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 14#define PERL_REGCOMP_H_ 15 16#ifndef RE_PESSIMISTIC_PARENS 17/* Define this to 1 if you want to enable a really aggressive and 18 * inefficient paren cleanup during backtracking which should ensure 19 * correctness. Doing so should fix any bugs related to backreferences, 20 * at the cost of saving and restoring paren state far more than we 21 * necessarily must. 22 * 23 * When it is set to 0 we try to optimize away unnecessary save/restore 24 * operations which could potentially introduce bugs. We should pass our 25 * test suite with this as 0, but setting it to 1 might fix cases we do 26 * not currently test for. If setting this to 1 does fix a bug, then 27 * review the code related to storing and restoring paren state. 28 * 29 * See comment for VOLATILE_REF below for more details of a 30 * related case. 31 */ 32#define RE_PESSIMISTIC_PARENS 0 33#endif 34 35/* a VOLATILE_REF is a ref which is inside of a capturing group and it 36 * refers to the capturing group it is inside of or to a following capture 37 * group which might be affected by what this capture group matches, and 38 * thus the ref requires additional backtracking support. For example: 39 * 40 * "xa=xaaa" =~ /^(xa|=?\1a){2}\z/ 41 * 42 * should not match. In older perls the matching process would go like this: 43 * 44 * Iter 1: "xa" matches in capture group. 45 * Iter 2: "xa" does not match, goes to next alternation. 46 * "=" matches in =? 47 * Bifurcates here (= might not match) 48 * "xa" matches via \1 from previous iteration 49 * "a" matches via "a" at end of second alternation 50 * # at this point $1 is "=xaa" 51 * \z does not match -> backtracks. 52 * Backtracks to Iter 2 "=?" Bifurcation point where we have NOT matched "=" 53 * "=xaa" matches via \1 (as $1 has not been reset) 54 * "a" matches via "a" at end of second alternation 55 * "\z" does match. -> Pattern matches overall. 56 * 57 * What should happen and now does happen instead is: 58 * 59 * Backtracks to Iter 2 "=?" Bifurcation point where we have NOT matched "=", 60 * \1 does not match as it is "xa" (as $1 was reset when backtracked) 61 * and the current character in the string is an "=" 62 * 63 * The fact that \1 in this case is marked as a VOLATILE_REF is what ensures 64 * that we reset the capture buffer properly. 65 * 66 * See 59db194299c94c6707095797c3df0e2f67ff82b2 67 * and 38508ce8fc3a1bd12a3bb65e9d4ceb9b396a18db 68 * for more details. 69 */ 70#define VOLATILE_REF 1 71 72#include "regcharclass.h" 73 74/* Convert branch sequences to more efficient trie ops? */ 75#define PERL_ENABLE_TRIE_OPTIMISATION 1 76 77/* Be really aggressive about optimising patterns with trie sequences? */ 78#define PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION 1 79 80/* Should the optimiser take positive assertions into account? */ 81#define PERL_ENABLE_POSITIVE_ASSERTION_STUDY 0 82 83/* Not for production use: */ 84#define PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS 0 85 86/* 87 * Structure for regexp "program". This is essentially a linear encoding 88 * of a nondeterministic finite-state machine (aka syntax charts or 89 * "railroad normal form" in parsing technology). Each node is an opcode 90 * plus a "next" pointer, possibly plus an operand. "Next" pointers of 91 * all nodes except BRANCH implement concatenation; a "next" pointer with 92 * a BRANCH on both ends of it is connecting two alternatives. (Here we 93 * have one of the subtle syntax dependencies: an individual BRANCH (as 94 * opposed to a collection of them) is never concatenated with anything 95 * because of operator precedence.) The operand of some types of node is 96 * a literal string; for others, it is a node leading into a sub-FSM. In 97 * particular, the operand of a BRANCH node is the first node of the branch. 98 * (NB this is *not* a tree structure: the tail of the branch connects 99 * to the thing following the set of BRANCHes.) The opcodes are defined 100 * in regnodes.h which is generated from regcomp.sym by regcomp.pl. 101 */ 102 103/* 104 * A node is one char of opcode followed by two chars of "next" pointer. 105 * "Next" pointers are stored as two 8-bit pieces, high order first. The 106 * value is a positive offset from the opcode of the node containing it. 107 * An operand, if any, simply follows the node. (Note that much of the 108 * code generation knows about this implicit relationship.) 109 * 110 * Using two bytes for the "next" pointer is vast overkill for most things, 111 * but allows patterns to get big without disasters. 112 * 113 * [The "next" pointer is always aligned on an even 114 * boundary, and reads the offset directly as a short.] 115 */ 116 117/* This is the stuff that used to live in regexp.h that was truly 118 private to the engine itself. It now lives here. */ 119 120typedef struct regexp_internal { 121 regnode *regstclass; /* Optional startclass as identified or constructed 122 by the optimiser */ 123 struct reg_data *data; /* Additional miscellaneous data used by the program. 124 Used to make it easier to clone and free arbitrary 125 data that the regops need. Often the ARG field of 126 a regop is an index into this structure. NOTE the 127 0th element of this structure is NEVER used and is 128 strictly reserved for internal purposes. */ 129 struct reg_code_blocks *code_blocks;/* positions of literal (?{}) */ 130 U32 proglen; /* size of the compiled program in regnodes */ 131 U32 name_list_idx; /* Optional data index of an array of paren names, 132 only valid when RXp_PAREN_NAMES(prog) is true, 133 0 means "no value" like any other index into the 134 data array.*/ 135 regnode program[1]; /* Unwarranted chumminess with compiler. */ 136} regexp_internal; 137 138#define RXi_SET(x,y) (x)->pprivate = (void*)(y) 139#define RXi_GET(x) ((regexp_internal *)((x)->pprivate)) 140#define RXi_GET_DECL(r,ri) regexp_internal *ri = RXi_GET(r) 141#define RXi_GET_DECL_NULL(r,ri) regexp_internal *ri = (r) ? RXi_GET(r) : NULL 142/* 143 * Flags stored in regexp->intflags 144 * These are used only internally to the regexp engine 145 * 146 * See regexp.h for flags used externally to the regexp engine 147 */ 148#define RXp_INTFLAGS(rx) ((rx)->intflags) 149#define RX_INTFLAGS(prog) RXp_INTFLAGS(ReANY(prog)) 150 151#define PREGf_SKIP 0x00000001 152#define PREGf_IMPLICIT 0x00000002 /* Converted .* to ^.* */ 153#define PREGf_NAUGHTY 0x00000004 /* how exponential is this pattern? */ 154#define PREGf_VERBARG_SEEN 0x00000008 155#define PREGf_CUTGROUP_SEEN 0x00000010 156#define PREGf_USE_RE_EVAL 0x00000020 /* compiled with "use re 'eval'" */ 157/* these used to be extflags, but are now intflags */ 158#define PREGf_NOSCAN 0x00000040 159 /* spare */ 160#define PREGf_GPOS_SEEN 0x00000100 161#define PREGf_GPOS_FLOAT 0x00000200 162 163#define PREGf_ANCH_MBOL 0x00000400 164#define PREGf_ANCH_SBOL 0x00000800 165#define PREGf_ANCH_GPOS 0x00001000 166#define PREGf_RECURSE_SEEN 0x00002000 167#define PREGf_PESSIMIZE_SEEN 0x00004000 168 169#define PREGf_ANCH \ 170 ( PREGf_ANCH_SBOL | PREGf_ANCH_GPOS | PREGf_ANCH_MBOL ) 171 172/* this is where the old regcomp.h started */ 173 174 175/* Define the various regnode structures. These all should be a multiple 176 * of 32 bits large, and they should by and large correspond with each other 177 * in terms of naming, etc. Things can and will break in subtle ways if you 178 * change things without care. If you look at regexp.h you will see it 179 * contains this: 180 * 181 * union regnode_head { 182 * struct { 183 * union { 184 * U8 flags; 185 * U8 str_len_u8; 186 * U8 first_byte; 187 * } u_8; 188 * U8 type; 189 * U16 next_off; 190 * } data; 191 * U32 data_u32; 192 * }; 193 * 194 * struct regnode { 195 * union regnode_head head; 196 * }; 197 * 198 * Which really is a complicated and alignment friendly version of 199 * 200 * struct { 201 * U8 flags; 202 * U8 type; 203 * U16 next_off; 204 * }; 205 * 206 * This structure is the base unit of elements in the regexp program. 207 * When we increment our way through the program we increment by the 208 * size of this structure (32 bits), and in all cases where regnode 209 * sizing is considered it is in units of this structure. All regnodes 210 * have a union regnode_head as their first parameter. 211 * 212 * This implies that no regnode style structure should contain 64 bit 213 * aligned members. Since the base regnode is 32 bits any member might 214 * not be 64 bit aligned no matter how you might try to pad out the 215 * struct itself (the regnode_ssc is special in this regard as it is 216 * never used in a program directly). If you want to store 64 bit 217 * members you need to store them specially. The struct regnode_p and the 218 * ARGp() and ARGp_SET() macros and related inline functions provide an example 219 * solution. Note they deal with a slightly more complicated problem than simple 220 * alignment, as pointers may be 32 bits or 64 bits depending on platform, 221 * but they illustrate the pattern to follow if you want to put a 64 bit value 222 * into a regnode. 223 224 * NOTE: Ideally we do not put pointers into the regnodes in a program. Instead 225 * we put them in the "data" part of the regexp structure and store the index into 226 * the data in the pointers in the regnode. This allows the pointer to be handled 227 * properly during clone/free operations (eg refcount bookkeeping). See S_add_data(), 228 * Perl_regdupe_internal(), Perl_regfree_internal() in regcomp.c for how the data 229 * array can be used, the letters 'arsSu' all refer to different types of SV that 230 * we already have support for in the data array. 231 */ 232 233union regnode_arg { 234 I32 i32; 235 U32 u32; 236 struct { 237 U16 u16a; 238 U16 u16b; 239 } hi_lo; 240}; 241 242 243struct regnode_string { 244 union regnode_head head; 245 char string[1]; 246}; 247 248struct regnode_lstring { /* Constructed this way to keep the string aligned. */ 249 union regnode_head head; 250 U32 str_len_u32; /* Only 18 bits allowed before would overflow 'next_off' */ 251 char string[1]; 252}; 253 254struct regnode_anyofhs { /* Constructed this way to keep the string aligned. */ 255 union regnode_head head; 256 union regnode_arg arg1; 257 char string[1]; 258}; 259 260/* Argument bearing node - workhorse, ARG1u() is often used for the data field 261 * Can store either a signed 32 bit value via ARG1i() or unsigned 32 bit value 262 * via ARG1u(), or two unsigned 16 bit values via ARG1a() or ARG1b() 263 */ 264struct regnode_1 { 265 union regnode_head head; 266 union regnode_arg arg1; 267}; 268 269/* Node whose argument is 'SV *'. This needs to be used very carefully in 270 * situations where pointers won't become invalid because of, say re-mallocs. 271 * 272 * Note that this regnode type is problematic and should not be used or copied 273 * and will be removed in the future. Pointers should be stored in the data[] 274 * array and an index into the data array stored in the regnode, which allows the 275 * pointers to be handled properly during clone/free operations on the regexp 276 * data structure. As a byproduct it also saves space, often we use a 16 bit 277 * member to store indexes into the data[] array. 278 * 279 * Also note that the weird storage here is because regnodes are 32 bit aligned, 280 * which means we cannot have a 64 bit aligned member. To make things more annoying 281 * the size of a pointer may vary by platform. Thus we use a character array, and 282 * then use inline functions to copy the data in or out. 283 * */ 284struct regnode_p { 285 union regnode_head head; 286 char arg1_sv_ptr_bytes[sizeof(SV *)]; 287}; 288 289/* "Two Node" - similar to a regnode_1 but with space for an extra 32 290 * bit value, or two 16 bit valus. The first fields must match regnode_1. 291 * Extra field can be accessed as (U32)ARG2u() (I32)ARG2i() or (U16)ARG2a() 292 * and (U16)ARG2b() */ 293struct regnode_2 { 294 union regnode_head head; 295 union regnode_arg arg1; 296 union regnode_arg arg2; 297}; 298 299/* "Three Node" - similar to a regnode_2 but with space for an additional 300 * 32 bit value, or two 16 bit values. The first fields must match regnode_2. 301 * The extra field can be accessed as (U32)ARG3u() (I32)ARG3i() or (U16)ARG3a() 302 * and (U16)ARG3b(). 303 * Currently used for the CURLY style regops used to represent quantifers, 304 * storing the min and of the quantifier via ARG1i() and ARG2i(), along with 305 * ARG3a() and ARG3b() which are used to store information about the number of 306 * parens before and inside the quantified expression. */ 307struct regnode_3 { 308 union regnode_head head; 309 union regnode_arg arg1; 310 union regnode_arg arg2; 311 union regnode_arg arg3; 312}; 313 314#define REGNODE_BBM_BITMAP_LEN \ 315 /* 6 info bits requires 64 bits; 5 => 32 */ \ 316 ((1 << (UTF_CONTINUATION_BYTE_INFO_BITS)) / CHARBITS) 317 318/* Used for matching any two-byte UTF-8 character whose start byte is known. 319 * The array is a bitmap capable of representing any possible continuation 320 * byte. */ 321struct regnode_bbm { 322 union regnode_head head; 323 U8 bitmap[REGNODE_BBM_BITMAP_LEN]; 324}; 325 326#define ANYOF_BITMAP_SIZE (NUM_ANYOF_CODE_POINTS / CHARBITS) 327 328/* Note that these form structs which are supersets of the next smaller one, by 329 * appending fields. Alignment problems can occur if one of those optional 330 * fields requires stricter alignment than the base struct. And formal 331 * parameters that can really be two or more of the structs should be 332 * declared as the smallest one it could be. See commit message for 333 * 7dcac5f6a5195002b55c935ee1d67f67e1df280b. Regnode allocation is done 334 * without regard to alignment, and changing it to would also require changing 335 * the code that inserts and deletes regnodes. The basic single-argument 336 * regnode has a U32, which is what reganode() allocates as a unit. Therefore 337 * no field can require stricter alignment than U32. */ 338 339/* also used by trie */ 340struct regnode_charclass { 341 union regnode_head head; 342 union regnode_arg arg1; 343 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */ 344}; 345 346/* has runtime (locale) \d, \w, ..., [:posix:] classes */ 347struct regnode_charclass_posixl { 348 union regnode_head head; 349 union regnode_arg arg1; 350 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */ 351 U32 classflags; /* and run-time */ 352}; 353 354/* A synthetic start class (SSC); is a regnode_charclass_posixl_fold, plus an 355 * extra SV*, used only during regex construction and which is not used by the 356 * main machinery in regexec.c and which does not get embedded in the final compiled 357 * regex program. 358 * 359 * Because it does not get embedded it does not have to comply with the alignment 360 * and sizing constraints required for a normal regnode structure: it MAY contain 361 * pointers or members of whatever size needed and the compiler will do the right 362 * thing. (Every other regnode type is 32 bit aligned.) 363 * 364 * Note that the 'next_off' field is unused, as the SSC stands alone, so there is 365 * never a next node. 366 */ 367struct regnode_ssc { 368 union regnode_head head; 369 union regnode_arg arg1; 370 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */ 371 U32 classflags; /* ... and run-time */ 372 373 /* Auxiliary, only used during construction; NULL afterwards: list of code 374 * points matched */ 375 SV* invlist; 376}; 377 378/* We take advantage of 'next_off' not otherwise being used in the SSC by 379 * actually using it: by setting it to 1. This allows us to test and 380 * distinguish between an SSC and other ANYOF node types, as 'next_off' cannot 381 * otherwise be 1, because it is the offset to the next regnode expressed in 382 * units of regnodes. Since an ANYOF node contains extra fields, it adds up 383 * to 12 regnode units on 32-bit systems, (hence the minimum this can be (if 384 * not 0) is 11 there. Even if things get tightly packed on a 64-bit system, 385 * it still would be more than 1. */ 386#define set_ANYOF_SYNTHETIC(n) \ 387 STMT_START{ \ 388 OP(n) = ANYOF; \ 389 NEXT_OFF(n) = 1; \ 390 } STMT_END 391 392#define is_ANYOF_SYNTHETIC(n) (REGNODE_TYPE(OP(n)) == ANYOF && NEXT_OFF(n) == 1) 393 394/* XXX fix this description. 395 Impose a limit of REG_INFTY on various pattern matching operations 396 to limit stack growth and to avoid "infinite" recursions. 397*/ 398/* The default size for REG_INFTY is I32_MAX, which is the same as UINT_MAX 399 (see perl.h). Unfortunately I32 isn't necessarily 32 bits (see handy.h). 400 On the Cray C90, or Cray T90, I32_MAX is considerably larger than it 401 might be elsewhere. To limit stack growth to reasonable sizes, supply a 402 smaller default. 403 --Andy Dougherty 11 June 1998 404 --Amended by Yves Orton 15 Jan 2023 405*/ 406#if INTSIZE > 4 407# ifndef REG_INFTY 408# define REG_INFTY nBIT_IMAX(32) 409# endif 410#endif 411 412#ifndef REG_INFTY 413# define REG_INFTY I32_MAX 414#endif 415 416#define ARG_VALUE(arg) (arg) 417#define ARG__SET(arg,val) ((arg) = (val)) 418 419#undef ARG 420#undef ARG1 421#undef ARG2 422 423/* convention: each arg is is 32 bits, with the "u" suffix 424 * being unsigned 32 bits, the "i" suffix being signed 32 bits, 425 * and the "a" and "b" suffixes being unsigned 16 bit fields. 426 * 427 * We provide all 4 macros for each case for consistency, even 428 * though they arent all used. 429 */ 430 431#define ARG1u(p) ARG_VALUE(ARG1u_LOC(p)) 432#define ARG1i(p) ARG_VALUE(ARG1i_LOC(p)) 433#define ARG1a(p) ARG_VALUE(ARG1a_LOC(p)) 434#define ARG1b(p) ARG_VALUE(ARG1b_LOC(p)) 435 436#define ARG2u(p) ARG_VALUE(ARG2u_LOC(p)) 437#define ARG2i(p) ARG_VALUE(ARG2i_LOC(p)) 438#define ARG2a(p) ARG_VALUE(ARG2a_LOC(p)) 439#define ARG2b(p) ARG_VALUE(ARG2b_LOC(p)) 440 441#define ARG3u(p) ARG_VALUE(ARG3u_LOC(p)) 442#define ARG3i(p) ARG_VALUE(ARG3i_LOC(p)) 443#define ARG3a(p) ARG_VALUE(ARG3a_LOC(p)) 444#define ARG3b(p) ARG_VALUE(ARG3b_LOC(p)) 445 446#define ARGp(p) ARGp_VALUE_inline(p) 447 448#define ARG1u_SET(p, val) ARG__SET(ARG1u_LOC(p), (val)) 449#define ARG1i_SET(p, val) ARG__SET(ARG1i_LOC(p), (val)) 450#define ARG1a_SET(p, val) ARG__SET(ARG1a_LOC(p), (val)) 451#define ARG1b_SET(p, val) ARG__SET(ARG1b_LOC(p), (val)) 452 453#define ARG2u_SET(p, val) ARG__SET(ARG2u_LOC(p), (val)) 454#define ARG2i_SET(p, val) ARG__SET(ARG2i_LOC(p), (val)) 455#define ARG2a_SET(p, val) ARG__SET(ARG2a_LOC(p), (val)) 456#define ARG2b_SET(p, val) ARG__SET(ARG2b_LOC(p), (val)) 457 458#define ARG3u_SET(p, val) ARG__SET(ARG3u_LOC(p), (val)) 459#define ARG3i_SET(p, val) ARG__SET(ARG3i_LOC(p), (val)) 460#define ARG3a_SET(p, val) ARG__SET(ARG3a_LOC(p), (val)) 461#define ARG3b_SET(p, val) ARG__SET(ARG3b_LOC(p), (val)) 462 463#define ARGp_SET(p, val) ARGp_SET_inline((p),(val)) 464 465/* the following define was set to 0xde in 075abff3 466 * as part of some linting logic. I have set it to 0 467 * as otherwise in every place where we /might/ set flags 468 * we have to set it 0 explicitly, which duplicates 469 * assignments and IMO adds an unacceptable level of 470 * surprise to working in the regex engine. If this 471 * is changed from 0 then at the very least make sure 472 * that SBOL for /^/ sets the flags to 0 explicitly. 473 * -- Yves */ 474 475#define NODE_ALIGN(node) 476#define SIZE_ALIGN NODE_ALIGN 477 478#undef OP 479#undef OPERAND 480#undef STRING 481#undef NEXT_OFF 482#undef NODE_ALIGN 483 484#define NEXT_OFF(p) ((p)->head.data.next_off) 485#define OP(p) ((p)->head.data.type) 486#define STR_LEN_U8(p) ((p)->head.data.u_8.str_len_u8) 487#define FIRST_BYTE(p) ((p)->head.data.u_8.first_byte) 488#define FLAGS(p) ((p)->head.data.u_8.flags) /* Caution: Doesn't apply to all \ 489 regnode types. For some, it's the \ 490 character set of the regnode */ 491#define STR_LENs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \ 492 STR_LEN_U8((struct regnode_string *)p)) 493#define STRINGs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \ 494 ((struct regnode_string *)p)->string) 495#define OPERANDs(p) STRINGs(p) 496 497#define PARNO(p) ARG1u(p) /* APPLIES for OPEN and CLOSE only */ 498 499#define NODE_ALIGN_FILL(node) (FLAGS(node) = 0) 500 501/* Long strings. Currently limited to length 18 bits, which handles a 262000 502 * byte string. The limiting factor is the 16 bit 'next_off' field, which 503 * points to the next regnode, so the furthest away it can be is 2**16. On 504 * most architectures, regnodes are 2**2 bytes long, so that yields 2**18 505 * bytes. Should a longer string be desired, we could increase it to 26 bits 506 * fairly easily, by changing this node to have longj type which causes the ARG 507 * field to be used for the link to the next regnode (although code would have 508 * to be changed to account for this), and then use a combination of the flags 509 * and next_off fields for the length. To get 34 bit length, also change the 510 * node to be an ARG2L, using the second 32 bit field for the length, and not 511 * using the flags nor next_off fields at all. One could have an llstring node 512 * and even an lllstring type. */ 513#define STR_LENl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 514 (((struct regnode_lstring *)p)->str_len_u32)) 515#define STRINGl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 516 (((struct regnode_lstring *)p)->string)) 517#define OPERANDl(p) STRINGl(p) 518 519#define STR_LEN(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 520 ? STR_LENl(p) : STR_LENs(p)) 521#define STRING(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 522 ? STRINGl(p) : STRINGs(p)) 523#define OPERAND(p) STRING(p) 524 525/* The number of (smallest) regnode equivalents that a string of length l bytes 526 * occupies - Used by the REGNODE_AFTER() macros and functions. */ 527#define STR_SZ(l) (((l) + sizeof(regnode) - 1) / sizeof(regnode)) 528 529#define setSTR_LEN(p,v) \ 530 STMT_START{ \ 531 if (OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \ 532 ((struct regnode_lstring *)(p))->str_len_u32 = (v); \ 533 else \ 534 STR_LEN_U8((struct regnode_string *)(p)) = (v); \ 535 } STMT_END 536 537#define ANYOFR_BASE_BITS 20 538#define ANYOFRbase(p) (ARG1u(p) & nBIT_MASK(ANYOFR_BASE_BITS)) 539#define ANYOFRdelta(p) (ARG1u(p) >> ANYOFR_BASE_BITS) 540 541#undef NODE_ALIGN 542#undef ARG_LOC 543 544#define NODE_ALIGN(node) 545#define ARGp_BYTES_LOC(p) (((struct regnode_p *)p)->arg1_sv_ptr_bytes) 546#define ARG1u_LOC(p) (((struct regnode_1 *)p)->arg1.u32) 547#define ARG1i_LOC(p) (((struct regnode_1 *)p)->arg1.i32) 548#define ARG1a_LOC(p) (((struct regnode_1 *)p)->arg1.hi_lo.u16a) 549#define ARG1b_LOC(p) (((struct regnode_1 *)p)->arg1.hi_lo.u16b) 550#define ARG2u_LOC(p) (((struct regnode_2 *)p)->arg2.u32) 551#define ARG2i_LOC(p) (((struct regnode_2 *)p)->arg2.i32) 552#define ARG2a_LOC(p) (((struct regnode_2 *)p)->arg2.hi_lo.u16a) 553#define ARG2b_LOC(p) (((struct regnode_2 *)p)->arg2.hi_lo.u16b) 554#define ARG3u_LOC(p) (((struct regnode_3 *)p)->arg3.u32) 555#define ARG3i_LOC(p) (((struct regnode_3 *)p)->arg3.i32) 556#define ARG3a_LOC(p) (((struct regnode_3 *)p)->arg3.hi_lo.u16a) 557#define ARG3b_LOC(p) (((struct regnode_3 *)p)->arg3.hi_lo.u16b) 558 559/* These should no longer be used directly in most cases. Please use 560 * the REGNODE_AFTER() macros instead. */ 561#define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */ 562 563/* Core macros for computing "the regnode after this one". See also 564 * Perl_regnode_after() in reginline.h 565 * 566 * At the struct level regnodes are a linked list, with each node pointing 567 * at the next (via offsets), usually via the C<next_off> field in the 568 * structure. Where there is a need for a node to have two children the 569 * immediate physical successor of the node in the compiled program is used 570 * to represent one of them. A good example is the BRANCH construct, 571 * consider the pattern C</head(?:[ab]foo|[cd]bar)tail/> 572 * 573 * 1: EXACT <head> (3) 574 * 3: BRANCH (8) 575 * 4: ANYOFR[ab] (6) 576 * 6: EXACT <foo> (14) 577 * 8: BRANCH (FAIL) 578 * 9: ANYOFR[cd] (11) 579 * 11: EXACT <bar> (14) 580 * 13: TAIL (14) 581 * 14: EXACT <tail> (16) 582 * 16: END (0) 583 * 584 * The numbers in parens at the end of each line show the "next_off" value 585 * for that regnode in the program. We can see that the C<next_off> of 586 * the first BRANCH node (#3) is the second BRANCH node (#8), and indicates 587 * where execution should go if the regnodes *following* the BRANCH node fail 588 * to accept the input string. Thus to find the "next BRANCH" we would do 589 * C<Perl_regnext()> and follow the C<next_off> pointer, and to find 590 * the "BRANCHes contents" we would use C<REGNODE_AFTER()>. 591 * 592 * Be aware that C<REGNODE_AFTER()> is not guaranteed to give a *useful* 593 * result once the regex peephole optimizer has run (it will be correct 594 * however!). By the time code in regexec.c executes various regnodes 595 * may have been optimized out of the C<next_off> chain. An example 596 * can be seen above, node 13 will never be reached during execution 597 * flow as it has been stitched out of the C<next_off> chain. Both 6 and 598 * 11 would have pointed at it during compilation, but it exists only to 599 * facilitate the construction of the BRANCH structure and is effectively 600 * a NOOP, and thus the optimizer adjusts the links so it is skipped 601 * from execution time flow. In regexec.c it is only safe to use 602 * REGNODE_AFTER() on specific node types. 603 * 604 * Conversely during compilation C<Perl_regnext()> may not work properly 605 * as the C<next_off> may not be known until "later", (such as in the 606 * case of BRANCH nodes) and thus in regcomp.c the REGNODE_AFTER() macro 607 * is used very heavily instead. 608 * 609 * There are several variants of the REGNODE_AFTER_xxx() macros which 610 * are intended for use in different situations depending on how 611 * confident the code is about what type of node it is trying to find a 612 * successor for. 613 * 614 * So for instance if you know you are dealing with a known node type of 615 * constant size then you should use REGNODE_AFTER_type(n,TYPE). 616 * 617 * If you have a regnode pointer and you know you are dealing with a 618 * regnode type of constant size and you have already extracted its 619 * opcode use: REGNODE_AFTER_opcode(n,OPCODE). 620 * 621 * If you have a regnode and you know it is variable size then you 622 * you can produce optimized code by using REGNODE_AFTER_varies(n). 623 * 624 * If you have a regnode pointer and nothing else use: REGNODE_AFTER(n) 625 * This is the safest option and wraps C<Perl_regnode_after()>. It 626 * should produce the correct result regardless of its argument. The 627 * other options only produce correct results under specific 628 * constraints. 629 */ 630#define REGNODE_AFTER_PLUS(p,extra) ((p) + NODE_STEP_REGNODE + (extra)) 631/* under DEBUGGING we check that all REGNODE_AFTER optimized macros did the 632 * same thing that Perl_regnode_after() would have done. Note that when 633 * not compiled under DEBUGGING the assert_() macro is empty. Thus we 634 * don't have to implement different versions for DEBUGGING and not DEBUGGING, 635 * and explains why all the macros use REGNODE_AFTER_PLUS_DEBUG() under the 636 * hood. */ 637#define REGNODE_AFTER_PLUS_DEBUG(p,extra) \ 638 (assert_(check_regnode_after(p,extra)) REGNODE_AFTER_PLUS((p),(extra))) 639 640/* find the regnode after this p by using the opcode we previously extracted 641 * with OP(p) */ 642#define REGNODE_AFTER_opcode(p,op) REGNODE_AFTER_PLUS_DEBUG((p),REGNODE_ARG_LEN(op)) 643 644/* find the regnode after this p by using the size of the struct associated with 645 * the opcode for p. use this when you *know* that p is pointer to a given type*/ 646#define REGNODE_AFTER_type(p,t) REGNODE_AFTER_PLUS_DEBUG((p),EXTRA_SIZE(t)) 647 648/* find the regnode after this p by using OP(p) to find the regnode type of p */ 649#define REGNODE_AFTER_varies(p) regnode_after(p,TRUE) 650 651/* find the regnode after this p by using OP(p) to find the regnode type of p */ 652#define REGNODE_AFTER(p) regnode_after(p,FALSE) 653 654 655/* REGNODE_BEFORE() is trickier to deal with in terms of validation, execution. 656 * All the places that use it assume that p will be one struct regnode large. 657 * So to validate it we do the math to go backwards and then validate that the 658 * type of regnode we landed on is actually one regnode large. In theory if 659 * things go wrong the opcode should be illegal or say the item should be larger 660 * than it is, etc. */ 661#define REGNODE_BEFORE_BASE(p) ((p) - NODE_STEP_REGNODE) 662#define REGNODE_BEFORE_BASE_DEBUG(p) \ 663 (assert_(check_regnode_after(REGNODE_BEFORE_BASE(p),0)) REGNODE_BEFORE_BASE(p)) 664#define REGNODE_BEFORE(p) REGNODE_BEFORE_BASE_DEBUG(p) 665 666#define FILL_NODE(offset, op) \ 667 STMT_START { \ 668 OP(REGNODE_p(offset)) = op; \ 669 NEXT_OFF(REGNODE_p(offset)) = 0; \ 670 } STMT_END 671#define FILL_ADVANCE_NODE(offset, op) \ 672 STMT_START { \ 673 FILL_NODE(offset, op); \ 674 (offset)++; \ 675 } STMT_END 676#define FILL_ADVANCE_NODE_ARG1u(offset, op, arg) \ 677 STMT_START { \ 678 ARG1u_SET(REGNODE_p(offset), arg); \ 679 FILL_ADVANCE_NODE(offset, op); \ 680 /* This is used generically for other operations \ 681 * that have a longer argument */ \ 682 (offset) += REGNODE_ARG_LEN(op); \ 683 } STMT_END 684#define FILL_ADVANCE_NODE_ARGp(offset, op, arg) \ 685 STMT_START { \ 686 ARGp_SET(REGNODE_p(offset), arg); \ 687 FILL_ADVANCE_NODE(offset, op); \ 688 (offset) += REGNODE_ARG_LEN(op); \ 689 } STMT_END 690#define FILL_ADVANCE_NODE_2ui_ARG(offset, op, arg1, arg2) \ 691 STMT_START { \ 692 ARG1u_SET(REGNODE_p(offset), arg1); \ 693 ARG2i_SET(REGNODE_p(offset), arg2); \ 694 FILL_ADVANCE_NODE(offset, op); \ 695 (offset) += 2; \ 696 } STMT_END 697 698/* define these after we define the normal macros, so we can use 699 * ARGp_BYTES_LOC(n) */ 700 701static inline SV * 702ARGp_VALUE_inline(struct regnode *node) { 703 SV *ptr; 704 memcpy(&ptr, ARGp_BYTES_LOC(node), sizeof(ptr)); 705 706 return ptr; 707} 708 709static inline void 710ARGp_SET_inline(struct regnode *node, SV *ptr) { 711 memcpy(ARGp_BYTES_LOC(node), &ptr, sizeof(ptr)); 712} 713 714#define REG_MAGIC 0234 715 716/* An ANYOF node matches a single code point based on specified criteria. It 717 * now comes in several styles, but originally it was just a 256 element 718 * bitmap, indexed by the code point (which was always just a byte). If the 719 * corresponding bit for a code point is 1, the code point matches; if 0, it 720 * doesn't match (complemented if inverted). This worked fine before Unicode 721 * existed, but making a bit map long enough to accommodate a bit for every 722 * possible Unicode code point is prohibitively large. Therefore it is made 723 * much much smaller, and an inversion list is created to handle code points 724 * not represented by the bitmap. (It is now possible to compile the bitmap to 725 * a larger size to avoid the slower inversion list lookup for however big the 726 * bitmap is set to, but this is rarely done). If the bitmap is sufficient to 727 * specify all possible matches (with nothing outside it matching), no 728 * inversion list is needed nor included, and the argument to the ANYOF node is 729 * set to the following: */ 730 731#define ANYOF_MATCHES_ALL_OUTSIDE_BITMAP_VALUE U32_MAX 732#define ANYOF_MATCHES_ALL_OUTSIDE_BITMAP(node) \ 733 (ARG1u(node) == ANYOF_MATCHES_ALL_OUTSIDE_BITMAP_VALUE) 734 735#define ANYOF_MATCHES_NONE_OUTSIDE_BITMAP_VALUE \ 736 /* Assumes ALL is odd */ (ANYOF_MATCHES_ALL_OUTSIDE_BITMAP_VALUE - 1) 737#define ANYOF_MATCHES_NONE_OUTSIDE_BITMAP(node) \ 738 (ARG1u(node) == ANYOF_MATCHES_NONE_OUTSIDE_BITMAP_VALUE) 739 740#define ANYOF_ONLY_HAS_BITMAP_MASK ANYOF_MATCHES_NONE_OUTSIDE_BITMAP_VALUE 741#define ANYOF_ONLY_HAS_BITMAP(node) \ 742 ((ARG1u(node) & ANYOF_ONLY_HAS_BITMAP_MASK) == ANYOF_ONLY_HAS_BITMAP_MASK) 743 744#define ANYOF_HAS_AUX(node) (! ANYOF_ONLY_HAS_BITMAP(node)) 745 746/* There are also ANYOFM nodes, used when the bit patterns representing the 747 * matched code points happen to be such that they can be checked by ANDing 748 * with a mask. The regex compiler looks for and silently optimizes to using 749 * this node type in the few cases where it works out. The eight octal digits 750 * form such a group. These nodes are simple and fast and no further 751 * discussion is needed here. 752 * 753 * And, there are ANYOFH-ish nodes which match only code points that aren't in 754 * the bitmap (the H stands for High). These are common for expressing 755 * Unicode properties concerning non-Latin scripts. They dispense with the 756 * bitmap altogether and don't need any of the flags discussed below. 757 * 758 * And, there are ANYOFR-ish nodes which match within a single range. 759 * 760 * When there is a need to specify what matches outside the bitmap, it is done 761 * by allocating an AV as part of the pattern's compiled form, and the argument 762 * to the node instead of being ANYOF_ONLY_HAS_BITMAP, points to that AV. 763 * 764 * (Actually, that is an oversimplification. The AV is placed into the 765 * pattern's struct reg_data, and what is stored in the node's argument field 766 * is its index into that struct. And the inversion list is just one element, 767 * the zeroth, of the AV.) 768 * 769 * There are certain situations where a single inversion list can't handle all 770 * the complexity. These are dealt with by having extra elements in the AV, by 771 * specifying flag bits in the ANYOF node, and/or special code. As an example, 772 * there are instances where what the ANYOF node matches is not completely 773 * known until runtime. In these cases, a flag is set, and the bitmap has a 1 774 * for the code points which are known at compile time to be 1, and a 0 for the 775 * ones that are known to be 0, or require runtime resolution. Some missing 776 * information can be found by merely seeing if the pattern is UTF-8 or not; 777 * other cases require looking at the extra elements in the AV. 778 * 779 * There are 5 cases where the bitmap is insufficient. These are specified by 780 * flags in the node's flags field. We could use five bits to represent the 5 781 * cases, but to save flags bits (which are perennially in short supply), we 782 * play some games. The cases are: 783 * 784 * 1) As already mentioned, if some code points outside the bitmap match, and 785 * some do not, an inversion list is specified to indicate which ones. 786 * 787 * 2) Under /d rules, it can happen that code points that are in the upper 788 * latin1 range (\x80-\xFF or their equivalents on EBCDIC platforms) match 789 * only if the runtime target string being matched against is UTF-8. For 790 * example /[\w[:punct:]]/d. This happens only for certain posix classes, 791 * and all such ones also have above-bitmap matches. 792 * 793 * Note that /d rules are no longer encouraged; 'use 5.14' or higher 794 * deselects them. But they are still supported, and a flag is required 795 * so that they can be properly handled. But it can be a shared flag: see 796 * 4) below. 797 * 798 * 3) Also under /d rules, something like /[\Wfoo]/ will match everything in 799 * the \x80-\xFF range, unless the string being matched against is UTF-8. 800 * An inversion list could be created for this case, but this is 801 * relatively common, and it turns out that it's all or nothing: if any 802 * one of these code points matches, they all do. Hence a single bit 803 * suffices. We use a shared flag that doesn't take up space by itself: 804 * ANYOFD_NON_UTF8_MATCHES_ALL_NON_ASCII__shared. This also means there 805 * is an inversion list for the things that don't fit into the bitmap. 806 * 807 * 4) A user-defined \p{} property may not have been defined by the time the 808 * regex is compiled. In this case, we don't know until runtime what it 809 * will match, so we have to assume it could match anything, including 810 * code points that ordinarily would be in the bitmap. A flag bit is 811 * necessary to indicate this, though we can use the 812 * ANYOF_HAS_EXTRA_RUNTIME_MATCHES flag, along with the node not being 813 * ANYOFD. The information required to construct the property is stored 814 * in the AV pointed to by the node's argument. This case is quite 815 * uncommon in the field, and the /(?[...])/ construct is a better way to 816 * accomplish what this feature does. 817 * 818 * 5) /[foo]/il may have folds that are only valid if the runtime locale is a 819 * UTF-8 one. The ANYOF_HAS_EXTRA_RUNTIME_MATCHES flag can also be used 820 * for these. The list is stored in a different element of the AV, so its 821 * existence differentiates this case from that of 4), along with the node 822 * being ANYOFL, with the ANYOFL_FOLD flag being set. There are a few 823 * additional folds valid only if the UTF-8 locale is a Turkic one which 824 * is tested for explicitly. 825 * 826 * Note that the user-defined property flag and the /il flag can affect whether 827 * an ASCII character matches in the bitmap or not. 828 * 829 * And this still isn't the end of the story. In some cases, warnings are 830 * supposed to be raised when matching certain categories of code points in the 831 * target string. Flags are set to indicate this. This adds up to a bunch of 832 * flags required, and we only have 8 available. That is why we share some. 833 * At the moment, there are two spare flag bits, but this could be increased by 834 * various tricks: 835 * 836 * ANYOF_MATCHES_POSIXL is redundant with the node type ANYOFPOSIXL. That flag 837 * could be removed, but at the expense of having to write extra code, which 838 * would take up space, and writing this turns out to be not hard, but not 839 * trivial. 840 * 841 * If this is done, an extension would be to make all ANYOFL nodes contain the 842 * extra 32 bits that ANYOFPOSIXL ones do, doubling each instance's size. The 843 * posix flags only occupy 30 bits, so the ANYOFL_FOLD and 844 * ANYOFL_UTF8_LOCALE_REQD bits could be moved to that extra space, but it 845 * would also mean extra instructions, as there are currently places in the 846 * code that assume those two bits are zero. 847 * 848 * Some flags are not used in synthetic start class (SSC) nodes, so could be 849 * shared should new flags be needed for SSCs, like SSC_MATCHES_EMPTY_STRING 850 * now. */ 851 852/* If this is set, the result of the match should be complemented. regexec.c 853 * is expecting this to be in the low bit. Never in an SSC */ 854#define ANYOF_INVERT 0x01 855 856/* For the SSC node only, which cannot be inverted, so is shared with that bit. 857 * This is used only during regex compilation. */ 858#define SSC_MATCHES_EMPTY_STRING ANYOF_INVERT 859 860/* Set if this is a regnode_charclass_posixl vs a regnode_charclass. This 861 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale 862 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved 863 * at compile-time. Only set under /l; can be in SSC */ 864#define ANYOF_MATCHES_POSIXL 0x02 865 866/* The fold is calculated and stored in the bitmap where possible at compile 867 * time. However under locale, the actual folding varies depending on 868 * what the locale is at the time of execution, so it has to be deferred until 869 * then. Only set under /l; never in an SSC */ 870#define ANYOFL_FOLD 0x04 871 872/* Warn if the runtime locale isn't a UTF-8 one (and the generated node assumes 873 * a UTF-8 locale. */ 874#define ANYOFL_UTF8_LOCALE_REQD 0x08 875 876/* Spare: Be sure to change ANYOF_FLAGS_ALL if this gets used 0x10 */ 877 878/* Spare: Be sure to change ANYOF_FLAGS_ALL if this gets used 0x20 */ 879 880/* Shared bit that indicates that there are potential additional matches stored 881 * outside the bitmap, as pointed to by the AV given by the node's argument. 882 * The node type is used at runtime (in conjunction with this flag and other 883 * information available then) to decide if the flag should be acted upon. 884 * This extra information is needed because of at least one of the following 885 * three reasons. 886 * Under /d and the matched string is in UTF-8, it means the ANYOFD node 887 * matches more things than in the bitmap. Those things will be any 888 * code point too high for the bitmap, but crucially, any non-ASCII 889 * characters that match iff when using Unicode rules. These all are 890 * < 256. 891 * 892 * Under /l and ANYOFL_FOLD is set, this flag may indicate there are 893 * potential matches valid only if the locale is a UTF-8 one. If so, 894 * a list of them is stored in the AV. 895 * 896 * For any non-ANYOFD node, there may be a user-defined property that 897 * wasn't yet defined at the time the regex was compiled, and so must 898 * be looked up at runtime, The information required to do so will 899 * also be in the AV. 900 * 901 * Note that an ANYOFL node may contain both a user-defined property, and 902 * folds not always valid. The important thing is that there is an AV to 903 * look at. */ 904#define ANYOF_HAS_EXTRA_RUNTIME_MATCHES 0x40 905 906/* Shared bit: 907 * Under /d it means the ANYOFD node matches all non-ASCII Latin1 908 * characters when the target string is not in utf8. 909 * When not under /d, it means the ANYOF node should raise a warning if 910 * matching against an above-Unicode code point. 911 * (These uses are mutually exclusive because the warning requires a \p{}, and 912 * \p{} implies /u which deselects /d). An SSC node only has this bit set if 913 * what is meant is the warning. The names are to make sure that you are 914 * cautioned about its shared nature */ 915#define ANYOFD_NON_UTF8_MATCHES_ALL_NON_ASCII__shared 0x80 916#define ANYOF_WARN_SUPER__shared 0x80 917 918#define ANYOF_FLAGS_ALL ((U8) ~(0x10|0x20)) 919 920#define ANYOF_LOCALE_FLAGS ( ANYOFL_FOLD \ 921 | ANYOF_MATCHES_POSIXL \ 922 | ANYOFL_UTF8_LOCALE_REQD) 923 924/* These are the flags that apply to both regular ANYOF nodes and synthetic 925 * start class nodes during construction of the SSC. During finalization of 926 * the SSC, other of the flags may get added to it */ 927#define ANYOF_COMMON_FLAGS 0 928 929/* Character classes for node->classflags of ANYOF */ 930/* Should be synchronized with a table in regprop() */ 931/* 2n should be the normal one, paired with its complement at 2n+1 */ 932 933#define ANYOF_ALPHA ((CC_ALPHA_) * 2) 934#define ANYOF_NALPHA ((ANYOF_ALPHA) + 1) 935#define ANYOF_ALPHANUMERIC ((CC_ALPHANUMERIC_) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */ 936#define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1) 937#define ANYOF_ASCII ((CC_ASCII_) * 2) 938#define ANYOF_NASCII ((ANYOF_ASCII) + 1) 939#define ANYOF_BLANK ((CC_BLANK_) * 2) /* GNU extension: space and tab: non-vertical space */ 940#define ANYOF_NBLANK ((ANYOF_BLANK) + 1) 941#define ANYOF_CASED ((CC_CASED_) * 2) /* Pseudo class for [:lower:] or 942 [:upper:] under /i */ 943#define ANYOF_NCASED ((ANYOF_CASED) + 1) 944#define ANYOF_CNTRL ((CC_CNTRL_) * 2) 945#define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1) 946#define ANYOF_DIGIT ((CC_DIGIT_) * 2) /* \d */ 947#define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1) 948#define ANYOF_GRAPH ((CC_GRAPH_) * 2) 949#define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1) 950#define ANYOF_LOWER ((CC_LOWER_) * 2) 951#define ANYOF_NLOWER ((ANYOF_LOWER) + 1) 952#define ANYOF_PRINT ((CC_PRINT_) * 2) 953#define ANYOF_NPRINT ((ANYOF_PRINT) + 1) 954#define ANYOF_PUNCT ((CC_PUNCT_) * 2) 955#define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1) 956#define ANYOF_SPACE ((CC_SPACE_) * 2) /* \s */ 957#define ANYOF_NSPACE ((ANYOF_SPACE) + 1) 958#define ANYOF_UPPER ((CC_UPPER_) * 2) 959#define ANYOF_NUPPER ((ANYOF_UPPER) + 1) 960#define ANYOF_WORDCHAR ((CC_WORDCHAR_) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */ 961#define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1) 962#define ANYOF_XDIGIT ((CC_XDIGIT_) * 2) 963#define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1) 964 965/* pseudo classes below this, not stored in the class bitmap, but used as flags 966 during compilation of char classes */ 967 968#define ANYOF_VERTWS ((CC_VERTSPACE_) * 2) 969#define ANYOF_NVERTWS ((ANYOF_VERTWS)+1) 970 971/* It is best if this is the last one, as all above it are stored as bits in a 972 * bitmap, and it isn't part of that bitmap */ 973#if CC_VERTSPACE_ != HIGHEST_REGCOMP_DOT_H_SYNC_ 974# error Problem with handy.h HIGHEST_REGCOMP_DOT_H_SYNC_ #define 975#endif 976 977#define ANYOF_POSIXL_MAX (ANYOF_VERTWS) /* So upper loop limit is written: 978 * '< ANYOF_MAX' 979 * Hence doesn't include VERTWS, as that 980 * is a pseudo class */ 981#define ANYOF_MAX ANYOF_POSIXL_MAX 982 983#if (ANYOF_POSIXL_MAX > 32) /* Must fit in 32-bit word */ 984# error Problem with handy.h CC_foo_ #defines 985#endif 986 987#define ANYOF_HORIZWS ((ANYOF_POSIXL_MAX)+2) /* = (ANYOF_NVERTWS + 1) */ 988#define ANYOF_NHORIZWS ((ANYOF_POSIXL_MAX)+3) 989 990#define ANYOF_UNIPROP ((ANYOF_POSIXL_MAX)+4) /* Used to indicate a Unicode 991 property: \p{} or \P{} */ 992 993/* Backward source code compatibility. */ 994 995#define ANYOF_ALNUML ANYOF_ALNUM 996#define ANYOF_NALNUML ANYOF_NALNUM 997#define ANYOF_SPACEL ANYOF_SPACE 998#define ANYOF_NSPACEL ANYOF_NSPACE 999#define ANYOF_ALNUM ANYOF_WORDCHAR 1000#define ANYOF_NALNUM ANYOF_NWORDCHAR 1001 1002/* Utility macros for the bitmap and classes of ANYOF */ 1003 1004#define BITMAP_BYTE(p, c) (( (U8*) (p)) [ ( ( (UV) (c)) >> 3) ] ) 1005#define BITMAP_BIT(c) (1U << ((c) & 7)) 1006#define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & BITMAP_BIT((U8)(c))) 1007 1008#define ANYOF_FLAGS(p) (FLAGS(p)) 1009 1010#define ANYOF_BIT(c) BITMAP_BIT(c) 1011 1012#define ANYOF_POSIXL_BITMAP(p) (((regnode_charclass_posixl*) (p))->classflags) 1013 1014#define POSIXL_SET(field, c) ((field) |= (1U << (c))) 1015#define ANYOF_POSIXL_SET(p, c) POSIXL_SET(ANYOF_POSIXL_BITMAP(p), (c)) 1016 1017#define POSIXL_CLEAR(field, c) ((field) &= ~ (1U <<(c))) 1018#define ANYOF_POSIXL_CLEAR(p, c) POSIXL_CLEAR(ANYOF_POSIXL_BITMAP(p), (c)) 1019 1020#define POSIXL_TEST(field, c) ((field) & (1U << (c))) 1021#define ANYOF_POSIXL_TEST(p, c) POSIXL_TEST(ANYOF_POSIXL_BITMAP(p), (c)) 1022 1023#define POSIXL_ZERO(field) STMT_START { (field) = 0; } STMT_END 1024#define ANYOF_POSIXL_ZERO(ret) POSIXL_ZERO(ANYOF_POSIXL_BITMAP(ret)) 1025 1026#define ANYOF_POSIXL_SET_TO_BITMAP(p, bits) \ 1027 STMT_START { ANYOF_POSIXL_BITMAP(p) = (bits); } STMT_END 1028 1029/* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */ 1030#define ANYOF_POSIXL_SETALL(ret) \ 1031 STMT_START { \ 1032 ANYOF_POSIXL_BITMAP(ret) = nBIT_MASK(ANYOF_POSIXL_MAX); \ 1033 } STMT_END 1034#define ANYOF_CLASS_SETALL(ret) ANYOF_POSIXL_SETALL(ret) 1035 1036#define ANYOF_POSIXL_TEST_ANY_SET(p) \ 1037 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) && ANYOF_POSIXL_BITMAP(p)) 1038#define ANYOF_CLASS_TEST_ANY_SET(p) ANYOF_POSIXL_TEST_ANY_SET(p) 1039 1040/* Since an SSC always has this field, we don't have to test for that; nor do 1041 * we want to because the bit isn't set for SSC during its construction */ 1042#define ANYOF_POSIXL_SSC_TEST_ANY_SET(p) \ 1043 cBOOL(((regnode_ssc*)(p))->classflags) 1044#define ANYOF_POSIXL_SSC_TEST_ALL_SET(p) /* Are all bits set? */ \ 1045 (((regnode_ssc*) (p))->classflags \ 1046 == nBIT_MASK(ANYOF_POSIXL_MAX)) 1047 1048#define ANYOF_POSIXL_TEST_ALL_SET(p) \ 1049 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \ 1050 && ANYOF_POSIXL_BITMAP(p) == nBIT_MASK(ANYOF_POSIXL_MAX)) 1051 1052#define ANYOF_POSIXL_OR(source, dest) STMT_START { (dest)->classflags |= (source)->classflags ; } STMT_END 1053#define ANYOF_CLASS_OR(source, dest) ANYOF_POSIXL_OR((source), (dest)) 1054 1055#define ANYOF_POSIXL_AND(source, dest) STMT_START { (dest)->classflags &= (source)->classflags ; } STMT_END 1056 1057#define ANYOF_BITMAP_ZERO(ret) Zero(((regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char) 1058#define ANYOF_BITMAP(p) ((regnode_charclass*)(p))->bitmap 1059#define ANYOF_BITMAP_BYTE(p, c) BITMAP_BYTE(ANYOF_BITMAP(p), c) 1060#define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c)) 1061#define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c)) 1062#define ANYOF_BITMAP_TEST(p, c) cBOOL(ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c)) 1063 1064#define ANYOF_BITMAP_SETALL(p) \ 1065 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE) 1066#define ANYOF_BITMAP_CLEARALL(p) \ 1067 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE) 1068 1069/* 1070 * Utility definitions. 1071 */ 1072#ifndef CHARMASK 1073# define UCHARAT(p) ((int)*(const U8*)(p)) 1074#else 1075# define UCHARAT(p) ((int)*(p)&CHARMASK) 1076#endif 1077 1078/* Number of regnode equivalents that 'guy' occupies beyond the size of the 1079 * smallest regnode. */ 1080#define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode)) 1081 1082#define REG_ZERO_LEN_SEEN 0x00000001 1083#define REG_LOOKBEHIND_SEEN 0x00000002 1084/* add a short form alias to keep the line length police happy */ 1085#define REG_LB_SEEN REG_LOOKBEHIND_SEEN 1086#define REG_GPOS_SEEN 0x00000004 1087/* spare */ 1088#define REG_RECURSE_SEEN 0x00000020 1089#define REG_TOP_LEVEL_BRANCHES_SEEN 0x00000040 1090#define REG_VERBARG_SEEN 0x00000080 1091#define REG_CUTGROUP_SEEN 0x00000100 1092#define REG_RUN_ON_COMMENT_SEEN 0x00000200 1093#define REG_UNFOLDED_MULTI_SEEN 0x00000400 1094/* spare */ 1095#define REG_UNBOUNDED_QUANTIFIER_SEEN 0x00001000 1096#define REG_PESSIMIZE_SEEN 0x00002000 1097 1098 1099START_EXTERN_C 1100 1101#ifdef PLUGGABLE_RE_EXTENSION 1102#include "re_nodes.h" 1103#else 1104#include "regnodes.h" 1105#endif 1106 1107#ifndef PLUGGABLE_RE_EXTENSION 1108#ifndef DOINIT 1109EXTCONST regexp_engine PL_core_reg_engine; 1110#else /* DOINIT */ 1111EXTCONST regexp_engine PL_core_reg_engine = { 1112 Perl_re_compile, 1113 Perl_regexec_flags, 1114 Perl_re_intuit_start, 1115 Perl_re_intuit_string, 1116 Perl_regfree_internal, 1117 Perl_reg_numbered_buff_fetch, 1118 Perl_reg_numbered_buff_store, 1119 Perl_reg_numbered_buff_length, 1120 Perl_reg_named_buff, 1121 Perl_reg_named_buff_iter, 1122 Perl_reg_qr_package, 1123#if defined(USE_ITHREADS) 1124 Perl_regdupe_internal, 1125#endif 1126 Perl_re_op_compile 1127}; 1128#endif /* DOINIT */ 1129#endif /* PLUGGABLE_RE_EXTENSION */ 1130 1131 1132END_EXTERN_C 1133 1134 1135/* .what is a character array with one character for each member of .data 1136 * The character describes the function of the corresponding .data item: 1137 * a - AV for paren_name_list under DEBUGGING 1138 * f - start-class data for regstclass optimization 1139 * l - start op for literal (?{EVAL}) item 1140 * L - start op for literal (?{EVAL}) item, with separate CV (qr//) 1141 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/ 1142 * s - inversion list for Unicode-style character class, and the 1143 * multicharacter strings resulting from casefolding the single-character 1144 * entries in the character class 1145 * t - trie struct 1146 * u - trie struct's widecharmap (a HV, so can't share, must dup) 1147 * also used for revcharmap and words under DEBUGGING 1148 * T - aho-trie struct 1149 * S - sv for named capture lookup 1150 * 20010712 mjd@plover.com 1151 * (Remember to update re_dup() and pregfree() if you add any items.) 1152 */ 1153struct reg_data { 1154 U32 count; 1155 U8 *what; 1156 void* data[1]; 1157}; 1158 1159/* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses 1160 anchored* and float* via array indexes 0 and 1. */ 1161#define anchored_substr substrs->data[0].substr 1162#define anchored_utf8 substrs->data[0].utf8_substr 1163#define anchored_offset substrs->data[0].min_offset 1164#define anchored_end_shift substrs->data[0].end_shift 1165 1166#define float_substr substrs->data[1].substr 1167#define float_utf8 substrs->data[1].utf8_substr 1168#define float_min_offset substrs->data[1].min_offset 1169#define float_max_offset substrs->data[1].max_offset 1170#define float_end_shift substrs->data[1].end_shift 1171 1172#define check_substr substrs->data[2].substr 1173#define check_utf8 substrs->data[2].utf8_substr 1174#define check_offset_min substrs->data[2].min_offset 1175#define check_offset_max substrs->data[2].max_offset 1176#define check_end_shift substrs->data[2].end_shift 1177 1178#define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr) 1179#define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8) 1180#define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr) 1181#define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8) 1182 1183/* trie related stuff */ 1184 1185/* a transition record for the state machine. the 1186 check field determines which state "owns" the 1187 transition. the char the transition is for is 1188 determined by offset from the owning states base 1189 field. the next field determines which state 1190 is to be transitioned to if any. 1191*/ 1192struct _reg_trie_trans { 1193 U32 next; 1194 U32 check; 1195}; 1196 1197/* a transition list element for the list based representation */ 1198struct _reg_trie_trans_list_elem { 1199 U16 forid; 1200 U32 newstate; 1201}; 1202typedef struct _reg_trie_trans_list_elem reg_trie_trans_le; 1203 1204/* a state for compressed nodes. base is an offset 1205 into an array of reg_trie_trans array. If wordnum is 1206 nonzero the state is accepting. if base is zero then 1207 the state has no children (and will be accepting) 1208*/ 1209struct _reg_trie_state { 1210 U16 wordnum; 1211 union { 1212 U32 base; 1213 reg_trie_trans_le* list; 1214 } trans; 1215}; 1216 1217/* info per word; indexed by wordnum */ 1218typedef struct { 1219 U16 prev; /* previous word in acceptance chain; eg in 1220 * zzz|abc|ab/ after matching the chars abc, the 1221 * accepted word is #2, and the previous accepted 1222 * word is #3 */ 1223 U32 len; /* how many chars long is this word? */ 1224 U32 accept; /* accept state for this word */ 1225} reg_trie_wordinfo; 1226 1227 1228typedef struct _reg_trie_state reg_trie_state; 1229typedef struct _reg_trie_trans reg_trie_trans; 1230 1231 1232/* anything in here that needs to be freed later 1233 should be dealt with in pregfree. 1234 refcount is first in both this and _reg_ac_data to allow a space 1235 optimisation in Perl_regdupe. */ 1236struct _reg_trie_data { 1237 U32 refcount; /* number of times this trie is referenced */ 1238 U32 lasttrans; /* last valid transition element */ 1239 U16 *charmap; /* byte to charid lookup array */ 1240 reg_trie_state *states; /* state data */ 1241 reg_trie_trans *trans; /* array of transition elements */ 1242 char *bitmap; /* stclass bitmap */ 1243 U16 *jump; /* optional 1 indexed array of offsets before tail 1244 for the node following a given word. */ 1245 U16 *j_before_paren; /* optional 1 indexed array of parno reset data 1246 for the given jump. */ 1247 U16 *j_after_paren; /* optional 1 indexed array of parno reset data 1248 for the given jump. */ 1249 1250 reg_trie_wordinfo *wordinfo; /* array of info per word */ 1251 U16 uniquecharcount; /* unique chars in trie (width of trans table) */ 1252 U32 startstate; /* initial state - used for common prefix optimisation */ 1253 STRLEN minlen; /* minimum length of words in trie - build/opt only? */ 1254 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */ 1255 U32 prefixlen; /* #chars in common prefix */ 1256 U32 statecount; /* Build only - number of states in the states array 1257 (including the unused zero state) */ 1258 U32 wordcount; /* Build only */ 1259 U16 before_paren; 1260 U16 after_paren; 1261#ifdef DEBUGGING 1262 STRLEN charcount; /* Build only */ 1263#endif 1264}; 1265/* There is one (3 under DEBUGGING) pointers that logically belong in this 1266 structure, but are held outside as they need duplication on thread cloning, 1267 whereas the rest of the structure can be read only: 1268 HV *widecharmap; code points > 255 to charid 1269#ifdef DEBUGGING 1270 AV *words; Array of words contained in trie, for dumping 1271 AV *revcharmap; Map of each charid back to its character representation 1272#endif 1273*/ 1274 1275#define TRIE_WORDS_OFFSET 2 1276 1277typedef struct _reg_trie_data reg_trie_data; 1278 1279/* refcount is first in both this and _reg_trie_data to allow a space 1280 optimisation in Perl_regdupe. */ 1281struct _reg_ac_data { 1282 U32 refcount; 1283 U32 trie; 1284 U32 *fail; 1285 reg_trie_state *states; 1286}; 1287typedef struct _reg_ac_data reg_ac_data; 1288 1289/* ANY_BIT doesn't use the structure, so we can borrow it here. 1290 This is simpler than refactoring all of it as wed end up with 1291 three different sets... */ 1292 1293#define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap) 1294#define TRIE_BITMAP_BYTE(p, c) BITMAP_BYTE(TRIE_BITMAP(p), c) 1295#define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c)) 1296#define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c)) 1297#define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c)) 1298 1299#define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC) 1300#define IS_TRIE_AC(op) ((op)>=AHOCORASICK) 1301 1302/* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */ 1303#define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1) 1304#define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state)) 1305#define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state)) 1306 1307#ifdef DEBUGGING 1308#define TRIE_CHARCOUNT(trie) ((trie)->charcount) 1309#else 1310#define TRIE_CHARCOUNT(trie) (trie_charcount) 1311#endif 1312 1313#define RE_TRIE_MAXBUF_INIT 65536 1314#define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF" 1315#define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS" 1316 1317#define RE_COMPILE_RECURSION_INIT 1000 1318#define RE_COMPILE_RECURSION_LIMIT "\022E_COMPILE_RECURSION_LIMIT" 1319 1320/* 1321 1322RE_DEBUG_FLAGS is used to control what debug output is emitted 1323its divided into three groups of options, some of which interact. 1324The three groups are: Compile, Execute, Extra. There is room for a 1325further group, as currently only the low three bytes are used. 1326 1327 Compile Options: 1328 1329 PARSE 1330 PEEP 1331 TRIE 1332 PROGRAM 1333 1334 Execute Options: 1335 1336 INTUIT 1337 MATCH 1338 TRIE 1339 1340 Extra Options 1341 1342 TRIE 1343 1344If you modify any of these make sure you make corresponding changes to 1345re.pm, especially to the documentation. 1346 1347*/ 1348 1349 1350/* Compile */ 1351#define RE_DEBUG_COMPILE_MASK 0x0000FF 1352#define RE_DEBUG_COMPILE_PARSE 0x000001 1353#define RE_DEBUG_COMPILE_OPTIMISE 0x000002 1354#define RE_DEBUG_COMPILE_TRIE 0x000004 1355#define RE_DEBUG_COMPILE_DUMP 0x000008 1356#define RE_DEBUG_COMPILE_FLAGS 0x000010 1357#define RE_DEBUG_COMPILE_TEST 0x000020 1358 1359/* Execute */ 1360#define RE_DEBUG_EXECUTE_MASK 0x00FF00 1361#define RE_DEBUG_EXECUTE_INTUIT 0x000100 1362#define RE_DEBUG_EXECUTE_MATCH 0x000200 1363#define RE_DEBUG_EXECUTE_TRIE 0x000400 1364 1365/* Extra */ 1366#define RE_DEBUG_EXTRA_MASK 0x3FF0000 1367#define RE_DEBUG_EXTRA_TRIE 0x0010000 1368#define RE_DEBUG_EXTRA_STATE 0x0080000 1369#define RE_DEBUG_EXTRA_OPTIMISE 0x0100000 1370#define RE_DEBUG_EXTRA_BUFFERS 0x0400000 1371#define RE_DEBUG_EXTRA_GPOS 0x0800000 1372#define RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE 0x1000000 1373#define RE_DEBUG_EXTRA_WILDCARD 0x2000000 1374/* combined */ 1375#define RE_DEBUG_EXTRA_STACK 0x0280000 1376 1377#define RE_DEBUG_FLAG(x) (re_debug_flags & (x)) 1378/* Compile */ 1379#define DEBUG_COMPILE_r(x) DEBUG_r( \ 1380 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_MASK)) x ) 1381#define DEBUG_PARSE_r(x) DEBUG_r( \ 1382 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_PARSE)) x ) 1383#define DEBUG_OPTIMISE_r(x) DEBUG_r( \ 1384 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE)) x ) 1385#define DEBUG_DUMP_r(x) DEBUG_r( \ 1386 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_DUMP)) x ) 1387#define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \ 1388 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TRIE)) x ) 1389#define DEBUG_FLAGS_r(x) DEBUG_r( \ 1390 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_FLAGS)) x ) 1391#define DEBUG_TEST_r(x) DEBUG_r( \ 1392 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TEST)) x ) 1393/* Execute */ 1394#define DEBUG_EXECUTE_r(x) DEBUG_r( \ 1395 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_MASK)) x ) 1396#define DEBUG_INTUIT_r(x) DEBUG_r( \ 1397 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_INTUIT)) x ) 1398#define DEBUG_MATCH_r(x) DEBUG_r( \ 1399 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_MATCH)) x ) 1400#define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \ 1401 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_TRIE)) x ) 1402 1403/* Extra */ 1404#define DEBUG_EXTRA_r(x) DEBUG_r( \ 1405 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_MASK)) x ) 1406#define DEBUG_STATE_r(x) DEBUG_r( \ 1407 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_STATE)) x ) 1408#define DEBUG_STACK_r(x) DEBUG_r( \ 1409 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_STACK)) x ) 1410#define DEBUG_BUFFERS_r(x) DEBUG_r( \ 1411 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_BUFFERS)) x ) 1412 1413#define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \ 1414 if (DEBUG_v_TEST || ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \ 1415 RE_DEBUG_FLAG(RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE))) x ) 1416#define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \ 1417 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_TRIE)) x ) 1418#define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \ 1419 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_TRIE)) x ) 1420 1421#define DEBUG_TRIE_r(x) DEBUG_r( \ 1422 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TRIE \ 1423 | RE_DEBUG_EXECUTE_TRIE )) x ) 1424#define DEBUG_GPOS_r(x) DEBUG_r( \ 1425 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_GPOS)) x ) 1426 1427#define DEBUG_DUMP_PRE_OPTIMIZE_r(x) DEBUG_r( \ 1428 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE)) x ) 1429 1430/* initialization */ 1431/* Get the debug flags for code not in regcomp.c nor regexec.c. This doesn't 1432 * initialize the variable if it isn't already there, instead it just assumes 1433 * the flags are 0 */ 1434#define DECLARE_AND_GET_RE_DEBUG_FLAGS_NON_REGEX \ 1435 volatile IV re_debug_flags = 0; PERL_UNUSED_VAR(re_debug_flags); \ 1436 STMT_START { \ 1437 SV * re_debug_flags_sv = NULL; \ 1438 /* get_sv() can return NULL during global destruction. */ \ 1439 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \ 1440 if (re_debug_flags_sv && SvIOK(re_debug_flags_sv)) \ 1441 re_debug_flags=SvIV(re_debug_flags_sv); \ 1442 } STMT_END 1443 1444 1445#ifdef DEBUGGING 1446 1447/* For use in regcomp.c and regexec.c, Get the debug flags, and initialize to 1448 * the defaults if not done already */ 1449#define DECLARE_AND_GET_RE_DEBUG_FLAGS \ 1450 volatile IV re_debug_flags = 0; PERL_UNUSED_VAR(re_debug_flags); \ 1451 DEBUG_r({ \ 1452 SV * re_debug_flags_sv = NULL; \ 1453 /* get_sv() can return NULL during global destruction. */ \ 1454 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \ 1455 if (re_debug_flags_sv) { \ 1456 if (!SvIOK(re_debug_flags_sv)) /* If doesn't exist set to default */\ 1457 sv_setuv(re_debug_flags_sv, \ 1458 /* These defaults should be kept in sync with re.pm */ \ 1459 RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \ 1460 re_debug_flags=SvIV(re_debug_flags_sv); \ 1461 } \ 1462 }) 1463 1464#define isDEBUG_WILDCARD (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_WILDCARD)) 1465 1466#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \ 1467 const char * const rpv = \ 1468 pv_pretty((dsv), (pv), (l), (m), \ 1469 PL_colors[(c1)],PL_colors[(c2)], \ 1470 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \ 1471 const int rlen = SvCUR(dsv) 1472 1473/* This is currently unsed in the core */ 1474#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \ 1475 const char * const rpv = \ 1476 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \ 1477 PL_colors[(c1)],PL_colors[(c2)], \ 1478 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ) 1479 1480#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \ 1481 const char * const rpv = \ 1482 pv_pretty((dsv), (pv), (l), (m), \ 1483 PL_colors[0], PL_colors[1], \ 1484 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \ 1485 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \ 1486 ) 1487 1488#define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0)) 1489#define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "") 1490 1491#else /* if not DEBUGGING */ 1492 1493#define DECLARE_AND_GET_RE_DEBUG_FLAGS dNOOP 1494#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) dNOOP 1495#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) 1496#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) dNOOP 1497#define RE_SV_DUMPLEN(ItEm) 1498#define RE_SV_TAIL(ItEm) 1499#define isDEBUG_WILDCARD 0 1500 1501#endif /* DEBUG RELATED DEFINES */ 1502 1503#define FIRST_NON_ASCII_DECIMAL_DIGIT 0x660 /* ARABIC_INDIC_DIGIT_ZERO */ 1504 1505typedef enum { 1506 TRADITIONAL_BOUND = CC_WORDCHAR_, 1507 GCB_BOUND, 1508 LB_BOUND, 1509 SB_BOUND, 1510 WB_BOUND 1511} bound_type; 1512 1513/* This unpacks the FLAGS field of ANYOF[HR]x nodes. The value it contains 1514 * gives the strict lower bound for the UTF-8 start byte of any code point 1515 * matchable by the node, and a loose upper bound as well. 1516 * 1517 * The low bound is stored as 0xC0 + ((the upper 6 bits) >> 2) 1518 * The loose upper bound is determined from the lowest 2 bits and the low bound 1519 * (called x) as follows: 1520 * 1521 * 11 The upper limit of the range can be as much as (EF - x) / 8 1522 * 10 The upper limit of the range can be as much as (EF - x) / 4 1523 * 01 The upper limit of the range can be as much as (EF - x) / 2 1524 * 00 The upper limit of the range can be as much as EF 1525 * 1526 * For motivation of this design, see commit message in 1527 * 3146c00a633e9cbed741e10146662fbcedfdb8d3 */ 1528#ifdef EBCDIC 1529# define MAX_ANYOF_HRx_BYTE 0xF4 1530#else 1531# define MAX_ANYOF_HRx_BYTE 0xEF 1532#endif 1533#define LOWEST_ANYOF_HRx_BYTE(b) (((b) >> 2) + 0xC0) 1534#define HIGHEST_ANYOF_HRx_BYTE(b) \ 1535 (LOWEST_ANYOF_HRx_BYTE(b) \ 1536 + ((MAX_ANYOF_HRx_BYTE - LOWEST_ANYOF_HRx_BYTE(b)) >> ((b) & 3))) 1537 1538#if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) 1539# define GET_REGCLASS_AUX_DATA(a,b,c,d,e,f) get_regclass_aux_data(a,b,c,d,e,f) 1540#else 1541# define GET_REGCLASS_AUX_DATA(a,b,c,d,e,f) get_re_gclass_aux_data(a,b,c,d,e,f) 1542#endif 1543 1544#define REGNODE_TYPE(node) (PL_regnode_info[(node)].type) 1545#define REGNODE_OFF_BY_ARG(node) (PL_regnode_info[(node)].off_by_arg) 1546#define REGNODE_ARG_LEN(node) (PL_regnode_info[(node)].arg_len) 1547#define REGNODE_ARG_LEN_VARIES(node) (PL_regnode_info[(node)].arg_len_varies) 1548#define REGNODE_NAME(node) (PL_regnode_name[(node)]) 1549 1550#if defined(PERL_IN_REGEX_ENGINE) 1551#include "reginline.h" 1552#endif 1553 1554#define EVAL_OPTIMISTIC_FLAG 128 1555#define EVAL_FLAGS_MASK (EVAL_OPTIMISTIC_FLAG-1) 1556 1557 1558 1559#endif /* PERL_REGCOMP_H_ */ 1560 1561/* 1562 * ex: set ts=8 sts=4 sw=4 et: 1563 */ 1564