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
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.
34 /* #define REG_ALLOW_MINMOD_SUSPEND */
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:
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]
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
56 * [regmust is now supplied always. The tests that use regmust have a
57 * heuristic that disables the test if it usually matches.]
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.]
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.
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.)
90 * Using two bytes for the "next" pointer is vast overkill for most things,
91 * but allows patterns to get big without disasters.
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
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. */
104 typedef struct regexp_internal {
105 int name_list_idx; /* Optional data index of an array of paren names */
107 U32 *offsets; /* offset annotations 20001228 MJD
108 data about mapping the program to the
110 offsets[0] is proglen when this is used
115 regnode *regstclass; /* Optional startclass as identified or constructed
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. */
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)
130 * Flags stored in regexp->intflags
131 * These are used only internally to the regexp engine
133 * See regexp.h for flags used externally to the regexp engine
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'" */
143 /* this is where the old regcomp.h started */
145 struct regnode_string {
152 /* Argument bearing node - workhorse,
153 arg1 is often for the data field */
161 /* Similar to a regnode_1 but with an extra signed argument */
170 /* 'Two field' -- Two 16 bit unsigned args */
180 #define ANYOF_BITMAP_SIZE 32 /* 256 b/(8 b/B) */
181 #define ANYOF_CLASSBITMAP_SIZE 4 /* up to 32 (8*4) named classes */
183 /* also used by trie */
184 struct regnode_charclass {
188 U32 arg1; /* used as ptr in S_regclass */
189 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */
192 /* has runtime (locale) \d, \w, ..., [:posix:] classes */
193 struct regnode_charclass_class {
194 U8 flags; /* ANYOF_CLASS bit must go here */
197 U32 arg1; /* used as ptr in S_regclass */
198 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time */
199 char classflags[ANYOF_CLASSBITMAP_SIZE]; /* and run-time */
202 /* XXX fix this description.
203 Impose a limit of REG_INFTY on various pattern matching operations
204 to limit stack growth and to avoid "infinite" recursions.
206 /* The default size for REG_INFTY is I16_MAX, which is the same as
207 SHORT_MAX (see perl.h). Unfortunately I16 isn't necessarily 16 bits
208 (see handy.h). On the Cray C90, sizeof(short)==4 and hence I16_MAX is
209 ((1<<31)-1), while on the Cray T90, sizeof(short)==8 and I16_MAX is
210 ((1<<63)-1). To limit stack growth to reasonable sizes, supply a
212 --Andy Dougherty 11 June 1998
216 # define REG_INFTY ((1<<15)-1)
221 # define REG_INFTY I16_MAX
224 #define ARG_VALUE(arg) (arg)
225 #define ARG__SET(arg,val) ((arg) = (val))
231 #define ARG(p) ARG_VALUE(ARG_LOC(p))
232 #define ARG1(p) ARG_VALUE(ARG1_LOC(p))
233 #define ARG2(p) ARG_VALUE(ARG2_LOC(p))
234 #define ARG2L(p) ARG_VALUE(ARG2L_LOC(p))
236 #define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val))
237 #define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val))
238 #define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val))
239 #define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val))
244 #define NEXT_OFF(p) ((p)->next_off)
245 #define NODE_ALIGN(node)
246 #define NODE_ALIGN_FILL(node) ((node)->flags = 0xde) /* deadbeef */
248 #define SIZE_ALIGN NODE_ALIGN
255 #define OP(p) ((p)->type)
256 #define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \
257 regnode types. For some, it's the \
258 character set of the regnode */
259 #define OPERAND(p) (((struct regnode_string *)p)->string)
260 #define MASK(p) ((char*)OPERAND(p))
261 #define STR_LEN(p) (((struct regnode_string *)p)->str_len)
262 #define STRING(p) (((struct regnode_string *)p)->string)
263 #define STR_SZ(l) ((l + sizeof(regnode) - 1) / sizeof(regnode))
264 #define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p))+1)
271 #define NODE_ALIGN(node)
272 #define ARG_LOC(p) (((struct regnode_1 *)p)->arg1)
273 #define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1)
274 #define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2)
275 #define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2)
277 #define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */
278 #define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2)
280 #define NODE_STEP_B 4
282 #define NEXTOPER(p) ((p) + NODE_STEP_REGNODE)
283 #define PREVOPER(p) ((p) - NODE_STEP_REGNODE)
285 #define FILL_ADVANCE_NODE(ptr, op) STMT_START { \
286 (ptr)->type = op; (ptr)->next_off = 0; (ptr)++; } STMT_END
287 #define FILL_ADVANCE_NODE_ARG(ptr, op, arg) STMT_START { \
288 ARG_SET(ptr, arg); FILL_ADVANCE_NODE(ptr, op); (ptr) += 1; } STMT_END
290 #define REG_MAGIC 0234
292 #define SIZE_ONLY (RExC_emit == &PL_regdummy)
294 /* If the bitmap doesn't fully represent what this ANYOF node can match, the
295 * ARG is set to this special value (since 0, 1, ... are legal, but will never
296 * reach this high). */
297 #define ANYOF_NONBITMAP_EMPTY ((U32) -1)
299 /* The information used to be stored as as combination of the ANYOF_UTF8 and
300 * ANYOF_NONBITMAP_NON_UTF8 bits in the flags field, but was moved out of there
301 * to free up a bit for other uses. This tries to hide the change from
302 * existing code as much as possible. Now, the data structure that goes in ARG
303 * is not allocated unless it is needed, and that is what is used to determine
304 * if there is something outside the bitmap. The code now assumes that if
305 * that structure exists, that any UTF-8 encoded string should be tried against
306 * it, but a non-UTF8-encoded string will be tried only if the
307 * ANYOF_NONBITMAP_NON_UTF8 bit is also set. */
308 #define ANYOF_NONBITMAP(node) (ARG(node) != ANYOF_NONBITMAP_EMPTY)
310 /* Flags for node->flags of ANYOF. These are in short supply, so some games
311 * are done to share them, as described below. If necessary, the ANYOF_LOCALE
312 * and ANYOF_CLASS bits could be shared with a space penalty for locale nodes,
313 * but this isn't quite so easy, as the optimizer also uses ANYOF_CLASS.
314 * Another option would be to push them into new nodes. E.g. there could be an
315 * ANYOF_LOCALE node that would be in place of the flag of the same name.
316 * Once the planned change to compile all the above-latin1 code points is done,
317 * then the UNICODE_ALL bit can be freed up, with a small performance penalty.
318 * If flags need to be added that are applicable to the synthetic start class
319 * only, with some work, they could be put in the next-node field, or in an
320 * unused bit of the classflags field. */
322 #define ANYOF_LOCALE 0x01 /* /l modifier */
324 /* The fold is calculated and stored in the bitmap where possible at compile
325 * time. However there are two cases where it isn't possible. These share
326 * this bit: 1) under locale, where the actual folding varies depending on
327 * what the locale is at the time of execution; and 2) where the folding is
328 * specified in a swash, not the bitmap, such as characters which aren't
329 * specified in the bitmap, or properties that aren't looked at at compile time
331 #define ANYOF_LOC_NONBITMAP_FOLD 0x02
333 #define ANYOF_INVERT 0x04
335 /* Set if this is a struct regnode_charclass_class vs a regnode_charclass. This
336 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale
337 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved
339 #define ANYOF_CLASS 0x08
340 #define ANYOF_LARGE ANYOF_CLASS /* Same; name retained for back compat */
342 /* EOS, meaning that it can match an empty string too, is used for the
343 * synthetic start class only. */
344 #define ANYOF_EOS 0x10
346 /* ? Is this node the synthetic start class (ssc). This bit is shared with
347 * ANYOF_EOS, as the latter is used only for the ssc, and then not used by
348 * regexec.c. And, the code is structured so that if it is set, the ssc is
349 * not used, so it is guaranteed to be 0 for the ssc by the time regexec.c
350 * gets executed, and 0 for a non-ssc ANYOF node, as it only ever gets set for
351 * a potential ssc candidate. Thus setting it to 1 after it has been
352 * determined that the ssc will be used is not ambiguous */
353 #define ANYOF_IS_SYNTHETIC ANYOF_EOS
355 /* Can match something outside the bitmap that isn't in utf8 */
356 #define ANYOF_NONBITMAP_NON_UTF8 0x20
358 /* Matches every code point 0x100 and above*/
359 #define ANYOF_UNICODE_ALL 0x40
361 /* Match all Latin1 characters that aren't ASCII when the target string is not
363 #define ANYOF_NON_UTF8_LATIN1_ALL 0x80
365 #define ANYOF_FLAGS_ALL 0xff
367 /* These are the flags that ANYOF_INVERT being set or not doesn't affect
368 * whether they are operative or not. e.g., the node still has LOCALE
369 * regardless of being inverted; whereas ANYOF_UNICODE_ALL means something
370 * different if inverted */
371 #define INVERSION_UNAFFECTED_FLAGS (ANYOF_LOCALE \
372 |ANYOF_LOC_NONBITMAP_FOLD \
375 |ANYOF_NONBITMAP_NON_UTF8)
377 /* Character classes for node->classflags of ANYOF */
378 /* Should be synchronized with a table in regprop() */
379 /* 2n should pair with 2n+1 */
381 #define ANYOF_ALNUM 0 /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */
382 #define ANYOF_NALNUM 1
383 #define ANYOF_SPACE 2 /* \s */
384 #define ANYOF_NSPACE 3
385 #define ANYOF_DIGIT 4 /* \d */
386 #define ANYOF_NDIGIT 5
387 #define ANYOF_ALNUMC 6 /* [[:alnum:]] isalnum(3), utf8::IsAlnum, ALNUMC */
388 #define ANYOF_NALNUMC 7
389 #define ANYOF_ALPHA 8
390 #define ANYOF_NALPHA 9
391 #define ANYOF_ASCII 10
392 #define ANYOF_NASCII 11
393 #define ANYOF_CNTRL 12
394 #define ANYOF_NCNTRL 13
395 #define ANYOF_GRAPH 14
396 #define ANYOF_NGRAPH 15
397 #define ANYOF_LOWER 16
398 #define ANYOF_NLOWER 17
399 #define ANYOF_PRINT 18
400 #define ANYOF_NPRINT 19
401 #define ANYOF_PUNCT 20
402 #define ANYOF_NPUNCT 21
403 #define ANYOF_UPPER 22
404 #define ANYOF_NUPPER 23
405 #define ANYOF_XDIGIT 24
406 #define ANYOF_NXDIGIT 25
407 #define ANYOF_PSXSPC 26 /* POSIX space: \s plus the vertical tab */
408 #define ANYOF_NPSXSPC 27
409 #define ANYOF_BLANK 28 /* GNU extension: space and tab: non-vertical space */
410 #define ANYOF_NBLANK 29
414 /* pseudo classes, not stored in the class bitmap, but used as flags
415 during compilation of char classes */
417 #define ANYOF_VERTWS (ANYOF_MAX+1)
418 #define ANYOF_NVERTWS (ANYOF_MAX+2)
419 #define ANYOF_HORIZWS (ANYOF_MAX+3)
420 #define ANYOF_NHORIZWS (ANYOF_MAX+4)
422 /* Backward source code compatibility. */
424 #define ANYOF_ALNUML ANYOF_ALNUM
425 #define ANYOF_NALNUML ANYOF_NALNUM
426 #define ANYOF_SPACEL ANYOF_SPACE
427 #define ANYOF_NSPACEL ANYOF_NSPACE
429 /* Utility macros for the bitmap and classes of ANYOF */
431 #define ANYOF_SIZE (sizeof(struct regnode_charclass))
432 #define ANYOF_CLASS_SIZE (sizeof(struct regnode_charclass_class))
434 #define ANYOF_FLAGS(p) ((p)->flags)
436 #define ANYOF_BIT(c) (1 << ((c) & 7))
438 #define ANYOF_CLASS_BYTE(p, c) (((struct regnode_charclass_class*)(p))->classflags[((c) >> 3) & 3])
439 #define ANYOF_CLASS_SET(p, c) (ANYOF_CLASS_BYTE(p, c) |= ANYOF_BIT(c))
440 #define ANYOF_CLASS_CLEAR(p, c) (ANYOF_CLASS_BYTE(p, c) &= ~ANYOF_BIT(c))
441 #define ANYOF_CLASS_TEST(p, c) (ANYOF_CLASS_BYTE(p, c) & ANYOF_BIT(c))
443 #define ANYOF_CLASS_ZERO(ret) Zero(((struct regnode_charclass_class*)(ret))->classflags, ANYOF_CLASSBITMAP_SIZE, char)
444 #define ANYOF_CLASS_SETALL(ret) \
445 memset (((struct regnode_charclass_class*)(ret))->classflags, 255, ANYOF_CLASSBITMAP_SIZE)
446 #define ANYOF_BITMAP_ZERO(ret) Zero(((struct regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char)
448 #define ANYOF_BITMAP(p) (((struct regnode_charclass*)(p))->bitmap)
449 #define ANYOF_BITMAP_BYTE(p, c) (ANYOF_BITMAP(p)[(((U8)(c)) >> 3) & 31])
450 #define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c))
451 #define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c))
452 #define ANYOF_BITMAP_TEST(p, c) (ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c))
454 #define ANYOF_BITMAP_SETALL(p) \
455 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE)
456 #define ANYOF_BITMAP_CLEARALL(p) \
457 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE)
458 /* Check that all 256 bits are all set. Used in S_cl_is_anything() */
459 #define ANYOF_BITMAP_TESTALLSET(p) /* Assumes sizeof(p) == 32 */ \
460 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)
462 #define ANYOF_SKIP ((ANYOF_SIZE - 1)/sizeof(regnode))
463 #define ANYOF_CLASS_SKIP ((ANYOF_CLASS_SIZE - 1)/sizeof(regnode))
465 #if ANYOF_CLASSBITMAP_SIZE != 4
466 # error ANYOF_CLASSBITMAP_SIZE is expected to be 4
468 #define ANYOF_CLASS_TEST_ANY_SET(p) ((ANYOF_FLAGS(p) & ANYOF_CLASS) \
469 && memNE (((struct regnode_charclass_class*)(p))->classflags, \
470 "\0\0\0\0", ANYOF_CLASSBITMAP_SIZE))
471 /*#define ANYOF_CLASS_ADD_SKIP (ANYOF_CLASS_SKIP - ANYOF_SKIP)
476 * Utility definitions.
479 # define UCHARAT(p) ((int)*(const U8*)(p))
481 # define UCHARAT(p) ((int)*(p)&CHARMASK)
484 #define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode))
486 #define REG_SEEN_ZERO_LEN 0x00000001
487 #define REG_SEEN_LOOKBEHIND 0x00000002
488 #define REG_SEEN_GPOS 0x00000004
490 #define REG_SEEN_CANY 0x00000010
491 #define REG_SEEN_SANY REG_SEEN_CANY /* src bckwrd cmpt */
492 #define REG_SEEN_RECURSE 0x00000020
493 #define REG_TOP_LEVEL_BRANCHES 0x00000040
494 #define REG_SEEN_VERBARG 0x00000080
495 #define REG_SEEN_CUTGROUP 0x00000100
496 #define REG_SEEN_RUN_ON_COMMENT 0x00000200
497 #define REG_SEEN_EXACTF_SHARP_S 0x00000400
501 #ifdef PLUGGABLE_RE_EXTENSION
502 #include "re_nodes.h"
504 #include "regnodes.h"
507 #ifndef PLUGGABLE_RE_EXTENSION
509 EXTCONST regexp_engine PL_core_reg_engine;
511 EXTCONST regexp_engine PL_core_reg_engine = {
514 Perl_re_intuit_start,
515 Perl_re_intuit_string,
516 Perl_regfree_internal,
517 Perl_reg_numbered_buff_fetch,
518 Perl_reg_numbered_buff_store,
519 Perl_reg_numbered_buff_length,
521 Perl_reg_named_buff_iter,
523 #if defined(USE_ITHREADS)
524 Perl_regdupe_internal,
529 #endif /* PLUGGABLE_RE_EXTENSION */
535 /* .what is a character array with one character for each member of .data
536 * The character describes the function of the corresponding .data item:
537 * a - AV for paren_name_list under DEBUGGING
538 * f - start-class data for regstclass optimization
539 * l - start op for literal (?{EVAL}) item
540 * L - start op for literal (?{EVAL}) item, with separate CV (qr//)
541 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/
542 * s - swash for Unicode-style character class, and the multicharacter
543 * strings resulting from casefolding the single-character entries
544 * in the character class
546 * u - trie struct's widecharmap (a HV, so can't share, must dup)
547 * also used for revcharmap and words under DEBUGGING
548 * T - aho-trie struct
549 * S - sv for named capture lookup
550 * 20010712 mjd@plover.com
551 * (Remember to update re_dup() and pregfree() if you add any items.)
559 /* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses
560 anchored* and float* via array indexes 0 and 1. */
561 #define anchored_substr substrs->data[0].substr
562 #define anchored_utf8 substrs->data[0].utf8_substr
563 #define anchored_offset substrs->data[0].min_offset
564 #define anchored_end_shift substrs->data[0].end_shift
566 #define float_substr substrs->data[1].substr
567 #define float_utf8 substrs->data[1].utf8_substr
568 #define float_min_offset substrs->data[1].min_offset
569 #define float_max_offset substrs->data[1].max_offset
570 #define float_end_shift substrs->data[1].end_shift
572 #define check_substr substrs->data[2].substr
573 #define check_utf8 substrs->data[2].utf8_substr
574 #define check_offset_min substrs->data[2].min_offset
575 #define check_offset_max substrs->data[2].max_offset
576 #define check_end_shift substrs->data[2].end_shift
578 #define RX_ANCHORED_SUBSTR(rx) (((struct regexp *)SvANY(rx))->anchored_substr)
579 #define RX_ANCHORED_UTF8(rx) (((struct regexp *)SvANY(rx))->anchored_utf8)
580 #define RX_FLOAT_SUBSTR(rx) (((struct regexp *)SvANY(rx))->float_substr)
581 #define RX_FLOAT_UTF8(rx) (((struct regexp *)SvANY(rx))->float_utf8)
583 /* trie related stuff */
585 /* a transition record for the state machine. the
586 check field determines which state "owns" the
587 transition. the char the transition is for is
588 determined by offset from the owning states base
589 field. the next field determines which state
590 is to be transitioned to if any.
592 struct _reg_trie_trans {
597 /* a transition list element for the list based representation */
598 struct _reg_trie_trans_list_elem {
602 typedef struct _reg_trie_trans_list_elem reg_trie_trans_le;
604 /* a state for compressed nodes. base is an offset
605 into an array of reg_trie_trans array. If wordnum is
606 nonzero the state is accepting. if base is zero then
607 the state has no children (and will be accepting)
609 struct _reg_trie_state {
613 reg_trie_trans_le* list;
617 /* info per word; indexed by wordnum */
619 U16 prev; /* previous word in acceptance chain; eg in
620 * zzz|abc|ab/ after matching the chars abc, the
621 * accepted word is #2, and the previous accepted
623 U32 len; /* how many chars long is this word? */
624 U32 accept; /* accept state for this word */
628 typedef struct _reg_trie_state reg_trie_state;
629 typedef struct _reg_trie_trans reg_trie_trans;
632 /* anything in here that needs to be freed later
633 should be dealt with in pregfree.
634 refcount is first in both this and _reg_ac_data to allow a space
635 optimisation in Perl_regdupe. */
636 struct _reg_trie_data {
637 U32 refcount; /* number of times this trie is referenced */
638 U32 lasttrans; /* last valid transition element */
639 U16 *charmap; /* byte to charid lookup array */
640 reg_trie_state *states; /* state data */
641 reg_trie_trans *trans; /* array of transition elements */
642 char *bitmap; /* stclass bitmap */
643 U16 *jump; /* optional 1 indexed array of offsets before tail
644 for the node following a given word. */
645 reg_trie_wordinfo *wordinfo; /* array of info per word */
646 U16 uniquecharcount; /* unique chars in trie (width of trans table) */
647 U32 startstate; /* initial state - used for common prefix optimisation */
648 STRLEN minlen; /* minimum length of words in trie - build/opt only? */
649 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */
650 U32 prefixlen; /* #chars in common prefix */
651 U32 statecount; /* Build only - number of states in the states array
652 (including the unused zero state) */
653 U32 wordcount; /* Build only */
655 STRLEN charcount; /* Build only */
658 /* There is one (3 under DEBUGGING) pointers that logically belong in this
659 structure, but are held outside as they need duplication on thread cloning,
660 whereas the rest of the structure can be read only:
661 HV *widecharmap; code points > 255 to charid
663 AV *words; Array of words contained in trie, for dumping
664 AV *revcharmap; Map of each charid back to its character representation
668 #define TRIE_WORDS_OFFSET 2
670 typedef struct _reg_trie_data reg_trie_data;
672 /* refcount is first in both this and _reg_trie_data to allow a space
673 optimisation in Perl_regdupe. */
674 struct _reg_ac_data {
678 reg_trie_state *states;
680 typedef struct _reg_ac_data reg_ac_data;
682 /* ANY_BIT doesn't use the structure, so we can borrow it here.
683 This is simpler than refactoring all of it as wed end up with
684 three different sets... */
686 #define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap)
687 #define TRIE_BITMAP_BYTE(p, c) (TRIE_BITMAP(p)[(((U8)(c)) >> 3) & 31])
688 #define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c))
689 #define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c))
690 #define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
692 #define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC)
693 #define IS_TRIE_AC(op) ((op)>=AHOCORASICK)
696 #define BITMAP_BYTE(p, c) (((U8*)p)[(((U8)(c)) >> 3) & 31])
697 #define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
699 /* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */
700 #define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1)
701 #define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state))
702 #define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state))
705 #define TRIE_CHARCOUNT(trie) ((trie)->charcount)
707 #define TRIE_CHARCOUNT(trie) (trie_charcount)
710 #define RE_TRIE_MAXBUF_INIT 65536
711 #define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF"
712 #define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS"
716 RE_DEBUG_FLAGS is used to control what debug output is emitted
717 its divided into three groups of options, some of which interact.
718 The three groups are: Compile, Execute, Extra. There is room for a
719 further group, as currently only the low three bytes are used.
740 If you modify any of these make sure you make corresponding changes to
741 re.pm, especially to the documentation.
747 #define RE_DEBUG_COMPILE_MASK 0x0000FF
748 #define RE_DEBUG_COMPILE_PARSE 0x000001
749 #define RE_DEBUG_COMPILE_OPTIMISE 0x000002
750 #define RE_DEBUG_COMPILE_TRIE 0x000004
751 #define RE_DEBUG_COMPILE_DUMP 0x000008
752 #define RE_DEBUG_COMPILE_FLAGS 0x000010
755 #define RE_DEBUG_EXECUTE_MASK 0x00FF00
756 #define RE_DEBUG_EXECUTE_INTUIT 0x000100
757 #define RE_DEBUG_EXECUTE_MATCH 0x000200
758 #define RE_DEBUG_EXECUTE_TRIE 0x000400
761 #define RE_DEBUG_EXTRA_MASK 0xFF0000
762 #define RE_DEBUG_EXTRA_TRIE 0x010000
763 #define RE_DEBUG_EXTRA_OFFSETS 0x020000
764 #define RE_DEBUG_EXTRA_OFFDEBUG 0x040000
765 #define RE_DEBUG_EXTRA_STATE 0x080000
766 #define RE_DEBUG_EXTRA_OPTIMISE 0x100000
767 #define RE_DEBUG_EXTRA_BUFFERS 0x400000
768 #define RE_DEBUG_EXTRA_GPOS 0x800000
770 #define RE_DEBUG_EXTRA_STACK 0x280000
772 #define RE_DEBUG_FLAG(x) (re_debug_flags & x)
774 #define DEBUG_COMPILE_r(x) DEBUG_r( \
775 if (re_debug_flags & RE_DEBUG_COMPILE_MASK) x )
776 #define DEBUG_PARSE_r(x) DEBUG_r( \
777 if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x )
778 #define DEBUG_OPTIMISE_r(x) DEBUG_r( \
779 if (re_debug_flags & RE_DEBUG_COMPILE_OPTIMISE) x )
780 #define DEBUG_PARSE_r(x) DEBUG_r( \
781 if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x )
782 #define DEBUG_DUMP_r(x) DEBUG_r( \
783 if (re_debug_flags & RE_DEBUG_COMPILE_DUMP) x )
784 #define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \
785 if (re_debug_flags & RE_DEBUG_COMPILE_TRIE) x )
786 #define DEBUG_FLAGS_r(x) DEBUG_r( \
787 if (re_debug_flags & RE_DEBUG_COMPILE_FLAGS) x )
789 #define DEBUG_EXECUTE_r(x) DEBUG_r( \
790 if (re_debug_flags & RE_DEBUG_EXECUTE_MASK) x )
791 #define DEBUG_INTUIT_r(x) DEBUG_r( \
792 if (re_debug_flags & RE_DEBUG_EXECUTE_INTUIT) x )
793 #define DEBUG_MATCH_r(x) DEBUG_r( \
794 if (re_debug_flags & RE_DEBUG_EXECUTE_MATCH) x )
795 #define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \
796 if (re_debug_flags & RE_DEBUG_EXECUTE_TRIE) x )
799 #define DEBUG_EXTRA_r(x) DEBUG_r( \
800 if (re_debug_flags & RE_DEBUG_EXTRA_MASK) x )
801 #define DEBUG_OFFSETS_r(x) DEBUG_r( \
802 if (re_debug_flags & RE_DEBUG_EXTRA_OFFSETS) x )
803 #define DEBUG_STATE_r(x) DEBUG_r( \
804 if (re_debug_flags & RE_DEBUG_EXTRA_STATE) x )
805 #define DEBUG_STACK_r(x) DEBUG_r( \
806 if (re_debug_flags & RE_DEBUG_EXTRA_STACK) x )
807 #define DEBUG_BUFFERS_r(x) DEBUG_r( \
808 if (re_debug_flags & RE_DEBUG_EXTRA_BUFFERS) x )
810 #define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \
811 if ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \
812 (re_debug_flags & (RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE)) ) x )
813 #define MJD_OFFSET_DEBUG(x) DEBUG_r( \
814 if (re_debug_flags & RE_DEBUG_EXTRA_OFFDEBUG) \
815 Perl_warn_nocontext x )
816 #define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \
817 if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x )
818 #define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \
819 if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x )
821 #define DEBUG_TRIE_r(x) DEBUG_r( \
822 if (re_debug_flags & (RE_DEBUG_COMPILE_TRIE \
823 | RE_DEBUG_EXECUTE_TRIE )) x )
824 #define DEBUG_GPOS_r(x) DEBUG_r( \
825 if (re_debug_flags & RE_DEBUG_EXTRA_GPOS) x )
828 /* get_sv() can return NULL during global destruction. re_debug_flags can get
829 * clobbered by a longjmp, so must be initialized */
830 #define GET_RE_DEBUG_FLAGS DEBUG_r({ \
831 SV * re_debug_flags_sv = NULL; \
832 re_debug_flags = 0; \
833 re_debug_flags_sv = get_sv(RE_DEBUG_FLAGS, 1); \
834 if (re_debug_flags_sv) { \
835 if (!SvIOK(re_debug_flags_sv)) \
836 sv_setuv(re_debug_flags_sv, RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \
837 re_debug_flags=SvIV(re_debug_flags_sv); \
843 #define GET_RE_DEBUG_FLAGS_DECL VOL IV re_debug_flags \
844 PERL_UNUSED_DECL = 0; GET_RE_DEBUG_FLAGS;
846 #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \
847 const char * const rpv = \
848 pv_pretty((dsv), (pv), (l), (m), \
849 PL_colors[(c1)],PL_colors[(c2)], \
850 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \
851 const int rlen = SvCUR(dsv)
853 #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \
854 const char * const rpv = \
855 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \
856 PL_colors[(c1)],PL_colors[(c2)], \
857 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) )
859 #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \
860 const char * const rpv = \
861 pv_pretty((dsv), (pv), (l), (m), \
862 PL_colors[0], PL_colors[1], \
863 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \
864 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \
867 #define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0))
868 #define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "")
870 #else /* if not DEBUGGING */
872 #define GET_RE_DEBUG_FLAGS_DECL
873 #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2)
874 #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m)
875 #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m)
876 #define RE_SV_DUMPLEN(ItEm)
877 #define RE_SV_TAIL(ItEm)
879 #endif /* DEBUG RELATED DEFINES */
883 * c-indentation-style: bsd
885 * indent-tabs-mode: nil
888 * ex: set ts=8 sts=4 sw=4 et: