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 11 #if ! defined(PERL_REGCOMP_H_) && ( defined(PERL_CORE) \ 12 || defined(PERL_EXT_RE_BUILD)) 13 #define PERL_REGCOMP_H_ 14 15 #include "regcharclass.h" 16 17 /* Convert branch sequences to more efficient trie ops? */ 18 #define PERL_ENABLE_TRIE_OPTIMISATION 1 19 20 /* Be really aggressive about optimising patterns with trie sequences? */ 21 #define PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION 1 22 23 /* Should the optimiser take positive assertions into account? */ 24 #define PERL_ENABLE_POSITIVE_ASSERTION_STUDY 0 25 26 /* Not for production use: */ 27 #define PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS 0 28 29 /* 30 * Structure for regexp "program". This is essentially a linear encoding 31 * of a nondeterministic finite-state machine (aka syntax charts or 32 * "railroad normal form" in parsing technology). Each node is an opcode 33 * plus a "next" pointer, possibly plus an operand. "Next" pointers of 34 * all nodes except BRANCH implement concatenation; a "next" pointer with 35 * a BRANCH on both ends of it is connecting two alternatives. (Here we 36 * have one of the subtle syntax dependencies: an individual BRANCH (as 37 * opposed to a collection of them) is never concatenated with anything 38 * because of operator precedence.) The operand of some types of node is 39 * a literal string; for others, it is a node leading into a sub-FSM. In 40 * particular, the operand of a BRANCH node is the first node of the branch. 41 * (NB this is *not* a tree structure: the tail of the branch connects 42 * to the thing following the set of BRANCHes.) The opcodes are defined 43 * in regnodes.h which is generated from regcomp.sym by regcomp.pl. 44 */ 45 46 /* 47 * A node is one char of opcode followed by two chars of "next" pointer. 48 * "Next" pointers are stored as two 8-bit pieces, high order first. The 49 * value is a positive offset from the opcode of the node containing it. 50 * An operand, if any, simply follows the node. (Note that much of the 51 * code generation knows about this implicit relationship.) 52 * 53 * Using two bytes for the "next" pointer is vast overkill for most things, 54 * but allows patterns to get big without disasters. 55 * 56 * [The "next" pointer is always aligned on an even 57 * boundary, and reads the offset directly as a short.] 58 */ 59 60 /* This is the stuff that used to live in regexp.h that was truly 61 private to the engine itself. It now lives here. */ 62 63 typedef struct regexp_internal { 64 regnode *regstclass; /* Optional startclass as identified or constructed 65 by the optimiser */ 66 struct reg_data *data; /* Additional miscellaneous data used by the program. 67 Used to make it easier to clone and free arbitrary 68 data that the regops need. Often the ARG field of 69 a regop is an index into this structure. NOTE the 70 0th element of this structure is NEVER used and is 71 strictly reserved for internal purposes. */ 72 struct reg_code_blocks *code_blocks;/* positions of literal (?{}) */ 73 U32 proglen; /* size of the compiled program in regnodes */ 74 U32 name_list_idx; /* Optional data index of an array of paren names, 75 only valid when RXp_PAREN_NAMES(prog) is true, 76 0 means "no value" like any other index into the 77 data array.*/ 78 regnode program[1]; /* Unwarranted chumminess with compiler. */ 79 } regexp_internal; 80 81 #define RXi_SET(x,y) (x)->pprivate = (void*)(y) 82 #define RXi_GET(x) ((regexp_internal *)((x)->pprivate)) 83 #define RXi_GET_DECL(r,ri) regexp_internal *ri = RXi_GET(r) 84 /* 85 * Flags stored in regexp->intflags 86 * These are used only internally to the regexp engine 87 * 88 * See regexp.h for flags used externally to the regexp engine 89 */ 90 #define RXp_INTFLAGS(rx) ((rx)->intflags) 91 #define RX_INTFLAGS(prog) RXp_INTFLAGS(ReANY(prog)) 92 93 #define PREGf_SKIP 0x00000001 94 #define PREGf_IMPLICIT 0x00000002 /* Converted .* to ^.* */ 95 #define PREGf_NAUGHTY 0x00000004 /* how exponential is this pattern? */ 96 #define PREGf_VERBARG_SEEN 0x00000008 97 #define PREGf_CUTGROUP_SEEN 0x00000010 98 #define PREGf_USE_RE_EVAL 0x00000020 /* compiled with "use re 'eval'" */ 99 /* these used to be extflags, but are now intflags */ 100 #define PREGf_NOSCAN 0x00000040 101 /* spare */ 102 #define PREGf_GPOS_SEEN 0x00000100 103 #define PREGf_GPOS_FLOAT 0x00000200 104 105 #define PREGf_ANCH_MBOL 0x00000400 106 #define PREGf_ANCH_SBOL 0x00000800 107 #define PREGf_ANCH_GPOS 0x00001000 108 #define PREGf_RECURSE_SEEN 0x00002000 109 110 #define PREGf_ANCH \ 111 ( PREGf_ANCH_SBOL | PREGf_ANCH_GPOS | PREGf_ANCH_MBOL ) 112 113 /* this is where the old regcomp.h started */ 114 115 116 /* Define the various regnode structures. These all should be a multiple 117 * of 32 bits large, and they should by and large correspond with each other 118 * in terms of naming, etc. Things can and will break in subtle ways if you 119 * change things without care. If you look at regexp.h you will see it 120 * contains this: 121 * 122 * struct regnode { 123 * U8 flags; 124 * U8 type; 125 * U16 next_off; 126 * }; 127 * 128 * This structure is the base unit of elements in the regexp program. When 129 * we increment our way through the program we increment by the size of this 130 * structure, and in all cases where regnode sizing is considered it is in 131 * units of this structure. 132 * 133 * This implies that no regnode style structure should contain 64 bit 134 * aligned members. Since the base regnode is 32 bits any member might 135 * not be 64 bit aligned no matter how you might try to pad out the 136 * struct itself (the regnode_ssc is special in this regard as it is 137 * never used in a program directly). If you want to store 64 bit 138 * members you need to store them specially. The struct regnode_p and the 139 * ARGp() and ARGp_SET() macros and related inline functions provide an example 140 * solution. Note they deal with a slightly more complicated problem than simple 141 * alignment, as pointers may be 32 bits or 64 bits depending on platform, 142 * but they illustrate the pattern to follow if you want to put a 64 bit value 143 * into a regnode. 144 145 * NOTE: Ideally we do not put pointers into the regnodes in a program. Instead 146 * we put them in the "data" part of the regexp structure and store the index into 147 * the data in the pointers in the regnode. This allows the pointer to be handled 148 * properly during clone/free operations (eg refcount bookkeeping). See S_add_data(), 149 * Perl_regdupe_internal(), Perl_regfree_internal() in regcomp.c for how the data 150 * array can be used, the letters 'arsSu' all refer to different types of SV that 151 * we already have support for in the data array. 152 */ 153 154 struct regnode_string { 155 U8 str_len; 156 U8 type; 157 U16 next_off; 158 char string[1]; 159 }; 160 161 struct regnode_lstring { /* Constructed this way to keep the string aligned. */ 162 U8 flags; 163 U8 type; 164 U16 next_off; 165 U32 str_len; /* Only 18 bits allowed before would overflow 'next_off' */ 166 char string[1]; 167 }; 168 169 struct regnode_anyofhs { /* Constructed this way to keep the string aligned. */ 170 U8 str_len; 171 U8 type; 172 U16 next_off; 173 U32 arg1; /* set by set_ANYOF_arg() */ 174 char string[1]; 175 }; 176 177 /* Argument bearing node - workhorse, 178 arg1 is often for the data field */ 179 struct regnode_1 { 180 U8 flags; 181 U8 type; 182 U16 next_off; 183 U32 arg1; 184 }; 185 186 /* Node whose argument is 'SV *'. This needs to be used very carefully in 187 * situations where pointers won't become invalid because of, say re-mallocs. 188 * 189 * Note that this regnode type is problematic and should not be used or copied 190 * and will be removed in the future. Pointers should be stored in the data[] 191 * array and an index into the data array stored in the regnode, which allows the 192 * pointers to be handled properly during clone/free operations on the regexp 193 * data structure. As a byproduct it also saves space, often we use a 16 bit 194 * member to store indexes into the data[] array. 195 * 196 * Also note that the weird storage here is because regnodes are 32 bit aligned, 197 * which means we cannot have a 64 bit aligned member. To make things more annoying 198 * the size of a pointer may vary by platform. Thus we use a character array, and 199 * then use inline functions to copy the data in or out. 200 * */ 201 struct regnode_p { 202 U8 flags; 203 U8 type; 204 U16 next_off; 205 char arg1_sv_ptr_bytes[sizeof(SV *)]; 206 }; 207 208 /* Similar to a regnode_1 but with an extra signed argument */ 209 struct regnode_2L { 210 U8 flags; 211 U8 type; 212 U16 next_off; 213 U32 arg1; 214 I32 arg2; 215 }; 216 217 /* 'Two field' -- Two 16 bit unsigned args */ 218 struct regnode_2 { 219 U8 flags; 220 U8 type; 221 U16 next_off; 222 U16 arg1; 223 U16 arg2; 224 }; 225 226 #define ANYOF_BITMAP_SIZE (NUM_ANYOF_CODE_POINTS / 8) /* 8 bits/Byte */ 227 228 /* Note that these form structs which are supersets of the next smaller one, by 229 * appending fields. Alignment problems can occur if one of those optional 230 * fields requires stricter alignment than the base struct. And formal 231 * parameters that can really be two or more of the structs should be 232 * declared as the smallest one it could be. See commit message for 233 * 7dcac5f6a5195002b55c935ee1d67f67e1df280b. Regnode allocation is done 234 * without regard to alignment, and changing it to would also require changing 235 * the code that inserts and deletes regnodes. The basic single-argument 236 * regnode has a U32, which is what reganode() allocates as a unit. Therefore 237 * no field can require stricter alignment than U32. */ 238 239 /* also used by trie */ 240 struct regnode_charclass { 241 U8 flags; 242 U8 type; 243 U16 next_off; 244 U32 arg1; /* set by set_ANYOF_arg() */ 245 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */ 246 }; 247 248 /* has runtime (locale) \d, \w, ..., [:posix:] classes */ 249 struct regnode_charclass_posixl { 250 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */ 251 U8 type; 252 U16 next_off; 253 U32 arg1; 254 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */ 255 U32 classflags; /* and run-time */ 256 }; 257 258 /* A synthetic start class (SSC); is a regnode_charclass_posixl_fold, plus an 259 * extra SV*, used only during regex construction and which is not used by the 260 * main machinery in regexec.c and which does not get embedded in the final compiled 261 * regex program. 262 * 263 * Because it does not get embedded it does not have to comply with the alignment 264 * and sizing constraints required for a normal regnode structure: it MAY contain 265 * pointers or members of whatever size needed and the compiler will do the right 266 * thing. (Every other regnode type is 32 bit aligned.) 267 * 268 * Note that the 'next_off' field is unused, as the SSC stands alone, so there is 269 * never a next node. 270 */ 271 struct regnode_ssc { 272 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */ 273 U8 type; 274 U16 next_off; 275 U32 arg1; 276 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */ 277 U32 classflags; /* ... and run-time */ 278 279 /* Auxiliary, only used during construction; NULL afterwards: list of code 280 * points matched */ 281 SV* invlist; 282 }; 283 284 /* We take advantage of 'next_off' not otherwise being used in the SSC by 285 * actually using it: by setting it to 1. This allows us to test and 286 * distinguish between an SSC and other ANYOF node types, as 'next_off' cannot 287 * otherwise be 1, because it is the offset to the next regnode expressed in 288 * units of regnodes. Since an ANYOF node contains extra fields, it adds up 289 * to 12 regnode units on 32-bit systems, (hence the minimum this can be (if 290 * not 0) is 11 there. Even if things get tightly packed on a 64-bit system, 291 * it still would be more than 1. */ 292 #define set_ANYOF_SYNTHETIC(n) STMT_START{ OP(n) = ANYOF; \ 293 NEXT_OFF(n) = 1; \ 294 } STMT_END 295 #define is_ANYOF_SYNTHETIC(n) (PL_regkind[OP(n)] == ANYOF && NEXT_OFF(n) == 1) 296 297 /* XXX fix this description. 298 Impose a limit of REG_INFTY on various pattern matching operations 299 to limit stack growth and to avoid "infinite" recursions. 300 */ 301 /* The default size for REG_INFTY is U16_MAX, which is the same as 302 USHORT_MAX (see perl.h). Unfortunately U16 isn't necessarily 16 bits 303 (see handy.h). On the Cray C90, sizeof(short)==4 and hence U16_MAX is 304 ((1<<32)-1), while on the Cray T90, sizeof(short)==8 and U16_MAX is 305 ((1<<64)-1). To limit stack growth to reasonable sizes, supply a 306 smaller default. 307 --Andy Dougherty 11 June 1998 308 */ 309 #if SHORTSIZE > 2 310 # ifndef REG_INFTY 311 # define REG_INFTY nBIT_UMAX(16) 312 # endif 313 #endif 314 315 #ifndef REG_INFTY 316 # define REG_INFTY U16_MAX 317 #endif 318 319 #define ARG_VALUE(arg) (arg) 320 #define ARG__SET(arg,val) ((arg) = (val)) 321 322 #undef ARG 323 #undef ARG1 324 #undef ARG2 325 326 #define ARG(p) ARG_VALUE(ARG_LOC(p)) 327 #define ARGp(p) ARGp_VALUE_inline(p) 328 #define ARG1(p) ARG_VALUE(ARG1_LOC(p)) 329 #define ARG2(p) ARG_VALUE(ARG2_LOC(p)) 330 #define ARG2L(p) ARG_VALUE(ARG2L_LOC(p)) 331 332 #define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val)) 333 #define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val)) 334 #define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val)) 335 #define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val)) 336 #define ARGp_SET(p, val) ARGp_SET_inline((p),(val)) 337 338 #undef NEXT_OFF 339 #undef NODE_ALIGN 340 341 #define NEXT_OFF(p) ((p)->next_off) 342 #define NODE_ALIGN(node) 343 /* the following define was set to 0xde in 075abff3 344 * as part of some linting logic. I have set it to 0 345 * as otherwise in every place where we /might/ set flags 346 * we have to set it 0 explicitly, which duplicates 347 * assignments and IMO adds an unacceptable level of 348 * surprise to working in the regex engine. If this 349 * is changed from 0 then at the very least make sure 350 * that SBOL for /^/ sets the flags to 0 explicitly. 351 * -- Yves */ 352 #define NODE_ALIGN_FILL(node) ((node)->flags = 0) 353 354 #define SIZE_ALIGN NODE_ALIGN 355 356 #undef OP 357 #undef OPERAND 358 #undef STRING 359 360 #define OP(p) ((p)->type) 361 #define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \ 362 regnode types. For some, it's the \ 363 character set of the regnode */ 364 #define STR_LENs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \ 365 ((struct regnode_string *)p)->str_len) 366 #define STRINGs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \ 367 ((struct regnode_string *)p)->string) 368 #define OPERANDs(p) STRINGs(p) 369 370 /* Long strings. Currently limited to length 18 bits, which handles a 262000 371 * byte string. The limiting factor is the 16 bit 'next_off' field, which 372 * points to the next regnode, so the furthest away it can be is 2**16. On 373 * most architectures, regnodes are 2**2 bytes long, so that yields 2**18 374 * bytes. Should a longer string be desired, we could increase it to 26 bits 375 * fairly easily, by changing this node to have longj type which causes the ARG 376 * field to be used for the link to the next regnode (although code would have 377 * to be changed to account for this), and then use a combination of the flags 378 * and next_off fields for the length. To get 34 bit length, also change the 379 * node to be an ARG2L, using the second 32 bit field for the length, and not 380 * using the flags nor next_off fields at all. One could have an llstring node 381 * and even an lllstring type. */ 382 #define STR_LENl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 383 (((struct regnode_lstring *)p)->str_len)) 384 #define STRINGl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 385 (((struct regnode_lstring *)p)->string)) 386 #define OPERANDl(p) STRINGl(p) 387 388 #define STR_LEN(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 389 ? STR_LENl(p) : STR_LENs(p)) 390 #define STRING(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 391 ? STRINGl(p) : STRINGs(p)) 392 #define OPERAND(p) STRING(p) 393 394 /* The number of (smallest) regnode equivalents that a string of length l bytes 395 * occupies */ 396 #define STR_SZ(l) (((l) + sizeof(regnode) - 1) / sizeof(regnode)) 397 398 /* The number of (smallest) regnode equivalents that the EXACTISH node 'p' 399 * occupies */ 400 #define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p)) + 1 + regarglen[(p)->type]) 401 402 #define setSTR_LEN(p,v) \ 403 STMT_START{ \ 404 if (OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 405 ((struct regnode_lstring *)(p))->str_len = (v); \ 406 else \ 407 ((struct regnode_string *)(p))->str_len = (v); \ 408 } STMT_END 409 410 #define ANYOFR_BASE_BITS 20 411 #define ANYOFRbase(p) (ARG(p) & nBIT_MASK(ANYOFR_BASE_BITS)) 412 #define ANYOFRdelta(p) (ARG(p) >> ANYOFR_BASE_BITS) 413 414 #undef NODE_ALIGN 415 #undef ARG_LOC 416 #undef NEXTOPER 417 #undef PREVOPER 418 419 #define NODE_ALIGN(node) 420 #define ARG_LOC(p) (((struct regnode_1 *)p)->arg1) 421 #define ARGp_BYTES_LOC(p) (((struct regnode_p *)p)->arg1_sv_ptr_bytes) 422 #define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1) 423 #define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2) 424 #define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2) 425 426 #define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */ 427 #define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2) 428 429 #define NEXTOPER(p) ((p) + NODE_STEP_REGNODE) 430 #define PREVOPER(p) ((p) - NODE_STEP_REGNODE) 431 432 #define FILL_NODE(offset, op) \ 433 STMT_START { \ 434 OP(REGNODE_p(offset)) = op; \ 435 NEXT_OFF(REGNODE_p(offset)) = 0; \ 436 } STMT_END 437 #define FILL_ADVANCE_NODE(offset, op) \ 438 STMT_START { \ 439 FILL_NODE(offset, op); \ 440 (offset)++; \ 441 } STMT_END 442 #define FILL_ADVANCE_NODE_ARG(offset, op, arg) \ 443 STMT_START { \ 444 ARG_SET(REGNODE_p(offset), arg); \ 445 FILL_ADVANCE_NODE(offset, op); \ 446 /* This is used generically for other operations \ 447 * that have a longer argument */ \ 448 (offset) += regarglen[op]; \ 449 } STMT_END 450 #define FILL_ADVANCE_NODE_ARGp(offset, op, arg) \ 451 STMT_START { \ 452 ARGp_SET(REGNODE_p(offset), arg); \ 453 FILL_ADVANCE_NODE(offset, op); \ 454 (offset) += regarglen[op]; \ 455 } STMT_END 456 #define FILL_ADVANCE_NODE_2L_ARG(offset, op, arg1, arg2) \ 457 STMT_START { \ 458 ARG_SET(REGNODE_p(offset), arg1); \ 459 ARG2L_SET(REGNODE_p(offset), arg2); \ 460 FILL_ADVANCE_NODE(offset, op); \ 461 (offset) += 2; \ 462 } STMT_END 463 464 /* define these after we define the normal macros, so we can use 465 * ARGp_BYTES_LOC(n) */ 466 467 static inline SV * 468 ARGp_VALUE_inline(struct regnode *node) { 469 SV *ptr; 470 memcpy(&ptr, ARGp_BYTES_LOC(node), sizeof(ptr)); 471 472 return ptr; 473 } 474 475 static inline void 476 ARGp_SET_inline(struct regnode *node, SV *ptr) { 477 memcpy(ARGp_BYTES_LOC(node), &ptr, sizeof(ptr)); 478 } 479 480 #define REG_MAGIC 0234 481 482 /* An ANYOF node is basically a bitmap with the index being a code point. If 483 * the bit for that code point is 1, the code point matches; if 0, it doesn't 484 * match (complemented if inverted). There is an additional mechanism to deal 485 * with cases where the bitmap is insufficient in and of itself. This #define 486 * indicates if the bitmap does fully represent what this ANYOF node can match. 487 * The ARG is set to this special value (since 0, 1, ... are legal, but will 488 * never reach this high). */ 489 #define ANYOF_ONLY_HAS_BITMAP ((U32) -1) 490 491 /* When the bitmap isn't completely sufficient for handling the ANYOF node, 492 * flags (in node->flags of the ANYOF node) get set to indicate this. These 493 * are perennially in short supply. Beyond several cases where warnings need 494 * to be raised under certain circumstances, currently, there are six cases 495 * where the bitmap alone isn't sufficient. We could use six flags to 496 * represent the 6 cases, but to save flags bits, we play some games. The 497 * cases are: 498 * 499 * 1) The bitmap has a compiled-in very finite size. So something else needs 500 * to be used to specify if a code point that is too large for the bitmap 501 * actually matches. The mechanism currently is an inversion 502 * list. ANYOF_ONLY_HAS_BITMAP, described above, being TRUE indicates 503 * there are no matches of too-large code points. But if it is FALSE, 504 * then almost certainly there are matches too large for the bitmap. (The 505 * other cases, described below, either imply this one or are extremely 506 * rare in practice.) So we can just assume that a too-large code point 507 * will need something beyond the bitmap if ANYOF_ONLY_HAS_BITMAP is 508 * FALSE, instead of having a separate flag for this. 509 * 2) A subset of item 1) is if all possible code points outside the bitmap 510 * match. This is a common occurrence when the class is complemented, 511 * like /[^ij]/. Therefore a bit is reserved to indicate this, 512 * rather than having an inversion list created, 513 * ANYOF_MATCHES_ALL_ABOVE_BITMAP. 514 * 3) Under /d rules, it can happen that code points that are in the upper 515 * latin1 range (\x80-\xFF or their equivalents on EBCDIC platforms) match 516 * only if the runtime target string being matched against is UTF-8. For 517 * example /[\w[:punct:]]/d. This happens only for posix classes (with a 518 * couple of exceptions, like \d where it doesn't happen), and all such 519 * ones also have above-bitmap matches. Thus, 3) implies 1) as well. 520 * Note that /d rules are no longer encouraged; 'use 5.14' or higher 521 * deselects them. But a flag is required so that they can be properly 522 * handled. But it can be a shared flag: see 5) below. 523 * 4) Also under /d rules, something like /[\Wfoo]/ will match everything in 524 * the \x80-\xFF range, unless the string being matched against is UTF-8. 525 * An inversion list could be created for this case, but this is 526 * relatively common, and it turns out that it's all or nothing: if any 527 * one of these code points matches, they all do. Hence a single bit 528 * suffices. We use a shared flag that doesn't take up space by itself: 529 * ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER. This 530 * also implies 1), with one exception: [:^cntrl:]. 531 * 5) A user-defined \p{} property may not have been defined by the time the 532 * regex is compiled. In this case, we don't know until runtime what it 533 * will match, so we have to assume it could match anything, including 534 * code points that ordinarily would be in the bitmap. A flag bit is 535 * necessary to indicate this, though it can be shared with the item 3) 536 * flag, as that only occurs under /d, and this only occurs under non-d. 537 * This case is quite uncommon in the field, and the /(?[ ...])/ construct 538 * is a better way to accomplish what this feature does. This case also 539 * implies 1). 540 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 541 * is the shared flag. 542 * 6) /[foo]/il may have folds that are only valid if the runtime locale is a 543 * UTF-8 one. These are quite rare, so it would be good to avoid the 544 * expense of looking for them. But /l matching is slow anyway, and we've 545 * traditionally not worried too much about its performance. And this 546 * condition requires the ANYOFL_FOLD flag to be set, so testing for 547 * that flag would be sufficient to rule out most cases of this. So it is 548 * unclear if this should have a flag or not. But, this flag can be 549 * shared with another, so it doesn't occupy extra space. 550 * 551 * At the moment, there is one spare bit, but this could be increased by 552 * various tricks: 553 * 554 * If just one more bit is needed, as of this writing it seems to khw that the 555 * best choice would be to make ANYOF_MATCHES_ALL_ABOVE_BITMAP not a flag, but 556 * something like 557 * 558 * #define ANYOF_MATCHES_ALL_ABOVE_BITMAP ((U32) -2) 559 * 560 * and access it through the ARG like ANYOF_ONLY_HAS_BITMAP is. This flag is 561 * used by all ANYOF node types, and it could be used to avoid calling the 562 * handler function, as the macro REGINCLASS in regexec.c does now for other 563 * cases. 564 * 565 * Another possibility is based on the fact that ANYOF_MATCHES_POSIXL is 566 * redundant with the node type ANYOFPOSIXL. That flag could be removed, but 567 * at the expense of extra code in regexec.c. The flag has been retained 568 * because it allows us to see if we need to call reginclass, or just use the 569 * bitmap in one test. 570 * 571 * If this is done, an extension would be to make all ANYOFL nodes contain the 572 * extra 32 bits that ANYOFPOSIXL ones do. The posix flags only occupy 30 573 * bits, so the ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD flags 574 * and ANYOFL_FOLD could be moved to that extra space, but it would mean extra 575 * instructions, as there are currently places in the code that assume those 576 * two bits are zero. 577 * 578 * All told, 5 bits could be available for other uses if all of the above were 579 * done. 580 * 581 * Some flags are not used in synthetic start class (SSC) nodes, so could be 582 * shared should new flags be needed for SSCs, like SSC_MATCHES_EMPTY_STRING 583 * now. */ 584 585 /* If this is set, the result of the match should be complemented. regexec.c 586 * is expecting this to be in the low bit. Never in an SSC */ 587 #define ANYOF_INVERT 0x01 588 589 /* For the SSC node only, which cannot be inverted, so is shared with that bit. 590 * This is used only during regex compilation. */ 591 #define SSC_MATCHES_EMPTY_STRING ANYOF_INVERT 592 593 /* Set if this is a regnode_charclass_posixl vs a regnode_charclass. This 594 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale 595 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved 596 * at compile-time. Only set under /l; can be in SSC */ 597 #define ANYOF_MATCHES_POSIXL 0x02 598 599 /* The fold is calculated and stored in the bitmap where possible at compile 600 * time. However under locale, the actual folding varies depending on 601 * what the locale is at the time of execution, so it has to be deferred until 602 * then. Only set under /l; never in an SSC */ 603 #define ANYOFL_FOLD 0x04 604 605 /* Shared bit set only with ANYOFL and SSC nodes: 606 * If ANYOFL_FOLD is set, this flag indicates there are potential matches 607 * valid only if the locale is a UTF-8 one. 608 * If ANYOFL_FOLD is NOT set, this flag means to warn if the runtime locale 609 * isn't a UTF-8 one (and the generated node assumes a UTF-8 locale). 610 * None of INVERT, POSIXL, 611 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 612 * can be set. */ 613 #define ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD 0x08 614 615 /* Convenience macros for teasing apart the meanings when reading the above bit 616 * */ 617 #define ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags) \ 618 ((flags & ( ANYOFL_FOLD /* Both bits are set */ \ 619 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \ 620 == ( ANYOFL_FOLD \ 621 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) 622 623 #define ANYOFL_UTF8_LOCALE_REQD(flags) \ 624 ((flags & ( ANYOFL_FOLD /* Only REQD bit is set */ \ 625 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \ 626 == ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD) 627 628 /* Spare: Be sure to change ANYOF_FLAGS_ALL if this gets used 0x10 */ 629 630 /* If set, the node matches every code point NUM_ANYOF_CODE_POINTS and above. 631 * Can be in an SSC */ 632 #define ANYOF_MATCHES_ALL_ABOVE_BITMAP 0x20 633 634 /* Shared bit: 635 * Under /d it means the ANYOFD node matches more things if the target 636 * string is encoded in UTF-8; any such things will be non-ASCII, 637 * characters that are < 256, and can be accessed via the inversion 638 * list. 639 * When not under /d, it means the ANYOF node contains a user-defined 640 * property that wasn't yet defined at the time the regex was compiled, 641 * and so must be looked up at runtime, by creating an inversion list. 642 * (These uses are mutually exclusive because a user-defined property is 643 * specified by \p{}, and \p{} implies /u which deselects /d). The long macro 644 * name is to make sure that you are cautioned about its shared nature. Only 645 * the non-/d meaning can be in an SSC */ 646 #define ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 0x40 647 648 /* Shared bit: 649 * Under /d it means the ANYOFD node matches all non-ASCII Latin1 650 * characters when the target string is not in utf8. 651 * When not under /d, it means the ANYOF node should raise a warning if 652 * matching against an above-Unicode code point. 653 * (These uses are mutually exclusive because the warning requires a \p{}, and 654 * \p{} implies /u which deselects /d). An SSC node only has this bit set if 655 * what is meant is the warning. The long macro name is to make sure that you 656 * are cautioned about its shared nature */ 657 #define ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER 0x80 658 659 #define ANYOF_FLAGS_ALL ((U8) ~0x10) 660 661 #define ANYOF_LOCALE_FLAGS (ANYOFL_FOLD | ANYOF_MATCHES_POSIXL) 662 663 /* These are the flags that apply to both regular ANYOF nodes and synthetic 664 * start class nodes during construction of the SSC. During finalization of 665 * the SSC, other of the flags may get added to it */ 666 #define ANYOF_COMMON_FLAGS 0 667 668 /* Character classes for node->classflags of ANYOF */ 669 /* Should be synchronized with a table in regprop() */ 670 /* 2n should be the normal one, paired with its complement at 2n+1 */ 671 672 #define ANYOF_ALPHA ((_CC_ALPHA) * 2) 673 #define ANYOF_NALPHA ((ANYOF_ALPHA) + 1) 674 #define ANYOF_ALPHANUMERIC ((_CC_ALPHANUMERIC) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */ 675 #define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1) 676 #define ANYOF_ASCII ((_CC_ASCII) * 2) 677 #define ANYOF_NASCII ((ANYOF_ASCII) + 1) 678 #define ANYOF_BLANK ((_CC_BLANK) * 2) /* GNU extension: space and tab: non-vertical space */ 679 #define ANYOF_NBLANK ((ANYOF_BLANK) + 1) 680 #define ANYOF_CASED ((_CC_CASED) * 2) /* Pseudo class for [:lower:] or 681 [:upper:] under /i */ 682 #define ANYOF_NCASED ((ANYOF_CASED) + 1) 683 #define ANYOF_CNTRL ((_CC_CNTRL) * 2) 684 #define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1) 685 #define ANYOF_DIGIT ((_CC_DIGIT) * 2) /* \d */ 686 #define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1) 687 #define ANYOF_GRAPH ((_CC_GRAPH) * 2) 688 #define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1) 689 #define ANYOF_LOWER ((_CC_LOWER) * 2) 690 #define ANYOF_NLOWER ((ANYOF_LOWER) + 1) 691 #define ANYOF_PRINT ((_CC_PRINT) * 2) 692 #define ANYOF_NPRINT ((ANYOF_PRINT) + 1) 693 #define ANYOF_PUNCT ((_CC_PUNCT) * 2) 694 #define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1) 695 #define ANYOF_SPACE ((_CC_SPACE) * 2) /* \s */ 696 #define ANYOF_NSPACE ((ANYOF_SPACE) + 1) 697 #define ANYOF_UPPER ((_CC_UPPER) * 2) 698 #define ANYOF_NUPPER ((ANYOF_UPPER) + 1) 699 #define ANYOF_WORDCHAR ((_CC_WORDCHAR) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */ 700 #define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1) 701 #define ANYOF_XDIGIT ((_CC_XDIGIT) * 2) 702 #define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1) 703 704 /* pseudo classes below this, not stored in the class bitmap, but used as flags 705 during compilation of char classes */ 706 707 #define ANYOF_VERTWS ((_CC_VERTSPACE) * 2) 708 #define ANYOF_NVERTWS ((ANYOF_VERTWS)+1) 709 710 /* It is best if this is the last one, as all above it are stored as bits in a 711 * bitmap, and it isn't part of that bitmap */ 712 #if _CC_VERTSPACE != _HIGHEST_REGCOMP_DOT_H_SYNC 713 # error Problem with handy.h _HIGHEST_REGCOMP_DOT_H_SYNC #define 714 #endif 715 716 #define ANYOF_POSIXL_MAX (ANYOF_VERTWS) /* So upper loop limit is written: 717 * '< ANYOF_MAX' 718 * Hence doesn't include VERTWS, as that 719 * is a pseudo class */ 720 #define ANYOF_MAX ANYOF_POSIXL_MAX 721 722 #if (ANYOF_POSIXL_MAX > 32) /* Must fit in 32-bit word */ 723 # error Problem with handy.h _CC_foo #defines 724 #endif 725 726 #define ANYOF_HORIZWS ((ANYOF_POSIXL_MAX)+2) /* = (ANYOF_NVERTWS + 1) */ 727 #define ANYOF_NHORIZWS ((ANYOF_POSIXL_MAX)+3) 728 729 #define ANYOF_UNIPROP ((ANYOF_POSIXL_MAX)+4) /* Used to indicate a Unicode 730 property: \p{} or \P{} */ 731 732 /* Backward source code compatibility. */ 733 734 #define ANYOF_ALNUML ANYOF_ALNUM 735 #define ANYOF_NALNUML ANYOF_NALNUM 736 #define ANYOF_SPACEL ANYOF_SPACE 737 #define ANYOF_NSPACEL ANYOF_NSPACE 738 #define ANYOF_ALNUM ANYOF_WORDCHAR 739 #define ANYOF_NALNUM ANYOF_NWORDCHAR 740 741 /* Utility macros for the bitmap and classes of ANYOF */ 742 743 #define ANYOF_FLAGS(p) ((p)->flags) 744 745 #define ANYOF_BIT(c) (1U << ((c) & 7)) 746 747 #define ANYOF_POSIXL_BITMAP(p) (((regnode_charclass_posixl*) (p))->classflags) 748 749 #define POSIXL_SET(field, c) ((field) |= (1U << (c))) 750 #define ANYOF_POSIXL_SET(p, c) POSIXL_SET(ANYOF_POSIXL_BITMAP(p), (c)) 751 752 #define POSIXL_CLEAR(field, c) ((field) &= ~ (1U <<(c))) 753 #define ANYOF_POSIXL_CLEAR(p, c) POSIXL_CLEAR(ANYOF_POSIXL_BITMAP(p), (c)) 754 755 #define POSIXL_TEST(field, c) ((field) & (1U << (c))) 756 #define ANYOF_POSIXL_TEST(p, c) POSIXL_TEST(ANYOF_POSIXL_BITMAP(p), (c)) 757 758 #define POSIXL_ZERO(field) STMT_START { (field) = 0; } STMT_END 759 #define ANYOF_POSIXL_ZERO(ret) POSIXL_ZERO(ANYOF_POSIXL_BITMAP(ret)) 760 761 #define ANYOF_POSIXL_SET_TO_BITMAP(p, bits) \ 762 STMT_START { ANYOF_POSIXL_BITMAP(p) = (bits); } STMT_END 763 764 /* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */ 765 #define ANYOF_POSIXL_SETALL(ret) \ 766 STMT_START { \ 767 ANYOF_POSIXL_BITMAP(ret) = nBIT_MASK(ANYOF_POSIXL_MAX); \ 768 } STMT_END 769 #define ANYOF_CLASS_SETALL(ret) ANYOF_POSIXL_SETALL(ret) 770 771 #define ANYOF_POSIXL_TEST_ANY_SET(p) \ 772 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) && ANYOF_POSIXL_BITMAP(p)) 773 #define ANYOF_CLASS_TEST_ANY_SET(p) ANYOF_POSIXL_TEST_ANY_SET(p) 774 775 /* Since an SSC always has this field, we don't have to test for that; nor do 776 * we want to because the bit isn't set for SSC during its construction */ 777 #define ANYOF_POSIXL_SSC_TEST_ANY_SET(p) \ 778 cBOOL(((regnode_ssc*)(p))->classflags) 779 #define ANYOF_POSIXL_SSC_TEST_ALL_SET(p) /* Are all bits set? */ \ 780 (((regnode_ssc*) (p))->classflags \ 781 == nBIT_MASK(ANYOF_POSIXL_MAX)) 782 783 #define ANYOF_POSIXL_TEST_ALL_SET(p) \ 784 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \ 785 && ANYOF_POSIXL_BITMAP(p) == nBIT_MASK(ANYOF_POSIXL_MAX)) 786 787 #define ANYOF_POSIXL_OR(source, dest) STMT_START { (dest)->classflags |= (source)->classflags ; } STMT_END 788 #define ANYOF_CLASS_OR(source, dest) ANYOF_POSIXL_OR((source), (dest)) 789 790 #define ANYOF_POSIXL_AND(source, dest) STMT_START { (dest)->classflags &= (source)->classflags ; } STMT_END 791 792 #define ANYOF_BITMAP_ZERO(ret) Zero(((regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char) 793 #define ANYOF_BITMAP(p) ((regnode_charclass*)(p))->bitmap 794 #define ANYOF_BITMAP_BYTE(p, c) BITMAP_BYTE(ANYOF_BITMAP(p), c) 795 #define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c)) 796 #define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c)) 797 #define ANYOF_BITMAP_TEST(p, c) cBOOL(ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c)) 798 799 #define ANYOF_BITMAP_SETALL(p) \ 800 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE) 801 #define ANYOF_BITMAP_CLEARALL(p) \ 802 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE) 803 804 /* 805 * Utility definitions. 806 */ 807 #ifndef CHARMASK 808 # define UCHARAT(p) ((int)*(const U8*)(p)) 809 #else 810 # define UCHARAT(p) ((int)*(p)&CHARMASK) 811 #endif 812 813 /* Number of regnode equivalents that 'guy' occupies beyond the size of the 814 * smallest regnode. */ 815 #define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode)) 816 817 #define REG_ZERO_LEN_SEEN 0x00000001 818 #define REG_LOOKBEHIND_SEEN 0x00000002 819 /* add a short form alias to keep the line length police happy */ 820 #define REG_LB_SEEN REG_LOOKBEHIND_SEEN 821 #define REG_GPOS_SEEN 0x00000004 822 /* spare */ 823 #define REG_RECURSE_SEEN 0x00000020 824 #define REG_TOP_LEVEL_BRANCHES_SEEN 0x00000040 825 #define REG_VERBARG_SEEN 0x00000080 826 #define REG_CUTGROUP_SEEN 0x00000100 827 #define REG_RUN_ON_COMMENT_SEEN 0x00000200 828 #define REG_UNFOLDED_MULTI_SEEN 0x00000400 829 /* spare */ 830 #define REG_UNBOUNDED_QUANTIFIER_SEEN 0x00001000 831 832 833 START_EXTERN_C 834 835 #ifdef PLUGGABLE_RE_EXTENSION 836 #include "re_nodes.h" 837 #else 838 #include "regnodes.h" 839 #endif 840 841 #ifndef PLUGGABLE_RE_EXTENSION 842 #ifndef DOINIT 843 EXTCONST regexp_engine PL_core_reg_engine; 844 #else /* DOINIT */ 845 EXTCONST regexp_engine PL_core_reg_engine = { 846 Perl_re_compile, 847 Perl_regexec_flags, 848 Perl_re_intuit_start, 849 Perl_re_intuit_string, 850 Perl_regfree_internal, 851 Perl_reg_numbered_buff_fetch, 852 Perl_reg_numbered_buff_store, 853 Perl_reg_numbered_buff_length, 854 Perl_reg_named_buff, 855 Perl_reg_named_buff_iter, 856 Perl_reg_qr_package, 857 #if defined(USE_ITHREADS) 858 Perl_regdupe_internal, 859 #endif 860 Perl_re_op_compile 861 }; 862 #endif /* DOINIT */ 863 #endif /* PLUGGABLE_RE_EXTENSION */ 864 865 866 END_EXTERN_C 867 868 869 /* .what is a character array with one character for each member of .data 870 * The character describes the function of the corresponding .data item: 871 * a - AV for paren_name_list under DEBUGGING 872 * f - start-class data for regstclass optimization 873 * l - start op for literal (?{EVAL}) item 874 * L - start op for literal (?{EVAL}) item, with separate CV (qr//) 875 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/ 876 * s - inversion list for Unicode-style character class, and the 877 * multicharacter strings resulting from casefolding the single-character 878 * entries in the character class 879 * t - trie struct 880 * u - trie struct's widecharmap (a HV, so can't share, must dup) 881 * also used for revcharmap and words under DEBUGGING 882 * T - aho-trie struct 883 * S - sv for named capture lookup 884 * 20010712 mjd@plover.com 885 * (Remember to update re_dup() and pregfree() if you add any items.) 886 */ 887 struct reg_data { 888 U32 count; 889 U8 *what; 890 void* data[1]; 891 }; 892 893 /* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses 894 anchored* and float* via array indexes 0 and 1. */ 895 #define anchored_substr substrs->data[0].substr 896 #define anchored_utf8 substrs->data[0].utf8_substr 897 #define anchored_offset substrs->data[0].min_offset 898 #define anchored_end_shift substrs->data[0].end_shift 899 900 #define float_substr substrs->data[1].substr 901 #define float_utf8 substrs->data[1].utf8_substr 902 #define float_min_offset substrs->data[1].min_offset 903 #define float_max_offset substrs->data[1].max_offset 904 #define float_end_shift substrs->data[1].end_shift 905 906 #define check_substr substrs->data[2].substr 907 #define check_utf8 substrs->data[2].utf8_substr 908 #define check_offset_min substrs->data[2].min_offset 909 #define check_offset_max substrs->data[2].max_offset 910 #define check_end_shift substrs->data[2].end_shift 911 912 #define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr) 913 #define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8) 914 #define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr) 915 #define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8) 916 917 /* trie related stuff */ 918 919 /* a transition record for the state machine. the 920 check field determines which state "owns" the 921 transition. the char the transition is for is 922 determined by offset from the owning states base 923 field. the next field determines which state 924 is to be transitioned to if any. 925 */ 926 struct _reg_trie_trans { 927 U32 next; 928 U32 check; 929 }; 930 931 /* a transition list element for the list based representation */ 932 struct _reg_trie_trans_list_elem { 933 U16 forid; 934 U32 newstate; 935 }; 936 typedef struct _reg_trie_trans_list_elem reg_trie_trans_le; 937 938 /* a state for compressed nodes. base is an offset 939 into an array of reg_trie_trans array. If wordnum is 940 nonzero the state is accepting. if base is zero then 941 the state has no children (and will be accepting) 942 */ 943 struct _reg_trie_state { 944 U16 wordnum; 945 union { 946 U32 base; 947 reg_trie_trans_le* list; 948 } trans; 949 }; 950 951 /* info per word; indexed by wordnum */ 952 typedef struct { 953 U16 prev; /* previous word in acceptance chain; eg in 954 * zzz|abc|ab/ after matching the chars abc, the 955 * accepted word is #2, and the previous accepted 956 * word is #3 */ 957 U32 len; /* how many chars long is this word? */ 958 U32 accept; /* accept state for this word */ 959 } reg_trie_wordinfo; 960 961 962 typedef struct _reg_trie_state reg_trie_state; 963 typedef struct _reg_trie_trans reg_trie_trans; 964 965 966 /* anything in here that needs to be freed later 967 should be dealt with in pregfree. 968 refcount is first in both this and _reg_ac_data to allow a space 969 optimisation in Perl_regdupe. */ 970 struct _reg_trie_data { 971 U32 refcount; /* number of times this trie is referenced */ 972 U32 lasttrans; /* last valid transition element */ 973 U16 *charmap; /* byte to charid lookup array */ 974 reg_trie_state *states; /* state data */ 975 reg_trie_trans *trans; /* array of transition elements */ 976 char *bitmap; /* stclass bitmap */ 977 U16 *jump; /* optional 1 indexed array of offsets before tail 978 for the node following a given word. */ 979 reg_trie_wordinfo *wordinfo; /* array of info per word */ 980 U16 uniquecharcount; /* unique chars in trie (width of trans table) */ 981 U32 startstate; /* initial state - used for common prefix optimisation */ 982 STRLEN minlen; /* minimum length of words in trie - build/opt only? */ 983 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */ 984 U32 prefixlen; /* #chars in common prefix */ 985 U32 statecount; /* Build only - number of states in the states array 986 (including the unused zero state) */ 987 U32 wordcount; /* Build only */ 988 #ifdef DEBUGGING 989 STRLEN charcount; /* Build only */ 990 #endif 991 }; 992 /* There is one (3 under DEBUGGING) pointers that logically belong in this 993 structure, but are held outside as they need duplication on thread cloning, 994 whereas the rest of the structure can be read only: 995 HV *widecharmap; code points > 255 to charid 996 #ifdef DEBUGGING 997 AV *words; Array of words contained in trie, for dumping 998 AV *revcharmap; Map of each charid back to its character representation 999 #endif 1000 */ 1001 1002 #define TRIE_WORDS_OFFSET 2 1003 1004 typedef struct _reg_trie_data reg_trie_data; 1005 1006 /* refcount is first in both this and _reg_trie_data to allow a space 1007 optimisation in Perl_regdupe. */ 1008 struct _reg_ac_data { 1009 U32 refcount; 1010 U32 trie; 1011 U32 *fail; 1012 reg_trie_state *states; 1013 }; 1014 typedef struct _reg_ac_data reg_ac_data; 1015 1016 /* ANY_BIT doesn't use the structure, so we can borrow it here. 1017 This is simpler than refactoring all of it as wed end up with 1018 three different sets... */ 1019 1020 #define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap) 1021 #define TRIE_BITMAP_BYTE(p, c) BITMAP_BYTE(TRIE_BITMAP(p), c) 1022 #define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c)) 1023 #define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c)) 1024 #define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c)) 1025 1026 #define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC) 1027 #define IS_TRIE_AC(op) ((op)>=AHOCORASICK) 1028 1029 1030 #define BITMAP_BYTE(p, c) (( (U8*) p) [ ( ( (UV) (c)) >> 3) ] ) 1031 #define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c)) 1032 1033 /* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */ 1034 #define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1) 1035 #define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state)) 1036 #define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state)) 1037 1038 #ifdef DEBUGGING 1039 #define TRIE_CHARCOUNT(trie) ((trie)->charcount) 1040 #else 1041 #define TRIE_CHARCOUNT(trie) (trie_charcount) 1042 #endif 1043 1044 #define RE_TRIE_MAXBUF_INIT 65536 1045 #define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF" 1046 #define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS" 1047 1048 #define RE_COMPILE_RECURSION_INIT 1000 1049 #define RE_COMPILE_RECURSION_LIMIT "\022E_COMPILE_RECURSION_LIMIT" 1050 1051 /* 1052 1053 RE_DEBUG_FLAGS is used to control what debug output is emitted 1054 its divided into three groups of options, some of which interact. 1055 The three groups are: Compile, Execute, Extra. There is room for a 1056 further group, as currently only the low three bytes are used. 1057 1058 Compile Options: 1059 1060 PARSE 1061 PEEP 1062 TRIE 1063 PROGRAM 1064 1065 Execute Options: 1066 1067 INTUIT 1068 MATCH 1069 TRIE 1070 1071 Extra Options 1072 1073 TRIE 1074 1075 If you modify any of these make sure you make corresponding changes to 1076 re.pm, especially to the documentation. 1077 1078 */ 1079 1080 1081 /* Compile */ 1082 #define RE_DEBUG_COMPILE_MASK 0x0000FF 1083 #define RE_DEBUG_COMPILE_PARSE 0x000001 1084 #define RE_DEBUG_COMPILE_OPTIMISE 0x000002 1085 #define RE_DEBUG_COMPILE_TRIE 0x000004 1086 #define RE_DEBUG_COMPILE_DUMP 0x000008 1087 #define RE_DEBUG_COMPILE_FLAGS 0x000010 1088 #define RE_DEBUG_COMPILE_TEST 0x000020 1089 1090 /* Execute */ 1091 #define RE_DEBUG_EXECUTE_MASK 0x00FF00 1092 #define RE_DEBUG_EXECUTE_INTUIT 0x000100 1093 #define RE_DEBUG_EXECUTE_MATCH 0x000200 1094 #define RE_DEBUG_EXECUTE_TRIE 0x000400 1095 1096 /* Extra */ 1097 #define RE_DEBUG_EXTRA_MASK 0x3FF0000 1098 #define RE_DEBUG_EXTRA_TRIE 0x0010000 1099 #define RE_DEBUG_EXTRA_STATE 0x0080000 1100 #define RE_DEBUG_EXTRA_OPTIMISE 0x0100000 1101 #define RE_DEBUG_EXTRA_BUFFERS 0x0400000 1102 #define RE_DEBUG_EXTRA_GPOS 0x0800000 1103 #define RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE 0x1000000 1104 #define RE_DEBUG_EXTRA_WILDCARD 0x2000000 1105 /* combined */ 1106 #define RE_DEBUG_EXTRA_STACK 0x0280000 1107 1108 #define RE_DEBUG_FLAG(x) (re_debug_flags & (x)) 1109 /* Compile */ 1110 #define DEBUG_COMPILE_r(x) DEBUG_r( \ 1111 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_MASK)) x ) 1112 #define DEBUG_PARSE_r(x) DEBUG_r( \ 1113 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_PARSE)) x ) 1114 #define DEBUG_OPTIMISE_r(x) DEBUG_r( \ 1115 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE)) x ) 1116 #define DEBUG_DUMP_r(x) DEBUG_r( \ 1117 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_DUMP)) x ) 1118 #define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \ 1119 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TRIE)) x ) 1120 #define DEBUG_FLAGS_r(x) DEBUG_r( \ 1121 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_FLAGS)) x ) 1122 #define DEBUG_TEST_r(x) DEBUG_r( \ 1123 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TEST)) x ) 1124 /* Execute */ 1125 #define DEBUG_EXECUTE_r(x) DEBUG_r( \ 1126 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_MASK)) x ) 1127 #define DEBUG_INTUIT_r(x) DEBUG_r( \ 1128 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_INTUIT)) x ) 1129 #define DEBUG_MATCH_r(x) DEBUG_r( \ 1130 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_MATCH)) x ) 1131 #define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \ 1132 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_TRIE)) x ) 1133 1134 /* Extra */ 1135 #define DEBUG_EXTRA_r(x) DEBUG_r( \ 1136 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_MASK)) x ) 1137 #define DEBUG_STATE_r(x) DEBUG_r( \ 1138 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_STATE)) x ) 1139 #define DEBUG_STACK_r(x) DEBUG_r( \ 1140 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_STACK)) x ) 1141 #define DEBUG_BUFFERS_r(x) DEBUG_r( \ 1142 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_BUFFERS)) x ) 1143 1144 #define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \ 1145 if (DEBUG_v_TEST || ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \ 1146 RE_DEBUG_FLAG(RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE))) x ) 1147 #define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \ 1148 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_TRIE)) x ) 1149 #define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \ 1150 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_TRIE)) x ) 1151 1152 #define DEBUG_TRIE_r(x) DEBUG_r( \ 1153 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TRIE \ 1154 | RE_DEBUG_EXECUTE_TRIE )) x ) 1155 #define DEBUG_GPOS_r(x) DEBUG_r( \ 1156 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_GPOS)) x ) 1157 1158 #define DEBUG_DUMP_PRE_OPTIMIZE_r(x) DEBUG_r( \ 1159 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE)) x ) 1160 1161 /* initialization */ 1162 /* Get the debug flags for code not in regcomp.c nor regexec.c. This doesn't 1163 * initialize the variable if it isn't already there, instead it just assumes 1164 * the flags are 0 */ 1165 #define DECLARE_AND_GET_RE_DEBUG_FLAGS_NON_REGEX \ 1166 volatile IV re_debug_flags = 0; PERL_UNUSED_VAR(re_debug_flags); \ 1167 STMT_START { \ 1168 SV * re_debug_flags_sv = NULL; \ 1169 /* get_sv() can return NULL during global destruction. */ \ 1170 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \ 1171 if (re_debug_flags_sv && SvIOK(re_debug_flags_sv)) \ 1172 re_debug_flags=SvIV(re_debug_flags_sv); \ 1173 } STMT_END 1174 1175 1176 #ifdef DEBUGGING 1177 1178 /* For use in regcomp.c and regexec.c, Get the debug flags, and initialize to 1179 * the defaults if not done already */ 1180 #define DECLARE_AND_GET_RE_DEBUG_FLAGS \ 1181 volatile IV re_debug_flags = 0; PERL_UNUSED_VAR(re_debug_flags); \ 1182 DEBUG_r({ \ 1183 SV * re_debug_flags_sv = NULL; \ 1184 /* get_sv() can return NULL during global destruction. */ \ 1185 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \ 1186 if (re_debug_flags_sv) { \ 1187 if (!SvIOK(re_debug_flags_sv)) /* If doesnt exist set to default */\ 1188 sv_setuv(re_debug_flags_sv, \ 1189 /* These defaults should be kept in sync with re.pm */ \ 1190 RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \ 1191 re_debug_flags=SvIV(re_debug_flags_sv); \ 1192 } \ 1193 }) 1194 1195 #define isDEBUG_WILDCARD (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_WILDCARD)) 1196 1197 #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \ 1198 const char * const rpv = \ 1199 pv_pretty((dsv), (pv), (l), (m), \ 1200 PL_colors[(c1)],PL_colors[(c2)], \ 1201 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \ 1202 const int rlen = SvCUR(dsv) 1203 1204 /* This is currently unsed in the core */ 1205 #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \ 1206 const char * const rpv = \ 1207 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \ 1208 PL_colors[(c1)],PL_colors[(c2)], \ 1209 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ) 1210 1211 #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \ 1212 const char * const rpv = \ 1213 pv_pretty((dsv), (pv), (l), (m), \ 1214 PL_colors[0], PL_colors[1], \ 1215 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \ 1216 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \ 1217 ) 1218 1219 #define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0)) 1220 #define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "") 1221 1222 #else /* if not DEBUGGING */ 1223 1224 #define DECLARE_AND_GET_RE_DEBUG_FLAGS dNOOP 1225 #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) dNOOP 1226 #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) 1227 #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) dNOOP 1228 #define RE_SV_DUMPLEN(ItEm) 1229 #define RE_SV_TAIL(ItEm) 1230 #define isDEBUG_WILDCARD 0 1231 1232 #endif /* DEBUG RELATED DEFINES */ 1233 1234 #define FIRST_NON_ASCII_DECIMAL_DIGIT 0x660 /* ARABIC_INDIC_DIGIT_ZERO */ 1235 1236 typedef enum { 1237 TRADITIONAL_BOUND = _CC_WORDCHAR, 1238 GCB_BOUND, 1239 LB_BOUND, 1240 SB_BOUND, 1241 WB_BOUND 1242 } bound_type; 1243 1244 /* This unpacks the FLAGS field of ANYOF[HR]x nodes. The value it contains 1245 * gives the strict lower bound for the UTF-8 start byte of any code point 1246 * matchable by the node, and a loose upper bound as well. 1247 * 1248 * The low bound is stored as 0xC0 + ((the upper 6 bits) >> 2) 1249 * The loose upper bound is determined from the lowest 2 bits and the low bound 1250 * (called x) as follows: 1251 * 1252 * 11 The upper limit of the range can be as much as (EF - x) / 8 1253 * 10 The upper limit of the range can be as much as (EF - x) / 4 1254 * 01 The upper limit of the range can be as much as (EF - x) / 2 1255 * 00 The upper limit of the range can be as much as EF 1256 * 1257 * For motivation of this design, see commit message in 1258 * 3146c00a633e9cbed741e10146662fbcedfdb8d3 */ 1259 #ifdef EBCDIC 1260 # define MAX_ANYOF_HRx_BYTE 0xF4 1261 #else 1262 # define MAX_ANYOF_HRx_BYTE 0xEF 1263 #endif 1264 #define LOWEST_ANYOF_HRx_BYTE(b) (((b) >> 2) + 0xC0) 1265 #define HIGHEST_ANYOF_HRx_BYTE(b) \ 1266 (LOWEST_ANYOF_HRx_BYTE(b) \ 1267 + ((MAX_ANYOF_HRx_BYTE - LOWEST_ANYOF_HRx_BYTE(b)) >> ((b) & 3))) 1268 1269 #endif /* PERL_REGCOMP_H_ */ 1270 1271 /* 1272 * ex: set ts=8 sts=4 sw=4 et: 1273 */ 1274