| 1 | /* regcomp.h |
| 2 | * |
| 3 | * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, |
| 4 | * 2000, 2001, 2002, 2003, 2005 by Larry Wall and others |
| 5 | * |
| 6 | * You may distribute under the terms of either the GNU General Public |
| 7 | * License or the Artistic License, as specified in the README file. |
| 8 | * |
| 9 | */ |
| 10 | |
| 11 | typedef OP OP_4tree; /* Will be redefined later. */ |
| 12 | |
| 13 | |
| 14 | /* Convert branch sequences to more efficient trie ops? */ |
| 15 | #define PERL_ENABLE_TRIE_OPTIMISATION 1 |
| 16 | |
| 17 | /* Be really agressive about optimising patterns with trie sequences? */ |
| 18 | #define PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION 1 |
| 19 | |
| 20 | /* Should the optimiser take positive assertions into account? */ |
| 21 | #define PERL_ENABLE_POSITIVE_ASSERTION_STUDY 1 |
| 22 | |
| 23 | /* Not for production use: */ |
| 24 | #define PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS 0 |
| 25 | |
| 26 | /* Unless the next line is uncommented it is illegal to combine lazy |
| 27 | matching with possessive matching. Frankly it doesn't make much sense |
| 28 | to allow it as X*?+ matches nothing, X+?+ matches a single char only, |
| 29 | and X{min,max}?+ matches min times only. |
| 30 | */ |
| 31 | /* #define REG_ALLOW_MINMOD_SUSPEND */ |
| 32 | |
| 33 | /* |
| 34 | * The "internal use only" fields in regexp.h are present to pass info from |
| 35 | * compile to execute that permits the execute phase to run lots faster on |
| 36 | * simple cases. They are: |
| 37 | * |
| 38 | * regstart sv that must begin a match; NULL if none obvious |
| 39 | * reganch is the match anchored (at beginning-of-line only)? |
| 40 | * regmust string (pointer into program) that match must include, or NULL |
| 41 | * [regmust changed to SV* for bminstr()--law] |
| 42 | * regmlen length of regmust string |
| 43 | * [regmlen not used currently] |
| 44 | * |
| 45 | * Regstart and reganch permit very fast decisions on suitable starting points |
| 46 | * for a match, cutting down the work a lot. Regmust permits fast rejection |
| 47 | * of lines that cannot possibly match. The regmust tests are costly enough |
| 48 | * that pregcomp() supplies a regmust only if the r.e. contains something |
| 49 | * potentially expensive (at present, the only such thing detected is * or + |
| 50 | * at the start of the r.e., which can involve a lot of backup). Regmlen is |
| 51 | * supplied because the test in pregexec() needs it and pregcomp() is computing |
| 52 | * it anyway. |
| 53 | * [regmust is now supplied always. The tests that use regmust have a |
| 54 | * heuristic that disables the test if it usually matches.] |
| 55 | * |
| 56 | * [In fact, we now use regmust in many cases to locate where the search |
| 57 | * starts in the string, so if regback is >= 0, the regmust search is never |
| 58 | * wasted effort. The regback variable says how many characters back from |
| 59 | * where regmust matched is the earliest possible start of the match. |
| 60 | * For instance, /[a-z].foo/ has a regmust of 'foo' and a regback of 2.] |
| 61 | */ |
| 62 | |
| 63 | /* |
| 64 | * Structure for regexp "program". This is essentially a linear encoding |
| 65 | * of a nondeterministic finite-state machine (aka syntax charts or |
| 66 | * "railroad normal form" in parsing technology). Each node is an opcode |
| 67 | * plus a "next" pointer, possibly plus an operand. "Next" pointers of |
| 68 | * all nodes except BRANCH implement concatenation; a "next" pointer with |
| 69 | * a BRANCH on both ends of it is connecting two alternatives. (Here we |
| 70 | * have one of the subtle syntax dependencies: an individual BRANCH (as |
| 71 | * opposed to a collection of them) is never concatenated with anything |
| 72 | * because of operator precedence.) The operand of some types of node is |
| 73 | * a literal string; for others, it is a node leading into a sub-FSM. In |
| 74 | * particular, the operand of a BRANCH node is the first node of the branch. |
| 75 | * (NB this is *not* a tree structure: the tail of the branch connects |
| 76 | * to the thing following the set of BRANCHes.) The opcodes are defined |
| 77 | * in regnodes.h which is generated from regcomp.sym by regcomp.pl. |
| 78 | */ |
| 79 | |
| 80 | /* |
| 81 | * A node is one char of opcode followed by two chars of "next" pointer. |
| 82 | * "Next" pointers are stored as two 8-bit pieces, high order first. The |
| 83 | * value is a positive offset from the opcode of the node containing it. |
| 84 | * An operand, if any, simply follows the node. (Note that much of the |
| 85 | * code generation knows about this implicit relationship.) |
| 86 | * |
| 87 | * Using two bytes for the "next" pointer is vast overkill for most things, |
| 88 | * but allows patterns to get big without disasters. |
| 89 | * |
| 90 | * [The "next" pointer is always aligned on an even |
| 91 | * boundary, and reads the offset directly as a short. Also, there is no |
| 92 | * special test to reverse the sign of BACK pointers since the offset is |
| 93 | * stored negative.] |
| 94 | */ |
| 95 | |
| 96 | struct regnode_string { |
| 97 | U8 str_len; |
| 98 | U8 type; |
| 99 | U16 next_off; |
| 100 | char string[1]; |
| 101 | }; |
| 102 | |
| 103 | /* Argument bearing node - workhorse, |
| 104 | arg1 is often for the data field */ |
| 105 | struct regnode_1 { |
| 106 | U8 flags; |
| 107 | U8 type; |
| 108 | U16 next_off; |
| 109 | U32 arg1; |
| 110 | }; |
| 111 | |
| 112 | /* Similar to a regnode_1 but with an extra signed argument */ |
| 113 | struct regnode_2L { |
| 114 | U8 flags; |
| 115 | U8 type; |
| 116 | U16 next_off; |
| 117 | U32 arg1; |
| 118 | I32 arg2; |
| 119 | }; |
| 120 | |
| 121 | /* 'Two field' -- Two 16 bit unsigned args */ |
| 122 | struct regnode_2 { |
| 123 | U8 flags; |
| 124 | U8 type; |
| 125 | U16 next_off; |
| 126 | U16 arg1; |
| 127 | U16 arg2; |
| 128 | }; |
| 129 | |
| 130 | |
| 131 | #define ANYOF_BITMAP_SIZE 32 /* 256 b/(8 b/B) */ |
| 132 | #define ANYOF_CLASSBITMAP_SIZE 4 /* up to 32 (8*4) named classes */ |
| 133 | |
| 134 | /* also used by trie */ |
| 135 | struct regnode_charclass { |
| 136 | U8 flags; |
| 137 | U8 type; |
| 138 | U16 next_off; |
| 139 | U32 arg1; |
| 140 | char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */ |
| 141 | }; |
| 142 | |
| 143 | struct regnode_charclass_class { /* has [[:blah:]] classes */ |
| 144 | U8 flags; /* should have ANYOF_CLASS here */ |
| 145 | U8 type; |
| 146 | U16 next_off; |
| 147 | U32 arg1; |
| 148 | char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time */ |
| 149 | char classflags[ANYOF_CLASSBITMAP_SIZE]; /* and run-time */ |
| 150 | }; |
| 151 | |
| 152 | /* XXX fix this description. |
| 153 | Impose a limit of REG_INFTY on various pattern matching operations |
| 154 | to limit stack growth and to avoid "infinite" recursions. |
| 155 | */ |
| 156 | /* The default size for REG_INFTY is I16_MAX, which is the same as |
| 157 | SHORT_MAX (see perl.h). Unfortunately I16 isn't necessarily 16 bits |
| 158 | (see handy.h). On the Cray C90, sizeof(short)==4 and hence I16_MAX is |
| 159 | ((1<<31)-1), while on the Cray T90, sizeof(short)==8 and I16_MAX is |
| 160 | ((1<<63)-1). To limit stack growth to reasonable sizes, supply a |
| 161 | smaller default. |
| 162 | --Andy Dougherty 11 June 1998 |
| 163 | */ |
| 164 | #if SHORTSIZE > 2 |
| 165 | # ifndef REG_INFTY |
| 166 | # define REG_INFTY ((1<<15)-1) |
| 167 | # endif |
| 168 | #endif |
| 169 | |
| 170 | #ifndef REG_INFTY |
| 171 | # define REG_INFTY I16_MAX |
| 172 | #endif |
| 173 | |
| 174 | #define ARG_VALUE(arg) (arg) |
| 175 | #define ARG__SET(arg,val) ((arg) = (val)) |
| 176 | |
| 177 | #undef ARG |
| 178 | #undef ARG1 |
| 179 | #undef ARG2 |
| 180 | |
| 181 | #define ARG(p) ARG_VALUE(ARG_LOC(p)) |
| 182 | #define ARG1(p) ARG_VALUE(ARG1_LOC(p)) |
| 183 | #define ARG2(p) ARG_VALUE(ARG2_LOC(p)) |
| 184 | #define ARG2L(p) ARG_VALUE(ARG2L_LOC(p)) |
| 185 | |
| 186 | #define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val)) |
| 187 | #define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val)) |
| 188 | #define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val)) |
| 189 | #define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val)) |
| 190 | |
| 191 | #undef NEXT_OFF |
| 192 | #undef NODE_ALIGN |
| 193 | |
| 194 | #define NEXT_OFF(p) ((p)->next_off) |
| 195 | #define NODE_ALIGN(node) |
| 196 | #define NODE_ALIGN_FILL(node) ((node)->flags = 0xde) /* deadbeef */ |
| 197 | |
| 198 | #define SIZE_ALIGN NODE_ALIGN |
| 199 | |
| 200 | #undef OP |
| 201 | #undef OPERAND |
| 202 | #undef MASK |
| 203 | #undef STRING |
| 204 | |
| 205 | #define OP(p) ((p)->type) |
| 206 | #define OPERAND(p) (((struct regnode_string *)p)->string) |
| 207 | #define MASK(p) ((char*)OPERAND(p)) |
| 208 | #define STR_LEN(p) (((struct regnode_string *)p)->str_len) |
| 209 | #define STRING(p) (((struct regnode_string *)p)->string) |
| 210 | #define STR_SZ(l) ((l + sizeof(regnode) - 1) / sizeof(regnode)) |
| 211 | #define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p))+1) |
| 212 | |
| 213 | #undef NODE_ALIGN |
| 214 | #undef ARG_LOC |
| 215 | #undef NEXTOPER |
| 216 | #undef PREVOPER |
| 217 | |
| 218 | #define NODE_ALIGN(node) |
| 219 | #define ARG_LOC(p) (((struct regnode_1 *)p)->arg1) |
| 220 | #define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1) |
| 221 | #define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2) |
| 222 | #define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2) |
| 223 | |
| 224 | #define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */ |
| 225 | #define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2) |
| 226 | |
| 227 | #define NODE_STEP_B 4 |
| 228 | |
| 229 | #define NEXTOPER(p) ((p) + NODE_STEP_REGNODE) |
| 230 | #define PREVOPER(p) ((p) - NODE_STEP_REGNODE) |
| 231 | |
| 232 | #define FILL_ADVANCE_NODE(ptr, op) STMT_START { \ |
| 233 | (ptr)->type = op; (ptr)->next_off = 0; (ptr)++; } STMT_END |
| 234 | #define FILL_ADVANCE_NODE_ARG(ptr, op, arg) STMT_START { \ |
| 235 | ARG_SET(ptr, arg); FILL_ADVANCE_NODE(ptr, op); (ptr) += 1; } STMT_END |
| 236 | |
| 237 | #define REG_MAGIC 0234 |
| 238 | |
| 239 | #define SIZE_ONLY (RExC_emit == &PL_regdummy) |
| 240 | |
| 241 | /* Flags for node->flags of ANYOF */ |
| 242 | |
| 243 | #define ANYOF_CLASS 0x08 /* has [[:blah:]] classes */ |
| 244 | #define ANYOF_INVERT 0x04 |
| 245 | #define ANYOF_FOLD 0x02 |
| 246 | #define ANYOF_LOCALE 0x01 |
| 247 | |
| 248 | /* Used for regstclass only */ |
| 249 | #define ANYOF_EOS 0x10 /* Can match an empty string too */ |
| 250 | |
| 251 | /* There is a character or a range past 0xff */ |
| 252 | #define ANYOF_UNICODE 0x20 |
| 253 | #define ANYOF_UNICODE_ALL 0x40 /* Can match any char past 0xff */ |
| 254 | |
| 255 | /* size of node is large (includes class pointer) */ |
| 256 | #define ANYOF_LARGE 0x80 |
| 257 | |
| 258 | /* Are there any runtime flags on in this node? */ |
| 259 | #define ANYOF_RUNTIME(s) (ANYOF_FLAGS(s) & 0x0f) |
| 260 | |
| 261 | #define ANYOF_FLAGS_ALL 0xff |
| 262 | |
| 263 | /* Character classes for node->classflags of ANYOF */ |
| 264 | /* Should be synchronized with a table in regprop() */ |
| 265 | /* 2n should pair with 2n+1 */ |
| 266 | |
| 267 | #define ANYOF_ALNUM 0 /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */ |
| 268 | #define ANYOF_NALNUM 1 |
| 269 | #define ANYOF_SPACE 2 /* \s */ |
| 270 | #define ANYOF_NSPACE 3 |
| 271 | #define ANYOF_DIGIT 4 |
| 272 | #define ANYOF_NDIGIT 5 |
| 273 | #define ANYOF_ALNUMC 6 /* isalnum(3), utf8::IsAlnum, ALNUMC */ |
| 274 | #define ANYOF_NALNUMC 7 |
| 275 | #define ANYOF_ALPHA 8 |
| 276 | #define ANYOF_NALPHA 9 |
| 277 | #define ANYOF_ASCII 10 |
| 278 | #define ANYOF_NASCII 11 |
| 279 | #define ANYOF_CNTRL 12 |
| 280 | #define ANYOF_NCNTRL 13 |
| 281 | #define ANYOF_GRAPH 14 |
| 282 | #define ANYOF_NGRAPH 15 |
| 283 | #define ANYOF_LOWER 16 |
| 284 | #define ANYOF_NLOWER 17 |
| 285 | #define ANYOF_PRINT 18 |
| 286 | #define ANYOF_NPRINT 19 |
| 287 | #define ANYOF_PUNCT 20 |
| 288 | #define ANYOF_NPUNCT 21 |
| 289 | #define ANYOF_UPPER 22 |
| 290 | #define ANYOF_NUPPER 23 |
| 291 | #define ANYOF_XDIGIT 24 |
| 292 | #define ANYOF_NXDIGIT 25 |
| 293 | #define ANYOF_PSXSPC 26 /* POSIX space: \s plus the vertical tab */ |
| 294 | #define ANYOF_NPSXSPC 27 |
| 295 | #define ANYOF_BLANK 28 /* GNU extension: space and tab: non-vertical space */ |
| 296 | #define ANYOF_NBLANK 29 |
| 297 | |
| 298 | #define ANYOF_MAX 32 |
| 299 | |
| 300 | /* Backward source code compatibility. */ |
| 301 | |
| 302 | #define ANYOF_ALNUML ANYOF_ALNUM |
| 303 | #define ANYOF_NALNUML ANYOF_NALNUM |
| 304 | #define ANYOF_SPACEL ANYOF_SPACE |
| 305 | #define ANYOF_NSPACEL ANYOF_NSPACE |
| 306 | |
| 307 | /* Utility macros for the bitmap and classes of ANYOF */ |
| 308 | |
| 309 | #define ANYOF_SIZE (sizeof(struct regnode_charclass)) |
| 310 | #define ANYOF_CLASS_SIZE (sizeof(struct regnode_charclass_class)) |
| 311 | |
| 312 | #define ANYOF_FLAGS(p) ((p)->flags) |
| 313 | |
| 314 | #define ANYOF_BIT(c) (1 << ((c) & 7)) |
| 315 | |
| 316 | #define ANYOF_CLASS_BYTE(p, c) (((struct regnode_charclass_class*)(p))->classflags[((c) >> 3) & 3]) |
| 317 | #define ANYOF_CLASS_SET(p, c) (ANYOF_CLASS_BYTE(p, c) |= ANYOF_BIT(c)) |
| 318 | #define ANYOF_CLASS_CLEAR(p, c) (ANYOF_CLASS_BYTE(p, c) &= ~ANYOF_BIT(c)) |
| 319 | #define ANYOF_CLASS_TEST(p, c) (ANYOF_CLASS_BYTE(p, c) & ANYOF_BIT(c)) |
| 320 | |
| 321 | #define ANYOF_CLASS_ZERO(ret) Zero(((struct regnode_charclass_class*)(ret))->classflags, ANYOF_CLASSBITMAP_SIZE, char) |
| 322 | #define ANYOF_BITMAP_ZERO(ret) Zero(((struct regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char) |
| 323 | |
| 324 | #define ANYOF_BITMAP(p) (((struct regnode_charclass*)(p))->bitmap) |
| 325 | #define ANYOF_BITMAP_BYTE(p, c) (ANYOF_BITMAP(p)[(((U8)(c)) >> 3) & 31]) |
| 326 | #define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c)) |
| 327 | #define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c)) |
| 328 | #define ANYOF_BITMAP_TEST(p, c) (ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c)) |
| 329 | |
| 330 | #define ANYOF_BITMAP_SETALL(p) \ |
| 331 | memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE) |
| 332 | #define ANYOF_BITMAP_CLEARALL(p) \ |
| 333 | Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE) |
| 334 | /* Check that all 256 bits are all set. Used in S_cl_is_anything() */ |
| 335 | #define ANYOF_BITMAP_TESTALLSET(p) \ |
| 336 | memEQ (ANYOF_BITMAP(p), "\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377", ANYOF_BITMAP_SIZE) |
| 337 | |
| 338 | #define ANYOF_SKIP ((ANYOF_SIZE - 1)/sizeof(regnode)) |
| 339 | #define ANYOF_CLASS_SKIP ((ANYOF_CLASS_SIZE - 1)/sizeof(regnode)) |
| 340 | #define ANYOF_CLASS_ADD_SKIP (ANYOF_CLASS_SKIP - ANYOF_SKIP) |
| 341 | |
| 342 | |
| 343 | /* |
| 344 | * Utility definitions. |
| 345 | */ |
| 346 | #ifndef CHARMASK |
| 347 | # define UCHARAT(p) ((int)*(const U8*)(p)) |
| 348 | #else |
| 349 | # define UCHARAT(p) ((int)*(p)&CHARMASK) |
| 350 | #endif |
| 351 | |
| 352 | #define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode)) |
| 353 | |
| 354 | #define REG_SEEN_ZERO_LEN 0x00000001 |
| 355 | #define REG_SEEN_LOOKBEHIND 0x00000002 |
| 356 | #define REG_SEEN_GPOS 0x00000004 |
| 357 | #define REG_SEEN_EVAL 0x00000008 |
| 358 | #define REG_SEEN_CANY 0x00000010 |
| 359 | #define REG_SEEN_SANY REG_SEEN_CANY /* src bckwrd cmpt */ |
| 360 | #define REG_SEEN_RECURSE 0x00000020 |
| 361 | #define REG_TOP_LEVEL_BRANCHES 0x00000040 |
| 362 | #define REG_SEEN_VERBARG 0x00000080 |
| 363 | #define REG_SEEN_CUTGROUP 0x00000100 |
| 364 | |
| 365 | START_EXTERN_C |
| 366 | |
| 367 | #ifdef PLUGGABLE_RE_EXTENSION |
| 368 | #include "re_nodes.h" |
| 369 | #else |
| 370 | #include "regnodes.h" |
| 371 | #endif |
| 372 | |
| 373 | /* The following have no fixed length. U8 so we can do strchr() on it. */ |
| 374 | #ifndef DOINIT |
| 375 | EXTCONST U8 PL_varies[]; |
| 376 | #else |
| 377 | EXTCONST U8 PL_varies[] = { |
| 378 | BRANCH, BACK, STAR, PLUS, CURLY, CURLYX, REF, REFF, REFFL, |
| 379 | WHILEM, CURLYM, CURLYN, BRANCHJ, IFTHEN, SUSPEND, CLUMP, 0 |
| 380 | }; |
| 381 | #endif |
| 382 | |
| 383 | /* The following always have a length of 1. U8 we can do strchr() on it. */ |
| 384 | /* (Note that length 1 means "one character" under UTF8, not "one octet".) */ |
| 385 | #ifndef DOINIT |
| 386 | EXTCONST U8 PL_simple[]; |
| 387 | #else |
| 388 | EXTCONST U8 PL_simple[] = { |
| 389 | REG_ANY, SANY, CANY, |
| 390 | ANYOF, |
| 391 | ALNUM, ALNUML, |
| 392 | NALNUM, NALNUML, |
| 393 | SPACE, SPACEL, |
| 394 | NSPACE, NSPACEL, |
| 395 | DIGIT, NDIGIT, |
| 396 | 0 |
| 397 | }; |
| 398 | #endif |
| 399 | |
| 400 | #ifndef PLUGGABLE_RE_EXTENSION |
| 401 | #ifndef DOINIT |
| 402 | EXTCONST regexp_engine PL_core_reg_engine; |
| 403 | #else /* DOINIT */ |
| 404 | EXTCONST regexp_engine PL_core_reg_engine = { |
| 405 | Perl_pregcomp, |
| 406 | Perl_regexec_flags, |
| 407 | Perl_re_intuit_start, |
| 408 | Perl_re_intuit_string, |
| 409 | Perl_pregfree, |
| 410 | Perl_reg_stringify, |
| 411 | #if defined(USE_ITHREADS) |
| 412 | Perl_regdupe |
| 413 | #endif |
| 414 | }; |
| 415 | #endif /* DOINIT */ |
| 416 | #endif /* PLUGGABLE_RE_EXTENSION */ |
| 417 | |
| 418 | |
| 419 | END_EXTERN_C |
| 420 | |
| 421 | |
| 422 | /* .what is a character array with one character for each member of .data |
| 423 | * The character describes the function of the corresponding .data item: |
| 424 | * f - start-class data for regstclass optimization |
| 425 | * n - Root of op tree for (?{EVAL}) item |
| 426 | * o - Start op for (?{EVAL}) item |
| 427 | * p - Pad for (?{EVAL}) item |
| 428 | * s - swash for unicode-style character class, and the multicharacter |
| 429 | * strings resulting from casefolding the single-character entries |
| 430 | * in the character class |
| 431 | * t - trie struct |
| 432 | * u - trie struct's widecharmap (a HV, so can't share, must dup) |
| 433 | * T - aho-trie struct |
| 434 | * S - sv for named capture lookup |
| 435 | * 20010712 mjd@plover.com |
| 436 | * (Remember to update re_dup() and pregfree() if you add any items.) |
| 437 | */ |
| 438 | struct reg_data { |
| 439 | U32 count; |
| 440 | U8 *what; |
| 441 | void* data[1]; |
| 442 | }; |
| 443 | |
| 444 | struct reg_substr_datum { |
| 445 | I32 min_offset; |
| 446 | I32 max_offset; |
| 447 | SV *substr; /* non-utf8 variant */ |
| 448 | SV *utf8_substr; /* utf8 variant */ |
| 449 | I32 end_shift; |
| 450 | }; |
| 451 | |
| 452 | struct reg_substr_data { |
| 453 | struct reg_substr_datum data[3]; /* Actual array */ |
| 454 | }; |
| 455 | |
| 456 | #define anchored_substr substrs->data[0].substr |
| 457 | #define anchored_utf8 substrs->data[0].utf8_substr |
| 458 | #define anchored_offset substrs->data[0].min_offset |
| 459 | #define anchored_end_shift substrs->data[0].end_shift |
| 460 | |
| 461 | #define float_substr substrs->data[1].substr |
| 462 | #define float_utf8 substrs->data[1].utf8_substr |
| 463 | #define float_min_offset substrs->data[1].min_offset |
| 464 | #define float_max_offset substrs->data[1].max_offset |
| 465 | #define float_end_shift substrs->data[1].end_shift |
| 466 | |
| 467 | #define check_substr substrs->data[2].substr |
| 468 | #define check_utf8 substrs->data[2].utf8_substr |
| 469 | #define check_offset_min substrs->data[2].min_offset |
| 470 | #define check_offset_max substrs->data[2].max_offset |
| 471 | #define check_end_shift substrs->data[2].end_shift |
| 472 | |
| 473 | |
| 474 | |
| 475 | /* trie related stuff */ |
| 476 | |
| 477 | /* a transition record for the state machine. the |
| 478 | check field determines which state "owns" the |
| 479 | transition. the char the transition is for is |
| 480 | determined by offset from the owning states base |
| 481 | field. the next field determines which state |
| 482 | is to be transitioned to if any. |
| 483 | */ |
| 484 | struct _reg_trie_trans { |
| 485 | U32 next; |
| 486 | U32 check; |
| 487 | }; |
| 488 | |
| 489 | /* a transition list element for the list based representation */ |
| 490 | struct _reg_trie_trans_list_elem { |
| 491 | U16 forid; |
| 492 | U32 newstate; |
| 493 | }; |
| 494 | typedef struct _reg_trie_trans_list_elem reg_trie_trans_le; |
| 495 | |
| 496 | /* a state for compressed nodes. base is an offset |
| 497 | into an array of reg_trie_trans array. If wordnum is |
| 498 | nonzero the state is accepting. if base is zero then |
| 499 | the state has no children (and will be accepting) |
| 500 | */ |
| 501 | struct _reg_trie_state { |
| 502 | U16 wordnum; |
| 503 | union { |
| 504 | U32 base; |
| 505 | reg_trie_trans_le* list; |
| 506 | } trans; |
| 507 | }; |
| 508 | |
| 509 | |
| 510 | |
| 511 | typedef struct _reg_trie_state reg_trie_state; |
| 512 | typedef struct _reg_trie_trans reg_trie_trans; |
| 513 | |
| 514 | |
| 515 | /* anything in here that needs to be freed later |
| 516 | should be dealt with in pregfree. |
| 517 | refcount is first in both this and _reg_ac_data to allow a space |
| 518 | optimisation in Perl_regdupe. */ |
| 519 | struct _reg_trie_data { |
| 520 | U32 refcount; /* number of times this trie is referenced */ |
| 521 | U16 uniquecharcount; /* unique chars in trie (width of trans table) */ |
| 522 | U32 lasttrans; /* last valid transition element */ |
| 523 | U16 *charmap; /* byte to charid lookup array */ |
| 524 | reg_trie_state *states; /* state data */ |
| 525 | reg_trie_trans *trans; /* array of transition elements */ |
| 526 | char *bitmap; /* stclass bitmap */ |
| 527 | U32 startstate; /* initial state - used for common prefix optimisation */ |
| 528 | STRLEN minlen; /* minimum length of words in trie - build/opt only? */ |
| 529 | STRLEN maxlen; /* maximum length of words in trie - build/opt only? */ |
| 530 | U32 *wordlen; /* array of lengths of words */ |
| 531 | U16 *jump; /* optional 1 indexed array of offsets before tail |
| 532 | for the node following a given word. */ |
| 533 | U16 *nextword; /* optional 1 indexed array to support linked list |
| 534 | of duplicate wordnums */ |
| 535 | U32 statecount; /* Build only - number of states in the states array |
| 536 | (including the unused zero state) */ |
| 537 | U32 wordcount; /* Build only */ |
| 538 | #ifdef DEBUGGING |
| 539 | STRLEN charcount; /* Build only */ |
| 540 | AV *words; /* Array of words contained in trie, for dumping */ |
| 541 | AV *revcharmap; /* Map of each charid back to its character representation */ |
| 542 | #endif |
| 543 | }; |
| 544 | typedef struct _reg_trie_data reg_trie_data; |
| 545 | |
| 546 | /* refcount is first in both this and _reg_trie_data to allow a space |
| 547 | optimisation in Perl_regdupe. */ |
| 548 | struct _reg_ac_data { |
| 549 | U32 refcount; |
| 550 | U32 *fail; |
| 551 | reg_trie_state *states; |
| 552 | U32 trie; |
| 553 | }; |
| 554 | typedef struct _reg_ac_data reg_ac_data; |
| 555 | |
| 556 | /* ANY_BIT doesnt use the structure, so we can borrow it here. |
| 557 | This is simpler than refactoring all of it as wed end up with |
| 558 | three different sets... */ |
| 559 | |
| 560 | #define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap) |
| 561 | #define TRIE_BITMAP_BYTE(p, c) (TRIE_BITMAP(p)[(((U8)(c)) >> 3) & 31]) |
| 562 | #define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c)) |
| 563 | #define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c)) |
| 564 | #define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c)) |
| 565 | |
| 566 | #define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC) |
| 567 | #define IS_TRIE_AC(op) ((op)>=AHOCORASICK) |
| 568 | |
| 569 | |
| 570 | #define BITMAP_BYTE(p, c) (((U8*)p)[(((U8)(c)) >> 3) & 31]) |
| 571 | #define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c)) |
| 572 | |
| 573 | /* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */ |
| 574 | #define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1) |
| 575 | #define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state)) |
| 576 | #define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state)) |
| 577 | |
| 578 | #ifdef DEBUGGING |
| 579 | #define TRIE_CHARCOUNT(trie) ((trie)->charcount) |
| 580 | #define TRIE_REVCHARMAP(trie) ((trie)->revcharmap) |
| 581 | #else |
| 582 | #define TRIE_CHARCOUNT(trie) (trie_charcount) |
| 583 | #define TRIE_REVCHARMAP(trie) (trie_revcharmap) |
| 584 | #endif |
| 585 | |
| 586 | #define RE_TRIE_MAXBUF_INIT 65536 |
| 587 | #define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF" |
| 588 | #define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS" |
| 589 | |
| 590 | /* |
| 591 | |
| 592 | RE_DEBUG_FLAGS is used to control what debug output is emitted |
| 593 | its divided into three groups of options, some of which interact. |
| 594 | The three groups are: Compile, Execute, Extra. There is room for a |
| 595 | further group, as currently only the low three bytes are used. |
| 596 | |
| 597 | Compile Options: |
| 598 | |
| 599 | PARSE |
| 600 | PEEP |
| 601 | TRIE |
| 602 | PROGRAM |
| 603 | OFFSETS |
| 604 | |
| 605 | Execute Options: |
| 606 | |
| 607 | INTUIT |
| 608 | MATCH |
| 609 | TRIE |
| 610 | |
| 611 | Extra Options |
| 612 | |
| 613 | TRIE |
| 614 | OFFSETS |
| 615 | |
| 616 | If you modify any of these make sure you make corresponding changes to |
| 617 | re.pm, especially to the documentation. |
| 618 | |
| 619 | */ |
| 620 | |
| 621 | |
| 622 | /* Compile */ |
| 623 | #define RE_DEBUG_COMPILE_MASK 0x0000FF |
| 624 | #define RE_DEBUG_COMPILE_PARSE 0x000001 |
| 625 | #define RE_DEBUG_COMPILE_OPTIMISE 0x000002 |
| 626 | #define RE_DEBUG_COMPILE_TRIE 0x000004 |
| 627 | #define RE_DEBUG_COMPILE_DUMP 0x000008 |
| 628 | |
| 629 | /* Execute */ |
| 630 | #define RE_DEBUG_EXECUTE_MASK 0x00FF00 |
| 631 | #define RE_DEBUG_EXECUTE_INTUIT 0x000100 |
| 632 | #define RE_DEBUG_EXECUTE_MATCH 0x000200 |
| 633 | #define RE_DEBUG_EXECUTE_TRIE 0x000400 |
| 634 | |
| 635 | /* Extra */ |
| 636 | #define RE_DEBUG_EXTRA_MASK 0xFF0000 |
| 637 | #define RE_DEBUG_EXTRA_TRIE 0x010000 |
| 638 | #define RE_DEBUG_EXTRA_OFFSETS 0x020000 |
| 639 | #define RE_DEBUG_EXTRA_OFFDEBUG 0x040000 |
| 640 | #define RE_DEBUG_EXTRA_STATE 0x080000 |
| 641 | #define RE_DEBUG_EXTRA_OPTIMISE 0x100000 |
| 642 | /* combined */ |
| 643 | #define RE_DEBUG_EXTRA_STACK 0x280000 |
| 644 | |
| 645 | #define RE_DEBUG_FLAG(x) (re_debug_flags & x) |
| 646 | /* Compile */ |
| 647 | #define DEBUG_COMPILE_r(x) DEBUG_r( \ |
| 648 | if (re_debug_flags & RE_DEBUG_COMPILE_MASK) x ) |
| 649 | #define DEBUG_PARSE_r(x) DEBUG_r( \ |
| 650 | if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x ) |
| 651 | #define DEBUG_OPTIMISE_r(x) DEBUG_r( \ |
| 652 | if (re_debug_flags & RE_DEBUG_COMPILE_OPTIMISE) x ) |
| 653 | #define DEBUG_PARSE_r(x) DEBUG_r( \ |
| 654 | if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x ) |
| 655 | #define DEBUG_DUMP_r(x) DEBUG_r( \ |
| 656 | if (re_debug_flags & RE_DEBUG_COMPILE_DUMP) x ) |
| 657 | #define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \ |
| 658 | if (re_debug_flags & RE_DEBUG_COMPILE_TRIE) x ) |
| 659 | |
| 660 | /* Execute */ |
| 661 | #define DEBUG_EXECUTE_r(x) DEBUG_r( \ |
| 662 | if (re_debug_flags & RE_DEBUG_EXECUTE_MASK) x ) |
| 663 | #define DEBUG_INTUIT_r(x) DEBUG_r( \ |
| 664 | if (re_debug_flags & RE_DEBUG_EXECUTE_INTUIT) x ) |
| 665 | #define DEBUG_MATCH_r(x) DEBUG_r( \ |
| 666 | if (re_debug_flags & RE_DEBUG_EXECUTE_MATCH) x ) |
| 667 | #define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \ |
| 668 | if (re_debug_flags & RE_DEBUG_EXECUTE_TRIE) x ) |
| 669 | |
| 670 | /* Extra */ |
| 671 | #define DEBUG_EXTRA_r(x) DEBUG_r( \ |
| 672 | if (re_debug_flags & RE_DEBUG_EXTRA_MASK) x ) |
| 673 | #define DEBUG_OFFSETS_r(x) DEBUG_r( \ |
| 674 | if (re_debug_flags & RE_DEBUG_EXTRA_OFFSETS) x ) |
| 675 | #define DEBUG_STATE_r(x) DEBUG_r( \ |
| 676 | if (re_debug_flags & RE_DEBUG_EXTRA_STATE) x ) |
| 677 | #define DEBUG_STACK_r(x) DEBUG_r( \ |
| 678 | if (re_debug_flags & RE_DEBUG_EXTRA_STACK) x ) |
| 679 | #define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \ |
| 680 | if ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \ |
| 681 | (re_debug_flags & (RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE)) ) x ) |
| 682 | #define MJD_OFFSET_DEBUG(x) DEBUG_r( \ |
| 683 | if (re_debug_flags & RE_DEBUG_EXTRA_OFFDEBUG) \ |
| 684 | Perl_warn_nocontext x ) |
| 685 | #define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \ |
| 686 | if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x ) |
| 687 | #define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \ |
| 688 | if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x ) |
| 689 | |
| 690 | #define DEBUG_TRIE_r(x) DEBUG_r( \ |
| 691 | if (re_debug_flags & (RE_DEBUG_COMPILE_TRIE \ |
| 692 | | RE_DEBUG_EXECUTE_TRIE )) x ) |
| 693 | |
| 694 | /* initialization */ |
| 695 | /* get_sv() can return NULL during global destruction. */ |
| 696 | #define GET_RE_DEBUG_FLAGS DEBUG_r({ \ |
| 697 | SV * re_debug_flags_sv = NULL; \ |
| 698 | re_debug_flags_sv = get_sv(RE_DEBUG_FLAGS, 1); \ |
| 699 | if (re_debug_flags_sv) { \ |
| 700 | if (!SvIOK(re_debug_flags_sv)) \ |
| 701 | sv_setuv(re_debug_flags_sv, RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \ |
| 702 | re_debug_flags=SvIV(re_debug_flags_sv); \ |
| 703 | }\ |
| 704 | }) |
| 705 | |
| 706 | #ifdef DEBUGGING |
| 707 | |
| 708 | #define GET_RE_DEBUG_FLAGS_DECL IV re_debug_flags = 0; GET_RE_DEBUG_FLAGS; |
| 709 | |
| 710 | #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \ |
| 711 | const char * const rpv = \ |
| 712 | pv_pretty((dsv), (pv), (l), (m), \ |
| 713 | PL_colors[(c1)],PL_colors[(c2)], \ |
| 714 | ((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \ |
| 715 | const int rlen = SvCUR(dsv) |
| 716 | |
| 717 | #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \ |
| 718 | const char * const rpv = \ |
| 719 | pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \ |
| 720 | PL_colors[(c1)],PL_colors[(c2)], \ |
| 721 | ((isuni) ? PERL_PV_ESCAPE_UNI : 0) ) |
| 722 | |
| 723 | #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \ |
| 724 | const char * const rpv = \ |
| 725 | pv_pretty((dsv), (pv), (l), (m), \ |
| 726 | PL_colors[0], PL_colors[1], \ |
| 727 | ( PERL_PV_PRETTY_QUOTE | PERL_PV_PRETTY_ELIPSES | \ |
| 728 | ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \ |
| 729 | ) |
| 730 | |
| 731 | #define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0)) |
| 732 | #define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "") |
| 733 | |
| 734 | #else /* if not DEBUGGING */ |
| 735 | |
| 736 | #define GET_RE_DEBUG_FLAGS_DECL |
| 737 | #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) |
| 738 | #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) |
| 739 | #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) |
| 740 | #define RE_SV_DUMPLEN(ItEm) |
| 741 | #define RE_SV_TAIL(ItEm) |
| 742 | |
| 743 | #endif /* DEBUG RELATED DEFINES */ |
| 744 | |