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.
11 #if ! defined(PERL_REGCOMP_H_) && ( defined(PERL_CORE) \
12 || defined(PERL_EXT_RE_BUILD))
14 #define PERL_REGCOMP_H_
16 #include "regcharclass.h"
18 /* Convert branch sequences to more efficient trie ops? */
19 #define PERL_ENABLE_TRIE_OPTIMISATION 1
21 /* Be really aggressive about optimising patterns with trie sequences? */
22 #define PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION 1
24 /* Should the optimiser take positive assertions into account? */
25 #define PERL_ENABLE_POSITIVE_ASSERTION_STUDY 0
27 /* Not for production use: */
28 #define PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS 0
31 * Structure for regexp "program". This is essentially a linear encoding
32 * of a nondeterministic finite-state machine (aka syntax charts or
33 * "railroad normal form" in parsing technology). Each node is an opcode
34 * plus a "next" pointer, possibly plus an operand. "Next" pointers of
35 * all nodes except BRANCH implement concatenation; a "next" pointer with
36 * a BRANCH on both ends of it is connecting two alternatives. (Here we
37 * have one of the subtle syntax dependencies: an individual BRANCH (as
38 * opposed to a collection of them) is never concatenated with anything
39 * because of operator precedence.) The operand of some types of node is
40 * a literal string; for others, it is a node leading into a sub-FSM. In
41 * particular, the operand of a BRANCH node is the first node of the branch.
42 * (NB this is *not* a tree structure: the tail of the branch connects
43 * to the thing following the set of BRANCHes.) The opcodes are defined
44 * in regnodes.h which is generated from regcomp.sym by regcomp.pl.
48 * A node is one char of opcode followed by two chars of "next" pointer.
49 * "Next" pointers are stored as two 8-bit pieces, high order first. The
50 * value is a positive offset from the opcode of the node containing it.
51 * An operand, if any, simply follows the node. (Note that much of the
52 * code generation knows about this implicit relationship.)
54 * Using two bytes for the "next" pointer is vast overkill for most things,
55 * but allows patterns to get big without disasters.
57 * [The "next" pointer is always aligned on an even
58 * boundary, and reads the offset directly as a short.]
61 /* This is the stuff that used to live in regexp.h that was truly
62 private to the engine itself. It now lives here. */
64 typedef struct regexp_internal {
65 regnode *regstclass; /* Optional startclass as identified or constructed
67 struct reg_data *data; /* Additional miscellaneous data used by the program.
68 Used to make it easier to clone and free arbitrary
69 data that the regops need. Often the ARG field of
70 a regop is an index into this structure. NOTE the
71 0th element of this structure is NEVER used and is
72 strictly reserved for internal purposes. */
73 struct reg_code_blocks *code_blocks;/* positions of literal (?{}) */
74 U32 proglen; /* size of the compiled program in regnodes */
75 U32 name_list_idx; /* Optional data index of an array of paren names,
76 only valid when RXp_PAREN_NAMES(prog) is true,
77 0 means "no value" like any other index into the
79 regnode program[1]; /* Unwarranted chumminess with compiler. */
82 #define RXi_SET(x,y) (x)->pprivate = (void*)(y)
83 #define RXi_GET(x) ((regexp_internal *)((x)->pprivate))
84 #define RXi_GET_DECL(r,ri) regexp_internal *ri = RXi_GET(r)
85 #define RXi_GET_DECL_NULL(r,ri) regexp_internal *ri = (r) ? RXi_GET(r) : NULL
87 * Flags stored in regexp->intflags
88 * These are used only internally to the regexp engine
90 * See regexp.h for flags used externally to the regexp engine
92 #define RXp_INTFLAGS(rx) ((rx)->intflags)
93 #define RX_INTFLAGS(prog) RXp_INTFLAGS(ReANY(prog))
95 #define PREGf_SKIP 0x00000001
96 #define PREGf_IMPLICIT 0x00000002 /* Converted .* to ^.* */
97 #define PREGf_NAUGHTY 0x00000004 /* how exponential is this pattern? */
98 #define PREGf_VERBARG_SEEN 0x00000008
99 #define PREGf_CUTGROUP_SEEN 0x00000010
100 #define PREGf_USE_RE_EVAL 0x00000020 /* compiled with "use re 'eval'" */
101 /* these used to be extflags, but are now intflags */
102 #define PREGf_NOSCAN 0x00000040
104 #define PREGf_GPOS_SEEN 0x00000100
105 #define PREGf_GPOS_FLOAT 0x00000200
107 #define PREGf_ANCH_MBOL 0x00000400
108 #define PREGf_ANCH_SBOL 0x00000800
109 #define PREGf_ANCH_GPOS 0x00001000
110 #define PREGf_RECURSE_SEEN 0x00002000
113 ( PREGf_ANCH_SBOL | PREGf_ANCH_GPOS | PREGf_ANCH_MBOL )
115 /* this is where the old regcomp.h started */
118 /* Define the various regnode structures. These all should be a multiple
119 * of 32 bits large, and they should by and large correspond with each other
120 * in terms of naming, etc. Things can and will break in subtle ways if you
121 * change things without care. If you look at regexp.h you will see it
130 * This structure is the base unit of elements in the regexp program. When
131 * we increment our way through the program we increment by the size of this
132 * structure, and in all cases where regnode sizing is considered it is in
133 * units of this structure.
135 * This implies that no regnode style structure should contain 64 bit
136 * aligned members. Since the base regnode is 32 bits any member might
137 * not be 64 bit aligned no matter how you might try to pad out the
138 * struct itself (the regnode_ssc is special in this regard as it is
139 * never used in a program directly). If you want to store 64 bit
140 * members you need to store them specially. The struct regnode_p and the
141 * ARGp() and ARGp_SET() macros and related inline functions provide an example
142 * solution. Note they deal with a slightly more complicated problem than simple
143 * alignment, as pointers may be 32 bits or 64 bits depending on platform,
144 * but they illustrate the pattern to follow if you want to put a 64 bit value
147 * NOTE: Ideally we do not put pointers into the regnodes in a program. Instead
148 * we put them in the "data" part of the regexp structure and store the index into
149 * the data in the pointers in the regnode. This allows the pointer to be handled
150 * properly during clone/free operations (eg refcount bookkeeping). See S_add_data(),
151 * Perl_regdupe_internal(), Perl_regfree_internal() in regcomp.c for how the data
152 * array can be used, the letters 'arsSu' all refer to different types of SV that
153 * we already have support for in the data array.
156 struct regnode_string {
163 struct regnode_lstring { /* Constructed this way to keep the string aligned. */
167 U32 str_len; /* Only 18 bits allowed before would overflow 'next_off' */
171 struct regnode_anyofhs { /* Constructed this way to keep the string aligned. */
175 U32 arg1; /* set by set_ANYOF_arg() */
179 /* Argument bearing node - workhorse,
180 arg1 is often for the data field */
188 /* Node whose argument is 'SV *'. This needs to be used very carefully in
189 * situations where pointers won't become invalid because of, say re-mallocs.
191 * Note that this regnode type is problematic and should not be used or copied
192 * and will be removed in the future. Pointers should be stored in the data[]
193 * array and an index into the data array stored in the regnode, which allows the
194 * pointers to be handled properly during clone/free operations on the regexp
195 * data structure. As a byproduct it also saves space, often we use a 16 bit
196 * member to store indexes into the data[] array.
198 * Also note that the weird storage here is because regnodes are 32 bit aligned,
199 * which means we cannot have a 64 bit aligned member. To make things more annoying
200 * the size of a pointer may vary by platform. Thus we use a character array, and
201 * then use inline functions to copy the data in or out.
207 char arg1_sv_ptr_bytes[sizeof(SV *)];
210 /* Similar to a regnode_1 but with an extra signed argument */
219 /* 'Two field' -- Two 16 bit unsigned args */
228 #define REGNODE_BBM_BITMAP_LEN \
229 /* 6 info bits requires 64 bits; 5 => 32 */ \
230 ((1 << (UTF_CONTINUATION_BYTE_INFO_BITS)) / CHARBITS)
232 /* Used for matching any two-byte UTF-8 character whose start byte is known.
233 * The array is a bitmap capable of representing any possible continuation
239 U8 bitmap[REGNODE_BBM_BITMAP_LEN];
242 #define ANYOF_BITMAP_SIZE (NUM_ANYOF_CODE_POINTS / CHARBITS)
244 /* Note that these form structs which are supersets of the next smaller one, by
245 * appending fields. Alignment problems can occur if one of those optional
246 * fields requires stricter alignment than the base struct. And formal
247 * parameters that can really be two or more of the structs should be
248 * declared as the smallest one it could be. See commit message for
249 * 7dcac5f6a5195002b55c935ee1d67f67e1df280b. Regnode allocation is done
250 * without regard to alignment, and changing it to would also require changing
251 * the code that inserts and deletes regnodes. The basic single-argument
252 * regnode has a U32, which is what reganode() allocates as a unit. Therefore
253 * no field can require stricter alignment than U32. */
255 /* also used by trie */
256 struct regnode_charclass {
260 U32 arg1; /* set by set_ANYOF_arg() */
261 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */
264 /* has runtime (locale) \d, \w, ..., [:posix:] classes */
265 struct regnode_charclass_posixl {
266 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
270 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
271 U32 classflags; /* and run-time */
274 /* A synthetic start class (SSC); is a regnode_charclass_posixl_fold, plus an
275 * extra SV*, used only during regex construction and which is not used by the
276 * main machinery in regexec.c and which does not get embedded in the final compiled
279 * Because it does not get embedded it does not have to comply with the alignment
280 * and sizing constraints required for a normal regnode structure: it MAY contain
281 * pointers or members of whatever size needed and the compiler will do the right
282 * thing. (Every other regnode type is 32 bit aligned.)
284 * Note that the 'next_off' field is unused, as the SSC stands alone, so there is
288 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
292 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
293 U32 classflags; /* ... and run-time */
295 /* Auxiliary, only used during construction; NULL afterwards: list of code
300 /* We take advantage of 'next_off' not otherwise being used in the SSC by
301 * actually using it: by setting it to 1. This allows us to test and
302 * distinguish between an SSC and other ANYOF node types, as 'next_off' cannot
303 * otherwise be 1, because it is the offset to the next regnode expressed in
304 * units of regnodes. Since an ANYOF node contains extra fields, it adds up
305 * to 12 regnode units on 32-bit systems, (hence the minimum this can be (if
306 * not 0) is 11 there. Even if things get tightly packed on a 64-bit system,
307 * it still would be more than 1. */
308 #define set_ANYOF_SYNTHETIC(n) STMT_START{ OP(n) = ANYOF; \
311 #define is_ANYOF_SYNTHETIC(n) (REGNODE_TYPE(OP(n)) == ANYOF && NEXT_OFF(n) == 1)
313 /* XXX fix this description.
314 Impose a limit of REG_INFTY on various pattern matching operations
315 to limit stack growth and to avoid "infinite" recursions.
317 /* The default size for REG_INFTY is U16_MAX, which is the same as
318 USHORT_MAX (see perl.h). Unfortunately U16 isn't necessarily 16 bits
319 (see handy.h). On the Cray C90, sizeof(short)==4 and hence U16_MAX is
320 ((1<<32)-1), while on the Cray T90, sizeof(short)==8 and U16_MAX is
321 ((1<<64)-1). To limit stack growth to reasonable sizes, supply a
323 --Andy Dougherty 11 June 1998
327 # define REG_INFTY nBIT_UMAX(16)
332 # define REG_INFTY U16_MAX
335 #define ARG_VALUE(arg) (arg)
336 #define ARG__SET(arg,val) ((arg) = (val))
342 #define ARG(p) ARG_VALUE(ARG_LOC(p))
343 #define ARGp(p) ARGp_VALUE_inline(p)
344 #define ARG1(p) ARG_VALUE(ARG1_LOC(p))
345 #define ARG2(p) ARG_VALUE(ARG2_LOC(p))
346 #define ARG2L(p) ARG_VALUE(ARG2L_LOC(p))
348 #define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val))
349 #define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val))
350 #define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val))
351 #define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val))
352 #define ARGp_SET(p, val) ARGp_SET_inline((p),(val))
357 #define NEXT_OFF(p) ((p)->next_off)
358 #define NODE_ALIGN(node)
359 /* the following define was set to 0xde in 075abff3
360 * as part of some linting logic. I have set it to 0
361 * as otherwise in every place where we /might/ set flags
362 * we have to set it 0 explicitly, which duplicates
363 * assignments and IMO adds an unacceptable level of
364 * surprise to working in the regex engine. If this
365 * is changed from 0 then at the very least make sure
366 * that SBOL for /^/ sets the flags to 0 explicitly.
368 #define NODE_ALIGN_FILL(node) ((node)->flags = 0)
370 #define SIZE_ALIGN NODE_ALIGN
376 #define OP(p) ((p)->type)
377 #define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \
378 regnode types. For some, it's the \
379 character set of the regnode */
380 #define STR_LENs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \
381 ((struct regnode_string *)p)->str_len)
382 #define STRINGs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \
383 ((struct regnode_string *)p)->string)
384 #define OPERANDs(p) STRINGs(p)
386 /* Long strings. Currently limited to length 18 bits, which handles a 262000
387 * byte string. The limiting factor is the 16 bit 'next_off' field, which
388 * points to the next regnode, so the furthest away it can be is 2**16. On
389 * most architectures, regnodes are 2**2 bytes long, so that yields 2**18
390 * bytes. Should a longer string be desired, we could increase it to 26 bits
391 * fairly easily, by changing this node to have longj type which causes the ARG
392 * field to be used for the link to the next regnode (although code would have
393 * to be changed to account for this), and then use a combination of the flags
394 * and next_off fields for the length. To get 34 bit length, also change the
395 * node to be an ARG2L, using the second 32 bit field for the length, and not
396 * using the flags nor next_off fields at all. One could have an llstring node
397 * and even an lllstring type. */
398 #define STR_LENl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
399 (((struct regnode_lstring *)p)->str_len))
400 #define STRINGl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
401 (((struct regnode_lstring *)p)->string))
402 #define OPERANDl(p) STRINGl(p)
404 #define STR_LEN(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
405 ? STR_LENl(p) : STR_LENs(p))
406 #define STRING(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
407 ? STRINGl(p) : STRINGs(p))
408 #define OPERAND(p) STRING(p)
410 /* The number of (smallest) regnode equivalents that a string of length l bytes
411 * occupies - Used by the REGNODE_AFTER() macros and functions. */
412 #define STR_SZ(l) (((l) + sizeof(regnode) - 1) / sizeof(regnode))
414 #define setSTR_LEN(p,v) \
416 if (OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
417 ((struct regnode_lstring *)(p))->str_len = (v); \
419 ((struct regnode_string *)(p))->str_len = (v); \
422 #define ANYOFR_BASE_BITS 20
423 #define ANYOFRbase(p) (ARG(p) & nBIT_MASK(ANYOFR_BASE_BITS))
424 #define ANYOFRdelta(p) (ARG(p) >> ANYOFR_BASE_BITS)
429 #define NODE_ALIGN(node)
430 #define ARG_LOC(p) (((struct regnode_1 *)p)->arg1)
431 #define ARGp_BYTES_LOC(p) (((struct regnode_p *)p)->arg1_sv_ptr_bytes)
432 #define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1)
433 #define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2)
434 #define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2)
436 /* These should no longer be used directly in most cases. Please use
437 * the REGNODE_AFTER() macros instead. */
438 #define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */
439 #define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2)
441 /* Core macros for computing "the regnode after this one". See also
442 * Perl_regnode_after() in reginline.h
444 * At the struct level regnodes are a linked list, with each node pointing
445 * at the next (via offsets), usually via the C<next_off> field in the
446 * structure. Where there is a need for a node to have two children the
447 * immediate physical successor of the node in the compiled program is used
448 * to represent one of them. A good example is the BRANCH construct,
449 * consider the pattern C</head(?:[ab]foo|[cd]bar)tail/>
451 * 1: EXACT <head> (3)
454 * 6: EXACT <foo> (14)
457 * 11: EXACT <bar> (14)
459 * 14: EXACT <tail> (16)
462 * The numbers in parens at the end of each line show the "next_off" value
463 * for that regnode in the program. We can see that the C<next_off> of
464 * the first BRANCH node (#3) is the second BRANCH node (#8), and indicates
465 * where execution should go if the regnodes *following* the BRANCH node fail
466 * to accept the input string. Thus to find the "next BRANCH" we would do
467 * C<Perl_regnext()> and follow the C<next_off> pointer, and to find
468 * the "BRANCHes contents" we would use C<REGNODE_AFTER()>.
470 * Be aware that C<REGNODE_AFTER()> is not guaranteed to give a *useful*
471 * result once the regex peephole optimizer has run (it will be correct
472 * however!). By the time code in regexec.c executes various regnodes
473 * may have been optimized out of the the C<next_off> chain. An example
474 * can be seen above, node 13 will never be reached during execution
475 * flow as it has been stitched out of the C<next_off> chain. Both 6 and
476 * 11 would have pointed at it during compilation, but it exists only to
477 * facilitate the construction of the BRANCH structure and is effectively
478 * a NOOP, and thus the optimizer adjusts the links so it is skipped
479 * from execution time flow. In regexec.c it is only safe to use
480 * REGNODE_AFTER() on specific node types.
482 * Conversely during compilation C<Perl_regnext()> may not work properly
483 * as the C<next_off> may not be known until "later", (such as in the
484 * case of BRANCH nodes) and thus in regcomp.c the REGNODE_AFTER() macro
485 * is used very heavily instead.
487 * There are several variants of the REGNODE_AFTER_xxx() macros which
488 * are intended for use in different situations depending on how
489 * confident the code is about what type of node it is trying to find a
492 * So for instance if you know you are dealing with a known node type of
493 * constant size then you should use REGNODE_AFTER_type(n,TYPE).
495 * If you have a regnode pointer and you know you are dealing with a
496 * regnode type of constant size and you have already extracted its
497 * opcode use: REGNODE_AFTER_opcode(n,OPCODE).
499 * If you have a regnode and you know it is variable size then you
500 * you can produce optimized code by using REGNODE_AFTER_varies(n).
502 * If you have a regnode pointer and nothing else use: REGNODE_AFTER(n)
503 * This is the safest option and wraps C<Perl_regnode_after()>. It
504 * should produce the correct result regardless of its argument. The
505 * other options only produce correct results under specific
508 #define REGNODE_AFTER_PLUS(p,extra) ((p) + NODE_STEP_REGNODE + (extra))
509 /* under DEBUGGING we check that all REGNODE_AFTER optimized macros did the
510 * same thing that Perl_regnode_after() would have done. Note that when
511 * not compiled under DEBUGGING the assert_() macro is empty. Thus we
512 * don't have to implement different versions for DEBUGGING and not DEBUGGING,
513 * and explains why all the macros use REGNODE_AFTER_PLUS_DEBUG() under the
515 #define REGNODE_AFTER_PLUS_DEBUG(p,extra) \
516 (assert_(check_regnode_after(p,extra)) REGNODE_AFTER_PLUS((p),(extra)))
518 /* find the regnode after this p by using the opcode we previously extracted
520 #define REGNODE_AFTER_opcode(p,op) REGNODE_AFTER_PLUS_DEBUG((p),REGNODE_ARG_LEN(op))
522 /* find the regnode after this p by using the size of the struct associated with
523 * the opcode for p. use this when you *know* that p is pointer to a given type*/
524 #define REGNODE_AFTER_type(p,t) REGNODE_AFTER_PLUS_DEBUG((p),EXTRA_SIZE(t))
526 /* find the regnode after this p by using OP(p) to find the regnode type of p */
527 #define REGNODE_AFTER_varies(p) regnode_after(p,TRUE)
529 /* find the regnode after this p by using OP(p) to find the regnode type of p */
530 #define REGNODE_AFTER(p) regnode_after(p,FALSE)
533 /* REGNODE_BEFORE() is trickier to deal with in terms of validation, execution.
534 * All the places that use it assume that p will be one struct regnode large.
535 * So to validate it we do the math to go backwards and then validate that the
536 * type of regnode we landed on is actually one regnode large. In theory if
537 * things go wrong the opcode should be illegal or say the item should be larger
538 * than it is, etc. */
539 #define REGNODE_BEFORE_BASE(p) ((p) - NODE_STEP_REGNODE)
540 #define REGNODE_BEFORE_BASE_DEBUG(p) \
541 (assert_(check_regnode_after(REGNODE_BEFORE_BASE(p),0)) REGNODE_BEFORE_BASE(p))
542 #define REGNODE_BEFORE(p) REGNODE_BEFORE_BASE_DEBUG(p)
544 #define FILL_NODE(offset, op) \
546 OP(REGNODE_p(offset)) = op; \
547 NEXT_OFF(REGNODE_p(offset)) = 0; \
549 #define FILL_ADVANCE_NODE(offset, op) \
551 FILL_NODE(offset, op); \
554 #define FILL_ADVANCE_NODE_ARG(offset, op, arg) \
556 ARG_SET(REGNODE_p(offset), arg); \
557 FILL_ADVANCE_NODE(offset, op); \
558 /* This is used generically for other operations \
559 * that have a longer argument */ \
560 (offset) += REGNODE_ARG_LEN(op); \
562 #define FILL_ADVANCE_NODE_ARGp(offset, op, arg) \
564 ARGp_SET(REGNODE_p(offset), arg); \
565 FILL_ADVANCE_NODE(offset, op); \
566 (offset) += REGNODE_ARG_LEN(op); \
568 #define FILL_ADVANCE_NODE_2L_ARG(offset, op, arg1, arg2) \
570 ARG_SET(REGNODE_p(offset), arg1); \
571 ARG2L_SET(REGNODE_p(offset), arg2); \
572 FILL_ADVANCE_NODE(offset, op); \
576 /* define these after we define the normal macros, so we can use
577 * ARGp_BYTES_LOC(n) */
580 ARGp_VALUE_inline(struct regnode *node) {
582 memcpy(&ptr, ARGp_BYTES_LOC(node), sizeof(ptr));
588 ARGp_SET_inline(struct regnode *node, SV *ptr) {
589 memcpy(ARGp_BYTES_LOC(node), &ptr, sizeof(ptr));
592 #define REG_MAGIC 0234
594 /* An ANYOF node matches a single code point based on specified criteria. It
595 * now comes in several styles, but originally it was just a 256 element
596 * bitmap, indexed by the code point (which was always just a byte). If the
597 * corresponding bit for a code point is 1, the code point matches; if 0, it
598 * doesn't match (complemented if inverted). This worked fine before Unicode
599 * existed, but making a bit map long enough to accommodate a bit for every
600 * possible Unicode code point is prohibitively large. Therefore it is made
601 * much much smaller, and an inversion list is created to handle code points
602 * not represented by the bitmap. (It is now possible to compile the bitmap to
603 * a larger size to avoid the slower inversion list lookup for however big the
604 * bitmap is set to, but this is rarely done). If the bitmap is sufficient to
605 * specify all possible matches (with nothing outside it matching), no
606 * inversion list is needed nor included, and the argument to the ANYOF node is
607 * set to the following: */
609 #define ANYOF_MATCHES_ALL_OUTSIDE_BITMAP_VALUE U32_MAX
610 #define ANYOF_MATCHES_ALL_OUTSIDE_BITMAP(node) \
611 (ARG(node) == ANYOF_MATCHES_ALL_OUTSIDE_BITMAP_VALUE)
613 #define ANYOF_MATCHES_NONE_OUTSIDE_BITMAP_VALUE \
614 /* Assumes ALL is odd */ (ANYOF_MATCHES_ALL_OUTSIDE_BITMAP_VALUE - 1)
615 #define ANYOF_MATCHES_NONE_OUTSIDE_BITMAP(node) \
616 (ARG(node) == ANYOF_MATCHES_NONE_OUTSIDE_BITMAP_VALUE)
618 #define ANYOF_ONLY_HAS_BITMAP_MASK ANYOF_MATCHES_NONE_OUTSIDE_BITMAP_VALUE
619 #define ANYOF_ONLY_HAS_BITMAP(node) \
620 ((ARG(node) & ANYOF_ONLY_HAS_BITMAP_MASK) == ANYOF_ONLY_HAS_BITMAP_MASK)
622 #define ANYOF_HAS_AUX(node) (! ANYOF_ONLY_HAS_BITMAP(node))
624 /* There are also ANYOFM nodes, used when the bit patterns representing the
625 * matched code points happen to be such that they can be checked by ANDing
626 * with a mask. The regex compiler looks for and silently optimizes to using
627 * this node type in the few cases where it works out. The eight octal digits
628 * form such a group. These nodes are simple and fast and no further
629 * discussion is needed here.
631 * And, there are ANYOFH-ish nodes which match only code points that aren't in
632 * the bitmap (the H stands for High). These are common for expressing
633 * Unicode properties concerning non-Latin scripts. They dispense with the
634 * bitmap altogether and don't need any of the flags discussed below.
636 * And, there are ANYOFR-ish nodes which match within a single range.
638 * When there is a need to specify what matches outside the bitmap, it is done
639 * by allocating an AV as part of the pattern's compiled form, and the argument
640 * to the node instead of being ANYOF_ONLY_HAS_BITMAP, points to that AV.
642 * (Actually, that is an oversimplification. The AV is placed into the
643 * pattern's struct reg_data, and what is stored in the node's argument field
644 * is its index into that struct. And the inversion list is just one element,
645 * the zeroth, of the AV.)
647 * There are certain situations where a single inversion list can't handle all
648 * the complexity. These are dealt with by having extra elements in the AV, by
649 * specifying flag bits in the ANYOF node, and/or special code. As an example,
650 * there are instances where what the ANYOF node matches is not completely
651 * known until runtime. In these cases, a flag is set, and the bitmap has a 1
652 * for the code points which are known at compile time to be 1, and a 0 for the
653 * ones that are known to be 0, or require runtime resolution. Some missing
654 * information can be found by merely seeing if the pattern is UTF-8 or not;
655 * other cases require looking at the extra elements in the AV.
657 * There are 5 cases where the bitmap is insufficient. These are specified by
658 * flags in the node's flags field. We could use five bits to represent the 5
659 * cases, but to save flags bits (which are perennially in short supply), we
660 * play some games. The cases are:
662 * 1) As already mentioned, if some code points outside the bitmap match, and
663 * some do not, an inversion list is specified to indicate which ones.
665 * 2) Under /d rules, it can happen that code points that are in the upper
666 * latin1 range (\x80-\xFF or their equivalents on EBCDIC platforms) match
667 * only if the runtime target string being matched against is UTF-8. For
668 * example /[\w[:punct:]]/d. This happens only for certain posix classes,
669 * and all such ones also have above-bitmap matches.
671 * Note that /d rules are no longer encouraged; 'use 5.14' or higher
672 * deselects them. But they are still supported, and a flag is required
673 * so that they can be properly handled. But it can be a shared flag: see
676 * 3) Also under /d rules, something like /[\Wfoo]/ will match everything in
677 * the \x80-\xFF range, unless the string being matched against is UTF-8.
678 * An inversion list could be created for this case, but this is
679 * relatively common, and it turns out that it's all or nothing: if any
680 * one of these code points matches, they all do. Hence a single bit
681 * suffices. We use a shared flag that doesn't take up space by itself:
682 * ANYOFD_NON_UTF8_MATCHES_ALL_NON_ASCII__shared. This also means there
683 * is an inversion list for the things that don't fit into the bitmap.
685 * 4) A user-defined \p{} property may not have been defined by the time the
686 * regex is compiled. In this case, we don't know until runtime what it
687 * will match, so we have to assume it could match anything, including
688 * code points that ordinarily would be in the bitmap. A flag bit is
689 * necessary to indicate this, though we can use the
690 * ANYOF_HAS_EXTRA_RUNTIME_MATCHES flag, along with the node not being
691 * ANYOFD. The information required to construct the property is stored
692 * in the AV pointed to by the node's argument. This case is quite
693 * uncommon in the field, and the /(?[...])/ construct is a better way to
694 * accomplish what this feature does.
696 * 5) /[foo]/il may have folds that are only valid if the runtime locale is a
697 * UTF-8 one. The ANYOF_HAS_EXTRA_RUNTIME_MATCHES flag can also be used
698 * for these. The list is stored in a different element of the AV, so its
699 * existence differentiates this case from that of 4), along with the node
700 * being ANYOFL, with the ANYOFL_FOLD flag being set. There are a few
701 * additional folds valid only if the UTF-8 locale is a Turkic one which
702 * is tested for explicitly.
704 * Note that the user-defined property flag and the /il flag can affect whether
705 * an ASCII character matches in the bitmap or not.
707 * And this still isn't the end of the story. In some cases, warnings are
708 * supposed to be raised when matching certain categories of code points in the
709 * target string. Flags are set to indicate this. This adds up to a bunch of
710 * flags required, and we only have 8 available. That is why we share some.
711 * At the moment, there are two spare flag bits, but this could be increased by
714 * ANYOF_MATCHES_POSIXL is redundant with the node type ANYOFPOSIXL. That flag
715 * could be removed, but at the expense of having to write extra code, which
716 * would take up space, and writing this turns out to be not hard, but not
719 * If this is done, an extension would be to make all ANYOFL nodes contain the
720 * extra 32 bits that ANYOFPOSIXL ones do, doubling each instance's size. The
721 * posix flags only occupy 30 bits, so the ANYOFL_FOLD and
722 * ANYOFL_UTF8_LOCALE_REQD bits could be moved to that extra space, but it
723 * would also mean extra instructions, as there are currently places in the
724 * code that assume those two bits are zero.
726 * Some flags are not used in synthetic start class (SSC) nodes, so could be
727 * shared should new flags be needed for SSCs, like SSC_MATCHES_EMPTY_STRING
730 /* If this is set, the result of the match should be complemented. regexec.c
731 * is expecting this to be in the low bit. Never in an SSC */
732 #define ANYOF_INVERT 0x01
734 /* For the SSC node only, which cannot be inverted, so is shared with that bit.
735 * This is used only during regex compilation. */
736 #define SSC_MATCHES_EMPTY_STRING ANYOF_INVERT
738 /* Set if this is a regnode_charclass_posixl vs a regnode_charclass. This
739 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale
740 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved
741 * at compile-time. Only set under /l; can be in SSC */
742 #define ANYOF_MATCHES_POSIXL 0x02
744 /* The fold is calculated and stored in the bitmap where possible at compile
745 * time. However under locale, the actual folding varies depending on
746 * what the locale is at the time of execution, so it has to be deferred until
747 * then. Only set under /l; never in an SSC */
748 #define ANYOFL_FOLD 0x04
750 /* Warn if the runtime locale isn't a UTF-8 one (and the generated node assumes
752 #define ANYOFL_UTF8_LOCALE_REQD 0x08
754 /* Spare: Be sure to change ANYOF_FLAGS_ALL if this gets used 0x10 */
756 /* Spare: Be sure to change ANYOF_FLAGS_ALL if this gets used 0x20 */
758 /* Shared bit that indicates that there are potential additional matches stored
759 * outside the bitmap, as pointed to by the AV given by the node's argument.
760 * The node type is used at runtime (in conjunction with this flag and other
761 * information available then) to decide if the flag should be acted upon.
762 * This extra information is needed because of at least one of the following
764 * Under /d and the matched string is in UTF-8, it means the ANYOFD node
765 * matches more things than in the bitmap. Those things will be any
766 * code point too high for the bitmap, but crucially, any non-ASCII
767 * characters that match iff when using Unicode rules. These all are
770 * Under /l and ANYOFL_FOLD is set, this flag may indicate there are
771 * potential matches valid only if the locale is a UTF-8 one. If so,
772 * a list of them is stored in the AV.
774 * For any non-ANYOFD node, there may be a user-defined property that
775 * wasn't yet defined at the time the regex was compiled, and so must
776 * be looked up at runtime, The information required to do so will
779 * Note that an ANYOFL node may contain both a user-defined property, and
780 * folds not always valid. The important thing is that there is an AV to
782 #define ANYOF_HAS_EXTRA_RUNTIME_MATCHES 0x40
785 * Under /d it means the ANYOFD node matches all non-ASCII Latin1
786 * characters when the target string is not in utf8.
787 * When not under /d, it means the ANYOF node should raise a warning if
788 * matching against an above-Unicode code point.
789 * (These uses are mutually exclusive because the warning requires a \p{}, and
790 * \p{} implies /u which deselects /d). An SSC node only has this bit set if
791 * what is meant is the warning. The names are to make sure that you are
792 * cautioned about its shared nature */
793 #define ANYOFD_NON_UTF8_MATCHES_ALL_NON_ASCII__shared 0x80
794 #define ANYOF_WARN_SUPER__shared 0x80
796 #define ANYOF_FLAGS_ALL ((U8) ~(0x10|0x20))
798 #define ANYOF_LOCALE_FLAGS ( ANYOFL_FOLD \
799 | ANYOF_MATCHES_POSIXL \
800 | ANYOFL_UTF8_LOCALE_REQD)
802 /* These are the flags that apply to both regular ANYOF nodes and synthetic
803 * start class nodes during construction of the SSC. During finalization of
804 * the SSC, other of the flags may get added to it */
805 #define ANYOF_COMMON_FLAGS 0
807 /* Character classes for node->classflags of ANYOF */
808 /* Should be synchronized with a table in regprop() */
809 /* 2n should be the normal one, paired with its complement at 2n+1 */
811 #define ANYOF_ALPHA ((CC_ALPHA_) * 2)
812 #define ANYOF_NALPHA ((ANYOF_ALPHA) + 1)
813 #define ANYOF_ALPHANUMERIC ((CC_ALPHANUMERIC_) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */
814 #define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1)
815 #define ANYOF_ASCII ((CC_ASCII_) * 2)
816 #define ANYOF_NASCII ((ANYOF_ASCII) + 1)
817 #define ANYOF_BLANK ((CC_BLANK_) * 2) /* GNU extension: space and tab: non-vertical space */
818 #define ANYOF_NBLANK ((ANYOF_BLANK) + 1)
819 #define ANYOF_CASED ((CC_CASED_) * 2) /* Pseudo class for [:lower:] or
820 [:upper:] under /i */
821 #define ANYOF_NCASED ((ANYOF_CASED) + 1)
822 #define ANYOF_CNTRL ((CC_CNTRL_) * 2)
823 #define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1)
824 #define ANYOF_DIGIT ((CC_DIGIT_) * 2) /* \d */
825 #define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1)
826 #define ANYOF_GRAPH ((CC_GRAPH_) * 2)
827 #define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1)
828 #define ANYOF_LOWER ((CC_LOWER_) * 2)
829 #define ANYOF_NLOWER ((ANYOF_LOWER) + 1)
830 #define ANYOF_PRINT ((CC_PRINT_) * 2)
831 #define ANYOF_NPRINT ((ANYOF_PRINT) + 1)
832 #define ANYOF_PUNCT ((CC_PUNCT_) * 2)
833 #define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1)
834 #define ANYOF_SPACE ((CC_SPACE_) * 2) /* \s */
835 #define ANYOF_NSPACE ((ANYOF_SPACE) + 1)
836 #define ANYOF_UPPER ((CC_UPPER_) * 2)
837 #define ANYOF_NUPPER ((ANYOF_UPPER) + 1)
838 #define ANYOF_WORDCHAR ((CC_WORDCHAR_) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */
839 #define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1)
840 #define ANYOF_XDIGIT ((CC_XDIGIT_) * 2)
841 #define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1)
843 /* pseudo classes below this, not stored in the class bitmap, but used as flags
844 during compilation of char classes */
846 #define ANYOF_VERTWS ((CC_VERTSPACE_) * 2)
847 #define ANYOF_NVERTWS ((ANYOF_VERTWS)+1)
849 /* It is best if this is the last one, as all above it are stored as bits in a
850 * bitmap, and it isn't part of that bitmap */
851 #if CC_VERTSPACE_ != HIGHEST_REGCOMP_DOT_H_SYNC_
852 # error Problem with handy.h HIGHEST_REGCOMP_DOT_H_SYNC_ #define
855 #define ANYOF_POSIXL_MAX (ANYOF_VERTWS) /* So upper loop limit is written:
857 * Hence doesn't include VERTWS, as that
858 * is a pseudo class */
859 #define ANYOF_MAX ANYOF_POSIXL_MAX
861 #if (ANYOF_POSIXL_MAX > 32) /* Must fit in 32-bit word */
862 # error Problem with handy.h CC_foo_ #defines
865 #define ANYOF_HORIZWS ((ANYOF_POSIXL_MAX)+2) /* = (ANYOF_NVERTWS + 1) */
866 #define ANYOF_NHORIZWS ((ANYOF_POSIXL_MAX)+3)
868 #define ANYOF_UNIPROP ((ANYOF_POSIXL_MAX)+4) /* Used to indicate a Unicode
869 property: \p{} or \P{} */
871 /* Backward source code compatibility. */
873 #define ANYOF_ALNUML ANYOF_ALNUM
874 #define ANYOF_NALNUML ANYOF_NALNUM
875 #define ANYOF_SPACEL ANYOF_SPACE
876 #define ANYOF_NSPACEL ANYOF_NSPACE
877 #define ANYOF_ALNUM ANYOF_WORDCHAR
878 #define ANYOF_NALNUM ANYOF_NWORDCHAR
880 /* Utility macros for the bitmap and classes of ANYOF */
882 #define BITMAP_BYTE(p, c) (( (U8*) (p)) [ ( ( (UV) (c)) >> 3) ] )
883 #define BITMAP_BIT(c) (1U << ((c) & 7))
884 #define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & BITMAP_BIT((U8)(c)))
886 #define ANYOF_FLAGS(p) ((p)->flags)
888 #define ANYOF_BIT(c) BITMAP_BIT(c)
890 #define ANYOF_POSIXL_BITMAP(p) (((regnode_charclass_posixl*) (p))->classflags)
892 #define POSIXL_SET(field, c) ((field) |= (1U << (c)))
893 #define ANYOF_POSIXL_SET(p, c) POSIXL_SET(ANYOF_POSIXL_BITMAP(p), (c))
895 #define POSIXL_CLEAR(field, c) ((field) &= ~ (1U <<(c)))
896 #define ANYOF_POSIXL_CLEAR(p, c) POSIXL_CLEAR(ANYOF_POSIXL_BITMAP(p), (c))
898 #define POSIXL_TEST(field, c) ((field) & (1U << (c)))
899 #define ANYOF_POSIXL_TEST(p, c) POSIXL_TEST(ANYOF_POSIXL_BITMAP(p), (c))
901 #define POSIXL_ZERO(field) STMT_START { (field) = 0; } STMT_END
902 #define ANYOF_POSIXL_ZERO(ret) POSIXL_ZERO(ANYOF_POSIXL_BITMAP(ret))
904 #define ANYOF_POSIXL_SET_TO_BITMAP(p, bits) \
905 STMT_START { ANYOF_POSIXL_BITMAP(p) = (bits); } STMT_END
907 /* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */
908 #define ANYOF_POSIXL_SETALL(ret) \
910 ANYOF_POSIXL_BITMAP(ret) = nBIT_MASK(ANYOF_POSIXL_MAX); \
912 #define ANYOF_CLASS_SETALL(ret) ANYOF_POSIXL_SETALL(ret)
914 #define ANYOF_POSIXL_TEST_ANY_SET(p) \
915 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) && ANYOF_POSIXL_BITMAP(p))
916 #define ANYOF_CLASS_TEST_ANY_SET(p) ANYOF_POSIXL_TEST_ANY_SET(p)
918 /* Since an SSC always has this field, we don't have to test for that; nor do
919 * we want to because the bit isn't set for SSC during its construction */
920 #define ANYOF_POSIXL_SSC_TEST_ANY_SET(p) \
921 cBOOL(((regnode_ssc*)(p))->classflags)
922 #define ANYOF_POSIXL_SSC_TEST_ALL_SET(p) /* Are all bits set? */ \
923 (((regnode_ssc*) (p))->classflags \
924 == nBIT_MASK(ANYOF_POSIXL_MAX))
926 #define ANYOF_POSIXL_TEST_ALL_SET(p) \
927 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
928 && ANYOF_POSIXL_BITMAP(p) == nBIT_MASK(ANYOF_POSIXL_MAX))
930 #define ANYOF_POSIXL_OR(source, dest) STMT_START { (dest)->classflags |= (source)->classflags ; } STMT_END
931 #define ANYOF_CLASS_OR(source, dest) ANYOF_POSIXL_OR((source), (dest))
933 #define ANYOF_POSIXL_AND(source, dest) STMT_START { (dest)->classflags &= (source)->classflags ; } STMT_END
935 #define ANYOF_BITMAP_ZERO(ret) Zero(((regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char)
936 #define ANYOF_BITMAP(p) ((regnode_charclass*)(p))->bitmap
937 #define ANYOF_BITMAP_BYTE(p, c) BITMAP_BYTE(ANYOF_BITMAP(p), c)
938 #define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c))
939 #define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c))
940 #define ANYOF_BITMAP_TEST(p, c) cBOOL(ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c))
942 #define ANYOF_BITMAP_SETALL(p) \
943 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE)
944 #define ANYOF_BITMAP_CLEARALL(p) \
945 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE)
948 * Utility definitions.
951 # define UCHARAT(p) ((int)*(const U8*)(p))
953 # define UCHARAT(p) ((int)*(p)&CHARMASK)
956 /* Number of regnode equivalents that 'guy' occupies beyond the size of the
957 * smallest regnode. */
958 #define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode))
960 #define REG_ZERO_LEN_SEEN 0x00000001
961 #define REG_LOOKBEHIND_SEEN 0x00000002
962 /* add a short form alias to keep the line length police happy */
963 #define REG_LB_SEEN REG_LOOKBEHIND_SEEN
964 #define REG_GPOS_SEEN 0x00000004
966 #define REG_RECURSE_SEEN 0x00000020
967 #define REG_TOP_LEVEL_BRANCHES_SEEN 0x00000040
968 #define REG_VERBARG_SEEN 0x00000080
969 #define REG_CUTGROUP_SEEN 0x00000100
970 #define REG_RUN_ON_COMMENT_SEEN 0x00000200
971 #define REG_UNFOLDED_MULTI_SEEN 0x00000400
973 #define REG_UNBOUNDED_QUANTIFIER_SEEN 0x00001000
978 #ifdef PLUGGABLE_RE_EXTENSION
979 #include "re_nodes.h"
981 #include "regnodes.h"
984 #ifndef PLUGGABLE_RE_EXTENSION
986 EXTCONST regexp_engine PL_core_reg_engine;
988 EXTCONST regexp_engine PL_core_reg_engine = {
991 Perl_re_intuit_start,
992 Perl_re_intuit_string,
993 Perl_regfree_internal,
994 Perl_reg_numbered_buff_fetch,
995 Perl_reg_numbered_buff_store,
996 Perl_reg_numbered_buff_length,
998 Perl_reg_named_buff_iter,
1000 #if defined(USE_ITHREADS)
1001 Perl_regdupe_internal,
1006 #endif /* PLUGGABLE_RE_EXTENSION */
1012 /* .what is a character array with one character for each member of .data
1013 * The character describes the function of the corresponding .data item:
1014 * a - AV for paren_name_list under DEBUGGING
1015 * f - start-class data for regstclass optimization
1016 * l - start op for literal (?{EVAL}) item
1017 * L - start op for literal (?{EVAL}) item, with separate CV (qr//)
1018 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/
1019 * s - inversion list for Unicode-style character class, and the
1020 * multicharacter strings resulting from casefolding the single-character
1021 * entries in the character class
1023 * u - trie struct's widecharmap (a HV, so can't share, must dup)
1024 * also used for revcharmap and words under DEBUGGING
1025 * T - aho-trie struct
1026 * S - sv for named capture lookup
1027 * 20010712 mjd@plover.com
1028 * (Remember to update re_dup() and pregfree() if you add any items.)
1036 /* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses
1037 anchored* and float* via array indexes 0 and 1. */
1038 #define anchored_substr substrs->data[0].substr
1039 #define anchored_utf8 substrs->data[0].utf8_substr
1040 #define anchored_offset substrs->data[0].min_offset
1041 #define anchored_end_shift substrs->data[0].end_shift
1043 #define float_substr substrs->data[1].substr
1044 #define float_utf8 substrs->data[1].utf8_substr
1045 #define float_min_offset substrs->data[1].min_offset
1046 #define float_max_offset substrs->data[1].max_offset
1047 #define float_end_shift substrs->data[1].end_shift
1049 #define check_substr substrs->data[2].substr
1050 #define check_utf8 substrs->data[2].utf8_substr
1051 #define check_offset_min substrs->data[2].min_offset
1052 #define check_offset_max substrs->data[2].max_offset
1053 #define check_end_shift substrs->data[2].end_shift
1055 #define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr)
1056 #define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8)
1057 #define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr)
1058 #define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8)
1060 /* trie related stuff */
1062 /* a transition record for the state machine. the
1063 check field determines which state "owns" the
1064 transition. the char the transition is for is
1065 determined by offset from the owning states base
1066 field. the next field determines which state
1067 is to be transitioned to if any.
1069 struct _reg_trie_trans {
1074 /* a transition list element for the list based representation */
1075 struct _reg_trie_trans_list_elem {
1079 typedef struct _reg_trie_trans_list_elem reg_trie_trans_le;
1081 /* a state for compressed nodes. base is an offset
1082 into an array of reg_trie_trans array. If wordnum is
1083 nonzero the state is accepting. if base is zero then
1084 the state has no children (and will be accepting)
1086 struct _reg_trie_state {
1090 reg_trie_trans_le* list;
1094 /* info per word; indexed by wordnum */
1096 U16 prev; /* previous word in acceptance chain; eg in
1097 * zzz|abc|ab/ after matching the chars abc, the
1098 * accepted word is #2, and the previous accepted
1100 U32 len; /* how many chars long is this word? */
1101 U32 accept; /* accept state for this word */
1102 } reg_trie_wordinfo;
1105 typedef struct _reg_trie_state reg_trie_state;
1106 typedef struct _reg_trie_trans reg_trie_trans;
1109 /* anything in here that needs to be freed later
1110 should be dealt with in pregfree.
1111 refcount is first in both this and _reg_ac_data to allow a space
1112 optimisation in Perl_regdupe. */
1113 struct _reg_trie_data {
1114 U32 refcount; /* number of times this trie is referenced */
1115 U32 lasttrans; /* last valid transition element */
1116 U16 *charmap; /* byte to charid lookup array */
1117 reg_trie_state *states; /* state data */
1118 reg_trie_trans *trans; /* array of transition elements */
1119 char *bitmap; /* stclass bitmap */
1120 U16 *jump; /* optional 1 indexed array of offsets before tail
1121 for the node following a given word. */
1122 reg_trie_wordinfo *wordinfo; /* array of info per word */
1123 U16 uniquecharcount; /* unique chars in trie (width of trans table) */
1124 U32 startstate; /* initial state - used for common prefix optimisation */
1125 STRLEN minlen; /* minimum length of words in trie - build/opt only? */
1126 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */
1127 U32 prefixlen; /* #chars in common prefix */
1128 U32 statecount; /* Build only - number of states in the states array
1129 (including the unused zero state) */
1130 U32 wordcount; /* Build only */
1132 STRLEN charcount; /* Build only */
1135 /* There is one (3 under DEBUGGING) pointers that logically belong in this
1136 structure, but are held outside as they need duplication on thread cloning,
1137 whereas the rest of the structure can be read only:
1138 HV *widecharmap; code points > 255 to charid
1140 AV *words; Array of words contained in trie, for dumping
1141 AV *revcharmap; Map of each charid back to its character representation
1145 #define TRIE_WORDS_OFFSET 2
1147 typedef struct _reg_trie_data reg_trie_data;
1149 /* refcount is first in both this and _reg_trie_data to allow a space
1150 optimisation in Perl_regdupe. */
1151 struct _reg_ac_data {
1155 reg_trie_state *states;
1157 typedef struct _reg_ac_data reg_ac_data;
1159 /* ANY_BIT doesn't use the structure, so we can borrow it here.
1160 This is simpler than refactoring all of it as wed end up with
1161 three different sets... */
1163 #define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap)
1164 #define TRIE_BITMAP_BYTE(p, c) BITMAP_BYTE(TRIE_BITMAP(p), c)
1165 #define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c))
1166 #define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c))
1167 #define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
1169 #define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC)
1170 #define IS_TRIE_AC(op) ((op)>=AHOCORASICK)
1172 /* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */
1173 #define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1)
1174 #define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state))
1175 #define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state))
1178 #define TRIE_CHARCOUNT(trie) ((trie)->charcount)
1180 #define TRIE_CHARCOUNT(trie) (trie_charcount)
1183 #define RE_TRIE_MAXBUF_INIT 65536
1184 #define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF"
1185 #define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS"
1187 #define RE_COMPILE_RECURSION_INIT 1000
1188 #define RE_COMPILE_RECURSION_LIMIT "\022E_COMPILE_RECURSION_LIMIT"
1192 RE_DEBUG_FLAGS is used to control what debug output is emitted
1193 its divided into three groups of options, some of which interact.
1194 The three groups are: Compile, Execute, Extra. There is room for a
1195 further group, as currently only the low three bytes are used.
1214 If you modify any of these make sure you make corresponding changes to
1215 re.pm, especially to the documentation.
1221 #define RE_DEBUG_COMPILE_MASK 0x0000FF
1222 #define RE_DEBUG_COMPILE_PARSE 0x000001
1223 #define RE_DEBUG_COMPILE_OPTIMISE 0x000002
1224 #define RE_DEBUG_COMPILE_TRIE 0x000004
1225 #define RE_DEBUG_COMPILE_DUMP 0x000008
1226 #define RE_DEBUG_COMPILE_FLAGS 0x000010
1227 #define RE_DEBUG_COMPILE_TEST 0x000020
1230 #define RE_DEBUG_EXECUTE_MASK 0x00FF00
1231 #define RE_DEBUG_EXECUTE_INTUIT 0x000100
1232 #define RE_DEBUG_EXECUTE_MATCH 0x000200
1233 #define RE_DEBUG_EXECUTE_TRIE 0x000400
1236 #define RE_DEBUG_EXTRA_MASK 0x3FF0000
1237 #define RE_DEBUG_EXTRA_TRIE 0x0010000
1238 #define RE_DEBUG_EXTRA_STATE 0x0080000
1239 #define RE_DEBUG_EXTRA_OPTIMISE 0x0100000
1240 #define RE_DEBUG_EXTRA_BUFFERS 0x0400000
1241 #define RE_DEBUG_EXTRA_GPOS 0x0800000
1242 #define RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE 0x1000000
1243 #define RE_DEBUG_EXTRA_WILDCARD 0x2000000
1245 #define RE_DEBUG_EXTRA_STACK 0x0280000
1247 #define RE_DEBUG_FLAG(x) (re_debug_flags & (x))
1249 #define DEBUG_COMPILE_r(x) DEBUG_r( \
1250 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_MASK)) x )
1251 #define DEBUG_PARSE_r(x) DEBUG_r( \
1252 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_PARSE)) x )
1253 #define DEBUG_OPTIMISE_r(x) DEBUG_r( \
1254 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE)) x )
1255 #define DEBUG_DUMP_r(x) DEBUG_r( \
1256 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_DUMP)) x )
1257 #define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \
1258 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TRIE)) x )
1259 #define DEBUG_FLAGS_r(x) DEBUG_r( \
1260 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_FLAGS)) x )
1261 #define DEBUG_TEST_r(x) DEBUG_r( \
1262 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TEST)) x )
1264 #define DEBUG_EXECUTE_r(x) DEBUG_r( \
1265 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_MASK)) x )
1266 #define DEBUG_INTUIT_r(x) DEBUG_r( \
1267 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_INTUIT)) x )
1268 #define DEBUG_MATCH_r(x) DEBUG_r( \
1269 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_MATCH)) x )
1270 #define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \
1271 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_TRIE)) x )
1274 #define DEBUG_EXTRA_r(x) DEBUG_r( \
1275 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_MASK)) x )
1276 #define DEBUG_STATE_r(x) DEBUG_r( \
1277 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_STATE)) x )
1278 #define DEBUG_STACK_r(x) DEBUG_r( \
1279 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_STACK)) x )
1280 #define DEBUG_BUFFERS_r(x) DEBUG_r( \
1281 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_BUFFERS)) x )
1283 #define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \
1284 if (DEBUG_v_TEST || ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \
1285 RE_DEBUG_FLAG(RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE))) x )
1286 #define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \
1287 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_TRIE)) x )
1288 #define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \
1289 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_TRIE)) x )
1291 #define DEBUG_TRIE_r(x) DEBUG_r( \
1292 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TRIE \
1293 | RE_DEBUG_EXECUTE_TRIE )) x )
1294 #define DEBUG_GPOS_r(x) DEBUG_r( \
1295 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_GPOS)) x )
1297 #define DEBUG_DUMP_PRE_OPTIMIZE_r(x) DEBUG_r( \
1298 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE)) x )
1300 /* initialization */
1301 /* Get the debug flags for code not in regcomp.c nor regexec.c. This doesn't
1302 * initialize the variable if it isn't already there, instead it just assumes
1303 * the flags are 0 */
1304 #define DECLARE_AND_GET_RE_DEBUG_FLAGS_NON_REGEX \
1305 volatile IV re_debug_flags = 0; PERL_UNUSED_VAR(re_debug_flags); \
1307 SV * re_debug_flags_sv = NULL; \
1308 /* get_sv() can return NULL during global destruction. */ \
1309 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \
1310 if (re_debug_flags_sv && SvIOK(re_debug_flags_sv)) \
1311 re_debug_flags=SvIV(re_debug_flags_sv); \
1317 /* For use in regcomp.c and regexec.c, Get the debug flags, and initialize to
1318 * the defaults if not done already */
1319 #define DECLARE_AND_GET_RE_DEBUG_FLAGS \
1320 volatile IV re_debug_flags = 0; PERL_UNUSED_VAR(re_debug_flags); \
1322 SV * re_debug_flags_sv = NULL; \
1323 /* get_sv() can return NULL during global destruction. */ \
1324 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \
1325 if (re_debug_flags_sv) { \
1326 if (!SvIOK(re_debug_flags_sv)) /* If doesnt exist set to default */\
1327 sv_setuv(re_debug_flags_sv, \
1328 /* These defaults should be kept in sync with re.pm */ \
1329 RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \
1330 re_debug_flags=SvIV(re_debug_flags_sv); \
1334 #define isDEBUG_WILDCARD (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_WILDCARD))
1336 #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \
1337 const char * const rpv = \
1338 pv_pretty((dsv), (pv), (l), (m), \
1339 PL_colors[(c1)],PL_colors[(c2)], \
1340 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \
1341 const int rlen = SvCUR(dsv)
1343 /* This is currently unsed in the core */
1344 #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \
1345 const char * const rpv = \
1346 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \
1347 PL_colors[(c1)],PL_colors[(c2)], \
1348 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) )
1350 #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \
1351 const char * const rpv = \
1352 pv_pretty((dsv), (pv), (l), (m), \
1353 PL_colors[0], PL_colors[1], \
1354 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \
1355 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \
1358 #define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0))
1359 #define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "")
1361 #else /* if not DEBUGGING */
1363 #define DECLARE_AND_GET_RE_DEBUG_FLAGS dNOOP
1364 #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) dNOOP
1365 #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m)
1366 #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) dNOOP
1367 #define RE_SV_DUMPLEN(ItEm)
1368 #define RE_SV_TAIL(ItEm)
1369 #define isDEBUG_WILDCARD 0
1371 #endif /* DEBUG RELATED DEFINES */
1373 #define FIRST_NON_ASCII_DECIMAL_DIGIT 0x660 /* ARABIC_INDIC_DIGIT_ZERO */
1376 TRADITIONAL_BOUND = CC_WORDCHAR_,
1383 /* This unpacks the FLAGS field of ANYOF[HR]x nodes. The value it contains
1384 * gives the strict lower bound for the UTF-8 start byte of any code point
1385 * matchable by the node, and a loose upper bound as well.
1387 * The low bound is stored as 0xC0 + ((the upper 6 bits) >> 2)
1388 * The loose upper bound is determined from the lowest 2 bits and the low bound
1389 * (called x) as follows:
1391 * 11 The upper limit of the range can be as much as (EF - x) / 8
1392 * 10 The upper limit of the range can be as much as (EF - x) / 4
1393 * 01 The upper limit of the range can be as much as (EF - x) / 2
1394 * 00 The upper limit of the range can be as much as EF
1396 * For motivation of this design, see commit message in
1397 * 3146c00a633e9cbed741e10146662fbcedfdb8d3 */
1399 # define MAX_ANYOF_HRx_BYTE 0xF4
1401 # define MAX_ANYOF_HRx_BYTE 0xEF
1403 #define LOWEST_ANYOF_HRx_BYTE(b) (((b) >> 2) + 0xC0)
1404 #define HIGHEST_ANYOF_HRx_BYTE(b) \
1405 (LOWEST_ANYOF_HRx_BYTE(b) \
1406 + ((MAX_ANYOF_HRx_BYTE - LOWEST_ANYOF_HRx_BYTE(b)) >> ((b) & 3)))
1408 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
1409 # define GET_REGCLASS_AUX_DATA(a,b,c,d,e,f) get_regclass_aux_data(a,b,c,d,e,f)
1411 # define GET_REGCLASS_AUX_DATA(a,b,c,d,e,f) get_re_gclass_aux_data(a,b,c,d,e,f)
1414 #define REGNODE_TYPE(node) (PL_regnode_info[(node)].type)
1415 #define REGNODE_OFF_BY_ARG(node) (PL_regnode_info[(node)].off_by_arg)
1416 #define REGNODE_ARG_LEN(node) (PL_regnode_info[(node)].arg_len)
1417 #define REGNODE_ARG_LEN_VARIES(node) (PL_regnode_info[(node)].arg_len_varies)
1418 #define REGNODE_NAME(node) (PL_regnode_name[(node)])
1420 #if defined(PERL_IN_REGCOMP_C) || defined(PERL_IN_REGEXEC_C)
1421 #include "reginline.h"
1424 #endif /* PERL_REGCOMP_H_ */
1427 * ex: set ts=8 sts=4 sw=4 et: