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 /* Activate offsets code - set to if 1 to enable */ 30 #ifdef DEBUGGING 31 #define RE_TRACK_PATTERN_OFFSETS 32 #endif 33 34 /* 35 * Structure for regexp "program". This is essentially a linear encoding 36 * of a nondeterministic finite-state machine (aka syntax charts or 37 * "railroad normal form" in parsing technology). Each node is an opcode 38 * plus a "next" pointer, possibly plus an operand. "Next" pointers of 39 * all nodes except BRANCH implement concatenation; a "next" pointer with 40 * a BRANCH on both ends of it is connecting two alternatives. (Here we 41 * have one of the subtle syntax dependencies: an individual BRANCH (as 42 * opposed to a collection of them) is never concatenated with anything 43 * because of operator precedence.) The operand of some types of node is 44 * a literal string; for others, it is a node leading into a sub-FSM. In 45 * particular, the operand of a BRANCH node is the first node of the branch. 46 * (NB this is *not* a tree structure: the tail of the branch connects 47 * to the thing following the set of BRANCHes.) The opcodes are defined 48 * in regnodes.h which is generated from regcomp.sym by regcomp.pl. 49 */ 50 51 /* 52 * A node is one char of opcode followed by two chars of "next" pointer. 53 * "Next" pointers are stored as two 8-bit pieces, high order first. The 54 * value is a positive offset from the opcode of the node containing it. 55 * An operand, if any, simply follows the node. (Note that much of the 56 * code generation knows about this implicit relationship.) 57 * 58 * Using two bytes for the "next" pointer is vast overkill for most things, 59 * but allows patterns to get big without disasters. 60 * 61 * [The "next" pointer is always aligned on an even 62 * boundary, and reads the offset directly as a short.] 63 */ 64 65 /* This is the stuff that used to live in regexp.h that was truly 66 private to the engine itself. It now lives here. */ 67 68 typedef struct regexp_internal { 69 union { 70 U32 *offsets; /* offset annotations 20001228 MJD 71 data about mapping the program to the 72 string - 73 offsets[0] is proglen when this is used 74 */ 75 U32 proglen; 76 } u; 77 78 regnode *regstclass; /* Optional startclass as identified or constructed 79 by the optimiser */ 80 struct reg_data *data; /* Additional miscellaneous data used by the program. 81 Used to make it easier to clone and free arbitrary 82 data that the regops need. Often the ARG field of 83 a regop is an index into this structure */ 84 struct reg_code_blocks *code_blocks;/* positions of literal (?{}) */ 85 int name_list_idx; /* Optional data index of an array of paren names */ 86 regnode program[1]; /* Unwarranted chumminess with compiler. */ 87 } regexp_internal; 88 89 #define RXi_SET(x,y) (x)->pprivate = (void*)(y) 90 #define RXi_GET(x) ((regexp_internal *)((x)->pprivate)) 91 #define RXi_GET_DECL(r,ri) regexp_internal *ri = RXi_GET(r) 92 /* 93 * Flags stored in regexp->intflags 94 * These are used only internally to the regexp engine 95 * 96 * See regexp.h for flags used externally to the regexp engine 97 */ 98 #define RXp_INTFLAGS(rx) ((rx)->intflags) 99 #define RX_INTFLAGS(prog) RXp_INTFLAGS(ReANY(prog)) 100 101 #define PREGf_SKIP 0x00000001 102 #define PREGf_IMPLICIT 0x00000002 /* Converted .* to ^.* */ 103 #define PREGf_NAUGHTY 0x00000004 /* how exponential is this pattern? */ 104 #define PREGf_VERBARG_SEEN 0x00000008 105 #define PREGf_CUTGROUP_SEEN 0x00000010 106 #define PREGf_USE_RE_EVAL 0x00000020 /* compiled with "use re 'eval'" */ 107 /* these used to be extflags, but are now intflags */ 108 #define PREGf_NOSCAN 0x00000040 109 /* spare */ 110 #define PREGf_GPOS_SEEN 0x00000100 111 #define PREGf_GPOS_FLOAT 0x00000200 112 113 #define PREGf_ANCH_MBOL 0x00000400 114 #define PREGf_ANCH_SBOL 0x00000800 115 #define PREGf_ANCH_GPOS 0x00001000 116 #define PREGf_RECURSE_SEEN 0x00002000 117 118 #define PREGf_ANCH \ 119 ( PREGf_ANCH_SBOL | PREGf_ANCH_GPOS | PREGf_ANCH_MBOL ) 120 121 /* this is where the old regcomp.h started */ 122 123 struct regnode_string { 124 U8 str_len; 125 U8 type; 126 U16 next_off; 127 char string[1]; 128 }; 129 130 struct regnode_lstring { /* Constructed this way to keep the string aligned. */ 131 U8 flags; 132 U8 type; 133 U16 next_off; 134 U32 str_len; /* Only 18 bits allowed before would overflow 'next_off' */ 135 char string[1]; 136 }; 137 138 struct regnode_anyofhs { /* Constructed this way to keep the string aligned. */ 139 U8 str_len; 140 U8 type; 141 U16 next_off; 142 U32 arg1; /* set by set_ANYOF_arg() */ 143 char string[1]; 144 }; 145 146 /* Argument bearing node - workhorse, 147 arg1 is often for the data field */ 148 struct regnode_1 { 149 U8 flags; 150 U8 type; 151 U16 next_off; 152 U32 arg1; 153 }; 154 155 /* Node whose argument is 'SV *'. This needs to be used very carefully in 156 * situations where pointers won't become invalid because of, say re-mallocs */ 157 struct regnode_p { 158 U8 flags; 159 U8 type; 160 U16 next_off; 161 SV * arg1; 162 }; 163 164 /* Similar to a regnode_1 but with an extra signed argument */ 165 struct regnode_2L { 166 U8 flags; 167 U8 type; 168 U16 next_off; 169 U32 arg1; 170 I32 arg2; 171 }; 172 173 /* 'Two field' -- Two 16 bit unsigned args */ 174 struct regnode_2 { 175 U8 flags; 176 U8 type; 177 U16 next_off; 178 U16 arg1; 179 U16 arg2; 180 }; 181 182 #define ANYOF_BITMAP_SIZE (NUM_ANYOF_CODE_POINTS / 8) /* 8 bits/Byte */ 183 184 /* Note that these form structs which are supersets of the next smaller one, by 185 * appending fields. Alignment problems can occur if one of those optional 186 * fields requires stricter alignment than the base struct. And formal 187 * parameters that can really be two or more of the structs should be 188 * declared as the smallest one it could be. See commit message for 189 * 7dcac5f6a5195002b55c935ee1d67f67e1df280b. Regnode allocation is done 190 * without regard to alignment, and changing it to would also require changing 191 * the code that inserts and deletes regnodes. The basic single-argument 192 * regnode has a U32, which is what reganode() allocates as a unit. Therefore 193 * no field can require stricter alignment than U32. */ 194 195 /* also used by trie */ 196 struct regnode_charclass { 197 U8 flags; 198 U8 type; 199 U16 next_off; 200 U32 arg1; /* set by set_ANYOF_arg() */ 201 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */ 202 }; 203 204 /* has runtime (locale) \d, \w, ..., [:posix:] classes */ 205 struct regnode_charclass_posixl { 206 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */ 207 U8 type; 208 U16 next_off; 209 U32 arg1; 210 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */ 211 U32 classflags; /* and run-time */ 212 }; 213 214 /* A synthetic start class (SSC); is a regnode_charclass_posixl_fold, plus an 215 * extra SV*, used only during its construction and which is not used by 216 * regexec.c. Note that the 'next_off' field is unused, as the SSC stands 217 * alone, so there is never a next node. Also, there is no alignment issue, 218 * because these are declared or allocated as a complete unit so the compiler 219 * takes care of alignment. This is unlike the other regnodes which are 220 * allocated in terms of multiples of a single-argument regnode. SSC nodes can 221 * have a pointer field because there is no alignment issue, and because it is 222 * set to NULL after construction, before any cloning of the pattern */ 223 struct regnode_ssc { 224 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */ 225 U8 type; 226 U16 next_off; 227 U32 arg1; 228 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */ 229 U32 classflags; /* ... and run-time */ 230 231 /* Auxiliary, only used during construction; NULL afterwards: list of code 232 * points matched */ 233 SV* invlist; 234 }; 235 236 /* We take advantage of 'next_off' not otherwise being used in the SSC by 237 * actually using it: by setting it to 1. This allows us to test and 238 * distinguish between an SSC and other ANYOF node types, as 'next_off' cannot 239 * otherwise be 1, because it is the offset to the next regnode expressed in 240 * units of regnodes. Since an ANYOF node contains extra fields, it adds up 241 * to 12 regnode units on 32-bit systems, (hence the minimum this can be (if 242 * not 0) is 11 there. Even if things get tightly packed on a 64-bit system, 243 * it still would be more than 1. */ 244 #define set_ANYOF_SYNTHETIC(n) STMT_START{ OP(n) = ANYOF; \ 245 NEXT_OFF(n) = 1; \ 246 } STMT_END 247 #define is_ANYOF_SYNTHETIC(n) (PL_regkind[OP(n)] == ANYOF && NEXT_OFF(n) == 1) 248 249 /* XXX fix this description. 250 Impose a limit of REG_INFTY on various pattern matching operations 251 to limit stack growth and to avoid "infinite" recursions. 252 */ 253 /* The default size for REG_INFTY is U16_MAX, which is the same as 254 USHORT_MAX (see perl.h). Unfortunately U16 isn't necessarily 16 bits 255 (see handy.h). On the Cray C90, sizeof(short)==4 and hence U16_MAX is 256 ((1<<32)-1), while on the Cray T90, sizeof(short)==8 and U16_MAX is 257 ((1<<64)-1). To limit stack growth to reasonable sizes, supply a 258 smaller default. 259 --Andy Dougherty 11 June 1998 260 */ 261 #if SHORTSIZE > 2 262 # ifndef REG_INFTY 263 # define REG_INFTY nBIT_UMAX(16) 264 # endif 265 #endif 266 267 #ifndef REG_INFTY 268 # define REG_INFTY U16_MAX 269 #endif 270 271 #define ARG_VALUE(arg) (arg) 272 #define ARG__SET(arg,val) ((arg) = (val)) 273 274 #undef ARG 275 #undef ARG1 276 #undef ARG2 277 278 #define ARG(p) ARG_VALUE(ARG_LOC(p)) 279 #define ARGp(p) ARG_VALUE(ARGp_LOC(p)) 280 #define ARG1(p) ARG_VALUE(ARG1_LOC(p)) 281 #define ARG2(p) ARG_VALUE(ARG2_LOC(p)) 282 #define ARG2L(p) ARG_VALUE(ARG2L_LOC(p)) 283 284 #define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val)) 285 #define ARGp_SET(p, val) ARG__SET(ARGp_LOC(p), (val)) 286 #define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val)) 287 #define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val)) 288 #define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val)) 289 290 #undef NEXT_OFF 291 #undef NODE_ALIGN 292 293 #define NEXT_OFF(p) ((p)->next_off) 294 #define NODE_ALIGN(node) 295 /* the following define was set to 0xde in 075abff3 296 * as part of some linting logic. I have set it to 0 297 * as otherwise in every place where we /might/ set flags 298 * we have to set it 0 explicitly, which duplicates 299 * assignments and IMO adds an unacceptable level of 300 * surprise to working in the regex engine. If this 301 * is changed from 0 then at the very least make sure 302 * that SBOL for /^/ sets the flags to 0 explicitly. 303 * -- Yves */ 304 #define NODE_ALIGN_FILL(node) ((node)->flags = 0) 305 306 #define SIZE_ALIGN NODE_ALIGN 307 308 #undef OP 309 #undef OPERAND 310 #undef STRING 311 312 #define OP(p) ((p)->type) 313 #define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \ 314 regnode types. For some, it's the \ 315 character set of the regnode */ 316 #define STR_LENs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \ 317 ((struct regnode_string *)p)->str_len) 318 #define STRINGs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \ 319 ((struct regnode_string *)p)->string) 320 #define OPERANDs(p) STRINGs(p) 321 322 /* Long strings. Currently limited to length 18 bits, which handles a 262000 323 * byte string. The limiting factor is the 16 bit 'next_off' field, which 324 * points to the next regnode, so the furthest away it can be is 2**16. On 325 * most architectures, regnodes are 2**2 bytes long, so that yields 2**18 326 * bytes. Should a longer string be desired, we could increase it to 26 bits 327 * fairly easily, by changing this node to have longj type which causes the ARG 328 * field to be used for the link to the next regnode (although code would have 329 * to be changed to account for this), and then use a combination of the flags 330 * and next_off fields for the length. To get 34 bit length, also change the 331 * node to be an ARG2L, using the second 32 bit field for the length, and not 332 * using the flags nor next_off fields at all. One could have an llstring node 333 * and even an lllstring type. */ 334 #define STR_LENl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 335 (((struct regnode_lstring *)p)->str_len)) 336 #define STRINGl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 337 (((struct regnode_lstring *)p)->string)) 338 #define OPERANDl(p) STRINGl(p) 339 340 #define STR_LEN(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 341 ? STR_LENl(p) : STR_LENs(p)) 342 #define STRING(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 343 ? STRINGl(p) : STRINGs(p)) 344 #define OPERAND(p) STRING(p) 345 346 /* The number of (smallest) regnode equivalents that a string of length l bytes 347 * occupies */ 348 #define STR_SZ(l) (((l) + sizeof(regnode) - 1) / sizeof(regnode)) 349 350 /* The number of (smallest) regnode equivalents that the EXACTISH node 'p' 351 * occupies */ 352 #define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p)) + 1 + regarglen[(p)->type]) 353 354 #define setSTR_LEN(p,v) \ 355 STMT_START{ \ 356 if (OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 357 ((struct regnode_lstring *)(p))->str_len = (v); \ 358 else \ 359 ((struct regnode_string *)(p))->str_len = (v); \ 360 } STMT_END 361 362 #define ANYOFR_BASE_BITS 20 363 #define ANYOFRbase(p) (ARG(p) & nBIT_MASK(ANYOFR_BASE_BITS)) 364 #define ANYOFRdelta(p) (ARG(p) >> ANYOFR_BASE_BITS) 365 366 #undef NODE_ALIGN 367 #undef ARG_LOC 368 #undef NEXTOPER 369 #undef PREVOPER 370 371 #define NODE_ALIGN(node) 372 #define ARG_LOC(p) (((struct regnode_1 *)p)->arg1) 373 #define ARGp_LOC(p) (((struct regnode_p *)p)->arg1) 374 #define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1) 375 #define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2) 376 #define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2) 377 378 #define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */ 379 #define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2) 380 381 #define NEXTOPER(p) ((p) + NODE_STEP_REGNODE) 382 #define PREVOPER(p) ((p) - NODE_STEP_REGNODE) 383 384 #define FILL_NODE(offset, op) \ 385 STMT_START { \ 386 OP(REGNODE_p(offset)) = op; \ 387 NEXT_OFF(REGNODE_p(offset)) = 0; \ 388 } STMT_END 389 #define FILL_ADVANCE_NODE(offset, op) \ 390 STMT_START { \ 391 FILL_NODE(offset, op); \ 392 (offset)++; \ 393 } STMT_END 394 #define FILL_ADVANCE_NODE_ARG(offset, op, arg) \ 395 STMT_START { \ 396 ARG_SET(REGNODE_p(offset), arg); \ 397 FILL_ADVANCE_NODE(offset, op); \ 398 /* This is used generically for other operations \ 399 * that have a longer argument */ \ 400 (offset) += regarglen[op]; \ 401 } STMT_END 402 #define FILL_ADVANCE_NODE_ARGp(offset, op, arg) \ 403 STMT_START { \ 404 ARGp_SET(REGNODE_p(offset), arg); \ 405 FILL_ADVANCE_NODE(offset, op); \ 406 (offset) += regarglen[op]; \ 407 } STMT_END 408 #define FILL_ADVANCE_NODE_2L_ARG(offset, op, arg1, arg2) \ 409 STMT_START { \ 410 ARG_SET(REGNODE_p(offset), arg1); \ 411 ARG2L_SET(REGNODE_p(offset), arg2); \ 412 FILL_ADVANCE_NODE(offset, op); \ 413 (offset) += 2; \ 414 } STMT_END 415 416 #define REG_MAGIC 0234 417 418 /* An ANYOF node is basically a bitmap with the index being a code point. If 419 * the bit for that code point is 1, the code point matches; if 0, it doesn't 420 * match (complemented if inverted). There is an additional mechanism to deal 421 * with cases where the bitmap is insufficient in and of itself. This #define 422 * indicates if the bitmap does fully represent what this ANYOF node can match. 423 * The ARG is set to this special value (since 0, 1, ... are legal, but will 424 * never reach this high). */ 425 #define ANYOF_ONLY_HAS_BITMAP ((U32) -1) 426 427 /* When the bitmap isn't completely sufficient for handling the ANYOF node, 428 * flags (in node->flags of the ANYOF node) get set to indicate this. These 429 * are perennially in short supply. Beyond several cases where warnings need 430 * to be raised under certain circumstances, currently, there are six cases 431 * where the bitmap alone isn't sufficient. We could use six flags to 432 * represent the 6 cases, but to save flags bits, we play some games. The 433 * cases are: 434 * 435 * 1) The bitmap has a compiled-in very finite size. So something else needs 436 * to be used to specify if a code point that is too large for the bitmap 437 * actually matches. The mechanism currently is an inversion 438 * list. ANYOF_ONLY_HAS_BITMAP, described above, being TRUE indicates 439 * there are no matches of too-large code points. But if it is FALSE, 440 * then almost certainly there are matches too large for the bitmap. (The 441 * other cases, described below, either imply this one or are extremely 442 * rare in practice.) So we can just assume that a too-large code point 443 * will need something beyond the bitmap if ANYOF_ONLY_HAS_BITMAP is 444 * FALSE, instead of having a separate flag for this. 445 * 2) A subset of item 1) is if all possible code points outside the bitmap 446 * match. This is a common occurrence when the class is complemented, 447 * like /[^ij]/. Therefore a bit is reserved to indicate this, 448 * rather than having an inversion list created, 449 * ANYOF_MATCHES_ALL_ABOVE_BITMAP. 450 * 3) Under /d rules, it can happen that code points that are in the upper 451 * latin1 range (\x80-\xFF or their equivalents on EBCDIC platforms) match 452 * only if the runtime target string being matched against is UTF-8. For 453 * example /[\w[:punct:]]/d. This happens only for posix classes (with a 454 * couple of exceptions, like \d where it doesn't happen), and all such 455 * ones also have above-bitmap matches. Thus, 3) implies 1) as well. 456 * Note that /d rules are no longer encouraged; 'use 5.14' or higher 457 * deselects them. But a flag is required so that they can be properly 458 * handled. But it can be a shared flag: see 5) below. 459 * 4) Also under /d rules, something like /[\Wfoo]/ will match everything in 460 * the \x80-\xFF range, unless the string being matched against is UTF-8. 461 * An inversion list could be created for this case, but this is 462 * relatively common, and it turns out that it's all or nothing: if any 463 * one of these code points matches, they all do. Hence a single bit 464 * suffices. We use a shared flag that doesn't take up space by itself: 465 * ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER. This 466 * also implies 1), with one exception: [:^cntrl:]. 467 * 5) A user-defined \p{} property may not have been defined by the time the 468 * regex is compiled. In this case, we don't know until runtime what it 469 * will match, so we have to assume it could match anything, including 470 * code points that ordinarily would be in the bitmap. A flag bit is 471 * necessary to indicate this, though it can be shared with the item 3) 472 * flag, as that only occurs under /d, and this only occurs under non-d. 473 * This case is quite uncommon in the field, and the /(?[ ...])/ construct 474 * is a better way to accomplish what this feature does. This case also 475 * implies 1). 476 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 477 * is the shared flag. 478 * 6) /[foo]/il may have folds that are only valid if the runtime locale is a 479 * UTF-8 one. These are quite rare, so it would be good to avoid the 480 * expense of looking for them. But /l matching is slow anyway, and we've 481 * traditionally not worried too much about its performance. And this 482 * condition requires the ANYOFL_FOLD flag to be set, so testing for 483 * that flag would be sufficient to rule out most cases of this. So it is 484 * unclear if this should have a flag or not. But, this flag can be 485 * shared with another, so it doesn't occupy extra space. 486 * 487 * At the moment, there is one spare bit, but this could be increased by 488 * various tricks: 489 * 490 * If just one more bit is needed, as of this writing it seems to khw that the 491 * best choice would be to make ANYOF_MATCHES_ALL_ABOVE_BITMAP not a flag, but 492 * something like 493 * 494 * #define ANYOF_MATCHES_ALL_ABOVE_BITMAP ((U32) -2) 495 * 496 * and access it through the ARG like ANYOF_ONLY_HAS_BITMAP is. This flag is 497 * used by all ANYOF node types, and it could be used to avoid calling the 498 * handler function, as the macro REGINCLASS in regexec.c does now for other 499 * cases. 500 * 501 * Another possibility is based on the fact that ANYOF_MATCHES_POSIXL is 502 * redundant with the node type ANYOFPOSIXL. That flag could be removed, but 503 * at the expense of extra code in regexec.c. The flag has been retained 504 * because it allows us to see if we need to call reginclass, or just use the 505 * bitmap in one test. 506 * 507 * If this is done, an extension would be to make all ANYOFL nodes contain the 508 * extra 32 bits that ANYOFPOSIXL ones do. The posix flags only occupy 30 509 * bits, so the ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD flags 510 * and ANYOFL_FOLD could be moved to that extra space, but it would mean extra 511 * instructions, as there are currently places in the code that assume those 512 * two bits are zero. 513 * 514 * All told, 5 bits could be available for other uses if all of the above were 515 * done. 516 * 517 * Some flags are not used in synthetic start class (SSC) nodes, so could be 518 * shared should new flags be needed for SSCs, like SSC_MATCHES_EMPTY_STRING 519 * now. */ 520 521 /* If this is set, the result of the match should be complemented. regexec.c 522 * is expecting this to be in the low bit. Never in an SSC */ 523 #define ANYOF_INVERT 0x01 524 525 /* For the SSC node only, which cannot be inverted, so is shared with that bit. 526 * This is used only during regex compilation. */ 527 #define SSC_MATCHES_EMPTY_STRING ANYOF_INVERT 528 529 /* Set if this is a regnode_charclass_posixl vs a regnode_charclass. This 530 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale 531 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved 532 * at compile-time. Only set under /l; can be in SSC */ 533 #define ANYOF_MATCHES_POSIXL 0x02 534 535 /* The fold is calculated and stored in the bitmap where possible at compile 536 * time. However under locale, the actual folding varies depending on 537 * what the locale is at the time of execution, so it has to be deferred until 538 * then. Only set under /l; never in an SSC */ 539 #define ANYOFL_FOLD 0x04 540 541 /* Shared bit set only with ANYOFL and SSC nodes: 542 * If ANYOFL_FOLD is set, this flag indicates there are potential matches 543 * valid only if the locale is a UTF-8 one. 544 * If ANYOFL_FOLD is NOT set, this flag means to warn if the runtime locale 545 * isn't a UTF-8 one (and the generated node assumes a UTF-8 locale). 546 * None of INVERT, POSIXL, 547 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 548 * can be set. */ 549 #define ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD 0x08 550 551 /* Convenience macros for teasing apart the meanings when reading the above bit 552 * */ 553 #define ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags) \ 554 ((flags & ( ANYOFL_FOLD /* Both bits are set */ \ 555 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \ 556 == ( ANYOFL_FOLD \ 557 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) 558 559 #define ANYOFL_UTF8_LOCALE_REQD(flags) \ 560 ((flags & ( ANYOFL_FOLD /* Only REQD bit is set */ \ 561 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \ 562 == ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD) 563 564 /* Spare: Be sure to change ANYOF_FLAGS_ALL if this gets used 0x10 */ 565 566 /* If set, the node matches every code point NUM_ANYOF_CODE_POINTS and above. 567 * Can be in an SSC */ 568 #define ANYOF_MATCHES_ALL_ABOVE_BITMAP 0x20 569 570 /* Shared bit: 571 * Under /d it means the ANYOFD node matches more things if the target 572 * string is encoded in UTF-8; any such things will be non-ASCII, 573 * characters that are < 256, and can be accessed via the inversion 574 * list. 575 * When not under /d, it means the ANYOF node contains a user-defined 576 * property that wasn't yet defined at the time the regex was compiled, 577 * and so must be looked up at runtime, by creating an inversion list. 578 * (These uses are mutually exclusive because a user-defined property is 579 * specified by \p{}, and \p{} implies /u which deselects /d). The long macro 580 * name is to make sure that you are cautioned about its shared nature. Only 581 * the non-/d meaning can be in an SSC */ 582 #define ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 0x40 583 584 /* Shared bit: 585 * Under /d it means the ANYOFD node matches all non-ASCII Latin1 586 * characters when the target string is not in utf8. 587 * When not under /d, it means the ANYOF node should raise a warning if 588 * matching against an above-Unicode code point. 589 * (These uses are mutually exclusive because the warning requires a \p{}, and 590 * \p{} implies /u which deselects /d). An SSC node only has this bit set if 591 * what is meant is the warning. The long macro name is to make sure that you 592 * are cautioned about its shared nature */ 593 #define ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER 0x80 594 595 #define ANYOF_FLAGS_ALL ((U8) ~0x10) 596 597 #define ANYOF_LOCALE_FLAGS (ANYOFL_FOLD | ANYOF_MATCHES_POSIXL) 598 599 /* These are the flags that apply to both regular ANYOF nodes and synthetic 600 * start class nodes during construction of the SSC. During finalization of 601 * the SSC, other of the flags may get added to it */ 602 #define ANYOF_COMMON_FLAGS 0 603 604 /* Character classes for node->classflags of ANYOF */ 605 /* Should be synchronized with a table in regprop() */ 606 /* 2n should be the normal one, paired with its complement at 2n+1 */ 607 608 #define ANYOF_ALPHA ((_CC_ALPHA) * 2) 609 #define ANYOF_NALPHA ((ANYOF_ALPHA) + 1) 610 #define ANYOF_ALPHANUMERIC ((_CC_ALPHANUMERIC) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */ 611 #define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1) 612 #define ANYOF_ASCII ((_CC_ASCII) * 2) 613 #define ANYOF_NASCII ((ANYOF_ASCII) + 1) 614 #define ANYOF_BLANK ((_CC_BLANK) * 2) /* GNU extension: space and tab: non-vertical space */ 615 #define ANYOF_NBLANK ((ANYOF_BLANK) + 1) 616 #define ANYOF_CASED ((_CC_CASED) * 2) /* Pseudo class for [:lower:] or 617 [:upper:] under /i */ 618 #define ANYOF_NCASED ((ANYOF_CASED) + 1) 619 #define ANYOF_CNTRL ((_CC_CNTRL) * 2) 620 #define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1) 621 #define ANYOF_DIGIT ((_CC_DIGIT) * 2) /* \d */ 622 #define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1) 623 #define ANYOF_GRAPH ((_CC_GRAPH) * 2) 624 #define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1) 625 #define ANYOF_LOWER ((_CC_LOWER) * 2) 626 #define ANYOF_NLOWER ((ANYOF_LOWER) + 1) 627 #define ANYOF_PRINT ((_CC_PRINT) * 2) 628 #define ANYOF_NPRINT ((ANYOF_PRINT) + 1) 629 #define ANYOF_PUNCT ((_CC_PUNCT) * 2) 630 #define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1) 631 #define ANYOF_SPACE ((_CC_SPACE) * 2) /* \s */ 632 #define ANYOF_NSPACE ((ANYOF_SPACE) + 1) 633 #define ANYOF_UPPER ((_CC_UPPER) * 2) 634 #define ANYOF_NUPPER ((ANYOF_UPPER) + 1) 635 #define ANYOF_WORDCHAR ((_CC_WORDCHAR) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */ 636 #define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1) 637 #define ANYOF_XDIGIT ((_CC_XDIGIT) * 2) 638 #define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1) 639 640 /* pseudo classes below this, not stored in the class bitmap, but used as flags 641 during compilation of char classes */ 642 643 #define ANYOF_VERTWS ((_CC_VERTSPACE) * 2) 644 #define ANYOF_NVERTWS ((ANYOF_VERTWS)+1) 645 646 /* It is best if this is the last one, as all above it are stored as bits in a 647 * bitmap, and it isn't part of that bitmap */ 648 #if _CC_VERTSPACE != _HIGHEST_REGCOMP_DOT_H_SYNC 649 # error Problem with handy.h _HIGHEST_REGCOMP_DOT_H_SYNC #define 650 #endif 651 652 #define ANYOF_POSIXL_MAX (ANYOF_VERTWS) /* So upper loop limit is written: 653 * '< ANYOF_MAX' 654 * Hence doesn't include VERTWS, as that 655 * is a pseudo class */ 656 #define ANYOF_MAX ANYOF_POSIXL_MAX 657 658 #if (ANYOF_POSIXL_MAX > 32) /* Must fit in 32-bit word */ 659 # error Problem with handy.h _CC_foo #defines 660 #endif 661 662 #define ANYOF_HORIZWS ((ANYOF_POSIXL_MAX)+2) /* = (ANYOF_NVERTWS + 1) */ 663 #define ANYOF_NHORIZWS ((ANYOF_POSIXL_MAX)+3) 664 665 #define ANYOF_UNIPROP ((ANYOF_POSIXL_MAX)+4) /* Used to indicate a Unicode 666 property: \p{} or \P{} */ 667 668 /* Backward source code compatibility. */ 669 670 #define ANYOF_ALNUML ANYOF_ALNUM 671 #define ANYOF_NALNUML ANYOF_NALNUM 672 #define ANYOF_SPACEL ANYOF_SPACE 673 #define ANYOF_NSPACEL ANYOF_NSPACE 674 #define ANYOF_ALNUM ANYOF_WORDCHAR 675 #define ANYOF_NALNUM ANYOF_NWORDCHAR 676 677 /* Utility macros for the bitmap and classes of ANYOF */ 678 679 #define ANYOF_FLAGS(p) ((p)->flags) 680 681 #define ANYOF_BIT(c) (1U << ((c) & 7)) 682 683 #define ANYOF_POSIXL_BITMAP(p) (((regnode_charclass_posixl*) (p))->classflags) 684 685 #define POSIXL_SET(field, c) ((field) |= (1U << (c))) 686 #define ANYOF_POSIXL_SET(p, c) POSIXL_SET(ANYOF_POSIXL_BITMAP(p), (c)) 687 688 #define POSIXL_CLEAR(field, c) ((field) &= ~ (1U <<(c))) 689 #define ANYOF_POSIXL_CLEAR(p, c) POSIXL_CLEAR(ANYOF_POSIXL_BITMAP(p), (c)) 690 691 #define POSIXL_TEST(field, c) ((field) & (1U << (c))) 692 #define ANYOF_POSIXL_TEST(p, c) POSIXL_TEST(ANYOF_POSIXL_BITMAP(p), (c)) 693 694 #define POSIXL_ZERO(field) STMT_START { (field) = 0; } STMT_END 695 #define ANYOF_POSIXL_ZERO(ret) POSIXL_ZERO(ANYOF_POSIXL_BITMAP(ret)) 696 697 #define ANYOF_POSIXL_SET_TO_BITMAP(p, bits) \ 698 STMT_START { ANYOF_POSIXL_BITMAP(p) = (bits); } STMT_END 699 700 /* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */ 701 #define ANYOF_POSIXL_SETALL(ret) \ 702 STMT_START { \ 703 ANYOF_POSIXL_BITMAP(ret) = nBIT_MASK(ANYOF_POSIXL_MAX); \ 704 } STMT_END 705 #define ANYOF_CLASS_SETALL(ret) ANYOF_POSIXL_SETALL(ret) 706 707 #define ANYOF_POSIXL_TEST_ANY_SET(p) \ 708 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) && ANYOF_POSIXL_BITMAP(p)) 709 #define ANYOF_CLASS_TEST_ANY_SET(p) ANYOF_POSIXL_TEST_ANY_SET(p) 710 711 /* Since an SSC always has this field, we don't have to test for that; nor do 712 * we want to because the bit isn't set for SSC during its construction */ 713 #define ANYOF_POSIXL_SSC_TEST_ANY_SET(p) \ 714 cBOOL(((regnode_ssc*)(p))->classflags) 715 #define ANYOF_POSIXL_SSC_TEST_ALL_SET(p) /* Are all bits set? */ \ 716 (((regnode_ssc*) (p))->classflags \ 717 == nBIT_MASK(ANYOF_POSIXL_MAX)) 718 719 #define ANYOF_POSIXL_TEST_ALL_SET(p) \ 720 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \ 721 && ANYOF_POSIXL_BITMAP(p) == nBIT_MASK(ANYOF_POSIXL_MAX)) 722 723 #define ANYOF_POSIXL_OR(source, dest) STMT_START { (dest)->classflags |= (source)->classflags ; } STMT_END 724 #define ANYOF_CLASS_OR(source, dest) ANYOF_POSIXL_OR((source), (dest)) 725 726 #define ANYOF_POSIXL_AND(source, dest) STMT_START { (dest)->classflags &= (source)->classflags ; } STMT_END 727 728 #define ANYOF_BITMAP_ZERO(ret) Zero(((regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char) 729 #define ANYOF_BITMAP(p) ((regnode_charclass*)(p))->bitmap 730 #define ANYOF_BITMAP_BYTE(p, c) BITMAP_BYTE(ANYOF_BITMAP(p), c) 731 #define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c)) 732 #define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c)) 733 #define ANYOF_BITMAP_TEST(p, c) cBOOL(ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c)) 734 735 #define ANYOF_BITMAP_SETALL(p) \ 736 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE) 737 #define ANYOF_BITMAP_CLEARALL(p) \ 738 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE) 739 740 /* 741 * Utility definitions. 742 */ 743 #ifndef CHARMASK 744 # define UCHARAT(p) ((int)*(const U8*)(p)) 745 #else 746 # define UCHARAT(p) ((int)*(p)&CHARMASK) 747 #endif 748 749 /* Number of regnode equivalents that 'guy' occupies beyond the size of the 750 * smallest regnode. */ 751 #define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode)) 752 753 #define REG_ZERO_LEN_SEEN 0x00000001 754 #define REG_LOOKBEHIND_SEEN 0x00000002 755 #define REG_GPOS_SEEN 0x00000004 756 /* spare */ 757 #define REG_RECURSE_SEEN 0x00000020 758 #define REG_TOP_LEVEL_BRANCHES_SEEN 0x00000040 759 #define REG_VERBARG_SEEN 0x00000080 760 #define REG_CUTGROUP_SEEN 0x00000100 761 #define REG_RUN_ON_COMMENT_SEEN 0x00000200 762 #define REG_UNFOLDED_MULTI_SEEN 0x00000400 763 /* spare */ 764 #define REG_UNBOUNDED_QUANTIFIER_SEEN 0x00001000 765 766 767 START_EXTERN_C 768 769 #ifdef PLUGGABLE_RE_EXTENSION 770 #include "re_nodes.h" 771 #else 772 #include "regnodes.h" 773 #endif 774 775 #ifndef PLUGGABLE_RE_EXTENSION 776 #ifndef DOINIT 777 EXTCONST regexp_engine PL_core_reg_engine; 778 #else /* DOINIT */ 779 EXTCONST regexp_engine PL_core_reg_engine = { 780 Perl_re_compile, 781 Perl_regexec_flags, 782 Perl_re_intuit_start, 783 Perl_re_intuit_string, 784 Perl_regfree_internal, 785 Perl_reg_numbered_buff_fetch, 786 Perl_reg_numbered_buff_store, 787 Perl_reg_numbered_buff_length, 788 Perl_reg_named_buff, 789 Perl_reg_named_buff_iter, 790 Perl_reg_qr_package, 791 #if defined(USE_ITHREADS) 792 Perl_regdupe_internal, 793 #endif 794 Perl_re_op_compile 795 }; 796 #endif /* DOINIT */ 797 #endif /* PLUGGABLE_RE_EXTENSION */ 798 799 800 END_EXTERN_C 801 802 803 /* .what is a character array with one character for each member of .data 804 * The character describes the function of the corresponding .data item: 805 * a - AV for paren_name_list under DEBUGGING 806 * f - start-class data for regstclass optimization 807 * l - start op for literal (?{EVAL}) item 808 * L - start op for literal (?{EVAL}) item, with separate CV (qr//) 809 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/ 810 * s - inversion list for Unicode-style character class, and the 811 * multicharacter strings resulting from casefolding the single-character 812 * entries in the character class 813 * t - trie struct 814 * u - trie struct's widecharmap (a HV, so can't share, must dup) 815 * also used for revcharmap and words under DEBUGGING 816 * T - aho-trie struct 817 * S - sv for named capture lookup 818 * 20010712 mjd@plover.com 819 * (Remember to update re_dup() and pregfree() if you add any items.) 820 */ 821 struct reg_data { 822 U32 count; 823 U8 *what; 824 void* data[1]; 825 }; 826 827 /* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses 828 anchored* and float* via array indexes 0 and 1. */ 829 #define anchored_substr substrs->data[0].substr 830 #define anchored_utf8 substrs->data[0].utf8_substr 831 #define anchored_offset substrs->data[0].min_offset 832 #define anchored_end_shift substrs->data[0].end_shift 833 834 #define float_substr substrs->data[1].substr 835 #define float_utf8 substrs->data[1].utf8_substr 836 #define float_min_offset substrs->data[1].min_offset 837 #define float_max_offset substrs->data[1].max_offset 838 #define float_end_shift substrs->data[1].end_shift 839 840 #define check_substr substrs->data[2].substr 841 #define check_utf8 substrs->data[2].utf8_substr 842 #define check_offset_min substrs->data[2].min_offset 843 #define check_offset_max substrs->data[2].max_offset 844 #define check_end_shift substrs->data[2].end_shift 845 846 #define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr) 847 #define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8) 848 #define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr) 849 #define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8) 850 851 /* trie related stuff */ 852 853 /* a transition record for the state machine. the 854 check field determines which state "owns" the 855 transition. the char the transition is for is 856 determined by offset from the owning states base 857 field. the next field determines which state 858 is to be transitioned to if any. 859 */ 860 struct _reg_trie_trans { 861 U32 next; 862 U32 check; 863 }; 864 865 /* a transition list element for the list based representation */ 866 struct _reg_trie_trans_list_elem { 867 U16 forid; 868 U32 newstate; 869 }; 870 typedef struct _reg_trie_trans_list_elem reg_trie_trans_le; 871 872 /* a state for compressed nodes. base is an offset 873 into an array of reg_trie_trans array. If wordnum is 874 nonzero the state is accepting. if base is zero then 875 the state has no children (and will be accepting) 876 */ 877 struct _reg_trie_state { 878 U16 wordnum; 879 union { 880 U32 base; 881 reg_trie_trans_le* list; 882 } trans; 883 }; 884 885 /* info per word; indexed by wordnum */ 886 typedef struct { 887 U16 prev; /* previous word in acceptance chain; eg in 888 * zzz|abc|ab/ after matching the chars abc, the 889 * accepted word is #2, and the previous accepted 890 * word is #3 */ 891 U32 len; /* how many chars long is this word? */ 892 U32 accept; /* accept state for this word */ 893 } reg_trie_wordinfo; 894 895 896 typedef struct _reg_trie_state reg_trie_state; 897 typedef struct _reg_trie_trans reg_trie_trans; 898 899 900 /* anything in here that needs to be freed later 901 should be dealt with in pregfree. 902 refcount is first in both this and _reg_ac_data to allow a space 903 optimisation in Perl_regdupe. */ 904 struct _reg_trie_data { 905 U32 refcount; /* number of times this trie is referenced */ 906 U32 lasttrans; /* last valid transition element */ 907 U16 *charmap; /* byte to charid lookup array */ 908 reg_trie_state *states; /* state data */ 909 reg_trie_trans *trans; /* array of transition elements */ 910 char *bitmap; /* stclass bitmap */ 911 U16 *jump; /* optional 1 indexed array of offsets before tail 912 for the node following a given word. */ 913 reg_trie_wordinfo *wordinfo; /* array of info per word */ 914 U16 uniquecharcount; /* unique chars in trie (width of trans table) */ 915 U32 startstate; /* initial state - used for common prefix optimisation */ 916 STRLEN minlen; /* minimum length of words in trie - build/opt only? */ 917 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */ 918 U32 prefixlen; /* #chars in common prefix */ 919 U32 statecount; /* Build only - number of states in the states array 920 (including the unused zero state) */ 921 U32 wordcount; /* Build only */ 922 #ifdef DEBUGGING 923 STRLEN charcount; /* Build only */ 924 #endif 925 }; 926 /* There is one (3 under DEBUGGING) pointers that logically belong in this 927 structure, but are held outside as they need duplication on thread cloning, 928 whereas the rest of the structure can be read only: 929 HV *widecharmap; code points > 255 to charid 930 #ifdef DEBUGGING 931 AV *words; Array of words contained in trie, for dumping 932 AV *revcharmap; Map of each charid back to its character representation 933 #endif 934 */ 935 936 #define TRIE_WORDS_OFFSET 2 937 938 typedef struct _reg_trie_data reg_trie_data; 939 940 /* refcount is first in both this and _reg_trie_data to allow a space 941 optimisation in Perl_regdupe. */ 942 struct _reg_ac_data { 943 U32 refcount; 944 U32 trie; 945 U32 *fail; 946 reg_trie_state *states; 947 }; 948 typedef struct _reg_ac_data reg_ac_data; 949 950 /* ANY_BIT doesn't use the structure, so we can borrow it here. 951 This is simpler than refactoring all of it as wed end up with 952 three different sets... */ 953 954 #define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap) 955 #define TRIE_BITMAP_BYTE(p, c) BITMAP_BYTE(TRIE_BITMAP(p), c) 956 #define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c)) 957 #define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c)) 958 #define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c)) 959 960 #define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC) 961 #define IS_TRIE_AC(op) ((op)>=AHOCORASICK) 962 963 964 #define BITMAP_BYTE(p, c) (( (U8*) p) [ ( ( (UV) (c)) >> 3) ] ) 965 #define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c)) 966 967 /* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */ 968 #define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1) 969 #define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state)) 970 #define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state)) 971 972 #ifdef DEBUGGING 973 #define TRIE_CHARCOUNT(trie) ((trie)->charcount) 974 #else 975 #define TRIE_CHARCOUNT(trie) (trie_charcount) 976 #endif 977 978 #define RE_TRIE_MAXBUF_INIT 65536 979 #define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF" 980 #define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS" 981 982 #define RE_COMPILE_RECURSION_INIT 1000 983 #define RE_COMPILE_RECURSION_LIMIT "\022E_COMPILE_RECURSION_LIMIT" 984 985 /* 986 987 RE_DEBUG_FLAGS is used to control what debug output is emitted 988 its divided into three groups of options, some of which interact. 989 The three groups are: Compile, Execute, Extra. There is room for a 990 further group, as currently only the low three bytes are used. 991 992 Compile Options: 993 994 PARSE 995 PEEP 996 TRIE 997 PROGRAM 998 OFFSETS 999 1000 Execute Options: 1001 1002 INTUIT 1003 MATCH 1004 TRIE 1005 1006 Extra Options 1007 1008 TRIE 1009 OFFSETS 1010 1011 If you modify any of these make sure you make corresponding changes to 1012 re.pm, especially to the documentation. 1013 1014 */ 1015 1016 1017 /* Compile */ 1018 #define RE_DEBUG_COMPILE_MASK 0x0000FF 1019 #define RE_DEBUG_COMPILE_PARSE 0x000001 1020 #define RE_DEBUG_COMPILE_OPTIMISE 0x000002 1021 #define RE_DEBUG_COMPILE_TRIE 0x000004 1022 #define RE_DEBUG_COMPILE_DUMP 0x000008 1023 #define RE_DEBUG_COMPILE_FLAGS 0x000010 1024 #define RE_DEBUG_COMPILE_TEST 0x000020 1025 1026 /* Execute */ 1027 #define RE_DEBUG_EXECUTE_MASK 0x00FF00 1028 #define RE_DEBUG_EXECUTE_INTUIT 0x000100 1029 #define RE_DEBUG_EXECUTE_MATCH 0x000200 1030 #define RE_DEBUG_EXECUTE_TRIE 0x000400 1031 1032 /* Extra */ 1033 #define RE_DEBUG_EXTRA_MASK 0x3FF0000 1034 #define RE_DEBUG_EXTRA_TRIE 0x0010000 1035 #define RE_DEBUG_EXTRA_OFFSETS 0x0020000 1036 #define RE_DEBUG_EXTRA_OFFDEBUG 0x0040000 1037 #define RE_DEBUG_EXTRA_STATE 0x0080000 1038 #define RE_DEBUG_EXTRA_OPTIMISE 0x0100000 1039 #define RE_DEBUG_EXTRA_BUFFERS 0x0400000 1040 #define RE_DEBUG_EXTRA_GPOS 0x0800000 1041 #define RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE 0x1000000 1042 #define RE_DEBUG_EXTRA_WILDCARD 0x2000000 1043 /* combined */ 1044 #define RE_DEBUG_EXTRA_STACK 0x0280000 1045 1046 #define RE_DEBUG_FLAG(x) (re_debug_flags & (x)) 1047 /* Compile */ 1048 #define DEBUG_COMPILE_r(x) DEBUG_r( \ 1049 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_MASK)) x ) 1050 #define DEBUG_PARSE_r(x) DEBUG_r( \ 1051 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_PARSE)) x ) 1052 #define DEBUG_OPTIMISE_r(x) DEBUG_r( \ 1053 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE)) x ) 1054 #define DEBUG_DUMP_r(x) DEBUG_r( \ 1055 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_DUMP)) x ) 1056 #define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \ 1057 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TRIE)) x ) 1058 #define DEBUG_FLAGS_r(x) DEBUG_r( \ 1059 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_FLAGS)) x ) 1060 #define DEBUG_TEST_r(x) DEBUG_r( \ 1061 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TEST)) x ) 1062 /* Execute */ 1063 #define DEBUG_EXECUTE_r(x) DEBUG_r( \ 1064 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_MASK)) x ) 1065 #define DEBUG_INTUIT_r(x) DEBUG_r( \ 1066 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_INTUIT)) x ) 1067 #define DEBUG_MATCH_r(x) DEBUG_r( \ 1068 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_MATCH)) x ) 1069 #define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \ 1070 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_TRIE)) x ) 1071 1072 /* Extra */ 1073 #define DEBUG_EXTRA_r(x) DEBUG_r( \ 1074 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_MASK)) x ) 1075 #define DEBUG_OFFSETS_r(x) DEBUG_r( \ 1076 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_OFFSETS)) x ) 1077 #define DEBUG_STATE_r(x) DEBUG_r( \ 1078 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_STATE)) x ) 1079 #define DEBUG_STACK_r(x) DEBUG_r( \ 1080 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_STACK)) x ) 1081 #define DEBUG_BUFFERS_r(x) DEBUG_r( \ 1082 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_BUFFERS)) x ) 1083 1084 #define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \ 1085 if (DEBUG_v_TEST || ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \ 1086 RE_DEBUG_FLAG(RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE))) x ) 1087 #define MJD_OFFSET_DEBUG(x) DEBUG_r( \ 1088 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_OFFDEBUG)) \ 1089 Perl_warn_nocontext x ) 1090 #define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \ 1091 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_TRIE)) x ) 1092 #define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \ 1093 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_TRIE)) x ) 1094 1095 #define DEBUG_TRIE_r(x) DEBUG_r( \ 1096 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TRIE \ 1097 | RE_DEBUG_EXECUTE_TRIE )) x ) 1098 #define DEBUG_GPOS_r(x) DEBUG_r( \ 1099 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_GPOS)) x ) 1100 1101 #define DEBUG_DUMP_PRE_OPTIMIZE_r(x) DEBUG_r( \ 1102 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE)) x ) 1103 1104 /* initialization */ 1105 /* Get the debug flags for code not in regcomp.c nor regexec.c. This doesn't 1106 * initialize the variable if it isn't already there, instead it just assumes 1107 * the flags are 0 */ 1108 #define DECLARE_AND_GET_RE_DEBUG_FLAGS_NON_REGEX \ 1109 volatile IV re_debug_flags = 0; PERL_UNUSED_VAR(re_debug_flags); \ 1110 STMT_START { \ 1111 SV * re_debug_flags_sv = NULL; \ 1112 /* get_sv() can return NULL during global destruction. */ \ 1113 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \ 1114 if (re_debug_flags_sv && SvIOK(re_debug_flags_sv)) \ 1115 re_debug_flags=SvIV(re_debug_flags_sv); \ 1116 } STMT_END 1117 1118 1119 #ifdef DEBUGGING 1120 1121 /* For use in regcomp.c and regexec.c, Get the debug flags, and initialize to 1122 * the defaults if not done already */ 1123 #define DECLARE_AND_GET_RE_DEBUG_FLAGS \ 1124 volatile IV re_debug_flags = 0; PERL_UNUSED_VAR(re_debug_flags); \ 1125 DEBUG_r({ \ 1126 SV * re_debug_flags_sv = NULL; \ 1127 /* get_sv() can return NULL during global destruction. */ \ 1128 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \ 1129 if (re_debug_flags_sv) { \ 1130 if (!SvIOK(re_debug_flags_sv)) /* If doesnt exist set to default */\ 1131 sv_setuv(re_debug_flags_sv, \ 1132 /* These defaults should be kept in sync with re.pm */ \ 1133 RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \ 1134 re_debug_flags=SvIV(re_debug_flags_sv); \ 1135 } \ 1136 }) 1137 1138 #define isDEBUG_WILDCARD (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_WILDCARD)) 1139 1140 #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \ 1141 const char * const rpv = \ 1142 pv_pretty((dsv), (pv), (l), (m), \ 1143 PL_colors[(c1)],PL_colors[(c2)], \ 1144 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \ 1145 const int rlen = SvCUR(dsv) 1146 1147 /* This is currently unsed in the core */ 1148 #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \ 1149 const char * const rpv = \ 1150 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \ 1151 PL_colors[(c1)],PL_colors[(c2)], \ 1152 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ) 1153 1154 #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \ 1155 const char * const rpv = \ 1156 pv_pretty((dsv), (pv), (l), (m), \ 1157 PL_colors[0], PL_colors[1], \ 1158 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \ 1159 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \ 1160 ) 1161 1162 #define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0)) 1163 #define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "") 1164 1165 #else /* if not DEBUGGING */ 1166 1167 #define DECLARE_AND_GET_RE_DEBUG_FLAGS dNOOP 1168 #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) dNOOP 1169 #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) 1170 #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) dNOOP 1171 #define RE_SV_DUMPLEN(ItEm) 1172 #define RE_SV_TAIL(ItEm) 1173 #define isDEBUG_WILDCARD 0 1174 1175 #endif /* DEBUG RELATED DEFINES */ 1176 1177 #define FIRST_NON_ASCII_DECIMAL_DIGIT 0x660 /* ARABIC_INDIC_DIGIT_ZERO */ 1178 1179 typedef enum { 1180 TRADITIONAL_BOUND = _CC_WORDCHAR, 1181 GCB_BOUND, 1182 LB_BOUND, 1183 SB_BOUND, 1184 WB_BOUND 1185 } bound_type; 1186 1187 /* This unpacks the FLAGS field of ANYOF[HR]x nodes. The value it contains 1188 * gives the strict lower bound for the UTF-8 start byte of any code point 1189 * matchable by the node, and a loose upper bound as well. 1190 * 1191 * The low bound is stored as 0xC0 + ((the upper 6 bits) >> 2) 1192 * The loose upper bound is determined from the lowest 2 bits and the low bound 1193 * (called x) as follows: 1194 * 1195 * 11 The upper limit of the range can be as much as (EF - x) / 8 1196 * 10 The upper limit of the range can be as much as (EF - x) / 4 1197 * 01 The upper limit of the range can be as much as (EF - x) / 2 1198 * 00 The upper limit of the range can be as much as EF 1199 * 1200 * For motivation of this design, see commit message in 1201 * 3146c00a633e9cbed741e10146662fbcedfdb8d3 */ 1202 #ifdef EBCDIC 1203 # define MAX_ANYOF_HRx_BYTE 0xF4 1204 #else 1205 # define MAX_ANYOF_HRx_BYTE 0xEF 1206 #endif 1207 #define LOWEST_ANYOF_HRx_BYTE(b) (((b) >> 2) + 0xC0) 1208 #define HIGHEST_ANYOF_HRx_BYTE(b) \ 1209 (LOWEST_ANYOF_HRx_BYTE(b) \ 1210 + ((MAX_ANYOF_HRx_BYTE - LOWEST_ANYOF_HRx_BYTE(b)) >> ((b) & 3))) 1211 1212 #endif /* PERL_REGCOMP_H_ */ 1213 1214 /* 1215 * ex: set ts=8 sts=4 sw=4 et: 1216 */ 1217