code point is compiled in to Perl via a macro. This |
| 7006 | * code is written for making the loops as tight as possible. |
| 7007 | * It could be refactored to save space instead */ |
| 7008 | switch (classnum) { |
| 7009 | case _CC_ENUM_SPACE: /* XXX would require separate code |
| 7010 | if we revert the change of \v |
| 7011 | matching this */ |
| 7012 | /* FALL THROUGH */ |
| 7013 | case _CC_ENUM_PSXSPC: |
| 7014 | while (hardcount < max |
| 7015 | && scan < loceol |
| 7016 | && (to_complement ^ cBOOL(isSPACE_utf8(scan)))) |
| 7017 | { |
| 7018 | scan += UTF8SKIP(scan); |
| 7019 | hardcount++; |
| 7020 | } |
| 7021 | break; |
| 7022 | case _CC_ENUM_BLANK: |
| 7023 | while (hardcount < max |
| 7024 | && scan < loceol |
| 7025 | && (to_complement ^ cBOOL(isBLANK_utf8(scan)))) |
| 7026 | { |
| 7027 | scan += UTF8SKIP(scan); |
| 7028 | hardcount++; |
| 7029 | } |
| 7030 | break; |
| 7031 | case _CC_ENUM_XDIGIT: |
| 7032 | while (hardcount < max |
| 7033 | && scan < loceol |
| 7034 | && (to_complement ^ cBOOL(isXDIGIT_utf8(scan)))) |
| 7035 | { |
| 7036 | scan += UTF8SKIP(scan); |
| 7037 | hardcount++; |
| 7038 | } |
| 7039 | break; |
| 7040 | case _CC_ENUM_VERTSPACE: |
| 7041 | while (hardcount < max |
| 7042 | && scan < loceol |
| 7043 | && (to_complement ^ cBOOL(isVERTWS_utf8(scan)))) |
| 7044 | { |
| 7045 | scan += UTF8SKIP(scan); |
| 7046 | hardcount++; |
| 7047 | } |
| 7048 | break; |
| 7049 | case _CC_ENUM_CNTRL: |
| 7050 | while (hardcount < max |
| 7051 | && scan < loceol |
| 7052 | && (to_complement ^ cBOOL(isCNTRL_utf8(scan)))) |
| 7053 | { |
| 7054 | scan += UTF8SKIP(scan); |
| 7055 | hardcount++; |
| 7056 | } |
| 7057 | break; |
| 7058 | default: |
| 7059 | Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum); |
| 7060 | } |
| 7061 | } |
| 7062 | } |
| 7063 | break; |
| 7064 | |
| 7065 | found_above_latin1: /* Continuation of POSIXU and NPOSIXU */ |
| 7066 | |
| 7067 | /* Load the swash if not already present */ |
| 7068 | if (! PL_utf8_swash_ptrs[classnum]) { |
| 7069 | U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; |
| 7070 | PL_utf8_swash_ptrs[classnum] = _core_swash_init( |
| 7071 | "utf8", swash_property_names[classnum], |
| 7072 | &PL_sv_undef, 1, 0, NULL, &flags); |
| 7073 | } |
| 7074 | |
| 7075 | while (hardcount < max && scan < loceol |
| 7076 | && to_complement ^ cBOOL(_generic_utf8( |
| 7077 | classnum, |
| 7078 | scan, |
| 7079 | swash_fetch(PL_utf8_swash_ptrs[classnum], |
| 7080 | (U8 *) scan, |
| 7081 | TRUE)))) |
| 7082 | { |
| 7083 | scan += UTF8SKIP(scan); |
| 7084 | hardcount++; |
| 7085 | } |
| 7086 | break; |
| 7087 | |
| 7088 | case LNBREAK: |
| 7089 | if (utf8_target) { |
| 7090 | while (hardcount < max && scan < loceol && |
| 7091 | (c=is_LNBREAK_utf8_safe(scan, loceol))) { |
| 7092 | scan += c; |
| 7093 | hardcount++; |
| 7094 | } |
| 7095 | } else { |
| 7096 | /* LNBREAK can match one or two latin chars, which is ok, but we |
| 7097 | * have to use hardcount in this situation, and throw away the |
| 7098 | * adjustment to <loceol> done before the switch statement */ |
| 7099 | loceol = PL_regeol; |
| 7100 | while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) { |
| 7101 | scan+=c; |
| 7102 | hardcount++; |
| 7103 | } |
| 7104 | } |
| 7105 | break; |
| 7106 | |
| 7107 | case BOUND: |
| 7108 | case BOUNDA: |
| 7109 | case BOUNDL: |
| 7110 | case BOUNDU: |
| 7111 | case EOS: |
| 7112 | case GPOS: |
| 7113 | case KEEPS: |
| 7114 | case NBOUND: |
| 7115 | case NBOUNDA: |
| 7116 | case NBOUNDL: |
| 7117 | case NBOUNDU: |
| 7118 | case OPFAIL: |
| 7119 | case SBOL: |
| 7120 | case SEOL: |
| 7121 | /* These are all 0 width, so match right here or not at all. */ |
| 7122 | break; |
| 7123 | |
| 7124 | default: |
| 7125 | Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]); |
| 7126 | assert(0); /* NOTREACHED */ |
| 7127 | |
| 7128 | } |
| 7129 | |
| 7130 | if (hardcount) |
| 7131 | c = hardcount; |
| 7132 | else |
| 7133 | c = scan - *startposp; |
| 7134 | *startposp = scan; |
| 7135 | |
| 7136 | DEBUG_r({ |
| 7137 | GET_RE_DEBUG_FLAGS_DECL; |
| 7138 | DEBUG_EXECUTE_r({ |
| 7139 | SV * const prop = sv_newmortal(); |
| 7140 | regprop(prog, prop, p); |
| 7141 | PerlIO_printf(Perl_debug_log, |
| 7142 | "%*s %s can match %"IVdf" times out of %"IVdf"...\n", |
| 7143 | REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max); |
| 7144 | }); |
| 7145 | }); |
| 7146 | |
| 7147 | return(c); |
| 7148 | } |
| 7149 | |
| 7150 | |
| 7151 | #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) |
| 7152 | /* |
| 7153 | - regclass_swash - prepare the utf8 swash. Wraps the shared core version to |
| 7154 | create a copy so that changes the caller makes won't change the shared one. |
| 7155 | If <altsvp> is non-null, will return NULL in it, for back-compat. |
| 7156 | */ |
| 7157 | SV * |
| 7158 | Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp) |
| 7159 | { |
| 7160 | PERL_ARGS_ASSERT_REGCLASS_SWASH; |
| 7161 | |
| 7162 | if (altsvp) { |
| 7163 | *altsvp = NULL; |
| 7164 | } |
| 7165 | |
| 7166 | return newSVsv(core_regclass_swash(prog, node, doinit, listsvp)); |
| 7167 | } |
| 7168 | #endif |
| 7169 | |
| 7170 | STATIC SV * |
| 7171 | S_core_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp) |
| 7172 | { |
| 7173 | /* Returns the swash for the input 'node' in the regex 'prog'. |
| 7174 | * If <doinit> is true, will attempt to create the swash if not already |
| 7175 | * done. |
| 7176 | * If <listsvp> is non-null, will return the swash initialization string in |
| 7177 | * it. |
| 7178 | * Tied intimately to how regcomp.c sets up the data structure */ |
| 7179 | |
| 7180 | dVAR; |
| 7181 | SV *sw = NULL; |
| 7182 | SV *si = NULL; |
| 7183 | SV* invlist = NULL; |
| 7184 | |
| 7185 | RXi_GET_DECL(prog,progi); |
| 7186 | const struct reg_data * const data = prog ? progi->data : NULL; |
| 7187 | |
| 7188 | PERL_ARGS_ASSERT_CORE_REGCLASS_SWASH; |
| 7189 | |
| 7190 | assert(ANYOF_NONBITMAP(node)); |
| 7191 | |
| 7192 | if (data && data->count) { |
| 7193 | const U32 n = ARG(node); |
| 7194 | |
| 7195 | if (data->what[n] == 's') { |
| 7196 | SV * const rv = MUTABLE_SV(data->data[n]); |
| 7197 | AV * const av = MUTABLE_AV(SvRV(rv)); |
| 7198 | SV **const ary = AvARRAY(av); |
| 7199 | U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; |
| 7200 | |
| 7201 | si = *ary; /* ary[0] = the string to initialize the swash with */ |
| 7202 | |
| 7203 | /* Elements 2 and 3 are either both present or both absent. [2] is |
| 7204 | * any inversion list generated at compile time; [3] indicates if |
| 7205 | * that inversion list has any user-defined properties in it. */ |
| 7206 | if (av_len(av) >= 2) { |
| 7207 | invlist = ary[2]; |
| 7208 | if (SvUV(ary[3])) { |
| 7209 | swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY; |
| 7210 | } |
| 7211 | } |
| 7212 | else { |
| 7213 | invlist = NULL; |
| 7214 | } |
| 7215 | |
| 7216 | /* Element [1] is reserved for the set-up swash. If already there, |
| 7217 | * return it; if not, create it and store it there */ |
| 7218 | if (SvROK(ary[1])) { |
| 7219 | sw = ary[1]; |
| 7220 | } |
| 7221 | else if (si && doinit) { |
| 7222 | |
| 7223 | sw = _core_swash_init("utf8", /* the utf8 package */ |
| 7224 | "", /* nameless */ |
| 7225 | si, |
| 7226 | 1, /* binary */ |
| 7227 | 0, /* not from tr/// */ |
| 7228 | invlist, |
| 7229 | &swash_init_flags); |
| 7230 | (void)av_store(av, 1, sw); |
| 7231 | } |
| 7232 | } |
| 7233 | } |
| 7234 | |
| 7235 | if (listsvp) { |
| 7236 | SV* matches_string = newSVpvn("", 0); |
| 7237 | |
| 7238 | /* Use the swash, if any, which has to have incorporated into it all |
| 7239 | * possibilities */ |
| 7240 | if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL) |
| 7241 | && (si && si != &PL_sv_undef)) |
| 7242 | { |
| 7243 | |
| 7244 | /* If no swash, use the input initialization string, if available */ |
| 7245 | sv_catsv(matches_string, si); |
| 7246 | } |
| 7247 | |
| 7248 | /* Add the inversion list to whatever we have. This may have come from |
| 7249 | * the swash, or from an input parameter */ |
| 7250 | if (invlist) { |
| 7251 | sv_catsv(matches_string, _invlist_contents(invlist)); |
| 7252 | } |
| 7253 | *listsvp = matches_string; |
| 7254 | } |
| 7255 | |
| 7256 | return sw; |
| 7257 | } |
| 7258 | |
| 7259 | /* |
| 7260 | - reginclass - determine if a character falls into a character class |
| 7261 | |
| 7262 | n is the ANYOF regnode |
| 7263 | p is the target string |
| 7264 | utf8_target tells whether p is in UTF-8. |
| 7265 | |
| 7266 | Returns true if matched; false otherwise. |
| 7267 | |
| 7268 | Note that this can be a synthetic start class, a combination of various |
| 7269 | nodes, so things you think might be mutually exclusive, such as locale, |
| 7270 | aren't. It can match both locale and non-locale |
| 7271 | |
| 7272 | */ |
| 7273 | |
| 7274 | STATIC bool |
| 7275 | S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const bool utf8_target) |
| 7276 | { |
| 7277 | dVAR; |
| 7278 | const char flags = ANYOF_FLAGS(n); |
| 7279 | bool match = FALSE; |
| 7280 | UV c = *p; |
| 7281 | |
| 7282 | PERL_ARGS_ASSERT_REGINCLASS; |
| 7283 | |
| 7284 | /* If c is not already the code point, get it. Note that |
| 7285 | * UTF8_IS_INVARIANT() works even if not in UTF-8 */ |
| 7286 | if (! UTF8_IS_INVARIANT(c) && utf8_target) { |
| 7287 | STRLEN c_len = 0; |
| 7288 | c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &c_len, |
| 7289 | (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) |
| 7290 | | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY); |
| 7291 | /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for |
| 7292 | * UTF8_ALLOW_FFFF */ |
| 7293 | if (c_len == (STRLEN)-1) |
| 7294 | Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)"); |
| 7295 | } |
| 7296 | |
| 7297 | /* If this character is potentially in the bitmap, check it */ |
| 7298 | if (c < 256) { |
| 7299 | if (ANYOF_BITMAP_TEST(n, c)) |
| 7300 | match = TRUE; |
| 7301 | else if (flags & ANYOF_NON_UTF8_LATIN1_ALL |
| 7302 | && ! utf8_target |
| 7303 | && ! isASCII(c)) |
| 7304 | { |
| 7305 | match = TRUE; |
| 7306 | } |
| 7307 | else if (flags & ANYOF_LOCALE) { |
| 7308 | RXp_MATCH_TAINTED_on(prog); |
| 7309 | |
| 7310 | if ((flags & ANYOF_LOC_FOLD) |
| 7311 | && ANYOF_BITMAP_TEST(n, PL_fold_locale[c])) |
| 7312 | { |
| 7313 | match = TRUE; |
| 7314 | } |
| 7315 | else if (ANYOF_CLASS_TEST_ANY_SET(n)) { |
| 7316 | |
| 7317 | /* The data structure is arranged so bits 0, 2, 4, ... are set |
| 7318 | * if the class includes the Posix character class given by |
| 7319 | * bit/2; and 1, 3, 5, ... are set if the class includes the |
| 7320 | * complemented Posix class given by int(bit/2). So we loop |
| 7321 | * through the bits, each time changing whether we complement |
| 7322 | * the result or not. Suppose for the sake of illustration |
| 7323 | * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0 |
| 7324 | * is set, it means there is a match for this ANYOF node if the |
| 7325 | * character is in the class given by the expression (0 / 2 = 0 |
| 7326 | * = \w). If it is in that class, isFOO_lc() will return 1, |
| 7327 | * and since 'to_complement' is 0, the result will stay TRUE, |
| 7328 | * and we exit the loop. Suppose instead that bit 0 is 0, but |
| 7329 | * bit 1 is 1. That means there is a match if the character |
| 7330 | * matches \W. We won't bother to call isFOO_lc() on bit 0, |
| 7331 | * but will on bit 1. On the second iteration 'to_complement' |
| 7332 | * will be 1, so the exclusive or will reverse things, so we |
| 7333 | * are testing for \W. On the third iteration, 'to_complement' |
| 7334 | * will be 0, and we would be testing for \s; the fourth |
| 7335 | * iteration would test for \S, etc. |
| 7336 | * |
| 7337 | * Note that this code assumes that all the classes are closed |
| 7338 | * under folding. For example, if a character matches \w, then |
| 7339 | * its fold does too; and vice versa. This should be true for |
| 7340 | * any well-behaved locale for all the currently defined Posix |
| 7341 | * classes, except for :lower: and :upper:, which are handled |
| 7342 | * by the pseudo-class :cased: which matches if either of the |
| 7343 | * other two does. To get rid of this assumption, an outer |
| 7344 | * loop could be used below to iterate over both the source |
| 7345 | * character, and its fold (if different) */ |
| 7346 | |
| 7347 | int count = 0; |
| 7348 | int to_complement = 0; |
| 7349 | while (count < ANYOF_MAX) { |
| 7350 | if (ANYOF_CLASS_TEST(n, count) |
| 7351 | && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c))) |
| 7352 | { |
| 7353 | match = TRUE; |
| 7354 | break; |
| 7355 | } |
| 7356 | count++; |
| 7357 | to_complement ^= 1; |
| 7358 | } |
| 7359 | } |
| 7360 | } |
| 7361 | } |
| 7362 | |
| 7363 | /* If the bitmap didn't (or couldn't) match, and something outside the |
| 7364 | * bitmap could match, try that. Locale nodes specify completely the |
| 7365 | * behavior of code points in the bit map (otherwise, a utf8 target would |
| 7366 | * cause them to be treated as Unicode and not locale), except in |
| 7367 | * the very unlikely event when this node is a synthetic start class, which |
| 7368 | * could be a combination of locale and non-locale nodes. So allow locale |
| 7369 | * to match for the synthetic start class, which will give a false |
| 7370 | * positive that will be resolved when the match is done again as not part |
| 7371 | * of the synthetic start class */ |
| 7372 | if (!match) { |
| 7373 | if (utf8_target && (flags & ANYOF_UNICODE_ALL) && c >= 256) { |
| 7374 | match = TRUE; /* Everything above 255 matches */ |
| 7375 | } |
| 7376 | else if (ANYOF_NONBITMAP(n) |
| 7377 | && ((flags & ANYOF_NONBITMAP_NON_UTF8) |
| 7378 | || (utf8_target |
| 7379 | && (c >=256 |
| 7380 | || (! (flags & ANYOF_LOCALE)) |
| 7381 | || OP(n) == ANYOF_SYNTHETIC)))) |
| 7382 | { |
| 7383 | SV * const sw = core_regclass_swash(prog, n, TRUE, 0); |
| 7384 | if (sw) { |
| 7385 | U8 * utf8_p; |
| 7386 | if (utf8_target) { |
| 7387 | utf8_p = (U8 *) p; |
| 7388 | } else { /* Convert to utf8 */ |
| 7389 | STRLEN len = 1; |
| 7390 | utf8_p = bytes_to_utf8(p, &len); |
| 7391 | } |
| 7392 | |
| 7393 | if (swash_fetch(sw, utf8_p, TRUE)) { |
| 7394 | match = TRUE; |
| 7395 | } |
| 7396 | |
| 7397 | /* If we allocated a string above, free it */ |
| 7398 | if (! utf8_target) Safefree(utf8_p); |
| 7399 | } |
| 7400 | } |
| 7401 | |
| 7402 | if (UNICODE_IS_SUPER(c) |
| 7403 | && OP(n) == ANYOF_WARN_SUPER |
| 7404 | && ckWARN_d(WARN_NON_UNICODE)) |
| 7405 | { |
| 7406 | Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE), |
| 7407 | "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", c); |
| 7408 | } |
| 7409 | } |
| 7410 | |
| 7411 | /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */ |
| 7412 | return cBOOL(flags & ANYOF_INVERT) ^ match; |
| 7413 | } |
| 7414 | |
| 7415 | STATIC U8 * |
| 7416 | S_reghop3(U8 *s, I32 off, const U8* lim) |
| 7417 | { |
| 7418 | /* return the position 'off' UTF-8 characters away from 's', forward if |
| 7419 | * 'off' >= 0, backwards if negative. But don't go outside of position |
| 7420 | * 'lim', which better be < s if off < 0 */ |
| 7421 | |
| 7422 | dVAR; |
| 7423 | |
| 7424 | PERL_ARGS_ASSERT_REGHOP3; |
| 7425 | |
| 7426 | if (off >= 0) { |
| 7427 | while (off-- && s < lim) { |
| 7428 | /* XXX could check well-formedness here */ |
| 7429 | s += UTF8SKIP(s); |
| 7430 | } |
| 7431 | } |
| 7432 | else { |
| 7433 | while (off++ && s > lim) { |
| 7434 | s--; |
| 7435 | if (UTF8_IS_CONTINUED(*s)) { |
| 7436 | while (s > lim && UTF8_IS_CONTINUATION(*s)) |
| 7437 | s--; |
| 7438 | } |
| 7439 | /* XXX could check well-formedness here */ |
| 7440 | } |
| 7441 | } |
| 7442 | return s; |
| 7443 | } |
| 7444 | |
| 7445 | #ifdef XXX_dmq |
| 7446 | /* there are a bunch of places where we use two reghop3's that should |
| 7447 | be replaced with this routine. but since thats not done yet |
| 7448 | we ifdef it out - dmq |
| 7449 | */ |
| 7450 | STATIC U8 * |
| 7451 | S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim) |
| 7452 | { |
| 7453 | dVAR; |
| 7454 | |
| 7455 | PERL_ARGS_ASSERT_REGHOP4; |
| 7456 | |
| 7457 | if (off >= 0) { |
| 7458 | while (off-- && s < rlim) { |
| 7459 | /* XXX could check well-formedness here */ |
| 7460 | s += UTF8SKIP(s); |
| 7461 | } |
| 7462 | } |
| 7463 | else { |
| 7464 | while (off++ && s > llim) { |
| 7465 | s--; |
| 7466 | if (UTF8_IS_CONTINUED(*s)) { |
| 7467 | while (s > llim && UTF8_IS_CONTINUATION(*s)) |
| 7468 | s--; |
| 7469 | } |
| 7470 | /* XXX could check well-formedness here */ |
| 7471 | } |
| 7472 | } |
| 7473 | return s; |
| 7474 | } |
| 7475 | #endif |
| 7476 | |
| 7477 | STATIC U8 * |
| 7478 | S_reghopmaybe3(U8* s, I32 off, const U8* lim) |
| 7479 | { |
| 7480 | dVAR; |
| 7481 | |
| 7482 | PERL_ARGS_ASSERT_REGHOPMAYBE3; |
| 7483 | |
| 7484 | if (off >= 0) { |
| 7485 | while (off-- && s < lim) { |
| 7486 | /* XXX could check well-formedness here */ |
| 7487 | s += UTF8SKIP(s); |
| 7488 | } |
| 7489 | if (off >= 0) |
| 7490 | return NULL; |
| 7491 | } |
| 7492 | else { |
| 7493 | while (off++ && s > lim) { |
| 7494 | s--; |
| 7495 | if (UTF8_IS_CONTINUED(*s)) { |
| 7496 | while (s > lim && UTF8_IS_CONTINUATION(*s)) |
| 7497 | s--; |
| 7498 | } |
| 7499 | /* XXX could check well-formedness here */ |
| 7500 | } |
| 7501 | if (off <= 0) |
| 7502 | return NULL; |
| 7503 | } |
| 7504 | return s; |
| 7505 | } |
| 7506 | |
| 7507 | static void |
| 7508 | restore_pos(pTHX_ void *arg) |
| 7509 | { |
| 7510 | dVAR; |
| 7511 | regexp * const rex = (regexp *)arg; |
| 7512 | if (PL_reg_state.re_state_eval_setup_done) { |
| 7513 | if (PL_reg_oldsaved) { |
| 7514 | rex->subbeg = PL_reg_oldsaved; |
| 7515 | rex->sublen = PL_reg_oldsavedlen; |
| 7516 | rex->suboffset = PL_reg_oldsavedoffset; |
| 7517 | rex->subcoffset = PL_reg_oldsavedcoffset; |
| 7518 | #ifdef PERL_ANY_COW |
| 7519 | rex->saved_copy = PL_nrs; |
| 7520 | #endif |
| 7521 | RXp_MATCH_COPIED_on(rex); |
| 7522 | } |
| 7523 | PL_reg_magic->mg_len = PL_reg_oldpos; |
| 7524 | PL_reg_state.re_state_eval_setup_done = FALSE; |
| 7525 | PL_curpm = PL_reg_oldcurpm; |
| 7526 | } |
| 7527 | } |
| 7528 | |
| 7529 | STATIC void |
| 7530 | S_to_utf8_substr(pTHX_ regexp *prog) |
| 7531 | { |
| 7532 | /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile |
| 7533 | * on the converted value */ |
| 7534 | |
| 7535 | int i = 1; |
| 7536 | |
| 7537 | PERL_ARGS_ASSERT_TO_UTF8_SUBSTR; |
| 7538 | |
| 7539 | do { |
| 7540 | if (prog->substrs->data[i].substr |
| 7541 | && !prog->substrs->data[i].utf8_substr) { |
| 7542 | SV* const sv = newSVsv(prog->substrs->data[i].substr); |
| 7543 | prog->substrs->data[i].utf8_substr = sv; |
| 7544 | sv_utf8_upgrade(sv); |
| 7545 | if (SvVALID(prog->substrs->data[i].substr)) { |
| 7546 | if (SvTAIL(prog->substrs->data[i].substr)) { |
| 7547 | /* Trim the trailing \n that fbm_compile added last |
| 7548 | time. */ |
| 7549 | SvCUR_set(sv, SvCUR(sv) - 1); |
| 7550 | /* Whilst this makes the SV technically "invalid" (as its |
| 7551 | buffer is no longer followed by "\0") when fbm_compile() |
| 7552 | adds the "\n" back, a "\0" is restored. */ |
| 7553 | fbm_compile(sv, FBMcf_TAIL); |
| 7554 | } else |
| 7555 | fbm_compile(sv, 0); |
| 7556 | } |
| 7557 | if (prog->substrs->data[i].substr == prog->check_substr) |
| 7558 | prog->check_utf8 = sv; |
| 7559 | } |
| 7560 | } while (i--); |
| 7561 | } |
| 7562 | |
| 7563 | STATIC bool |
| 7564 | S_to_byte_substr(pTHX_ regexp *prog) |
| 7565 | { |
| 7566 | /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile |
| 7567 | * on the converted value; returns FALSE if can't be converted. */ |
| 7568 | |
| 7569 | dVAR; |
| 7570 | int i = 1; |
| 7571 | |
| 7572 | PERL_ARGS_ASSERT_TO_BYTE_SUBSTR; |
| 7573 | |
| 7574 | do { |
| 7575 | if (prog->substrs->data[i].utf8_substr |
| 7576 | && !prog->substrs->data[i].substr) { |
| 7577 | SV* sv = newSVsv(prog->substrs->data[i].utf8_substr); |
| 7578 | if (! sv_utf8_downgrade(sv, TRUE)) { |
| 7579 | return FALSE; |
| 7580 | } |
| 7581 | if (SvVALID(prog->substrs->data[i].utf8_substr)) { |
| 7582 | if (SvTAIL(prog->substrs->data[i].utf8_substr)) { |
| 7583 | /* Trim the trailing \n that fbm_compile added last |
| 7584 | time. */ |
| 7585 | SvCUR_set(sv, SvCUR(sv) - 1); |
| 7586 | fbm_compile(sv, FBMcf_TAIL); |
| 7587 | } else |
| 7588 | fbm_compile(sv, 0); |
| 7589 | } |
| 7590 | prog->substrs->data[i].substr = sv; |
| 7591 | if (prog->substrs->data[i].utf8_substr == prog->check_utf8) |
| 7592 | prog->check_substr = sv; |
| 7593 | } |
| 7594 | } while (i--); |
| 7595 | |
| 7596 | return TRUE; |
| 7597 | } |
| 7598 | |
| 7599 | /* |
| 7600 | * Local variables: |
| 7601 | * c-indentation-style: bsd |
| 7602 | * c-basic-offset: 4 |
| 7603 | * indent-tabs-mode: nil |
| 7604 | * End: |
| 7605 | * |
| 7606 | * ex: set ts=8 sts=4 sw=4 et: |
| 7607 | */ |