| 1 | /* regcomp.h |
| 2 | * |
| 3 | * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, |
| 4 | * 2000, 2001, 2002, 2003, 2005, 2006, 2007, 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 | #include "regcharclass.h" |
| 11 | |
| 12 | /* Convert branch sequences to more efficient trie ops? */ |
| 13 | #define PERL_ENABLE_TRIE_OPTIMISATION 1 |
| 14 | |
| 15 | /* Be really aggressive about optimising patterns with trie sequences? */ |
| 16 | #define PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION 1 |
| 17 | |
| 18 | /* Should the optimiser take positive assertions into account? */ |
| 19 | #define PERL_ENABLE_POSITIVE_ASSERTION_STUDY 0 |
| 20 | |
| 21 | /* Not for production use: */ |
| 22 | #define PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS 0 |
| 23 | |
| 24 | /* Activate offsets code - set to if 1 to enable */ |
| 25 | #ifdef DEBUGGING |
| 26 | #define RE_TRACK_PATTERN_OFFSETS |
| 27 | #endif |
| 28 | |
| 29 | /* |
| 30 | * The "internal use only" fields in regexp.h are present to pass info from |
| 31 | * compile to execute that permits the execute phase to run lots faster on |
| 32 | * simple cases. They are: |
| 33 | * |
| 34 | * regstart sv that must begin a match; NULL if none obvious |
| 35 | * reganch is the match anchored (at beginning-of-line only)? |
| 36 | * regmust string (pointer into program) that match must include, or NULL |
| 37 | * [regmust changed to SV* for bminstr()--law] |
| 38 | * regmlen length of regmust string |
| 39 | * [regmlen not used currently] |
| 40 | * |
| 41 | * Regstart and reganch permit very fast decisions on suitable starting points |
| 42 | * for a match, cutting down the work a lot. Regmust permits fast rejection |
| 43 | * of lines that cannot possibly match. The regmust tests are costly enough |
| 44 | * that pregcomp() supplies a regmust only if the r.e. contains something |
| 45 | * potentially expensive (at present, the only such thing detected is * or + |
| 46 | * at the start of the r.e., which can involve a lot of backup). Regmlen is |
| 47 | * supplied because the test in pregexec() needs it and pregcomp() is computing |
| 48 | * it anyway. |
| 49 | * [regmust is now supplied always. The tests that use regmust have a |
| 50 | * heuristic that disables the test if it usually matches.] |
| 51 | * |
| 52 | * [In fact, we now use regmust in many cases to locate where the search |
| 53 | * starts in the string, so if regback is >= 0, the regmust search is never |
| 54 | * wasted effort. The regback variable says how many characters back from |
| 55 | * where regmust matched is the earliest possible start of the match. |
| 56 | * For instance, /[a-z].foo/ has a regmust of 'foo' and a regback of 2.] |
| 57 | */ |
| 58 | |
| 59 | /* |
| 60 | * Structure for regexp "program". This is essentially a linear encoding |
| 61 | * of a nondeterministic finite-state machine (aka syntax charts or |
| 62 | * "railroad normal form" in parsing technology). Each node is an opcode |
| 63 | * plus a "next" pointer, possibly plus an operand. "Next" pointers of |
| 64 | * all nodes except BRANCH implement concatenation; a "next" pointer with |
| 65 | * a BRANCH on both ends of it is connecting two alternatives. (Here we |
| 66 | * have one of the subtle syntax dependencies: an individual BRANCH (as |
| 67 | * opposed to a collection of them) is never concatenated with anything |
| 68 | * because of operator precedence.) The operand of some types of node is |
| 69 | * a literal string; for others, it is a node leading into a sub-FSM. In |
| 70 | * particular, the operand of a BRANCH node is the first node of the branch. |
| 71 | * (NB this is *not* a tree structure: the tail of the branch connects |
| 72 | * to the thing following the set of BRANCHes.) The opcodes are defined |
| 73 | * in regnodes.h which is generated from regcomp.sym by regcomp.pl. |
| 74 | */ |
| 75 | |
| 76 | /* |
| 77 | * A node is one char of opcode followed by two chars of "next" pointer. |
| 78 | * "Next" pointers are stored as two 8-bit pieces, high order first. The |
| 79 | * value is a positive offset from the opcode of the node containing it. |
| 80 | * An operand, if any, simply follows the node. (Note that much of the |
| 81 | * code generation knows about this implicit relationship.) |
| 82 | * |
| 83 | * Using two bytes for the "next" pointer is vast overkill for most things, |
| 84 | * but allows patterns to get big without disasters. |
| 85 | * |
| 86 | * [The "next" pointer is always aligned on an even |
| 87 | * boundary, and reads the offset directly as a short.] |
| 88 | */ |
| 89 | |
| 90 | /* This is the stuff that used to live in regexp.h that was truly |
| 91 | private to the engine itself. It now lives here. */ |
| 92 | |
| 93 | typedef struct regexp_internal { |
| 94 | int name_list_idx; /* Optional data index of an array of paren names */ |
| 95 | union { |
| 96 | U32 *offsets; /* offset annotations 20001228 MJD |
| 97 | data about mapping the program to the |
| 98 | string - |
| 99 | offsets[0] is proglen when this is used |
| 100 | */ |
| 101 | U32 proglen; |
| 102 | } u; |
| 103 | |
| 104 | regnode *regstclass; /* Optional startclass as identified or constructed |
| 105 | by the optimiser */ |
| 106 | struct reg_data *data; /* Additional miscellaneous data used by the program. |
| 107 | Used to make it easier to clone and free arbitrary |
| 108 | data that the regops need. Often the ARG field of |
| 109 | a regop is an index into this structure */ |
| 110 | struct reg_code_block *code_blocks;/* positions of literal (?{}) */ |
| 111 | int num_code_blocks; /* size of code_blocks[] */ |
| 112 | regnode program[1]; /* Unwarranted chumminess with compiler. */ |
| 113 | } regexp_internal; |
| 114 | |
| 115 | #define RXi_SET(x,y) (x)->pprivate = (void*)(y) |
| 116 | #define RXi_GET(x) ((regexp_internal *)((x)->pprivate)) |
| 117 | #define RXi_GET_DECL(r,ri) regexp_internal *ri = RXi_GET(r) |
| 118 | /* |
| 119 | * Flags stored in regexp->intflags |
| 120 | * These are used only internally to the regexp engine |
| 121 | * |
| 122 | * See regexp.h for flags used externally to the regexp engine |
| 123 | */ |
| 124 | #define RXp_INTFLAGS(rx) ((rx)->intflags) |
| 125 | #define RX_INTFLAGS(prog) RXp_INTFLAGS(ReANY(prog)) |
| 126 | |
| 127 | #define PREGf_SKIP 0x00000001 |
| 128 | #define PREGf_IMPLICIT 0x00000002 /* Converted .* to ^.* */ |
| 129 | #define PREGf_NAUGHTY 0x00000004 /* how exponential is this pattern? */ |
| 130 | #define PREGf_VERBARG_SEEN 0x00000008 |
| 131 | #define PREGf_CUTGROUP_SEEN 0x00000010 |
| 132 | #define PREGf_USE_RE_EVAL 0x00000020 /* compiled with "use re 'eval'" */ |
| 133 | /* these used to be extflags, but are now intflags */ |
| 134 | #define PREGf_NOSCAN 0x00000040 |
| 135 | /* spare */ |
| 136 | #define PREGf_GPOS_SEEN 0x00000100 |
| 137 | #define PREGf_GPOS_FLOAT 0x00000200 |
| 138 | |
| 139 | #define PREGf_ANCH_MBOL 0x00000400 |
| 140 | #define PREGf_ANCH_SBOL 0x00000800 |
| 141 | #define PREGf_ANCH_GPOS 0x00001000 |
| 142 | #define PREGf_RECURSE_SEEN 0x00002000 |
| 143 | |
| 144 | #define PREGf_ANCH \ |
| 145 | ( PREGf_ANCH_SBOL | PREGf_ANCH_GPOS | PREGf_ANCH_MBOL ) |
| 146 | |
| 147 | /* this is where the old regcomp.h started */ |
| 148 | |
| 149 | struct regnode_string { |
| 150 | U8 str_len; |
| 151 | U8 type; |
| 152 | U16 next_off; |
| 153 | char string[1]; |
| 154 | }; |
| 155 | |
| 156 | /* Argument bearing node - workhorse, |
| 157 | arg1 is often for the data field */ |
| 158 | struct regnode_1 { |
| 159 | U8 flags; |
| 160 | U8 type; |
| 161 | U16 next_off; |
| 162 | U32 arg1; |
| 163 | }; |
| 164 | |
| 165 | /* Similar to a regnode_1 but with an extra signed argument */ |
| 166 | struct regnode_2L { |
| 167 | U8 flags; |
| 168 | U8 type; |
| 169 | U16 next_off; |
| 170 | U32 arg1; |
| 171 | I32 arg2; |
| 172 | }; |
| 173 | |
| 174 | /* 'Two field' -- Two 16 bit unsigned args */ |
| 175 | struct regnode_2 { |
| 176 | U8 flags; |
| 177 | U8 type; |
| 178 | U16 next_off; |
| 179 | U16 arg1; |
| 180 | U16 arg2; |
| 181 | }; |
| 182 | |
| 183 | /* This give the number of code points that can be in the bitmap of an ANYOF |
| 184 | * node. The shift number must currently be one of: 8..12. It can't be less |
| 185 | * than 8 (256) because some code relies on it being at least that. Above 12 |
| 186 | * (4096), and you start running into warnings that some data structure widths |
| 187 | * have been exceeded, though the test suite as of this writing still passes |
| 188 | * for up through 16, which is as high as anyone would ever want to go, |
| 189 | * encompassing all of the Unicode BMP, and thus including all the economically |
| 190 | * important world scripts. At 12 most of them are: including Arabic, |
| 191 | * Cyrillic, Greek, Hebrew, Indian subcontinent, Latin, and Thai; but not Han, |
| 192 | * Japanese, nor Korean. (The regarglen structure in regnodes.h is a U8, and |
| 193 | * the trie types TRIEC and AHOCORASICKC are larger than U8 for shift values |
| 194 | * below above 12.) Be sure to benchmark before changing, as larger sizes do |
| 195 | * significantly slow down the test suite */ |
| 196 | #define NUM_ANYOF_CODE_POINTS (1 << 8) |
| 197 | |
| 198 | #define ANYOF_BITMAP_SIZE (NUM_ANYOF_CODE_POINTS / 8) /* 8 bits/Byte */ |
| 199 | |
| 200 | /* Note that these form structs which are supersets of the next smaller one, by |
| 201 | * appending fields. Alignment problems can occur if one of those optional |
| 202 | * fields requires stricter alignment than the base struct. And formal |
| 203 | * parameters that can really be two or more of the structs should be |
| 204 | * declared as the smallest one it could be. See commit message for |
| 205 | * 7dcac5f6a5195002b55c935ee1d67f67e1df280b. Regnode allocation is done |
| 206 | * without regard to alignment, and changing it to would also require changing |
| 207 | * the code that inserts and deletes regnodes. The basic single-argument |
| 208 | * regnode has a U32, which is what reganode() allocates as a unit. Therefore |
| 209 | * no field can require stricter alignment than U32. */ |
| 210 | |
| 211 | /* also used by trie */ |
| 212 | struct regnode_charclass { |
| 213 | U8 flags; |
| 214 | U8 type; |
| 215 | U16 next_off; |
| 216 | U32 arg1; /* set by set_ANYOF_arg() */ |
| 217 | char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */ |
| 218 | }; |
| 219 | |
| 220 | /* has runtime (locale) \d, \w, ..., [:posix:] classes */ |
| 221 | struct regnode_charclass_class { |
| 222 | U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */ |
| 223 | U8 type; |
| 224 | U16 next_off; |
| 225 | U32 arg1; |
| 226 | char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */ |
| 227 | U32 classflags; /* and run-time */ |
| 228 | }; |
| 229 | |
| 230 | /* A synthetic start class (SSC); is a regnode_charclass_posixl_fold, plus an |
| 231 | * extra SV*, used only during its construction and which is not used by |
| 232 | * regexec.c. Note that the 'next_off' field is unused, as the SSC stands |
| 233 | * alone, so there is never a next node. Also, there is no alignment issue, |
| 234 | * because these are declared or allocated as a complete unit so the compiler |
| 235 | * takes care of alignment. This is unlike the other regnodes which are |
| 236 | * allocated in terms of multiples of a single-argument regnode. SSC nodes can |
| 237 | * have a pointer field because there is no alignment issue, and because it is |
| 238 | * set to NULL after construction, before any cloning of the pattern */ |
| 239 | struct regnode_ssc { |
| 240 | U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */ |
| 241 | U8 type; |
| 242 | U16 next_off; |
| 243 | U32 arg1; |
| 244 | char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */ |
| 245 | U32 classflags; /* ... and run-time */ |
| 246 | |
| 247 | /* Auxiliary, only used during construction; NULL afterwards: list of code |
| 248 | * points matched */ |
| 249 | SV* invlist; |
| 250 | }; |
| 251 | |
| 252 | /* We take advantage of 'next_off' not otherwise being used in the SSC by |
| 253 | * actually using it: by setting it to 1. This allows us to test and |
| 254 | * distinguish between an SSC and other ANYOF node types, as 'next_off' cannot |
| 255 | * otherwise be 1, because it is the offset to the next regnode expressed in |
| 256 | * units of regnodes. Since an ANYOF node contains extra fields, it adds up |
| 257 | * to 12 regnode units on 32-bit systems, (hence the minimum this can be (if |
| 258 | * not 0) is 11 there. Even if things get tightly packed on a 64-bit system, |
| 259 | * it still would be more than 1. */ |
| 260 | #define set_ANYOF_SYNTHETIC(n) STMT_START{ OP(n) = ANYOF; \ |
| 261 | NEXT_OFF(n) = 1; \ |
| 262 | } STMT_END |
| 263 | #define is_ANYOF_SYNTHETIC(n) (PL_regkind[OP(n)] == ANYOF && NEXT_OFF(n) == 1) |
| 264 | |
| 265 | /* XXX fix this description. |
| 266 | Impose a limit of REG_INFTY on various pattern matching operations |
| 267 | to limit stack growth and to avoid "infinite" recursions. |
| 268 | */ |
| 269 | /* The default size for REG_INFTY is I16_MAX, which is the same as |
| 270 | SHORT_MAX (see perl.h). Unfortunately I16 isn't necessarily 16 bits |
| 271 | (see handy.h). On the Cray C90, sizeof(short)==4 and hence I16_MAX is |
| 272 | ((1<<31)-1), while on the Cray T90, sizeof(short)==8 and I16_MAX is |
| 273 | ((1<<63)-1). To limit stack growth to reasonable sizes, supply a |
| 274 | smaller default. |
| 275 | --Andy Dougherty 11 June 1998 |
| 276 | */ |
| 277 | #if SHORTSIZE > 2 |
| 278 | # ifndef REG_INFTY |
| 279 | # define REG_INFTY ((1<<15)-1) |
| 280 | # endif |
| 281 | #endif |
| 282 | |
| 283 | #ifndef REG_INFTY |
| 284 | # define REG_INFTY I16_MAX |
| 285 | #endif |
| 286 | |
| 287 | #define ARG_VALUE(arg) (arg) |
| 288 | #define ARG__SET(arg,val) ((arg) = (val)) |
| 289 | |
| 290 | #undef ARG |
| 291 | #undef ARG1 |
| 292 | #undef ARG2 |
| 293 | |
| 294 | #define ARG(p) ARG_VALUE(ARG_LOC(p)) |
| 295 | #define ARG1(p) ARG_VALUE(ARG1_LOC(p)) |
| 296 | #define ARG2(p) ARG_VALUE(ARG2_LOC(p)) |
| 297 | #define ARG2L(p) ARG_VALUE(ARG2L_LOC(p)) |
| 298 | |
| 299 | #define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val)) |
| 300 | #define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val)) |
| 301 | #define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val)) |
| 302 | #define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val)) |
| 303 | |
| 304 | #undef NEXT_OFF |
| 305 | #undef NODE_ALIGN |
| 306 | |
| 307 | #define NEXT_OFF(p) ((p)->next_off) |
| 308 | #define NODE_ALIGN(node) |
| 309 | /* the following define was set to 0xde in 075abff3 |
| 310 | * as part of some linting logic. I have set it to 0 |
| 311 | * as otherwise in every place where we /might/ set flags |
| 312 | * we have to set it 0 explicitly, which duplicates |
| 313 | * assignments and IMO adds an unacceptable level of |
| 314 | * surprise to working in the regex engine. If this |
| 315 | * is changed from 0 then at the very least make sure |
| 316 | * that SBOL for /^/ sets the flags to 0 explicitly. |
| 317 | * -- Yves */ |
| 318 | #define NODE_ALIGN_FILL(node) ((node)->flags = 0) |
| 319 | |
| 320 | #define SIZE_ALIGN NODE_ALIGN |
| 321 | |
| 322 | #undef OP |
| 323 | #undef OPERAND |
| 324 | #undef MASK |
| 325 | #undef STRING |
| 326 | |
| 327 | #define OP(p) ((p)->type) |
| 328 | #define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \ |
| 329 | regnode types. For some, it's the \ |
| 330 | character set of the regnode */ |
| 331 | #define OPERAND(p) (((struct regnode_string *)p)->string) |
| 332 | #define MASK(p) ((char*)OPERAND(p)) |
| 333 | #define STR_LEN(p) (((struct regnode_string *)p)->str_len) |
| 334 | #define STRING(p) (((struct regnode_string *)p)->string) |
| 335 | #define STR_SZ(l) ((l + sizeof(regnode) - 1) / sizeof(regnode)) |
| 336 | #define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p))+1) |
| 337 | |
| 338 | #undef NODE_ALIGN |
| 339 | #undef ARG_LOC |
| 340 | #undef NEXTOPER |
| 341 | #undef PREVOPER |
| 342 | |
| 343 | #define NODE_ALIGN(node) |
| 344 | #define ARG_LOC(p) (((struct regnode_1 *)p)->arg1) |
| 345 | #define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1) |
| 346 | #define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2) |
| 347 | #define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2) |
| 348 | |
| 349 | #define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */ |
| 350 | #define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2) |
| 351 | |
| 352 | #define NODE_STEP_B 4 |
| 353 | |
| 354 | #define NEXTOPER(p) ((p) + NODE_STEP_REGNODE) |
| 355 | #define PREVOPER(p) ((p) - NODE_STEP_REGNODE) |
| 356 | |
| 357 | #define FILL_ADVANCE_NODE(ptr, op) STMT_START { \ |
| 358 | (ptr)->type = op; (ptr)->next_off = 0; (ptr)++; } STMT_END |
| 359 | #define FILL_ADVANCE_NODE_ARG(ptr, op, arg) STMT_START { \ |
| 360 | ARG_SET(ptr, arg); FILL_ADVANCE_NODE(ptr, op); (ptr) += 1; } STMT_END |
| 361 | #define FILL_ADVANCE_NODE_2L_ARG(ptr, op, arg1, arg2) \ |
| 362 | STMT_START { \ |
| 363 | ARG_SET(ptr, arg1); \ |
| 364 | ARG2L_SET(ptr, arg2); \ |
| 365 | FILL_ADVANCE_NODE(ptr, op); \ |
| 366 | (ptr) += 2; \ |
| 367 | } STMT_END |
| 368 | |
| 369 | #define REG_MAGIC 0234 |
| 370 | |
| 371 | #define SIZE_ONLY (RExC_emit == (regnode *) & RExC_emit_dummy) |
| 372 | #define PASS1 SIZE_ONLY |
| 373 | #define PASS2 (! SIZE_ONLY) |
| 374 | |
| 375 | /* An ANYOF node is basically a bitmap with the index being a code point. If |
| 376 | * the bit for that code point is 1, the code point matches; if 0, it doesn't |
| 377 | * match (complemented if inverted). There is an additional mechanism to deal |
| 378 | * with cases where the bitmap is insufficient in and of itself. This #define |
| 379 | * indicates if the bitmap does fully represent what this ANYOF node can match. |
| 380 | * The ARG is set to this special value (since 0, 1, ... are legal, but will |
| 381 | * never reach this high). */ |
| 382 | #define ANYOF_ONLY_HAS_BITMAP ((U32) -1) |
| 383 | |
| 384 | /* When the bimap isn't completely sufficient for handling the ANYOF node, |
| 385 | * flags (in node->flags of the ANYOF node) get set to indicate this. These |
| 386 | * are perennially in short supply. Beyond several cases where warnings need |
| 387 | * to be raised under certain circumstances, currently, there are six cases |
| 388 | * where the bitmap alone isn't sufficient. We could use six flags to |
| 389 | * represent the 6 cases, but to save flags bits, we play some games. The |
| 390 | * cases are: |
| 391 | * |
| 392 | * 1) The bitmap has a compiled-in very finite size. So something else needs |
| 393 | * to be used to specify if a code point that is too large for the bitmap |
| 394 | * actually matches. The mechanism currently is a swash or inversion |
| 395 | * list. ANYOF_ONLY_HAS_BITMAP, described above, being TRUE indicates |
| 396 | * there are no matches of too-large code points. But if it is FALSE, |
| 397 | * then almost certainly there are matches too large for the bitmap. (The |
| 398 | * other cases, described below, either imply this one or are extremely |
| 399 | * rare in practice.) So we can just assume that a too-large code point |
| 400 | * will need something beyond the bitmap if ANYOF_ONLY_HAS_BITMAP is |
| 401 | * FALSE, instead of having a separate flag for this. |
| 402 | * 2) A subset of item 1) is if all possible code points outside the bitmap |
| 403 | * match. This is a common occurrence when the class is complemented, |
| 404 | * like /[^ij]/. Therefore a bit is reserved to indicate this, |
| 405 | * rather than having an expensive swash created, |
| 406 | * ANYOF_MATCHES_ALL_ABOVE_BITMAP. |
| 407 | * 3) Under /d rules, it can happen that code points that are in the upper |
| 408 | * latin1 range (\x80-\xFF or their equivalents on EBCDIC platforms) match |
| 409 | * only if the runtime target string being matched against is UTF-8. For |
| 410 | * example /[\w[:punct:]]/d. This happens only for posix classes (with a |
| 411 | * couple of exceptions, like \d where it doesn't happen), and all such |
| 412 | * ones also have above-bitmap matches. Thus, 3) implies 1) as well. |
| 413 | * Note that /d rules are no longer encouraged; 'use 5.14' or higher |
| 414 | * deselects them. But a flag is required so that they can be properly |
| 415 | * handled. But it can be a shared flag: see 5) below. |
| 416 | * 4) Also under /d rules, something like /[\Wfoo]/ will match everything in |
| 417 | * the \x80-\xFF range, unless the string being matched against is UTF-8. |
| 418 | * A swash could be created for this case, but this is relatively common, |
| 419 | * and it turns out that it's all or nothing: if any one of these code |
| 420 | * points matches, they all do. Hence a single bit suffices. We use a |
| 421 | * shared flag that doesn't take up space by itself: |
| 422 | * ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER. |
| 423 | * This also implies 1), with one exception: [:^cntrl:]. |
| 424 | * 5) A user-defined \p{} property may not have been defined by the time the |
| 425 | * regex is compiled. In this case, we don't know until runtime what it |
| 426 | * will match, so we have to assume it could match anything, including |
| 427 | * code points that ordinarily would be in the bitmap. A flag bit is |
| 428 | * necessary to indicate this , though it can be shared with the item 3) |
| 429 | * flag, as that only occurs under /d, and this only occurs under non-d. |
| 430 | * This case is quite uncommon in the field, and the /(?[ ...])/ construct |
| 431 | * is a better way to accomplish what this feature does. This case also |
| 432 | * implies 1). |
| 433 | * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP |
| 434 | * is the shared flag. |
| 435 | * 6) /[foo]/il may have folds that are only valid if the runtime locale is a |
| 436 | * UTF-8 one. These are quite rare, so it would be good to avoid the |
| 437 | * expense of looking for them. But /l matching is slow anyway, and we've |
| 438 | * traditionally not worried too much about its performance. And this |
| 439 | * condition requires the ANYOFL_FOLD flag to be set, so testing for |
| 440 | * that flag would be sufficient to rule out most cases of this. So it is |
| 441 | * unclear if this should have a flag or not. But, this flag can be |
| 442 | * shared with another, so it doesn't occupy extra space. |
| 443 | * |
| 444 | * At the moment, there is one spare bit, but this could be increased by |
| 445 | * various tricks. |
| 446 | * |
| 447 | * If just one more bit is needed, at this writing it seems to khw that the |
| 448 | * best choice would be to make ANYOF_MATCHES_ALL_ABOVE_BITMAP not a flag, but |
| 449 | * something like |
| 450 | * |
| 451 | * #define ANYOF_MATCHES_ALL_ABOVE_BITMAP ((U32) -2) |
| 452 | * |
| 453 | * and access it through the ARG like ANYOF_ONLY_HAS_BITMAP is. This flag is |
| 454 | * used by all ANYOF node types, and it could be used to avoid calling the |
| 455 | * handler function, as the macro REGINCLASS in regexec.c does now for other |
| 456 | * cases. |
| 457 | * |
| 458 | * Another possibility is to instead (or additionally) rename the ANYOF_POSIXL |
| 459 | * flag to be ANYOFL_LARGE, to mean that the ANYOF node has an extra 32 bits |
| 460 | * beyond what a regular one does. That's what it effectively means now, with |
| 461 | * the extra space all for the POSIX class flags. But those classes actually |
| 462 | * only occupy 30 bits, so the ANYOFL_FOLD and |
| 463 | * ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD flags could be moved |
| 464 | * to that extra space. The 30 bits in the extra word would indicate if a |
| 465 | * posix class should be looked up or not. The downside of this is that ANYOFL |
| 466 | * nodes with folding would always have to have the extra space allocated, even |
| 467 | * if they didn't use the 30 posix bits. There isn't an SSC problem as all |
| 468 | * SSCs are this large anyway. |
| 469 | * |
| 470 | * One could completely remove ANYOFL_LARGE and make all ANYOFL nodes large. |
| 471 | * REGINCLASS would have to be modified so that if the node type were this, it |
| 472 | * would call reginclass(), as the flag bit that indicates to do this now would |
| 473 | * be gone. |
| 474 | * |
| 475 | * All told, 5 bits could be available for other uses if all of the above were |
| 476 | * done. |
| 477 | * |
| 478 | * Some flags are not used in synthetic start class (SSC) nodes, so could be |
| 479 | * shared should new flags be needed for SSCs, like SSC_MATCHES_EMPTY_STRING |
| 480 | * now. */ |
| 481 | |
| 482 | /* If this is set, the result of the match should be complemented. regexec.c |
| 483 | * is expecting this to be in the low bit. Never in an SSC */ |
| 484 | #define ANYOF_INVERT 0x01 |
| 485 | |
| 486 | /* For the SSC node only, which cannot be inverted, so is shared with that bit. |
| 487 | * This is used only during regex compilation. */ |
| 488 | #define SSC_MATCHES_EMPTY_STRING ANYOF_INVERT |
| 489 | |
| 490 | /* Set if this is a regnode_charclass_posixl vs a regnode_charclass. This |
| 491 | * is used for runtime \d, \w, [:posix:], ..., which are used only in locale |
| 492 | * and the optimizer's synthetic start class. Non-locale \d, etc are resolved |
| 493 | * at compile-time. Only set under /l; can be in SSC */ |
| 494 | #define ANYOF_MATCHES_POSIXL 0x02 |
| 495 | |
| 496 | /* The fold is calculated and stored in the bitmap where possible at compile |
| 497 | * time. However under locale, the actual folding varies depending on |
| 498 | * what the locale is at the time of execution, so it has to be deferred until |
| 499 | * then. Only set under /l; never in an SSC */ |
| 500 | #define ANYOFL_FOLD 0x04 |
| 501 | |
| 502 | /* Shared bit set only with ANYOFL and SSC nodes: |
| 503 | * If ANYOFL_FOLD is set, this means there are potential matches valid |
| 504 | * only if the locale is a UTF-8 one. |
| 505 | * If ANYOFL_FOLD is NOT set, this means to warn if the runtime locale |
| 506 | * isn't a UTF-8 one (and the generated node assumes a UTF-8 locale). |
| 507 | * None of INVERT, POSIXL, |
| 508 | * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP |
| 509 | * can be set. */ |
| 510 | #define ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD 0x08 |
| 511 | |
| 512 | /* Convenience macros for teasing apart the meanings when reading the above bit |
| 513 | * */ |
| 514 | #define ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags) \ |
| 515 | ((flags & ( ANYOFL_FOLD /* Both bits are set */ \ |
| 516 | |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \ |
| 517 | == ( ANYOFL_FOLD \ |
| 518 | |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) |
| 519 | |
| 520 | #define ANYOFL_UTF8_LOCALE_REQD(flags) \ |
| 521 | ((flags & ( ANYOFL_FOLD /* Only REQD bit is set */ \ |
| 522 | |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \ |
| 523 | == ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD) |
| 524 | |
| 525 | /* Spare: 0x10 */ |
| 526 | |
| 527 | /* If set, the node matches every code point NUM_ANYOF_CODE_POINTS and above. |
| 528 | * Can be in an SSC */ |
| 529 | #define ANYOF_MATCHES_ALL_ABOVE_BITMAP 0x20 |
| 530 | |
| 531 | /* Shared bit: |
| 532 | * Under /d it means the ANYOFD node matches more things if the target |
| 533 | * string is encoded in UTF-8; any such things will be non-ASCII, |
| 534 | * characters that are < 256, and can be accessed via the swash. |
| 535 | * When not under /d, it means the ANYOF node contains a user-defined |
| 536 | * property that wasn't yet defined at the time the regex was compiled, |
| 537 | * and so must be looked up at runtime, by creating a swash |
| 538 | * (These uses are mutually exclusive because a user-defined property is |
| 539 | * specified by \p{}, and \p{} implies /u which deselects /d). The long macro |
| 540 | * name is to make sure that you are cautioned about its shared nature. Only |
| 541 | * the non-/d meaning can be in an SSC */ |
| 542 | #define ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 0x40 |
| 543 | |
| 544 | /* Shared bit: |
| 545 | * Under /d it means the ANYOFD node matches all non-ASCII Latin1 |
| 546 | * characters when the target string is not in utf8. |
| 547 | * When not under /d, it means the ANYOF node should raise a warning if |
| 548 | * matching against an above-Unicode code point. |
| 549 | * (These uses are mutually exclusive because the warning requires a \p{}, and |
| 550 | * \p{} implies /u which deselects /d). An SSC node only has this bit set if |
| 551 | * what is meant is the warning. The long macro name is to make sure that you |
| 552 | * are cautioned about its shared nature */ |
| 553 | #define ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER 0x80 |
| 554 | |
| 555 | #define ANYOF_FLAGS_ALL (0xff & ~0x10) |
| 556 | |
| 557 | #define ANYOF_LOCALE_FLAGS (ANYOFL_FOLD | ANYOF_MATCHES_POSIXL) |
| 558 | |
| 559 | /* These are the flags that apply to both regular ANYOF nodes and synthetic |
| 560 | * start class nodes during construction of the SSC. During finalization of |
| 561 | * the SSC, other of the flags may get added to it */ |
| 562 | #define ANYOF_COMMON_FLAGS 0 |
| 563 | |
| 564 | /* Character classes for node->classflags of ANYOF */ |
| 565 | /* Should be synchronized with a table in regprop() */ |
| 566 | /* 2n should be the normal one, paired with its complement at 2n+1 */ |
| 567 | |
| 568 | #define ANYOF_ALPHA ((_CC_ALPHA) * 2) |
| 569 | #define ANYOF_NALPHA ((ANYOF_ALPHA) + 1) |
| 570 | #define ANYOF_ALPHANUMERIC ((_CC_ALPHANUMERIC) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */ |
| 571 | #define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1) |
| 572 | #define ANYOF_ASCII ((_CC_ASCII) * 2) |
| 573 | #define ANYOF_NASCII ((ANYOF_ASCII) + 1) |
| 574 | #define ANYOF_BLANK ((_CC_BLANK) * 2) /* GNU extension: space and tab: non-vertical space */ |
| 575 | #define ANYOF_NBLANK ((ANYOF_BLANK) + 1) |
| 576 | #define ANYOF_CASED ((_CC_CASED) * 2) /* Pseudo class for [:lower:] or |
| 577 | [:upper:] under /i */ |
| 578 | #define ANYOF_NCASED ((ANYOF_CASED) + 1) |
| 579 | #define ANYOF_CNTRL ((_CC_CNTRL) * 2) |
| 580 | #define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1) |
| 581 | #define ANYOF_DIGIT ((_CC_DIGIT) * 2) /* \d */ |
| 582 | #define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1) |
| 583 | #define ANYOF_GRAPH ((_CC_GRAPH) * 2) |
| 584 | #define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1) |
| 585 | #define ANYOF_LOWER ((_CC_LOWER) * 2) |
| 586 | #define ANYOF_NLOWER ((ANYOF_LOWER) + 1) |
| 587 | #define ANYOF_PRINT ((_CC_PRINT) * 2) |
| 588 | #define ANYOF_NPRINT ((ANYOF_PRINT) + 1) |
| 589 | #define ANYOF_PUNCT ((_CC_PUNCT) * 2) |
| 590 | #define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1) |
| 591 | #define ANYOF_SPACE ((_CC_SPACE) * 2) /* \s */ |
| 592 | #define ANYOF_NSPACE ((ANYOF_SPACE) + 1) |
| 593 | #define ANYOF_UPPER ((_CC_UPPER) * 2) |
| 594 | #define ANYOF_NUPPER ((ANYOF_UPPER) + 1) |
| 595 | #define ANYOF_WORDCHAR ((_CC_WORDCHAR) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */ |
| 596 | #define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1) |
| 597 | #define ANYOF_XDIGIT ((_CC_XDIGIT) * 2) |
| 598 | #define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1) |
| 599 | |
| 600 | /* pseudo classes below this, not stored in the class bitmap, but used as flags |
| 601 | during compilation of char classes */ |
| 602 | |
| 603 | #define ANYOF_VERTWS ((_CC_VERTSPACE) * 2) |
| 604 | #define ANYOF_NVERTWS ((ANYOF_VERTWS)+1) |
| 605 | |
| 606 | /* It is best if this is the last one, as all above it are stored as bits in a |
| 607 | * bitmap, and it isn't part of that bitmap */ |
| 608 | #if _CC_VERTSPACE != _HIGHEST_REGCOMP_DOT_H_SYNC |
| 609 | # error Problem with handy.h _HIGHEST_REGCOMP_DOT_H_SYNC #define |
| 610 | #endif |
| 611 | |
| 612 | #define ANYOF_POSIXL_MAX (ANYOF_VERTWS) /* So upper loop limit is written: |
| 613 | * '< ANYOF_MAX' |
| 614 | * Hence doesn't include VERTWS, as that |
| 615 | * is a pseudo class */ |
| 616 | #define ANYOF_MAX ANYOF_POSIXL_MAX |
| 617 | |
| 618 | #if (ANYOF_POSIXL_MAX > 32) /* Must fit in 32-bit word */ |
| 619 | # error Problem with handy.h _CC_foo #defines |
| 620 | #endif |
| 621 | |
| 622 | #define ANYOF_HORIZWS ((ANYOF_POSIXL_MAX)+2) /* = (ANYOF_NVERTWS + 1) */ |
| 623 | #define ANYOF_NHORIZWS ((ANYOF_POSIXL_MAX)+3) |
| 624 | |
| 625 | #define ANYOF_UNIPROP ((ANYOF_POSIXL_MAX)+4) /* Used to indicate a Unicode |
| 626 | property: \p{} or \P{} */ |
| 627 | |
| 628 | /* Backward source code compatibility. */ |
| 629 | |
| 630 | #define ANYOF_ALNUML ANYOF_ALNUM |
| 631 | #define ANYOF_NALNUML ANYOF_NALNUM |
| 632 | #define ANYOF_SPACEL ANYOF_SPACE |
| 633 | #define ANYOF_NSPACEL ANYOF_NSPACE |
| 634 | #define ANYOF_ALNUM ANYOF_WORDCHAR |
| 635 | #define ANYOF_NALNUM ANYOF_NWORDCHAR |
| 636 | |
| 637 | /* Utility macros for the bitmap and classes of ANYOF */ |
| 638 | |
| 639 | #define ANYOF_FLAGS(p) ((p)->flags) |
| 640 | |
| 641 | #define ANYOF_BIT(c) (1U << ((c) & 7)) |
| 642 | |
| 643 | #define ANYOF_POSIXL_SET(p, c) (((regnode_charclass_posixl*) (p))->classflags |= (1U << (c))) |
| 644 | #define ANYOF_CLASS_SET(p, c) ANYOF_POSIXL_SET((p), (c)) |
| 645 | |
| 646 | #define ANYOF_POSIXL_CLEAR(p, c) (((regnode_charclass_posixl*) (p))->classflags &= ~ (1U <<(c))) |
| 647 | #define ANYOF_CLASS_CLEAR(p, c) ANYOF_POSIXL_CLEAR((p), (c)) |
| 648 | |
| 649 | #define ANYOF_POSIXL_TEST(p, c) (((regnode_charclass_posixl*) (p))->classflags & (1U << (c))) |
| 650 | #define ANYOF_CLASS_TEST(p, c) ANYOF_POSIXL_TEST((p), (c)) |
| 651 | |
| 652 | #define ANYOF_POSIXL_ZERO(ret) STMT_START { ((regnode_charclass_posixl*) (ret))->classflags = 0; } STMT_END |
| 653 | #define ANYOF_CLASS_ZERO(ret) ANYOF_POSIXL_ZERO(ret) |
| 654 | |
| 655 | /* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */ |
| 656 | #define ANYOF_POSIXL_SETALL(ret) STMT_START { ((regnode_charclass_posixl*) (ret))->classflags = ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1; } STMT_END |
| 657 | #define ANYOF_CLASS_SETALL(ret) ANYOF_POSIXL_SETALL(ret) |
| 658 | |
| 659 | #define ANYOF_POSIXL_TEST_ANY_SET(p) \ |
| 660 | ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \ |
| 661 | && (((regnode_charclass_posixl*)(p))->classflags)) |
| 662 | #define ANYOF_CLASS_TEST_ANY_SET(p) ANYOF_POSIXL_TEST_ANY_SET(p) |
| 663 | |
| 664 | /* Since an SSC always has this field, we don't have to test for that; nor do |
| 665 | * we want to because the bit isn't set for SSC during its construction */ |
| 666 | #define ANYOF_POSIXL_SSC_TEST_ANY_SET(p) \ |
| 667 | cBOOL(((regnode_ssc*)(p))->classflags) |
| 668 | #define ANYOF_POSIXL_SSC_TEST_ALL_SET(p) /* Are all bits set? */ \ |
| 669 | (((regnode_ssc*) (p))->classflags \ |
| 670 | == ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1) |
| 671 | |
| 672 | #define ANYOF_POSIXL_TEST_ALL_SET(p) \ |
| 673 | ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \ |
| 674 | && ((regnode_charclass_posixl*) (p))->classflags \ |
| 675 | == ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1) |
| 676 | |
| 677 | #define ANYOF_POSIXL_OR(source, dest) STMT_START { (dest)->classflags |= (source)->classflags ; } STMT_END |
| 678 | #define ANYOF_CLASS_OR(source, dest) ANYOF_POSIXL_OR((source), (dest)) |
| 679 | |
| 680 | #define ANYOF_POSIXL_AND(source, dest) STMT_START { (dest)->classflags &= (source)->classflags ; } STMT_END |
| 681 | |
| 682 | #define ANYOF_BITMAP_ZERO(ret) Zero(((regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char) |
| 683 | #define ANYOF_BITMAP(p) ((regnode_charclass*)(p))->bitmap |
| 684 | #define ANYOF_BITMAP_BYTE(p, c) BITMAP_BYTE(ANYOF_BITMAP(p), c) |
| 685 | #define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c)) |
| 686 | #define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c)) |
| 687 | #define ANYOF_BITMAP_TEST(p, c) cBOOL(ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c)) |
| 688 | |
| 689 | #define ANYOF_BITMAP_SETALL(p) \ |
| 690 | memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE) |
| 691 | #define ANYOF_BITMAP_CLEARALL(p) \ |
| 692 | Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE) |
| 693 | |
| 694 | #define ANYOF_SKIP EXTRA_SIZE(regnode_charclass) |
| 695 | #define ANYOF_POSIXL_SKIP EXTRA_SIZE(regnode_charclass_posixl) |
| 696 | |
| 697 | /* |
| 698 | * Utility definitions. |
| 699 | */ |
| 700 | #ifndef CHARMASK |
| 701 | # define UCHARAT(p) ((int)*(const U8*)(p)) |
| 702 | #else |
| 703 | # define UCHARAT(p) ((int)*(p)&CHARMASK) |
| 704 | #endif |
| 705 | |
| 706 | #define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode)) |
| 707 | |
| 708 | #define REG_ZERO_LEN_SEEN 0x00000001 |
| 709 | #define REG_LOOKBEHIND_SEEN 0x00000002 |
| 710 | #define REG_GPOS_SEEN 0x00000004 |
| 711 | /* spare */ |
| 712 | #define REG_RECURSE_SEEN 0x00000020 |
| 713 | #define REG_TOP_LEVEL_BRANCHES_SEEN 0x00000040 |
| 714 | #define REG_VERBARG_SEEN 0x00000080 |
| 715 | #define REG_CUTGROUP_SEEN 0x00000100 |
| 716 | #define REG_RUN_ON_COMMENT_SEEN 0x00000200 |
| 717 | #define REG_UNFOLDED_MULTI_SEEN 0x00000400 |
| 718 | /* spare */ |
| 719 | #define REG_UNBOUNDED_QUANTIFIER_SEEN 0x00001000 |
| 720 | |
| 721 | |
| 722 | START_EXTERN_C |
| 723 | |
| 724 | #ifdef PLUGGABLE_RE_EXTENSION |
| 725 | #include "re_nodes.h" |
| 726 | #else |
| 727 | #include "regnodes.h" |
| 728 | #endif |
| 729 | |
| 730 | #ifndef PLUGGABLE_RE_EXTENSION |
| 731 | #ifndef DOINIT |
| 732 | EXTCONST regexp_engine PL_core_reg_engine; |
| 733 | #else /* DOINIT */ |
| 734 | EXTCONST regexp_engine PL_core_reg_engine = { |
| 735 | Perl_re_compile, |
| 736 | Perl_regexec_flags, |
| 737 | Perl_re_intuit_start, |
| 738 | Perl_re_intuit_string, |
| 739 | Perl_regfree_internal, |
| 740 | Perl_reg_numbered_buff_fetch, |
| 741 | Perl_reg_numbered_buff_store, |
| 742 | Perl_reg_numbered_buff_length, |
| 743 | Perl_reg_named_buff, |
| 744 | Perl_reg_named_buff_iter, |
| 745 | Perl_reg_qr_package, |
| 746 | #if defined(USE_ITHREADS) |
| 747 | Perl_regdupe_internal, |
| 748 | #endif |
| 749 | Perl_re_op_compile |
| 750 | }; |
| 751 | #endif /* DOINIT */ |
| 752 | #endif /* PLUGGABLE_RE_EXTENSION */ |
| 753 | |
| 754 | |
| 755 | END_EXTERN_C |
| 756 | |
| 757 | |
| 758 | /* .what is a character array with one character for each member of .data |
| 759 | * The character describes the function of the corresponding .data item: |
| 760 | * a - AV for paren_name_list under DEBUGGING |
| 761 | * f - start-class data for regstclass optimization |
| 762 | * l - start op for literal (?{EVAL}) item |
| 763 | * L - start op for literal (?{EVAL}) item, with separate CV (qr//) |
| 764 | * r - pointer to an embedded code-containing qr, e.g. /ab$qr/ |
| 765 | * s - swash for Unicode-style character class, and the multicharacter |
| 766 | * strings resulting from casefolding the single-character entries |
| 767 | * in the character class |
| 768 | * t - trie struct |
| 769 | * u - trie struct's widecharmap (a HV, so can't share, must dup) |
| 770 | * also used for revcharmap and words under DEBUGGING |
| 771 | * T - aho-trie struct |
| 772 | * S - sv for named capture lookup |
| 773 | * 20010712 mjd@plover.com |
| 774 | * (Remember to update re_dup() and pregfree() if you add any items.) |
| 775 | */ |
| 776 | struct reg_data { |
| 777 | U32 count; |
| 778 | U8 *what; |
| 779 | void* data[1]; |
| 780 | }; |
| 781 | |
| 782 | /* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses |
| 783 | anchored* and float* via array indexes 0 and 1. */ |
| 784 | #define anchored_substr substrs->data[0].substr |
| 785 | #define anchored_utf8 substrs->data[0].utf8_substr |
| 786 | #define anchored_offset substrs->data[0].min_offset |
| 787 | #define anchored_end_shift substrs->data[0].end_shift |
| 788 | |
| 789 | #define float_substr substrs->data[1].substr |
| 790 | #define float_utf8 substrs->data[1].utf8_substr |
| 791 | #define float_min_offset substrs->data[1].min_offset |
| 792 | #define float_max_offset substrs->data[1].max_offset |
| 793 | #define float_end_shift substrs->data[1].end_shift |
| 794 | |
| 795 | #define check_substr substrs->data[2].substr |
| 796 | #define check_utf8 substrs->data[2].utf8_substr |
| 797 | #define check_offset_min substrs->data[2].min_offset |
| 798 | #define check_offset_max substrs->data[2].max_offset |
| 799 | #define check_end_shift substrs->data[2].end_shift |
| 800 | |
| 801 | #define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr) |
| 802 | #define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8) |
| 803 | #define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr) |
| 804 | #define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8) |
| 805 | |
| 806 | /* trie related stuff */ |
| 807 | |
| 808 | /* a transition record for the state machine. the |
| 809 | check field determines which state "owns" the |
| 810 | transition. the char the transition is for is |
| 811 | determined by offset from the owning states base |
| 812 | field. the next field determines which state |
| 813 | is to be transitioned to if any. |
| 814 | */ |
| 815 | struct _reg_trie_trans { |
| 816 | U32 next; |
| 817 | U32 check; |
| 818 | }; |
| 819 | |
| 820 | /* a transition list element for the list based representation */ |
| 821 | struct _reg_trie_trans_list_elem { |
| 822 | U16 forid; |
| 823 | U32 newstate; |
| 824 | }; |
| 825 | typedef struct _reg_trie_trans_list_elem reg_trie_trans_le; |
| 826 | |
| 827 | /* a state for compressed nodes. base is an offset |
| 828 | into an array of reg_trie_trans array. If wordnum is |
| 829 | nonzero the state is accepting. if base is zero then |
| 830 | the state has no children (and will be accepting) |
| 831 | */ |
| 832 | struct _reg_trie_state { |
| 833 | U16 wordnum; |
| 834 | union { |
| 835 | U32 base; |
| 836 | reg_trie_trans_le* list; |
| 837 | } trans; |
| 838 | }; |
| 839 | |
| 840 | /* info per word; indexed by wordnum */ |
| 841 | typedef struct { |
| 842 | U16 prev; /* previous word in acceptance chain; eg in |
| 843 | * zzz|abc|ab/ after matching the chars abc, the |
| 844 | * accepted word is #2, and the previous accepted |
| 845 | * word is #3 */ |
| 846 | U32 len; /* how many chars long is this word? */ |
| 847 | U32 accept; /* accept state for this word */ |
| 848 | } reg_trie_wordinfo; |
| 849 | |
| 850 | |
| 851 | typedef struct _reg_trie_state reg_trie_state; |
| 852 | typedef struct _reg_trie_trans reg_trie_trans; |
| 853 | |
| 854 | |
| 855 | /* anything in here that needs to be freed later |
| 856 | should be dealt with in pregfree. |
| 857 | refcount is first in both this and _reg_ac_data to allow a space |
| 858 | optimisation in Perl_regdupe. */ |
| 859 | struct _reg_trie_data { |
| 860 | U32 refcount; /* number of times this trie is referenced */ |
| 861 | U32 lasttrans; /* last valid transition element */ |
| 862 | U16 *charmap; /* byte to charid lookup array */ |
| 863 | reg_trie_state *states; /* state data */ |
| 864 | reg_trie_trans *trans; /* array of transition elements */ |
| 865 | char *bitmap; /* stclass bitmap */ |
| 866 | U16 *jump; /* optional 1 indexed array of offsets before tail |
| 867 | for the node following a given word. */ |
| 868 | reg_trie_wordinfo *wordinfo; /* array of info per word */ |
| 869 | U16 uniquecharcount; /* unique chars in trie (width of trans table) */ |
| 870 | U32 startstate; /* initial state - used for common prefix optimisation */ |
| 871 | STRLEN minlen; /* minimum length of words in trie - build/opt only? */ |
| 872 | STRLEN maxlen; /* maximum length of words in trie - build/opt only? */ |
| 873 | U32 prefixlen; /* #chars in common prefix */ |
| 874 | U32 statecount; /* Build only - number of states in the states array |
| 875 | (including the unused zero state) */ |
| 876 | U32 wordcount; /* Build only */ |
| 877 | #ifdef DEBUGGING |
| 878 | STRLEN charcount; /* Build only */ |
| 879 | #endif |
| 880 | }; |
| 881 | /* There is one (3 under DEBUGGING) pointers that logically belong in this |
| 882 | structure, but are held outside as they need duplication on thread cloning, |
| 883 | whereas the rest of the structure can be read only: |
| 884 | HV *widecharmap; code points > 255 to charid |
| 885 | #ifdef DEBUGGING |
| 886 | AV *words; Array of words contained in trie, for dumping |
| 887 | AV *revcharmap; Map of each charid back to its character representation |
| 888 | #endif |
| 889 | */ |
| 890 | |
| 891 | #define TRIE_WORDS_OFFSET 2 |
| 892 | |
| 893 | typedef struct _reg_trie_data reg_trie_data; |
| 894 | |
| 895 | /* refcount is first in both this and _reg_trie_data to allow a space |
| 896 | optimisation in Perl_regdupe. */ |
| 897 | struct _reg_ac_data { |
| 898 | U32 refcount; |
| 899 | U32 trie; |
| 900 | U32 *fail; |
| 901 | reg_trie_state *states; |
| 902 | }; |
| 903 | typedef struct _reg_ac_data reg_ac_data; |
| 904 | |
| 905 | /* ANY_BIT doesn't use the structure, so we can borrow it here. |
| 906 | This is simpler than refactoring all of it as wed end up with |
| 907 | three different sets... */ |
| 908 | |
| 909 | #define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap) |
| 910 | #define TRIE_BITMAP_BYTE(p, c) BITMAP_BYTE(TRIE_BITMAP(p), c) |
| 911 | #define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c)) |
| 912 | #define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c)) |
| 913 | #define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c)) |
| 914 | |
| 915 | #define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC) |
| 916 | #define IS_TRIE_AC(op) ((op)>=AHOCORASICK) |
| 917 | |
| 918 | |
| 919 | #define BITMAP_BYTE(p, c) (( (U8*) p) [ ( ( (UV) (c)) >> 3) ] ) |
| 920 | #define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c)) |
| 921 | |
| 922 | /* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */ |
| 923 | #define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1) |
| 924 | #define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state)) |
| 925 | #define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state)) |
| 926 | |
| 927 | #ifdef DEBUGGING |
| 928 | #define TRIE_CHARCOUNT(trie) ((trie)->charcount) |
| 929 | #else |
| 930 | #define TRIE_CHARCOUNT(trie) (trie_charcount) |
| 931 | #endif |
| 932 | |
| 933 | #define RE_TRIE_MAXBUF_INIT 65536 |
| 934 | #define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF" |
| 935 | #define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS" |
| 936 | |
| 937 | /* |
| 938 | |
| 939 | RE_DEBUG_FLAGS is used to control what debug output is emitted |
| 940 | its divided into three groups of options, some of which interact. |
| 941 | The three groups are: Compile, Execute, Extra. There is room for a |
| 942 | further group, as currently only the low three bytes are used. |
| 943 | |
| 944 | Compile Options: |
| 945 | |
| 946 | PARSE |
| 947 | PEEP |
| 948 | TRIE |
| 949 | PROGRAM |
| 950 | OFFSETS |
| 951 | |
| 952 | Execute Options: |
| 953 | |
| 954 | INTUIT |
| 955 | MATCH |
| 956 | TRIE |
| 957 | |
| 958 | Extra Options |
| 959 | |
| 960 | TRIE |
| 961 | OFFSETS |
| 962 | |
| 963 | If you modify any of these make sure you make corresponding changes to |
| 964 | re.pm, especially to the documentation. |
| 965 | |
| 966 | */ |
| 967 | |
| 968 | |
| 969 | /* Compile */ |
| 970 | #define RE_DEBUG_COMPILE_MASK 0x0000FF |
| 971 | #define RE_DEBUG_COMPILE_PARSE 0x000001 |
| 972 | #define RE_DEBUG_COMPILE_OPTIMISE 0x000002 |
| 973 | #define RE_DEBUG_COMPILE_TRIE 0x000004 |
| 974 | #define RE_DEBUG_COMPILE_DUMP 0x000008 |
| 975 | #define RE_DEBUG_COMPILE_FLAGS 0x000010 |
| 976 | #define RE_DEBUG_COMPILE_TEST 0x000020 |
| 977 | |
| 978 | /* Execute */ |
| 979 | #define RE_DEBUG_EXECUTE_MASK 0x00FF00 |
| 980 | #define RE_DEBUG_EXECUTE_INTUIT 0x000100 |
| 981 | #define RE_DEBUG_EXECUTE_MATCH 0x000200 |
| 982 | #define RE_DEBUG_EXECUTE_TRIE 0x000400 |
| 983 | |
| 984 | /* Extra */ |
| 985 | #define RE_DEBUG_EXTRA_MASK 0xFF0000 |
| 986 | #define RE_DEBUG_EXTRA_TRIE 0x010000 |
| 987 | #define RE_DEBUG_EXTRA_OFFSETS 0x020000 |
| 988 | #define RE_DEBUG_EXTRA_OFFDEBUG 0x040000 |
| 989 | #define RE_DEBUG_EXTRA_STATE 0x080000 |
| 990 | #define RE_DEBUG_EXTRA_OPTIMISE 0x100000 |
| 991 | #define RE_DEBUG_EXTRA_BUFFERS 0x400000 |
| 992 | #define RE_DEBUG_EXTRA_GPOS 0x800000 |
| 993 | /* combined */ |
| 994 | #define RE_DEBUG_EXTRA_STACK 0x280000 |
| 995 | |
| 996 | #define RE_DEBUG_FLAG(x) (re_debug_flags & x) |
| 997 | /* Compile */ |
| 998 | #define DEBUG_COMPILE_r(x) DEBUG_r( \ |
| 999 | if (re_debug_flags & RE_DEBUG_COMPILE_MASK) x ) |
| 1000 | #define DEBUG_PARSE_r(x) DEBUG_r( \ |
| 1001 | if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x ) |
| 1002 | #define DEBUG_OPTIMISE_r(x) DEBUG_r( \ |
| 1003 | if (re_debug_flags & RE_DEBUG_COMPILE_OPTIMISE) x ) |
| 1004 | #define DEBUG_PARSE_r(x) DEBUG_r( \ |
| 1005 | if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x ) |
| 1006 | #define DEBUG_DUMP_r(x) DEBUG_r( \ |
| 1007 | if (re_debug_flags & RE_DEBUG_COMPILE_DUMP) x ) |
| 1008 | #define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \ |
| 1009 | if (re_debug_flags & RE_DEBUG_COMPILE_TRIE) x ) |
| 1010 | #define DEBUG_FLAGS_r(x) DEBUG_r( \ |
| 1011 | if (re_debug_flags & RE_DEBUG_COMPILE_FLAGS) x ) |
| 1012 | #define DEBUG_TEST_r(x) DEBUG_r( \ |
| 1013 | if (re_debug_flags & RE_DEBUG_COMPILE_TEST) x ) |
| 1014 | /* Execute */ |
| 1015 | #define DEBUG_EXECUTE_r(x) DEBUG_r( \ |
| 1016 | if (re_debug_flags & RE_DEBUG_EXECUTE_MASK) x ) |
| 1017 | #define DEBUG_INTUIT_r(x) DEBUG_r( \ |
| 1018 | if (re_debug_flags & RE_DEBUG_EXECUTE_INTUIT) x ) |
| 1019 | #define DEBUG_MATCH_r(x) DEBUG_r( \ |
| 1020 | if (re_debug_flags & RE_DEBUG_EXECUTE_MATCH) x ) |
| 1021 | #define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \ |
| 1022 | if (re_debug_flags & RE_DEBUG_EXECUTE_TRIE) x ) |
| 1023 | |
| 1024 | /* Extra */ |
| 1025 | #define DEBUG_EXTRA_r(x) DEBUG_r( \ |
| 1026 | if (re_debug_flags & RE_DEBUG_EXTRA_MASK) x ) |
| 1027 | #define DEBUG_OFFSETS_r(x) DEBUG_r( \ |
| 1028 | if (re_debug_flags & RE_DEBUG_EXTRA_OFFSETS) x ) |
| 1029 | #define DEBUG_STATE_r(x) DEBUG_r( \ |
| 1030 | if (re_debug_flags & RE_DEBUG_EXTRA_STATE) x ) |
| 1031 | #define DEBUG_STACK_r(x) DEBUG_r( \ |
| 1032 | if (re_debug_flags & RE_DEBUG_EXTRA_STACK) x ) |
| 1033 | #define DEBUG_BUFFERS_r(x) DEBUG_r( \ |
| 1034 | if (re_debug_flags & RE_DEBUG_EXTRA_BUFFERS) x ) |
| 1035 | |
| 1036 | #define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \ |
| 1037 | if ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \ |
| 1038 | (re_debug_flags & (RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE)) ) x ) |
| 1039 | #define MJD_OFFSET_DEBUG(x) DEBUG_r( \ |
| 1040 | if (re_debug_flags & RE_DEBUG_EXTRA_OFFDEBUG) \ |
| 1041 | Perl_warn_nocontext x ) |
| 1042 | #define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \ |
| 1043 | if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x ) |
| 1044 | #define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \ |
| 1045 | if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x ) |
| 1046 | |
| 1047 | #define DEBUG_TRIE_r(x) DEBUG_r( \ |
| 1048 | if (re_debug_flags & (RE_DEBUG_COMPILE_TRIE \ |
| 1049 | | RE_DEBUG_EXECUTE_TRIE )) x ) |
| 1050 | #define DEBUG_GPOS_r(x) DEBUG_r( \ |
| 1051 | if (re_debug_flags & RE_DEBUG_EXTRA_GPOS) x ) |
| 1052 | |
| 1053 | /* initialization */ |
| 1054 | /* get_sv() can return NULL during global destruction. */ |
| 1055 | #define GET_RE_DEBUG_FLAGS DEBUG_r({ \ |
| 1056 | SV * re_debug_flags_sv = NULL; \ |
| 1057 | re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, 1) : NULL; \ |
| 1058 | if (re_debug_flags_sv) { \ |
| 1059 | if (!SvIOK(re_debug_flags_sv)) \ |
| 1060 | sv_setuv(re_debug_flags_sv, RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \ |
| 1061 | re_debug_flags=SvIV(re_debug_flags_sv); \ |
| 1062 | }\ |
| 1063 | }) |
| 1064 | |
| 1065 | #ifdef DEBUGGING |
| 1066 | |
| 1067 | #define GET_RE_DEBUG_FLAGS_DECL VOL IV re_debug_flags = 0; \ |
| 1068 | PERL_UNUSED_VAR(re_debug_flags); GET_RE_DEBUG_FLAGS; |
| 1069 | |
| 1070 | #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \ |
| 1071 | const char * const rpv = \ |
| 1072 | pv_pretty((dsv), (pv), (l), \ |
| 1073 | (PL_dump_re_max_len) ? PL_dump_re_max_len : (m), \ |
| 1074 | PL_colors[(c1)],PL_colors[(c2)], \ |
| 1075 | PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \ |
| 1076 | const int rlen = SvCUR(dsv) |
| 1077 | |
| 1078 | #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \ |
| 1079 | const char * const rpv = \ |
| 1080 | pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), \ |
| 1081 | (PL_dump_re_max_len) ? PL_dump_re_max_len : (m), \ |
| 1082 | PL_colors[(c1)],PL_colors[(c2)], \ |
| 1083 | PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ) |
| 1084 | |
| 1085 | #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \ |
| 1086 | const char * const rpv = \ |
| 1087 | pv_pretty((dsv), (pv), (l), \ |
| 1088 | (PL_dump_re_max_len) ? PL_dump_re_max_len : (m), \ |
| 1089 | PL_colors[0], PL_colors[1], \ |
| 1090 | ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \ |
| 1091 | ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \ |
| 1092 | ) |
| 1093 | |
| 1094 | #define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0)) |
| 1095 | #define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "") |
| 1096 | |
| 1097 | #else /* if not DEBUGGING */ |
| 1098 | |
| 1099 | #define GET_RE_DEBUG_FLAGS_DECL |
| 1100 | #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) |
| 1101 | #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) |
| 1102 | #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) |
| 1103 | #define RE_SV_DUMPLEN(ItEm) |
| 1104 | #define RE_SV_TAIL(ItEm) |
| 1105 | |
| 1106 | #endif /* DEBUG RELATED DEFINES */ |
| 1107 | |
| 1108 | typedef enum { |
| 1109 | TRADITIONAL_BOUND = _CC_WORDCHAR, |
| 1110 | GCB_BOUND, |
| 1111 | LB_BOUND, |
| 1112 | SB_BOUND, |
| 1113 | WB_BOUND |
| 1114 | } bound_type; |
| 1115 | |
| 1116 | /* |
| 1117 | * ex: set ts=8 sts=4 sw=4 et: |
| 1118 | */ |