1 /* regcomp.h 2 * 3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 4 * 2000, 2001, 2002, 2003, 2005, 2006, 2007, by Larry Wall and others 5 * 6 * You may distribute under the terms of either the GNU General Public 7 * License or the Artistic License, as specified in the README file. 8 * 9 */ 10 #include "regcharclass.h" 11 12 /* Convert branch sequences to more efficient trie ops? */ 13 #define PERL_ENABLE_TRIE_OPTIMISATION 1 14 15 /* Be really aggressive about optimising patterns with trie sequences? */ 16 #define PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION 1 17 18 /* Should the optimiser take positive assertions into account? */ 19 #define PERL_ENABLE_POSITIVE_ASSERTION_STUDY 0 20 21 /* Not for production use: */ 22 #define PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS 0 23 24 /* Activate offsets code - set to if 1 to enable */ 25 #ifdef DEBUGGING 26 #define RE_TRACK_PATTERN_OFFSETS 27 #endif 28 29 /* Unless the next line is uncommented it is illegal to combine lazy 30 matching with possessive matching. Frankly it doesn't make much sense 31 to allow it as X*?+ matches nothing, X+?+ matches a single char only, 32 and X{min,max}?+ matches min times only. 33 */ 34 /* #define REG_ALLOW_MINMOD_SUSPEND */ 35 36 /* 37 * The "internal use only" fields in regexp.h are present to pass info from 38 * compile to execute that permits the execute phase to run lots faster on 39 * simple cases. They are: 40 * 41 * regstart sv that must begin a match; NULL if none obvious 42 * reganch is the match anchored (at beginning-of-line only)? 43 * regmust string (pointer into program) that match must include, or NULL 44 * [regmust changed to SV* for bminstr()--law] 45 * regmlen length of regmust string 46 * [regmlen not used currently] 47 * 48 * Regstart and reganch permit very fast decisions on suitable starting points 49 * for a match, cutting down the work a lot. Regmust permits fast rejection 50 * of lines that cannot possibly match. The regmust tests are costly enough 51 * that pregcomp() supplies a regmust only if the r.e. contains something 52 * potentially expensive (at present, the only such thing detected is * or + 53 * at the start of the r.e., which can involve a lot of backup). Regmlen is 54 * supplied because the test in pregexec() needs it and pregcomp() is computing 55 * it anyway. 56 * [regmust is now supplied always. The tests that use regmust have a 57 * heuristic that disables the test if it usually matches.] 58 * 59 * [In fact, we now use regmust in many cases to locate where the search 60 * starts in the string, so if regback is >= 0, the regmust search is never 61 * wasted effort. The regback variable says how many characters back from 62 * where regmust matched is the earliest possible start of the match. 63 * For instance, /[a-z].foo/ has a regmust of 'foo' and a regback of 2.] 64 */ 65 66 /* 67 * Structure for regexp "program". This is essentially a linear encoding 68 * of a nondeterministic finite-state machine (aka syntax charts or 69 * "railroad normal form" in parsing technology). Each node is an opcode 70 * plus a "next" pointer, possibly plus an operand. "Next" pointers of 71 * all nodes except BRANCH implement concatenation; a "next" pointer with 72 * a BRANCH on both ends of it is connecting two alternatives. (Here we 73 * have one of the subtle syntax dependencies: an individual BRANCH (as 74 * opposed to a collection of them) is never concatenated with anything 75 * because of operator precedence.) The operand of some types of node is 76 * a literal string; for others, it is a node leading into a sub-FSM. In 77 * particular, the operand of a BRANCH node is the first node of the branch. 78 * (NB this is *not* a tree structure: the tail of the branch connects 79 * to the thing following the set of BRANCHes.) The opcodes are defined 80 * in regnodes.h which is generated from regcomp.sym by regcomp.pl. 81 */ 82 83 /* 84 * A node is one char of opcode followed by two chars of "next" pointer. 85 * "Next" pointers are stored as two 8-bit pieces, high order first. The 86 * value is a positive offset from the opcode of the node containing it. 87 * An operand, if any, simply follows the node. (Note that much of the 88 * code generation knows about this implicit relationship.) 89 * 90 * Using two bytes for the "next" pointer is vast overkill for most things, 91 * but allows patterns to get big without disasters. 92 * 93 * [The "next" pointer is always aligned on an even 94 * boundary, and reads the offset directly as a short. Also, there is no 95 * special test to reverse the sign of BACK pointers since the offset is 96 * stored negative.] 97 */ 98 99 /* This is the stuff that used to live in regexp.h that was truly 100 private to the engine itself. It now lives here. */ 101 102 103 104 typedef struct regexp_internal { 105 int name_list_idx; /* Optional data index of an array of paren names */ 106 union { 107 U32 *offsets; /* offset annotations 20001228 MJD 108 data about mapping the program to the 109 string - 110 offsets[0] is proglen when this is used 111 */ 112 U32 proglen; 113 } u; 114 115 regnode *regstclass; /* Optional startclass as identified or constructed 116 by the optimiser */ 117 struct reg_data *data; /* Additional miscellaneous data used by the program. 118 Used to make it easier to clone and free arbitrary 119 data that the regops need. Often the ARG field of 120 a regop is an index into this structure */ 121 struct reg_code_block *code_blocks;/* positions of literal (?{}) */ 122 int num_code_blocks; /* size of code_blocks[] */ 123 regnode program[1]; /* Unwarranted chumminess with compiler. */ 124 } regexp_internal; 125 126 #define RXi_SET(x,y) (x)->pprivate = (void*)(y) 127 #define RXi_GET(x) ((regexp_internal *)((x)->pprivate)) 128 #define RXi_GET_DECL(r,ri) regexp_internal *ri = RXi_GET(r) 129 /* 130 * Flags stored in regexp->intflags 131 * These are used only internally to the regexp engine 132 * 133 * See regexp.h for flags used externally to the regexp engine 134 */ 135 #define PREGf_SKIP 0x00000001 136 #define PREGf_IMPLICIT 0x00000002 /* Converted .* to ^.* */ 137 #define PREGf_NAUGHTY 0x00000004 /* how exponential is this pattern? */ 138 #define PREGf_VERBARG_SEEN 0x00000008 139 #define PREGf_CUTGROUP_SEEN 0x00000010 140 #define PREGf_USE_RE_EVAL 0x00000020 /* compiled with "use re 'eval'" */ 141 142 143 /* this is where the old regcomp.h started */ 144 145 struct regnode_string { 146 U8 str_len; 147 U8 type; 148 U16 next_off; 149 char string[1]; 150 }; 151 152 /* Argument bearing node - workhorse, 153 arg1 is often for the data field */ 154 struct regnode_1 { 155 U8 flags; 156 U8 type; 157 U16 next_off; 158 U32 arg1; 159 }; 160 161 /* Similar to a regnode_1 but with an extra signed argument */ 162 struct regnode_2L { 163 U8 flags; 164 U8 type; 165 U16 next_off; 166 U32 arg1; 167 I32 arg2; 168 }; 169 170 /* 'Two field' -- Two 16 bit unsigned args */ 171 struct regnode_2 { 172 U8 flags; 173 U8 type; 174 U16 next_off; 175 U16 arg1; 176 U16 arg2; 177 }; 178 179 180 #define ANYOF_BITMAP_SIZE 32 /* 256 b/(8 b/B) */ 181 182 /* also used by trie */ 183 struct regnode_charclass { 184 U8 flags; 185 U8 type; 186 U16 next_off; 187 U32 arg1; /* used as ptr in S_regclass */ 188 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */ 189 }; 190 191 /* has runtime (locale) \d, \w, ..., [:posix:] classes */ 192 struct regnode_charclass_class { 193 U8 flags; /* ANYOF_CLASS bit must go here */ 194 U8 type; 195 U16 next_off; 196 U32 arg1; /* used as ptr in S_regclass */ 197 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time */ 198 U32 classflags; /* and run-time */ 199 }; 200 201 /* XXX fix this description. 202 Impose a limit of REG_INFTY on various pattern matching operations 203 to limit stack growth and to avoid "infinite" recursions. 204 */ 205 /* The default size for REG_INFTY is I16_MAX, which is the same as 206 SHORT_MAX (see perl.h). Unfortunately I16 isn't necessarily 16 bits 207 (see handy.h). On the Cray C90, sizeof(short)==4 and hence I16_MAX is 208 ((1<<31)-1), while on the Cray T90, sizeof(short)==8 and I16_MAX is 209 ((1<<63)-1). To limit stack growth to reasonable sizes, supply a 210 smaller default. 211 --Andy Dougherty 11 June 1998 212 */ 213 #if SHORTSIZE > 2 214 # ifndef REG_INFTY 215 # define REG_INFTY ((1<<15)-1) 216 # endif 217 #endif 218 219 #ifndef REG_INFTY 220 # define REG_INFTY I16_MAX 221 #endif 222 223 #define ARG_VALUE(arg) (arg) 224 #define ARG__SET(arg,val) ((arg) = (val)) 225 226 #undef ARG 227 #undef ARG1 228 #undef ARG2 229 230 #define ARG(p) ARG_VALUE(ARG_LOC(p)) 231 #define ARG1(p) ARG_VALUE(ARG1_LOC(p)) 232 #define ARG2(p) ARG_VALUE(ARG2_LOC(p)) 233 #define ARG2L(p) ARG_VALUE(ARG2L_LOC(p)) 234 235 #define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val)) 236 #define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val)) 237 #define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val)) 238 #define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val)) 239 240 #undef NEXT_OFF 241 #undef NODE_ALIGN 242 243 #define NEXT_OFF(p) ((p)->next_off) 244 #define NODE_ALIGN(node) 245 #define NODE_ALIGN_FILL(node) ((node)->flags = 0xde) /* deadbeef */ 246 247 #define SIZE_ALIGN NODE_ALIGN 248 249 #undef OP 250 #undef OPERAND 251 #undef MASK 252 #undef STRING 253 254 #define OP(p) ((p)->type) 255 #define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \ 256 regnode types. For some, it's the \ 257 character set of the regnode */ 258 #define OPERAND(p) (((struct regnode_string *)p)->string) 259 #define MASK(p) ((char*)OPERAND(p)) 260 #define STR_LEN(p) (((struct regnode_string *)p)->str_len) 261 #define STRING(p) (((struct regnode_string *)p)->string) 262 #define STR_SZ(l) ((l + sizeof(regnode) - 1) / sizeof(regnode)) 263 #define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p))+1) 264 265 #undef NODE_ALIGN 266 #undef ARG_LOC 267 #undef NEXTOPER 268 #undef PREVOPER 269 270 #define NODE_ALIGN(node) 271 #define ARG_LOC(p) (((struct regnode_1 *)p)->arg1) 272 #define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1) 273 #define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2) 274 #define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2) 275 276 #define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */ 277 #define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2) 278 279 #define NODE_STEP_B 4 280 281 #define NEXTOPER(p) ((p) + NODE_STEP_REGNODE) 282 #define PREVOPER(p) ((p) - NODE_STEP_REGNODE) 283 284 #define FILL_ADVANCE_NODE(ptr, op) STMT_START { \ 285 (ptr)->type = op; (ptr)->next_off = 0; (ptr)++; } STMT_END 286 #define FILL_ADVANCE_NODE_ARG(ptr, op, arg) STMT_START { \ 287 ARG_SET(ptr, arg); FILL_ADVANCE_NODE(ptr, op); (ptr) += 1; } STMT_END 288 289 #define REG_MAGIC 0234 290 291 #define SIZE_ONLY (RExC_emit == &PL_regdummy) 292 293 /* If the bitmap doesn't fully represent what this ANYOF node can match, the 294 * ARG is set to this special value (since 0, 1, ... are legal, but will never 295 * reach this high). */ 296 #define ANYOF_NONBITMAP_EMPTY ((U32) -1) 297 298 /* The information used to be stored as as combination of the ANYOF_UTF8 and 299 * ANYOF_NONBITMAP_NON_UTF8 bits in the flags field, but was moved out of there 300 * to free up a bit for other uses. This tries to hide the change from 301 * existing code as much as possible. Now, the data structure that goes in ARG 302 * is not allocated unless it is needed, and that is what is used to determine 303 * if there is something outside the bitmap. The code now assumes that if 304 * that structure exists, that any UTF-8 encoded string should be tried against 305 * it, but a non-UTF8-encoded string will be tried only if the 306 * ANYOF_NONBITMAP_NON_UTF8 bit is also set. */ 307 #define ANYOF_NONBITMAP(node) (ARG(node) != ANYOF_NONBITMAP_EMPTY) 308 309 /* Flags for node->flags of ANYOF. These are in short supply, but there is one 310 * currently available. If more than this are needed, the ANYOF_LOCALE and 311 * ANYOF_CLASS bits could be shared, making a space penalty for all locale nodes. 312 * Also, the UNICODE_ALL bit could be freed up by resorting to creating a swash 313 * containing everything above 255. This introduces a performance penalty. 314 * Better would be to split it off into a separate node, which actually would 315 * improve performance a bit by allowing regexec.c to test for a UTF-8 316 * character being above 255 without having to call a function nor calculate 317 * its code point value. However, this solution might need to have a second 318 * node type, ANYOF_SYNTHETIC_ABOVE_LATIN1_ALL */ 319 320 #define ANYOF_LOCALE 0x01 /* /l modifier */ 321 322 /* The fold is calculated and stored in the bitmap where possible at compile 323 * time. However under locale, the actual folding varies depending on 324 * what the locale is at the time of execution, so it has to be deferred until 325 * then */ 326 #define ANYOF_LOC_FOLD 0x02 327 328 #define ANYOF_INVERT 0x04 329 330 /* Set if this is a struct regnode_charclass_class vs a regnode_charclass. This 331 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale 332 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved 333 * at compile-time. Could be shared with ANYOF_LOCALE, forcing all locale 334 * nodes to be large */ 335 #define ANYOF_CLASS 0x08 336 #define ANYOF_LARGE ANYOF_CLASS /* Same; name retained for back compat */ 337 338 /* Unused: 0x10. When using, be sure to change ANYOF_FLAGS_ALL below */ 339 340 /* Can match something outside the bitmap that isn't in utf8 */ 341 #define ANYOF_NONBITMAP_NON_UTF8 0x20 342 343 /* Matches every code point 0x100 and above*/ 344 #define ANYOF_UNICODE_ALL 0x40 345 346 /* Match all Latin1 characters that aren't ASCII when the target string is not 347 * in utf8. */ 348 #define ANYOF_NON_UTF8_LATIN1_ALL 0x80 349 350 #define ANYOF_FLAGS_ALL (0xff & ~0x10) 351 352 /* These are the flags that ANYOF_INVERT being set or not doesn't affect 353 * whether they are operative or not. e.g., the node still has LOCALE 354 * regardless of being inverted; whereas ANYOF_UNICODE_ALL means something 355 * different if inverted */ 356 #define INVERSION_UNAFFECTED_FLAGS (ANYOF_LOCALE \ 357 |ANYOF_LOC_FOLD \ 358 |ANYOF_CLASS \ 359 |ANYOF_NONBITMAP_NON_UTF8) 360 361 /* Character classes for node->classflags of ANYOF */ 362 /* Should be synchronized with a table in regprop() */ 363 /* 2n should be the normal one, paired with its complement at 2n+1 */ 364 365 #define ANYOF_ALPHA ((_CC_ALPHA) * 2) 366 #define ANYOF_NALPHA ((ANYOF_ALPHA) + 1) 367 #define ANYOF_ALPHANUMERIC ((_CC_ALPHANUMERIC) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */ 368 #define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1) 369 #define ANYOF_ASCII ((_CC_ASCII) * 2) 370 #define ANYOF_NASCII ((ANYOF_ASCII) + 1) 371 #define ANYOF_BLANK ((_CC_BLANK) * 2) /* GNU extension: space and tab: non-vertical space */ 372 #define ANYOF_NBLANK ((ANYOF_BLANK) + 1) 373 #define ANYOF_CASED ((_CC_CASED) * 2) /* Pseudo class for [:lower:] or 374 [:upper:] under /i */ 375 #define ANYOF_NCASED ((ANYOF_CASED) + 1) 376 #define ANYOF_CNTRL ((_CC_CNTRL) * 2) 377 #define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1) 378 #define ANYOF_DIGIT ((_CC_DIGIT) * 2) /* \d */ 379 #define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1) 380 #define ANYOF_GRAPH ((_CC_GRAPH) * 2) 381 #define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1) 382 #define ANYOF_LOWER ((_CC_LOWER) * 2) 383 #define ANYOF_NLOWER ((ANYOF_LOWER) + 1) 384 #define ANYOF_PRINT ((_CC_PRINT) * 2) 385 #define ANYOF_NPRINT ((ANYOF_PRINT) + 1) 386 #define ANYOF_PSXSPC ((_CC_PSXSPC) * 2) /* POSIX space: \s plus the vertical tab */ 387 #define ANYOF_NPSXSPC ((ANYOF_PSXSPC) + 1) 388 #define ANYOF_PUNCT ((_CC_PUNCT) * 2) 389 #define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1) 390 #define ANYOF_SPACE ((_CC_SPACE) * 2) /* \s */ 391 #define ANYOF_NSPACE ((ANYOF_SPACE) + 1) 392 #define ANYOF_UPPER ((_CC_UPPER) * 2) 393 #define ANYOF_NUPPER ((ANYOF_UPPER) + 1) 394 #define ANYOF_WORDCHAR ((_CC_WORDCHAR) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */ 395 #define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1) 396 #define ANYOF_XDIGIT ((_CC_XDIGIT) * 2) 397 #define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1) 398 399 /* pseudo classes below this, not stored in the class bitmap, but used as flags 400 during compilation of char classes */ 401 402 #define ANYOF_VERTWS ((_CC_VERTSPACE) * 2) 403 #define ANYOF_NVERTWS ((ANYOF_VERTWS)+1) 404 405 /* It is best if this is the last one, as all above it are stored as bits in a 406 * bitmap, and it isn't part of that bitmap */ 407 #if _CC_VERTSPACE != _HIGHEST_REGCOMP_DOT_H_SYNC 408 # error Problem with handy.h _HIGHEST_REGCOMP_DOT_H_SYNC #define 409 #endif 410 411 #define ANYOF_MAX (ANYOF_VERTWS) /* So upper loop limit is written: 412 * '< ANYOF_MAX' 413 * Hence doesn't include VERTWS, as that 414 * is a pseudo class */ 415 #if (ANYOF_MAX > 32) /* Must fit in 32-bit word */ 416 # error Problem with handy.h _CC_foo #defines 417 #endif 418 419 #define ANYOF_HORIZWS ((ANYOF_MAX)+2) /* = (ANYOF_NVERTWS + 1) */ 420 #define ANYOF_NHORIZWS ((ANYOF_MAX)+3) 421 422 #define ANYOF_UNIPROP ((ANYOF_MAX)+4) /* Used to indicate a Unicode 423 property: \p{} or \P{} */ 424 425 /* Backward source code compatibility. */ 426 427 #define ANYOF_ALNUML ANYOF_ALNUM 428 #define ANYOF_NALNUML ANYOF_NALNUM 429 #define ANYOF_SPACEL ANYOF_SPACE 430 #define ANYOF_NSPACEL ANYOF_NSPACE 431 #define ANYOF_ALNUM ANYOF_WORDCHAR 432 #define ANYOF_NALNUM ANYOF_NWORDCHAR 433 434 /* Utility macros for the bitmap and classes of ANYOF */ 435 436 #define ANYOF_SIZE (sizeof(struct regnode_charclass)) 437 #define ANYOF_CLASS_SIZE (sizeof(struct regnode_charclass_class)) 438 439 #define ANYOF_FLAGS(p) ((p)->flags) 440 441 #define ANYOF_BIT(c) (1 << ((c) & 7)) 442 443 #define ANYOF_CLASS_SET(p, c) (((struct regnode_charclass_class*) (p))->classflags |= (1U << (c))) 444 #define ANYOF_CLASS_CLEAR(p, c) (((struct regnode_charclass_class*) (p))->classflags &= ~ (1U <<(c))) 445 #define ANYOF_CLASS_TEST(p, c) (((struct regnode_charclass_class*) (p))->classflags & (1U << (c))) 446 447 #define ANYOF_CLASS_ZERO(ret) STMT_START { ((struct regnode_charclass_class*) (ret))->classflags = 0; } STMT_END 448 449 /* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */ 450 #define ANYOF_CLASS_SETALL(ret) STMT_START { ((struct regnode_charclass_class*) (ret))->classflags = ((1U << ((ANYOF_MAX) - 1))) - 1; } STMT_END 451 452 #define ANYOF_CLASS_OR(source, dest) STMT_START { (dest)->classflags |= source->classflags ; } STMT_END 453 454 #define ANYOF_BITMAP_ZERO(ret) Zero(((struct regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char) 455 #define ANYOF_BITMAP(p) (((struct regnode_charclass*)(p))->bitmap) 456 #define ANYOF_BITMAP_BYTE(p, c) (ANYOF_BITMAP(p)[(((U8)(c)) >> 3) & 31]) 457 #define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c)) 458 #define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c)) 459 #define ANYOF_BITMAP_TEST(p, c) (ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c)) 460 461 #define ANYOF_BITMAP_SETALL(p) \ 462 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE) 463 #define ANYOF_BITMAP_CLEARALL(p) \ 464 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE) 465 /* Check that all 256 bits are all set. Used in S_cl_is_anything() */ 466 #define ANYOF_BITMAP_TESTALLSET(p) /* Assumes sizeof(p) == 32 */ \ 467 memEQ (ANYOF_BITMAP(p), "\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377", ANYOF_BITMAP_SIZE) 468 469 #define ANYOF_SKIP ((ANYOF_SIZE - 1)/sizeof(regnode)) 470 #define ANYOF_CLASS_SKIP ((ANYOF_CLASS_SIZE - 1)/sizeof(regnode)) 471 472 #define ANYOF_CLASS_TEST_ANY_SET(p) \ 473 ((ANYOF_FLAGS(p) & ANYOF_CLASS) \ 474 && (((struct regnode_charclass_class*)(p))->classflags)) 475 /*#define ANYOF_CLASS_ADD_SKIP (ANYOF_CLASS_SKIP - ANYOF_SKIP) 476 * */ 477 478 479 /* 480 * Utility definitions. 481 */ 482 #ifndef CHARMASK 483 # define UCHARAT(p) ((int)*(const U8*)(p)) 484 #else 485 # define UCHARAT(p) ((int)*(p)&CHARMASK) 486 #endif 487 488 #define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode)) 489 490 #define REG_SEEN_ZERO_LEN 0x00000001 491 #define REG_SEEN_LOOKBEHIND 0x00000002 492 #define REG_SEEN_GPOS 0x00000004 493 /* spare */ 494 #define REG_SEEN_CANY 0x00000010 495 #define REG_SEEN_SANY REG_SEEN_CANY /* src bckwrd cmpt */ 496 #define REG_SEEN_RECURSE 0x00000020 497 #define REG_TOP_LEVEL_BRANCHES 0x00000040 498 #define REG_SEEN_VERBARG 0x00000080 499 #define REG_SEEN_CUTGROUP 0x00000100 500 #define REG_SEEN_RUN_ON_COMMENT 0x00000200 501 #define REG_SEEN_EXACTF_SHARP_S 0x00000400 502 503 START_EXTERN_C 504 505 #ifdef PLUGGABLE_RE_EXTENSION 506 #include "re_nodes.h" 507 #else 508 #include "regnodes.h" 509 #endif 510 511 #ifndef PLUGGABLE_RE_EXTENSION 512 #ifndef DOINIT 513 EXTCONST regexp_engine PL_core_reg_engine; 514 #else /* DOINIT */ 515 EXTCONST regexp_engine PL_core_reg_engine = { 516 Perl_re_compile, 517 Perl_regexec_flags, 518 Perl_re_intuit_start, 519 Perl_re_intuit_string, 520 Perl_regfree_internal, 521 Perl_reg_numbered_buff_fetch, 522 Perl_reg_numbered_buff_store, 523 Perl_reg_numbered_buff_length, 524 Perl_reg_named_buff, 525 Perl_reg_named_buff_iter, 526 Perl_reg_qr_package, 527 #if defined(USE_ITHREADS) 528 Perl_regdupe_internal, 529 #endif 530 Perl_re_op_compile 531 }; 532 #endif /* DOINIT */ 533 #endif /* PLUGGABLE_RE_EXTENSION */ 534 535 536 END_EXTERN_C 537 538 539 /* .what is a character array with one character for each member of .data 540 * The character describes the function of the corresponding .data item: 541 * a - AV for paren_name_list under DEBUGGING 542 * f - start-class data for regstclass optimization 543 * l - start op for literal (?{EVAL}) item 544 * L - start op for literal (?{EVAL}) item, with separate CV (qr//) 545 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/ 546 * s - swash for Unicode-style character class, and the multicharacter 547 * strings resulting from casefolding the single-character entries 548 * in the character class 549 * t - trie struct 550 * u - trie struct's widecharmap (a HV, so can't share, must dup) 551 * also used for revcharmap and words under DEBUGGING 552 * T - aho-trie struct 553 * S - sv for named capture lookup 554 * 20010712 mjd@plover.com 555 * (Remember to update re_dup() and pregfree() if you add any items.) 556 */ 557 struct reg_data { 558 U32 count; 559 U8 *what; 560 void* data[1]; 561 }; 562 563 /* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses 564 anchored* and float* via array indexes 0 and 1. */ 565 #define anchored_substr substrs->data[0].substr 566 #define anchored_utf8 substrs->data[0].utf8_substr 567 #define anchored_offset substrs->data[0].min_offset 568 #define anchored_end_shift substrs->data[0].end_shift 569 570 #define float_substr substrs->data[1].substr 571 #define float_utf8 substrs->data[1].utf8_substr 572 #define float_min_offset substrs->data[1].min_offset 573 #define float_max_offset substrs->data[1].max_offset 574 #define float_end_shift substrs->data[1].end_shift 575 576 #define check_substr substrs->data[2].substr 577 #define check_utf8 substrs->data[2].utf8_substr 578 #define check_offset_min substrs->data[2].min_offset 579 #define check_offset_max substrs->data[2].max_offset 580 #define check_end_shift substrs->data[2].end_shift 581 582 #define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr) 583 #define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8) 584 #define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr) 585 #define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8) 586 587 /* trie related stuff */ 588 589 /* a transition record for the state machine. the 590 check field determines which state "owns" the 591 transition. the char the transition is for is 592 determined by offset from the owning states base 593 field. the next field determines which state 594 is to be transitioned to if any. 595 */ 596 struct _reg_trie_trans { 597 U32 next; 598 U32 check; 599 }; 600 601 /* a transition list element for the list based representation */ 602 struct _reg_trie_trans_list_elem { 603 U16 forid; 604 U32 newstate; 605 }; 606 typedef struct _reg_trie_trans_list_elem reg_trie_trans_le; 607 608 /* a state for compressed nodes. base is an offset 609 into an array of reg_trie_trans array. If wordnum is 610 nonzero the state is accepting. if base is zero then 611 the state has no children (and will be accepting) 612 */ 613 struct _reg_trie_state { 614 U16 wordnum; 615 union { 616 U32 base; 617 reg_trie_trans_le* list; 618 } trans; 619 }; 620 621 /* info per word; indexed by wordnum */ 622 typedef struct { 623 U16 prev; /* previous word in acceptance chain; eg in 624 * zzz|abc|ab/ after matching the chars abc, the 625 * accepted word is #2, and the previous accepted 626 * word is #3 */ 627 U32 len; /* how many chars long is this word? */ 628 U32 accept; /* accept state for this word */ 629 } reg_trie_wordinfo; 630 631 632 typedef struct _reg_trie_state reg_trie_state; 633 typedef struct _reg_trie_trans reg_trie_trans; 634 635 636 /* anything in here that needs to be freed later 637 should be dealt with in pregfree. 638 refcount is first in both this and _reg_ac_data to allow a space 639 optimisation in Perl_regdupe. */ 640 struct _reg_trie_data { 641 U32 refcount; /* number of times this trie is referenced */ 642 U32 lasttrans; /* last valid transition element */ 643 U16 *charmap; /* byte to charid lookup array */ 644 reg_trie_state *states; /* state data */ 645 reg_trie_trans *trans; /* array of transition elements */ 646 char *bitmap; /* stclass bitmap */ 647 U16 *jump; /* optional 1 indexed array of offsets before tail 648 for the node following a given word. */ 649 reg_trie_wordinfo *wordinfo; /* array of info per word */ 650 U16 uniquecharcount; /* unique chars in trie (width of trans table) */ 651 U32 startstate; /* initial state - used for common prefix optimisation */ 652 STRLEN minlen; /* minimum length of words in trie - build/opt only? */ 653 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */ 654 U32 prefixlen; /* #chars in common prefix */ 655 U32 statecount; /* Build only - number of states in the states array 656 (including the unused zero state) */ 657 U32 wordcount; /* Build only */ 658 #ifdef DEBUGGING 659 STRLEN charcount; /* Build only */ 660 #endif 661 }; 662 /* There is one (3 under DEBUGGING) pointers that logically belong in this 663 structure, but are held outside as they need duplication on thread cloning, 664 whereas the rest of the structure can be read only: 665 HV *widecharmap; code points > 255 to charid 666 #ifdef DEBUGGING 667 AV *words; Array of words contained in trie, for dumping 668 AV *revcharmap; Map of each charid back to its character representation 669 #endif 670 */ 671 672 #define TRIE_WORDS_OFFSET 2 673 674 typedef struct _reg_trie_data reg_trie_data; 675 676 /* refcount is first in both this and _reg_trie_data to allow a space 677 optimisation in Perl_regdupe. */ 678 struct _reg_ac_data { 679 U32 refcount; 680 U32 trie; 681 U32 *fail; 682 reg_trie_state *states; 683 }; 684 typedef struct _reg_ac_data reg_ac_data; 685 686 /* ANY_BIT doesn't use the structure, so we can borrow it here. 687 This is simpler than refactoring all of it as wed end up with 688 three different sets... */ 689 690 #define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap) 691 #define TRIE_BITMAP_BYTE(p, c) (TRIE_BITMAP(p)[(((U8)(c)) >> 3) & 31]) 692 #define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c)) 693 #define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c)) 694 #define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c)) 695 696 #define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC) 697 #define IS_TRIE_AC(op) ((op)>=AHOCORASICK) 698 699 700 #define BITMAP_BYTE(p, c) (((U8*)p)[(((U8)(c)) >> 3) & 31]) 701 #define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c)) 702 703 /* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */ 704 #define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1) 705 #define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state)) 706 #define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state)) 707 708 #ifdef DEBUGGING 709 #define TRIE_CHARCOUNT(trie) ((trie)->charcount) 710 #else 711 #define TRIE_CHARCOUNT(trie) (trie_charcount) 712 #endif 713 714 #define RE_TRIE_MAXBUF_INIT 65536 715 #define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF" 716 #define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS" 717 718 /* 719 720 RE_DEBUG_FLAGS is used to control what debug output is emitted 721 its divided into three groups of options, some of which interact. 722 The three groups are: Compile, Execute, Extra. There is room for a 723 further group, as currently only the low three bytes are used. 724 725 Compile Options: 726 727 PARSE 728 PEEP 729 TRIE 730 PROGRAM 731 OFFSETS 732 733 Execute Options: 734 735 INTUIT 736 MATCH 737 TRIE 738 739 Extra Options 740 741 TRIE 742 OFFSETS 743 744 If you modify any of these make sure you make corresponding changes to 745 re.pm, especially to the documentation. 746 747 */ 748 749 750 /* Compile */ 751 #define RE_DEBUG_COMPILE_MASK 0x0000FF 752 #define RE_DEBUG_COMPILE_PARSE 0x000001 753 #define RE_DEBUG_COMPILE_OPTIMISE 0x000002 754 #define RE_DEBUG_COMPILE_TRIE 0x000004 755 #define RE_DEBUG_COMPILE_DUMP 0x000008 756 #define RE_DEBUG_COMPILE_FLAGS 0x000010 757 758 /* Execute */ 759 #define RE_DEBUG_EXECUTE_MASK 0x00FF00 760 #define RE_DEBUG_EXECUTE_INTUIT 0x000100 761 #define RE_DEBUG_EXECUTE_MATCH 0x000200 762 #define RE_DEBUG_EXECUTE_TRIE 0x000400 763 764 /* Extra */ 765 #define RE_DEBUG_EXTRA_MASK 0xFF0000 766 #define RE_DEBUG_EXTRA_TRIE 0x010000 767 #define RE_DEBUG_EXTRA_OFFSETS 0x020000 768 #define RE_DEBUG_EXTRA_OFFDEBUG 0x040000 769 #define RE_DEBUG_EXTRA_STATE 0x080000 770 #define RE_DEBUG_EXTRA_OPTIMISE 0x100000 771 #define RE_DEBUG_EXTRA_BUFFERS 0x400000 772 #define RE_DEBUG_EXTRA_GPOS 0x800000 773 /* combined */ 774 #define RE_DEBUG_EXTRA_STACK 0x280000 775 776 #define RE_DEBUG_FLAG(x) (re_debug_flags & x) 777 /* Compile */ 778 #define DEBUG_COMPILE_r(x) DEBUG_r( \ 779 if (re_debug_flags & RE_DEBUG_COMPILE_MASK) x ) 780 #define DEBUG_PARSE_r(x) DEBUG_r( \ 781 if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x ) 782 #define DEBUG_OPTIMISE_r(x) DEBUG_r( \ 783 if (re_debug_flags & RE_DEBUG_COMPILE_OPTIMISE) x ) 784 #define DEBUG_PARSE_r(x) DEBUG_r( \ 785 if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x ) 786 #define DEBUG_DUMP_r(x) DEBUG_r( \ 787 if (re_debug_flags & RE_DEBUG_COMPILE_DUMP) x ) 788 #define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \ 789 if (re_debug_flags & RE_DEBUG_COMPILE_TRIE) x ) 790 #define DEBUG_FLAGS_r(x) DEBUG_r( \ 791 if (re_debug_flags & RE_DEBUG_COMPILE_FLAGS) x ) 792 /* Execute */ 793 #define DEBUG_EXECUTE_r(x) DEBUG_r( \ 794 if (re_debug_flags & RE_DEBUG_EXECUTE_MASK) x ) 795 #define DEBUG_INTUIT_r(x) DEBUG_r( \ 796 if (re_debug_flags & RE_DEBUG_EXECUTE_INTUIT) x ) 797 #define DEBUG_MATCH_r(x) DEBUG_r( \ 798 if (re_debug_flags & RE_DEBUG_EXECUTE_MATCH) x ) 799 #define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \ 800 if (re_debug_flags & RE_DEBUG_EXECUTE_TRIE) x ) 801 802 /* Extra */ 803 #define DEBUG_EXTRA_r(x) DEBUG_r( \ 804 if (re_debug_flags & RE_DEBUG_EXTRA_MASK) x ) 805 #define DEBUG_OFFSETS_r(x) DEBUG_r( \ 806 if (re_debug_flags & RE_DEBUG_EXTRA_OFFSETS) x ) 807 #define DEBUG_STATE_r(x) DEBUG_r( \ 808 if (re_debug_flags & RE_DEBUG_EXTRA_STATE) x ) 809 #define DEBUG_STACK_r(x) DEBUG_r( \ 810 if (re_debug_flags & RE_DEBUG_EXTRA_STACK) x ) 811 #define DEBUG_BUFFERS_r(x) DEBUG_r( \ 812 if (re_debug_flags & RE_DEBUG_EXTRA_BUFFERS) x ) 813 814 #define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \ 815 if ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \ 816 (re_debug_flags & (RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE)) ) x ) 817 #define MJD_OFFSET_DEBUG(x) DEBUG_r( \ 818 if (re_debug_flags & RE_DEBUG_EXTRA_OFFDEBUG) \ 819 Perl_warn_nocontext x ) 820 #define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \ 821 if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x ) 822 #define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \ 823 if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x ) 824 825 #define DEBUG_TRIE_r(x) DEBUG_r( \ 826 if (re_debug_flags & (RE_DEBUG_COMPILE_TRIE \ 827 | RE_DEBUG_EXECUTE_TRIE )) x ) 828 #define DEBUG_GPOS_r(x) DEBUG_r( \ 829 if (re_debug_flags & RE_DEBUG_EXTRA_GPOS) x ) 830 831 /* initialization */ 832 /* get_sv() can return NULL during global destruction. */ 833 #define GET_RE_DEBUG_FLAGS DEBUG_r({ \ 834 SV * re_debug_flags_sv = NULL; \ 835 re_debug_flags_sv = get_sv(RE_DEBUG_FLAGS, 1); \ 836 if (re_debug_flags_sv) { \ 837 if (!SvIOK(re_debug_flags_sv)) \ 838 sv_setuv(re_debug_flags_sv, RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \ 839 re_debug_flags=SvIV(re_debug_flags_sv); \ 840 }\ 841 }) 842 843 #ifdef DEBUGGING 844 845 #define GET_RE_DEBUG_FLAGS_DECL VOL IV re_debug_flags = 0; \ 846 PERL_UNUSED_VAR(re_debug_flags); GET_RE_DEBUG_FLAGS; 847 848 #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \ 849 const char * const rpv = \ 850 pv_pretty((dsv), (pv), (l), (m), \ 851 PL_colors[(c1)],PL_colors[(c2)], \ 852 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \ 853 const int rlen = SvCUR(dsv) 854 855 #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \ 856 const char * const rpv = \ 857 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \ 858 PL_colors[(c1)],PL_colors[(c2)], \ 859 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ) 860 861 #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \ 862 const char * const rpv = \ 863 pv_pretty((dsv), (pv), (l), (m), \ 864 PL_colors[0], PL_colors[1], \ 865 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \ 866 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \ 867 ) 868 869 #define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0)) 870 #define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "") 871 872 #else /* if not DEBUGGING */ 873 874 #define GET_RE_DEBUG_FLAGS_DECL 875 #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) 876 #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) 877 #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) 878 #define RE_SV_DUMPLEN(ItEm) 879 #define RE_SV_TAIL(ItEm) 880 881 #endif /* DEBUG RELATED DEFINES */ 882 883 /* 884 * Local variables: 885 * c-indentation-style: bsd 886 * c-basic-offset: 4 887 * indent-tabs-mode: nil 888 * End: 889 * 890 * ex: set ts=8 sts=4 sw=4 et: 891 */ 892