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
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.
10 #include "regcharclass.h"
12 /* Convert branch sequences to more efficient trie ops? */
13 #define PERL_ENABLE_TRIE_OPTIMISATION 1
15 /* Be really aggressive about optimising patterns with trie sequences? */
16 #define PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION 1
18 /* Should the optimiser take positive assertions into account? */
19 #define PERL_ENABLE_POSITIVE_ASSERTION_STUDY 0
21 /* Not for production use: */
22 #define PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS 0
24 /* Activate offsets code - set to if 1 to enable */
26 #define RE_TRACK_PATTERN_OFFSETS
30 * The "internal use only" fields in regexp.h are present to pass info from
31 * compile to execute that permits the execute phase to run lots faster on
32 * simple cases. They are:
34 * regstart sv that must begin a match; NULL if none obvious
35 * reganch is the match anchored (at beginning-of-line only)?
36 * regmust string (pointer into program) that match must include, or NULL
37 * [regmust changed to SV* for bminstr()--law]
38 * regmlen length of regmust string
39 * [regmlen not used currently]
41 * Regstart and reganch permit very fast decisions on suitable starting points
42 * for a match, cutting down the work a lot. Regmust permits fast rejection
43 * of lines that cannot possibly match. The regmust tests are costly enough
44 * that pregcomp() supplies a regmust only if the r.e. contains something
45 * potentially expensive (at present, the only such thing detected is * or +
46 * at the start of the r.e., which can involve a lot of backup). Regmlen is
47 * supplied because the test in pregexec() needs it and pregcomp() is computing
49 * [regmust is now supplied always. The tests that use regmust have a
50 * heuristic that disables the test if it usually matches.]
52 * [In fact, we now use regmust in many cases to locate where the search
53 * starts in the string, so if regback is >= 0, the regmust search is never
54 * wasted effort. The regback variable says how many characters back from
55 * where regmust matched is the earliest possible start of the match.
56 * For instance, /[a-z].foo/ has a regmust of 'foo' and a regback of 2.]
60 * Structure for regexp "program". This is essentially a linear encoding
61 * of a nondeterministic finite-state machine (aka syntax charts or
62 * "railroad normal form" in parsing technology). Each node is an opcode
63 * plus a "next" pointer, possibly plus an operand. "Next" pointers of
64 * all nodes except BRANCH implement concatenation; a "next" pointer with
65 * a BRANCH on both ends of it is connecting two alternatives. (Here we
66 * have one of the subtle syntax dependencies: an individual BRANCH (as
67 * opposed to a collection of them) is never concatenated with anything
68 * because of operator precedence.) The operand of some types of node is
69 * a literal string; for others, it is a node leading into a sub-FSM. In
70 * particular, the operand of a BRANCH node is the first node of the branch.
71 * (NB this is *not* a tree structure: the tail of the branch connects
72 * to the thing following the set of BRANCHes.) The opcodes are defined
73 * in regnodes.h which is generated from regcomp.sym by regcomp.pl.
77 * A node is one char of opcode followed by two chars of "next" pointer.
78 * "Next" pointers are stored as two 8-bit pieces, high order first. The
79 * value is a positive offset from the opcode of the node containing it.
80 * An operand, if any, simply follows the node. (Note that much of the
81 * code generation knows about this implicit relationship.)
83 * Using two bytes for the "next" pointer is vast overkill for most things,
84 * but allows patterns to get big without disasters.
86 * [The "next" pointer is always aligned on an even
87 * boundary, and reads the offset directly as a short. Also, there is no
88 * special test to reverse the sign of BACK pointers since the offset is
92 /* This is the stuff that used to live in regexp.h that was truly
93 private to the engine itself. It now lives here. */
97 typedef struct regexp_internal {
98 int name_list_idx; /* Optional data index of an array of paren names */
100 U32 *offsets; /* offset annotations 20001228 MJD
101 data about mapping the program to the
103 offsets[0] is proglen when this is used
108 regnode *regstclass; /* Optional startclass as identified or constructed
110 struct reg_data *data; /* Additional miscellaneous data used by the program.
111 Used to make it easier to clone and free arbitrary
112 data that the regops need. Often the ARG field of
113 a regop is an index into this structure */
114 struct reg_code_block *code_blocks;/* positions of literal (?{}) */
115 int num_code_blocks; /* size of code_blocks[] */
116 regnode program[1]; /* Unwarranted chumminess with compiler. */
119 #define RXi_SET(x,y) (x)->pprivate = (void*)(y)
120 #define RXi_GET(x) ((regexp_internal *)((x)->pprivate))
121 #define RXi_GET_DECL(r,ri) regexp_internal *ri = RXi_GET(r)
123 * Flags stored in regexp->intflags
124 * These are used only internally to the regexp engine
126 * See regexp.h for flags used externally to the regexp engine
128 #define RXp_INTFLAGS(rx) ((rx)->intflags)
129 #define RX_INTFLAGS(prog) RXp_INTFLAGS(ReANY(prog))
131 #define PREGf_SKIP 0x00000001
132 #define PREGf_IMPLICIT 0x00000002 /* Converted .* to ^.* */
133 #define PREGf_NAUGHTY 0x00000004 /* how exponential is this pattern? */
134 #define PREGf_VERBARG_SEEN 0x00000008
135 #define PREGf_CUTGROUP_SEEN 0x00000010
136 #define PREGf_USE_RE_EVAL 0x00000020 /* compiled with "use re 'eval'" */
137 /* these used to be extflags, but are now intflags */
138 #define PREGf_NOSCAN 0x00000040
139 #define PREGf_CANY_SEEN 0x00000080
140 #define PREGf_GPOS_SEEN 0x00000100
141 #define PREGf_GPOS_FLOAT 0x00000200
143 #define PREGf_ANCH_BOL 0x00000400
144 #define PREGf_ANCH_MBOL 0x00000800
145 #define PREGf_ANCH_SBOL 0x00001000
146 #define PREGf_ANCH_GPOS 0x00002000
148 #define PREGf_ANCH (PREGf_ANCH_SBOL | PREGf_ANCH_GPOS | \
149 PREGf_ANCH_MBOL | PREGf_ANCH_BOL )
151 /* this is where the old regcomp.h started */
153 struct regnode_string {
160 /* Argument bearing node - workhorse,
161 arg1 is often for the data field */
169 /* Similar to a regnode_1 but with an extra signed argument */
178 /* 'Two field' -- Two 16 bit unsigned args */
187 #define NUM_ANYOF_CODE_POINTS 256
189 #define ANYOF_BITMAP_SIZE (NUM_ANYOF_CODE_POINTS / 8) /* 8 bits/Byte */
191 /* Note that these form structs which are supersets of the next smaller one, by
192 * appending fields. Alignment problems can occur if one of those optional
193 * fields requires stricter alignment than the base struct. And formal
194 * parameters that can really be two or more of the structs should be
195 * declared as the smallest one it could be. See commit message for
196 * 7dcac5f6a5195002b55c935ee1d67f67e1df280b. Regnode allocation is done
197 * without regard to alignment, and changing it to would also require changing
198 * the code that inserts and deletes regnodes. The basic single-argument
199 * regnode has a U32, which is what reganode() allocates as a unit. Therefore
200 * no field can require stricter alignment than U32. */
202 /* also used by trie */
203 struct regnode_charclass {
208 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */
211 /* has runtime (locale) \d, \w, ..., [:posix:] classes */
212 struct regnode_charclass_class {
213 U8 flags; /* ANYOF_POSIXL bit must go here */
217 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time */
218 U32 classflags; /* and run-time */
221 /* A synthetic start class (SSC); is a regnode_charclass_posixl_fold, plus an
222 * extra SV*, used only during its construction and which is not used by
223 * regexec.c. Note that the 'next_off' field is unused, as the SSC stands
224 * alone, so there is never a next node. Also, there is no alignment issue,
225 * becase these are declared or allocated as a complete unit so the compiler
226 * takes care of alignment. This is unlike the other regnodes which are
227 * allocated in terms of multiples of a single-argument regnode. SSC nodes can
228 * have a pointer field because there is no alignment issue, and because it is
229 * set to NULL after construction, before any cloning of the pattern */
231 U8 flags; /* ANYOF_POSIXL bit must go here */
235 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time */
236 U32 classflags; /* and run-time */
238 /* Auxiliary, only used during construction; NULL afterwards: list of code
243 /* We take advantage of 'next_off' not otherwise being used in the SSC by
244 * actually using it: by setting it to 1. This allows us to test and
245 * distinguish between an SSC and other ANYOF node types, as 'next_off' cannot
246 * otherwise be 1, because it is the offset to the next regnode expressed in
247 * units of regnodes. Since an ANYOF node contains extra fields, it adds up
248 * to 12 regnode units on 32-bit systems, (hence the minimum this can be (if
249 * not 0) is 11 there. Even if things get tightly packed on a 64-bit system,
250 * it still would be more than 1. */
251 #define set_ANYOF_SYNTHETIC(n) STMT_START{ OP(n) = ANYOF; \
254 #define is_ANYOF_SYNTHETIC(n) (OP(n) == ANYOF && NEXT_OFF(n) == 1)
256 /* XXX fix this description.
257 Impose a limit of REG_INFTY on various pattern matching operations
258 to limit stack growth and to avoid "infinite" recursions.
260 /* The default size for REG_INFTY is I16_MAX, which is the same as
261 SHORT_MAX (see perl.h). Unfortunately I16 isn't necessarily 16 bits
262 (see handy.h). On the Cray C90, sizeof(short)==4 and hence I16_MAX is
263 ((1<<31)-1), while on the Cray T90, sizeof(short)==8 and I16_MAX is
264 ((1<<63)-1). To limit stack growth to reasonable sizes, supply a
266 --Andy Dougherty 11 June 1998
270 # define REG_INFTY ((1<<15)-1)
275 # define REG_INFTY I16_MAX
278 #define ARG_VALUE(arg) (arg)
279 #define ARG__SET(arg,val) ((arg) = (val))
285 #define ARG(p) ARG_VALUE(ARG_LOC(p))
286 #define ARG1(p) ARG_VALUE(ARG1_LOC(p))
287 #define ARG2(p) ARG_VALUE(ARG2_LOC(p))
288 #define ARG2L(p) ARG_VALUE(ARG2L_LOC(p))
290 #define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val))
291 #define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val))
292 #define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val))
293 #define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val))
298 #define NEXT_OFF(p) ((p)->next_off)
299 #define NODE_ALIGN(node)
300 #define NODE_ALIGN_FILL(node) ((node)->flags = 0xde) /* deadbeef */
302 #define SIZE_ALIGN NODE_ALIGN
309 #define OP(p) ((p)->type)
310 #define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \
311 regnode types. For some, it's the \
312 character set of the regnode */
313 #define OPERAND(p) (((struct regnode_string *)p)->string)
314 #define MASK(p) ((char*)OPERAND(p))
315 #define STR_LEN(p) (((struct regnode_string *)p)->str_len)
316 #define STRING(p) (((struct regnode_string *)p)->string)
317 #define STR_SZ(l) ((l + sizeof(regnode) - 1) / sizeof(regnode))
318 #define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p))+1)
325 #define NODE_ALIGN(node)
326 #define ARG_LOC(p) (((struct regnode_1 *)p)->arg1)
327 #define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1)
328 #define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2)
329 #define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2)
331 #define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */
332 #define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2)
334 #define NODE_STEP_B 4
336 #define NEXTOPER(p) ((p) + NODE_STEP_REGNODE)
337 #define PREVOPER(p) ((p) - NODE_STEP_REGNODE)
339 #define FILL_ADVANCE_NODE(ptr, op) STMT_START { \
340 (ptr)->type = op; (ptr)->next_off = 0; (ptr)++; } STMT_END
341 #define FILL_ADVANCE_NODE_ARG(ptr, op, arg) STMT_START { \
342 ARG_SET(ptr, arg); FILL_ADVANCE_NODE(ptr, op); (ptr) += 1; } STMT_END
344 #define REG_MAGIC 0234
346 #define SIZE_ONLY (RExC_emit == (regnode *) & RExC_emit_dummy)
347 #define PASS1 SIZE_ONLY
348 #define PASS2 (! SIZE_ONLY)
350 /* If the bitmap doesn't fully represent what this ANYOF node can match, the
351 * ARG is set to this special value (since 0, 1, ... are legal, but will never
352 * reach this high). */
353 #define ANYOF_NONBITMAP_EMPTY ((U32) -1)
355 /* Flags for node->flags of ANYOF. These are in short supply, with none
356 * currently available. The ABOVE_LATIN1_ALL bit could be freed up
357 * by resorting to creating a swash containing everything above 255. This
358 * introduces a performance penalty. An option that wouldn't slow things down
359 * would be to split one of the two LOC flags out into a separate
360 * node, like what was done with ANYOF_NON_UTF8_NON_ASCII_ALL in commit
361 * 34fdef848b1687b91892ba55e9e0c3430e0770f6 (but which was reverted because it
362 * wasn't the best option available at the time), and using a LOC flag is
363 * probably better than that commit anyway. But it could be reinstated if we
364 * need a bit. The LOC flags are only for /l nodes; the reverted commit was
365 * only for /d, so there are no combinatorial issues. The LOC flag to use is
366 * probably the POSIXL one.
367 * Several flags are not used in synthetic start class (SSC) nodes, so could be
368 * shared should new flags be needed for SSCs, like ANYOF_EMPTY_STRING now. */
370 /* regexec.c is expecting this to be in the low bit */
371 #define ANYOF_INVERT 0x01
373 /* For the SSC node only, which cannot be inverted, so is shared with that bit.
374 * This means "Does this SSC match an empty string?" This is used only during
375 * regex compilation. */
376 #define ANYOF_EMPTY_STRING ANYOF_INVERT
378 /* Are there things that will match only if the target string is encoded in
379 * UTF-8? (This is not set if ANYOF_AOVE_LATIN1_ALL is set) */
380 #define ANYOF_UTF8 0x02
382 /* The fold is calculated and stored in the bitmap where possible at compile
383 * time. However under locale, the actual folding varies depending on
384 * what the locale is at the time of execution, so it has to be deferred until
386 #define ANYOF_LOC_FOLD 0x04
388 /* Set if this is a regnode_charclass_posixl vs a regnode_charclass. This
389 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale
390 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved
392 #define ANYOF_POSIXL 0x08
393 #define ANYOF_CLASS ANYOF_POSIXL
394 #define ANYOF_LARGE ANYOF_POSIXL
396 /* Should we raise a warning if matching against an above-Unicode code point?
398 #define ANYOF_WARN_SUPER 0x10
400 /* Can match something outside the bitmap that isn't in utf8 */
401 #define ANYOF_NONBITMAP_NON_UTF8 0x20
403 /* Matches every code point 0x100 and above*/
404 #define ANYOF_ABOVE_LATIN1_ALL 0x40
405 #define ANYOF_UNICODE_ALL ANYOF_ABOVE_LATIN1_ALL
407 /* Match all Latin1 characters that aren't ASCII when the target string is not
409 #define ANYOF_NON_UTF8_NON_ASCII_ALL 0x80
411 #define ANYOF_FLAGS_ALL (0xff)
413 #define ANYOF_LOCALE_FLAGS (ANYOF_LOC_FOLD | ANYOF_POSIXL)
415 /* These are the flags that apply to both regular ANYOF nodes and synthetic
416 * start class nodes during construction of the SSC. During finalization of
417 * the SSC, other of the flags could be added to it */
418 #define ANYOF_COMMON_FLAGS (ANYOF_WARN_SUPER|ANYOF_UTF8)
420 /* Character classes for node->classflags of ANYOF */
421 /* Should be synchronized with a table in regprop() */
422 /* 2n should be the normal one, paired with its complement at 2n+1 */
424 #define ANYOF_ALPHA ((_CC_ALPHA) * 2)
425 #define ANYOF_NALPHA ((ANYOF_ALPHA) + 1)
426 #define ANYOF_ALPHANUMERIC ((_CC_ALPHANUMERIC) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */
427 #define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1)
428 #define ANYOF_ASCII ((_CC_ASCII) * 2)
429 #define ANYOF_NASCII ((ANYOF_ASCII) + 1)
430 #define ANYOF_BLANK ((_CC_BLANK) * 2) /* GNU extension: space and tab: non-vertical space */
431 #define ANYOF_NBLANK ((ANYOF_BLANK) + 1)
432 #define ANYOF_CASED ((_CC_CASED) * 2) /* Pseudo class for [:lower:] or
433 [:upper:] under /i */
434 #define ANYOF_NCASED ((ANYOF_CASED) + 1)
435 #define ANYOF_CNTRL ((_CC_CNTRL) * 2)
436 #define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1)
437 #define ANYOF_DIGIT ((_CC_DIGIT) * 2) /* \d */
438 #define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1)
439 #define ANYOF_GRAPH ((_CC_GRAPH) * 2)
440 #define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1)
441 #define ANYOF_LOWER ((_CC_LOWER) * 2)
442 #define ANYOF_NLOWER ((ANYOF_LOWER) + 1)
443 #define ANYOF_PRINT ((_CC_PRINT) * 2)
444 #define ANYOF_NPRINT ((ANYOF_PRINT) + 1)
445 #define ANYOF_PSXSPC ((_CC_PSXSPC) * 2) /* POSIX space: \s plus the vertical tab */
446 #define ANYOF_NPSXSPC ((ANYOF_PSXSPC) + 1)
447 #define ANYOF_PUNCT ((_CC_PUNCT) * 2)
448 #define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1)
449 #define ANYOF_SPACE ((_CC_SPACE) * 2) /* \s */
450 #define ANYOF_NSPACE ((ANYOF_SPACE) + 1)
451 #define ANYOF_UPPER ((_CC_UPPER) * 2)
452 #define ANYOF_NUPPER ((ANYOF_UPPER) + 1)
453 #define ANYOF_WORDCHAR ((_CC_WORDCHAR) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */
454 #define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1)
455 #define ANYOF_XDIGIT ((_CC_XDIGIT) * 2)
456 #define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1)
458 /* pseudo classes below this, not stored in the class bitmap, but used as flags
459 during compilation of char classes */
461 #define ANYOF_VERTWS ((_CC_VERTSPACE) * 2)
462 #define ANYOF_NVERTWS ((ANYOF_VERTWS)+1)
464 /* It is best if this is the last one, as all above it are stored as bits in a
465 * bitmap, and it isn't part of that bitmap */
466 #if _CC_VERTSPACE != _HIGHEST_REGCOMP_DOT_H_SYNC
467 # error Problem with handy.h _HIGHEST_REGCOMP_DOT_H_SYNC #define
470 #define ANYOF_POSIXL_MAX (ANYOF_VERTWS) /* So upper loop limit is written:
472 * Hence doesn't include VERTWS, as that
473 * is a pseudo class */
474 #define ANYOF_MAX ANYOF_POSIXL_MAX
476 #if (ANYOF_POSIXL_MAX > 32) /* Must fit in 32-bit word */
477 # error Problem with handy.h _CC_foo #defines
480 #define ANYOF_HORIZWS ((ANYOF_POSIXL_MAX)+2) /* = (ANYOF_NVERTWS + 1) */
481 #define ANYOF_NHORIZWS ((ANYOF_POSIXL_MAX)+3)
483 #define ANYOF_UNIPROP ((ANYOF_POSIXL_MAX)+4) /* Used to indicate a Unicode
484 property: \p{} or \P{} */
486 /* Backward source code compatibility. */
488 #define ANYOF_ALNUML ANYOF_ALNUM
489 #define ANYOF_NALNUML ANYOF_NALNUM
490 #define ANYOF_SPACEL ANYOF_SPACE
491 #define ANYOF_NSPACEL ANYOF_NSPACE
492 #define ANYOF_ALNUM ANYOF_WORDCHAR
493 #define ANYOF_NALNUM ANYOF_NWORDCHAR
495 /* Utility macros for the bitmap and classes of ANYOF */
497 #define ANYOF_SIZE (sizeof(struct regnode_charclass))
498 #define ANYOF_POSIXL_SIZE (sizeof(regnode_charclass_posixl))
499 #define ANYOF_CLASS_SIZE ANYOF_POSIXL_SIZE
501 #define ANYOF_FLAGS(p) ((p)->flags)
503 #define ANYOF_BIT(c) (1 << ((c) & 7))
505 #define ANYOF_POSIXL_SET(p, c) (((regnode_charclass_posixl*) (p))->classflags |= (1U << (c)))
506 #define ANYOF_CLASS_SET(p, c) ANYOF_POSIXL_SET((p), (c))
508 #define ANYOF_POSIXL_CLEAR(p, c) (((regnode_charclass_posixl*) (p))->classflags &= ~ (1U <<(c)))
509 #define ANYOF_CLASS_CLEAR(p, c) ANYOF_POSIXL_CLEAR((p), (c))
511 #define ANYOF_POSIXL_TEST(p, c) (((regnode_charclass_posixl*) (p))->classflags & (1U << (c)))
512 #define ANYOF_CLASS_TEST(p, c) ANYOF_POSIXL_TEST((p), (c))
514 #define ANYOF_POSIXL_ZERO(ret) STMT_START { ((regnode_charclass_posixl*) (ret))->classflags = 0; } STMT_END
515 #define ANYOF_CLASS_ZERO(ret) ANYOF_POSIXL_ZERO(ret)
517 /* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */
518 #define ANYOF_POSIXL_SETALL(ret) STMT_START { ((regnode_charclass_posixl*) (ret))->classflags = ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1; } STMT_END
519 #define ANYOF_CLASS_SETALL(ret) ANYOF_POSIXL_SETALL(ret)
521 #define ANYOF_POSIXL_TEST_ANY_SET(p) \
522 ((ANYOF_FLAGS(p) & ANYOF_POSIXL) \
523 && (((regnode_charclass_posixl*)(p))->classflags))
524 #define ANYOF_CLASS_TEST_ANY_SET(p) ANYOF_POSIXL_TEST_ANY_SET(p)
526 /* Since an SSC always has this field, we don't have to test for that; nor do
527 * we want to because the bit isn't set for SSC during its construction */
528 #define ANYOF_POSIXL_SSC_TEST_ANY_SET(p) \
529 cBOOL(((regnode_ssc*)(p))->classflags)
530 #define ANYOF_POSIXL_SSC_TEST_ALL_SET(p) /* Are all bits set? */ \
531 (((regnode_ssc*) (p))->classflags \
532 == ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1)
534 #define ANYOF_POSIXL_TEST_ALL_SET(p) \
535 ((ANYOF_FLAGS(p) & ANYOF_POSIXL) \
536 && ((regnode_charclass_posixl*) (p))->classflags \
537 == ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1)
539 #define ANYOF_POSIXL_OR(source, dest) STMT_START { (dest)->classflags |= (source)->classflags ; } STMT_END
540 #define ANYOF_CLASS_OR(source, dest) ANYOF_POSIXL_OR((source), (dest))
542 #define ANYOF_POSIXL_AND(source, dest) STMT_START { (dest)->classflags &= (source)->classflags ; } STMT_END
544 #define ANYOF_BITMAP_ZERO(ret) Zero(((struct regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char)
545 #define ANYOF_BITMAP(p) (((struct regnode_charclass*)(p))->bitmap)
546 #define ANYOF_BITMAP_BYTE(p, c) (ANYOF_BITMAP(p)[(((U8)(c)) >> 3) & 31])
547 #define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c))
548 #define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c))
549 #define ANYOF_BITMAP_TEST(p, c) (ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c))
551 #define ANYOF_BITMAP_SETALL(p) \
552 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE)
553 #define ANYOF_BITMAP_CLEARALL(p) \
554 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE)
555 #if ANYOF_BITMAP_SIZE == 32
556 /* Check that all 256 bits are all set. */
557 # define ANYOF_BITMAP_TESTALLSET(p) /* Assumes sizeof(p) == 32 */ \
558 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)
560 # error Need to fix this if raise bitmap size. (As of this writing this macro is unused in the core)
563 #define ANYOF_SKIP ((ANYOF_SIZE - 1)/sizeof(regnode))
564 #define ANYOF_POSIXL_SKIP ((ANYOF_POSIXL_SIZE - 1)/sizeof(regnode))
565 #define ANYOF_CLASS_SKIP ANYOF_POSIXL_SKIP
568 * Utility definitions.
571 # define UCHARAT(p) ((int)*(const U8*)(p))
573 # define UCHARAT(p) ((int)*(p)&CHARMASK)
576 #define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode))
578 #define REG_ZERO_LEN_SEEN 0x00000001
579 #define REG_LOOKBEHIND_SEEN 0x00000002
580 #define REG_GPOS_SEEN 0x00000004
582 #define REG_CANY_SEEN 0x00000010
583 #define REG_RECURSE_SEEN 0x00000020
584 #define REG_TOP_LEVEL_BRANCHES_SEEN 0x00000040
585 #define REG_VERBARG_SEEN 0x00000080
586 #define REG_CUTGROUP_SEEN 0x00000100
587 #define REG_RUN_ON_COMMENT_SEEN 0x00000200
588 #define REG_UNFOLDED_MULTI_SEEN 0x00000400
589 #define REG_GOSTART_SEEN 0x00000800
590 #define REG_UNBOUNDED_QUANTIFIER_SEEN 0x00001000
595 #ifdef PLUGGABLE_RE_EXTENSION
596 #include "re_nodes.h"
598 #include "regnodes.h"
601 #ifndef PLUGGABLE_RE_EXTENSION
603 EXTCONST regexp_engine PL_core_reg_engine;
605 EXTCONST regexp_engine PL_core_reg_engine = {
608 Perl_re_intuit_start,
609 Perl_re_intuit_string,
610 Perl_regfree_internal,
611 Perl_reg_numbered_buff_fetch,
612 Perl_reg_numbered_buff_store,
613 Perl_reg_numbered_buff_length,
615 Perl_reg_named_buff_iter,
617 #if defined(USE_ITHREADS)
618 Perl_regdupe_internal,
623 #endif /* PLUGGABLE_RE_EXTENSION */
629 /* .what is a character array with one character for each member of .data
630 * The character describes the function of the corresponding .data item:
631 * a - AV for paren_name_list under DEBUGGING
632 * f - start-class data for regstclass optimization
633 * l - start op for literal (?{EVAL}) item
634 * L - start op for literal (?{EVAL}) item, with separate CV (qr//)
635 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/
636 * s - swash for Unicode-style character class, and the multicharacter
637 * strings resulting from casefolding the single-character entries
638 * in the character class
640 * u - trie struct's widecharmap (a HV, so can't share, must dup)
641 * also used for revcharmap and words under DEBUGGING
642 * T - aho-trie struct
643 * S - sv for named capture lookup
644 * 20010712 mjd@plover.com
645 * (Remember to update re_dup() and pregfree() if you add any items.)
653 /* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses
654 anchored* and float* via array indexes 0 and 1. */
655 #define anchored_substr substrs->data[0].substr
656 #define anchored_utf8 substrs->data[0].utf8_substr
657 #define anchored_offset substrs->data[0].min_offset
658 #define anchored_end_shift substrs->data[0].end_shift
660 #define float_substr substrs->data[1].substr
661 #define float_utf8 substrs->data[1].utf8_substr
662 #define float_min_offset substrs->data[1].min_offset
663 #define float_max_offset substrs->data[1].max_offset
664 #define float_end_shift substrs->data[1].end_shift
666 #define check_substr substrs->data[2].substr
667 #define check_utf8 substrs->data[2].utf8_substr
668 #define check_offset_min substrs->data[2].min_offset
669 #define check_offset_max substrs->data[2].max_offset
670 #define check_end_shift substrs->data[2].end_shift
672 #define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr)
673 #define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8)
674 #define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr)
675 #define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8)
677 /* trie related stuff */
679 /* a transition record for the state machine. the
680 check field determines which state "owns" the
681 transition. the char the transition is for is
682 determined by offset from the owning states base
683 field. the next field determines which state
684 is to be transitioned to if any.
686 struct _reg_trie_trans {
691 /* a transition list element for the list based representation */
692 struct _reg_trie_trans_list_elem {
696 typedef struct _reg_trie_trans_list_elem reg_trie_trans_le;
698 /* a state for compressed nodes. base is an offset
699 into an array of reg_trie_trans array. If wordnum is
700 nonzero the state is accepting. if base is zero then
701 the state has no children (and will be accepting)
703 struct _reg_trie_state {
707 reg_trie_trans_le* list;
711 /* info per word; indexed by wordnum */
713 U16 prev; /* previous word in acceptance chain; eg in
714 * zzz|abc|ab/ after matching the chars abc, the
715 * accepted word is #2, and the previous accepted
717 U32 len; /* how many chars long is this word? */
718 U32 accept; /* accept state for this word */
722 typedef struct _reg_trie_state reg_trie_state;
723 typedef struct _reg_trie_trans reg_trie_trans;
726 /* anything in here that needs to be freed later
727 should be dealt with in pregfree.
728 refcount is first in both this and _reg_ac_data to allow a space
729 optimisation in Perl_regdupe. */
730 struct _reg_trie_data {
731 U32 refcount; /* number of times this trie is referenced */
732 U32 lasttrans; /* last valid transition element */
733 U16 *charmap; /* byte to charid lookup array */
734 reg_trie_state *states; /* state data */
735 reg_trie_trans *trans; /* array of transition elements */
736 char *bitmap; /* stclass bitmap */
737 U16 *jump; /* optional 1 indexed array of offsets before tail
738 for the node following a given word. */
739 reg_trie_wordinfo *wordinfo; /* array of info per word */
740 U16 uniquecharcount; /* unique chars in trie (width of trans table) */
741 U32 startstate; /* initial state - used for common prefix optimisation */
742 STRLEN minlen; /* minimum length of words in trie - build/opt only? */
743 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */
744 U32 prefixlen; /* #chars in common prefix */
745 U32 statecount; /* Build only - number of states in the states array
746 (including the unused zero state) */
747 U32 wordcount; /* Build only */
749 STRLEN charcount; /* Build only */
752 /* There is one (3 under DEBUGGING) pointers that logically belong in this
753 structure, but are held outside as they need duplication on thread cloning,
754 whereas the rest of the structure can be read only:
755 HV *widecharmap; code points > 255 to charid
757 AV *words; Array of words contained in trie, for dumping
758 AV *revcharmap; Map of each charid back to its character representation
762 #define TRIE_WORDS_OFFSET 2
764 typedef struct _reg_trie_data reg_trie_data;
766 /* refcount is first in both this and _reg_trie_data to allow a space
767 optimisation in Perl_regdupe. */
768 struct _reg_ac_data {
772 reg_trie_state *states;
774 typedef struct _reg_ac_data reg_ac_data;
776 /* ANY_BIT doesn't use the structure, so we can borrow it here.
777 This is simpler than refactoring all of it as wed end up with
778 three different sets... */
780 #define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap)
781 #define TRIE_BITMAP_BYTE(p, c) (TRIE_BITMAP(p)[(((U8)(c)) >> 3) & 31])
782 #define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c))
783 #define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c))
784 #define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
786 #define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC)
787 #define IS_TRIE_AC(op) ((op)>=AHOCORASICK)
790 #define BITMAP_BYTE(p, c) (((U8*)p)[(((U8)(c)) >> 3) & 31])
791 #define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
793 /* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */
794 #define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1)
795 #define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state))
796 #define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state))
799 #define TRIE_CHARCOUNT(trie) ((trie)->charcount)
801 #define TRIE_CHARCOUNT(trie) (trie_charcount)
804 #define RE_TRIE_MAXBUF_INIT 65536
805 #define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF"
806 #define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS"
810 RE_DEBUG_FLAGS is used to control what debug output is emitted
811 its divided into three groups of options, some of which interact.
812 The three groups are: Compile, Execute, Extra. There is room for a
813 further group, as currently only the low three bytes are used.
834 If you modify any of these make sure you make corresponding changes to
835 re.pm, especially to the documentation.
841 #define RE_DEBUG_COMPILE_MASK 0x0000FF
842 #define RE_DEBUG_COMPILE_PARSE 0x000001
843 #define RE_DEBUG_COMPILE_OPTIMISE 0x000002
844 #define RE_DEBUG_COMPILE_TRIE 0x000004
845 #define RE_DEBUG_COMPILE_DUMP 0x000008
846 #define RE_DEBUG_COMPILE_FLAGS 0x000010
849 #define RE_DEBUG_EXECUTE_MASK 0x00FF00
850 #define RE_DEBUG_EXECUTE_INTUIT 0x000100
851 #define RE_DEBUG_EXECUTE_MATCH 0x000200
852 #define RE_DEBUG_EXECUTE_TRIE 0x000400
855 #define RE_DEBUG_EXTRA_MASK 0xFF0000
856 #define RE_DEBUG_EXTRA_TRIE 0x010000
857 #define RE_DEBUG_EXTRA_OFFSETS 0x020000
858 #define RE_DEBUG_EXTRA_OFFDEBUG 0x040000
859 #define RE_DEBUG_EXTRA_STATE 0x080000
860 #define RE_DEBUG_EXTRA_OPTIMISE 0x100000
861 #define RE_DEBUG_EXTRA_BUFFERS 0x400000
862 #define RE_DEBUG_EXTRA_GPOS 0x800000
864 #define RE_DEBUG_EXTRA_STACK 0x280000
866 #define RE_DEBUG_FLAG(x) (re_debug_flags & x)
868 #define DEBUG_COMPILE_r(x) DEBUG_r( \
869 if (re_debug_flags & RE_DEBUG_COMPILE_MASK) x )
870 #define DEBUG_PARSE_r(x) DEBUG_r( \
871 if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x )
872 #define DEBUG_OPTIMISE_r(x) DEBUG_r( \
873 if (re_debug_flags & RE_DEBUG_COMPILE_OPTIMISE) x )
874 #define DEBUG_PARSE_r(x) DEBUG_r( \
875 if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x )
876 #define DEBUG_DUMP_r(x) DEBUG_r( \
877 if (re_debug_flags & RE_DEBUG_COMPILE_DUMP) x )
878 #define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \
879 if (re_debug_flags & RE_DEBUG_COMPILE_TRIE) x )
880 #define DEBUG_FLAGS_r(x) DEBUG_r( \
881 if (re_debug_flags & RE_DEBUG_COMPILE_FLAGS) x )
883 #define DEBUG_EXECUTE_r(x) DEBUG_r( \
884 if (re_debug_flags & RE_DEBUG_EXECUTE_MASK) x )
885 #define DEBUG_INTUIT_r(x) DEBUG_r( \
886 if (re_debug_flags & RE_DEBUG_EXECUTE_INTUIT) x )
887 #define DEBUG_MATCH_r(x) DEBUG_r( \
888 if (re_debug_flags & RE_DEBUG_EXECUTE_MATCH) x )
889 #define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \
890 if (re_debug_flags & RE_DEBUG_EXECUTE_TRIE) x )
893 #define DEBUG_EXTRA_r(x) DEBUG_r( \
894 if (re_debug_flags & RE_DEBUG_EXTRA_MASK) x )
895 #define DEBUG_OFFSETS_r(x) DEBUG_r( \
896 if (re_debug_flags & RE_DEBUG_EXTRA_OFFSETS) x )
897 #define DEBUG_STATE_r(x) DEBUG_r( \
898 if (re_debug_flags & RE_DEBUG_EXTRA_STATE) x )
899 #define DEBUG_STACK_r(x) DEBUG_r( \
900 if (re_debug_flags & RE_DEBUG_EXTRA_STACK) x )
901 #define DEBUG_BUFFERS_r(x) DEBUG_r( \
902 if (re_debug_flags & RE_DEBUG_EXTRA_BUFFERS) x )
904 #define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \
905 if ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \
906 (re_debug_flags & (RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE)) ) x )
907 #define MJD_OFFSET_DEBUG(x) DEBUG_r( \
908 if (re_debug_flags & RE_DEBUG_EXTRA_OFFDEBUG) \
909 Perl_warn_nocontext x )
910 #define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \
911 if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x )
912 #define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \
913 if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x )
915 #define DEBUG_TRIE_r(x) DEBUG_r( \
916 if (re_debug_flags & (RE_DEBUG_COMPILE_TRIE \
917 | RE_DEBUG_EXECUTE_TRIE )) x )
918 #define DEBUG_GPOS_r(x) DEBUG_r( \
919 if (re_debug_flags & RE_DEBUG_EXTRA_GPOS) x )
922 /* get_sv() can return NULL during global destruction. */
923 #define GET_RE_DEBUG_FLAGS DEBUG_r({ \
924 SV * re_debug_flags_sv = NULL; \
925 re_debug_flags_sv = get_sv(RE_DEBUG_FLAGS, 1); \
926 if (re_debug_flags_sv) { \
927 if (!SvIOK(re_debug_flags_sv)) \
928 sv_setuv(re_debug_flags_sv, RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \
929 re_debug_flags=SvIV(re_debug_flags_sv); \
935 #define GET_RE_DEBUG_FLAGS_DECL VOL IV re_debug_flags = 0; \
936 PERL_UNUSED_VAR(re_debug_flags); GET_RE_DEBUG_FLAGS;
938 #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \
939 const char * const rpv = \
940 pv_pretty((dsv), (pv), (l), (m), \
941 PL_colors[(c1)],PL_colors[(c2)], \
942 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \
943 const int rlen = SvCUR(dsv)
945 #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \
946 const char * const rpv = \
947 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \
948 PL_colors[(c1)],PL_colors[(c2)], \
949 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) )
951 #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \
952 const char * const rpv = \
953 pv_pretty((dsv), (pv), (l), (m), \
954 PL_colors[0], PL_colors[1], \
955 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \
956 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \
959 #define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0))
960 #define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "")
962 #else /* if not DEBUGGING */
964 #define GET_RE_DEBUG_FLAGS_DECL
965 #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2)
966 #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m)
967 #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m)
968 #define RE_SV_DUMPLEN(ItEm)
969 #define RE_SV_TAIL(ItEm)
971 #endif /* DEBUG RELATED DEFINES */
975 * c-indentation-style: bsd
977 * indent-tabs-mode: nil
980 * ex: set ts=8 sts=4 sw=4 et: