This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c; ([?...]) are single char width
[perl5.git] / regcomp.h
CommitLineData
a0d0e21e 1/* regcomp.h
d6376244 2 *
4bb101f2 3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
663f364b 4 * 2000, 2001, 2002, 2003, 2005, 2006, 2007, by Larry Wall and others
d6376244
JH
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
a687059c 9 */
3dd7db29
JK
10
11#ifndef PERL_REGCOMP_H_
12#define PERL_REGCOMP_H_
13
e1d1eefb 14#include "regcharclass.h"
a687059c 15
58e23c8d 16/* Convert branch sequences to more efficient trie ops? */
07be1b83 17#define PERL_ENABLE_TRIE_OPTIMISATION 1
58e23c8d 18
486ec47a 19/* Be really aggressive about optimising patterns with trie sequences? */
07be1b83 20#define PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION 1
58e23c8d
YO
21
22/* Should the optimiser take positive assertions into account? */
68ba3a3f 23#define PERL_ENABLE_POSITIVE_ASSERTION_STUDY 0
58e23c8d
YO
24
25/* Not for production use: */
07be1b83 26#define PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS 0
58e23c8d 27
7122b237
YO
28/* Activate offsets code - set to if 1 to enable */
29#ifdef DEBUGGING
30#define RE_TRACK_PATTERN_OFFSETS
31#endif
32
a687059c
LW
33/*
34 * The "internal use only" fields in regexp.h are present to pass info from
35 * compile to execute that permits the execute phase to run lots faster on
36 * simple cases. They are:
37 *
bd61b366 38 * regstart sv that must begin a match; NULL if none obvious
a687059c
LW
39 * reganch is the match anchored (at beginning-of-line only)?
40 * regmust string (pointer into program) that match must include, or NULL
79072805 41 * [regmust changed to SV* for bminstr()--law]
a687059c
LW
42 * regmlen length of regmust string
43 * [regmlen not used currently]
44 *
45 * Regstart and reganch permit very fast decisions on suitable starting points
46 * for a match, cutting down the work a lot. Regmust permits fast rejection
47 * of lines that cannot possibly match. The regmust tests are costly enough
e50aee73 48 * that pregcomp() supplies a regmust only if the r.e. contains something
a687059c
LW
49 * potentially expensive (at present, the only such thing detected is * or +
50 * at the start of the r.e., which can involve a lot of backup). Regmlen is
e50aee73 51 * supplied because the test in pregexec() needs it and pregcomp() is computing
a687059c
LW
52 * it anyway.
53 * [regmust is now supplied always. The tests that use regmust have a
54 * heuristic that disables the test if it usually matches.]
55 *
56 * [In fact, we now use regmust in many cases to locate where the search
57 * starts in the string, so if regback is >= 0, the regmust search is never
58 * wasted effort. The regback variable says how many characters back from
59 * where regmust matched is the earliest possible start of the match.
60 * For instance, /[a-z].foo/ has a regmust of 'foo' and a regback of 2.]
61 */
62
63/*
64 * Structure for regexp "program". This is essentially a linear encoding
65 * of a nondeterministic finite-state machine (aka syntax charts or
66 * "railroad normal form" in parsing technology). Each node is an opcode
67 * plus a "next" pointer, possibly plus an operand. "Next" pointers of
68 * all nodes except BRANCH implement concatenation; a "next" pointer with
69 * a BRANCH on both ends of it is connecting two alternatives. (Here we
70 * have one of the subtle syntax dependencies: an individual BRANCH (as
71 * opposed to a collection of them) is never concatenated with anything
72 * because of operator precedence.) The operand of some types of node is
73 * a literal string; for others, it is a node leading into a sub-FSM. In
74 * particular, the operand of a BRANCH node is the first node of the branch.
75 * (NB this is *not* a tree structure: the tail of the branch connects
a3621e74
YO
76 * to the thing following the set of BRANCHes.) The opcodes are defined
77 * in regnodes.h which is generated from regcomp.sym by regcomp.pl.
a687059c
LW
78 */
79
a687059c
LW
80/*
81 * A node is one char of opcode followed by two chars of "next" pointer.
82 * "Next" pointers are stored as two 8-bit pieces, high order first. The
83 * value is a positive offset from the opcode of the node containing it.
84 * An operand, if any, simply follows the node. (Note that much of the
85 * code generation knows about this implicit relationship.)
86 *
87 * Using two bytes for the "next" pointer is vast overkill for most things,
88 * but allows patterns to get big without disasters.
89 *
e91177ed 90 * [The "next" pointer is always aligned on an even
62e6ef33 91 * boundary, and reads the offset directly as a short.]
a687059c 92 */
785a26d5
YO
93
94/* This is the stuff that used to live in regexp.h that was truly
95 private to the engine itself. It now lives here. */
96
785a26d5 97 typedef struct regexp_internal {
1f1031fe 98 int name_list_idx; /* Optional data index of an array of paren names */
7122b237
YO
99 union {
100 U32 *offsets; /* offset annotations 20001228 MJD
101 data about mapping the program to the
102 string -
103 offsets[0] is proglen when this is used
104 */
105 U32 proglen;
106 } u;
107
6d5c990f
RGS
108 regnode *regstclass; /* Optional startclass as identified or constructed
109 by the optimiser */
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 */
1acab4c5 114 struct reg_code_blocks *code_blocks;/* positions of literal (?{}) */
6d5c990f
RGS
115 regnode program[1]; /* Unwarranted chumminess with compiler. */
116} regexp_internal;
117
118#define RXi_SET(x,y) (x)->pprivate = (void*)(y)
119#define RXi_GET(x) ((regexp_internal *)((x)->pprivate))
120#define RXi_GET_DECL(r,ri) regexp_internal *ri = RXi_GET(r)
785a26d5
YO
121/*
122 * Flags stored in regexp->intflags
123 * These are used only internally to the regexp engine
124 *
125 * See regexp.h for flags used externally to the regexp engine
126 */
e3e400ec
YO
127#define RXp_INTFLAGS(rx) ((rx)->intflags)
128#define RX_INTFLAGS(prog) RXp_INTFLAGS(ReANY(prog))
129
785a26d5
YO
130#define PREGf_SKIP 0x00000001
131#define PREGf_IMPLICIT 0x00000002 /* Converted .* to ^.* */
132#define PREGf_NAUGHTY 0x00000004 /* how exponential is this pattern? */
133#define PREGf_VERBARG_SEEN 0x00000008
134#define PREGf_CUTGROUP_SEEN 0x00000010
732caac7 135#define PREGf_USE_RE_EVAL 0x00000020 /* compiled with "use re 'eval'" */
e3e400ec
YO
136/* these used to be extflags, but are now intflags */
137#define PREGf_NOSCAN 0x00000040
33c28ab2 138 /* spare */
8e1490ee
YO
139#define PREGf_GPOS_SEEN 0x00000100
140#define PREGf_GPOS_FLOAT 0x00000200
785a26d5 141
d3d47aac
YO
142#define PREGf_ANCH_MBOL 0x00000400
143#define PREGf_ANCH_SBOL 0x00000800
144#define PREGf_ANCH_GPOS 0x00001000
d5a00e4a 145#define PREGf_RECURSE_SEEN 0x00002000
8e1490ee 146
d3d47aac
YO
147#define PREGf_ANCH \
148 ( PREGf_ANCH_SBOL | PREGf_ANCH_GPOS | PREGf_ANCH_MBOL )
785a26d5
YO
149
150/* this is where the old regcomp.h started */
a687059c 151
c277df42 152struct regnode_string {
cd439c50 153 U8 str_len;
c277df42
IZ
154 U8 type;
155 U16 next_off;
cd439c50 156 char string[1];
c277df42 157};
a687059c 158
ae06e581
KW
159struct regnode_lstring { /* Constructed this way to keep the string aligned. */
160 U8 flags;
161 U8 type;
162 U16 next_off;
1d425dad 163 U32 str_len; /* Only 18 bits allowed before would overflow 'next_off' */
ae06e581
KW
164 char string[1];
165};
166
34924db0 167struct regnode_anyofhs { /* Constructed this way to keep the string aligned. */
53d42e43 168 U8 str_len;
34924db0
KW
169 U8 type;
170 U16 next_off;
171 U32 arg1; /* set by set_ANYOF_arg() */
172 char string[1];
173};
174
6bda09f9
YO
175/* Argument bearing node - workhorse,
176 arg1 is often for the data field */
c277df42
IZ
177struct regnode_1 {
178 U8 flags;
179 U8 type;
180 U16 next_off;
181 U32 arg1;
182};
183
6bda09f9
YO
184/* Similar to a regnode_1 but with an extra signed argument */
185struct regnode_2L {
186 U8 flags;
187 U8 type;
188 U16 next_off;
189 U32 arg1;
190 I32 arg2;
191};
192
193/* 'Two field' -- Two 16 bit unsigned args */
c277df42
IZ
194struct regnode_2 {
195 U8 flags;
196 U8 type;
197 U16 next_off;
198 U16 arg1;
199 U16 arg2;
200};
201
dcb20b36 202#define ANYOF_BITMAP_SIZE (NUM_ANYOF_CODE_POINTS / 8) /* 8 bits/Byte */
936ed897 203
1ee208c4
KW
204/* Note that these form structs which are supersets of the next smaller one, by
205 * appending fields. Alignment problems can occur if one of those optional
206 * fields requires stricter alignment than the base struct. And formal
207 * parameters that can really be two or more of the structs should be
208 * declared as the smallest one it could be. See commit message for
209 * 7dcac5f6a5195002b55c935ee1d67f67e1df280b. Regnode allocation is done
210 * without regard to alignment, and changing it to would also require changing
211 * the code that inserts and deletes regnodes. The basic single-argument
212 * regnode has a U32, which is what reganode() allocates as a unit. Therefore
213 * no field can require stricter alignment than U32. */
214
786e8c11 215/* also used by trie */
936ed897
IZ
216struct regnode_charclass {
217 U8 flags;
218 U8 type;
219 U16 next_off;
9da99c5e 220 U32 arg1; /* set by set_ANYOF_arg() */
19103b91 221 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */
936ed897
IZ
222};
223
bf3f2d85 224/* has runtime (locale) \d, \w, ..., [:posix:] classes */
ea461dd6 225struct regnode_charclass_posixl {
93e92956 226 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
936ed897
IZ
227 U8 type;
228 U16 next_off;
7e817188 229 U32 arg1;
03fa83ba 230 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
9cf78fa0 231 U32 classflags; /* and run-time */
936ed897
IZ
232};
233
85c8e306
KW
234/* A synthetic start class (SSC); is a regnode_charclass_posixl_fold, plus an
235 * extra SV*, used only during its construction and which is not used by
236 * regexec.c. Note that the 'next_off' field is unused, as the SSC stands
237 * alone, so there is never a next node. Also, there is no alignment issue,
b8fda935 238 * because these are declared or allocated as a complete unit so the compiler
85c8e306
KW
239 * takes care of alignment. This is unlike the other regnodes which are
240 * allocated in terms of multiples of a single-argument regnode. SSC nodes can
241 * have a pointer field because there is no alignment issue, and because it is
242 * set to NULL after construction, before any cloning of the pattern */
b8f7bb16 243struct regnode_ssc {
93e92956 244 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
b8f7bb16
KW
245 U8 type;
246 U16 next_off;
7e817188 247 U32 arg1;
03fa83ba 248 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
631e9161 249 U32 classflags; /* ... and run-time */
85c8e306
KW
250
251 /* Auxiliary, only used during construction; NULL afterwards: list of code
252 * points matched */
253 SV* invlist;
b8f7bb16
KW
254};
255
71068078
KW
256/* We take advantage of 'next_off' not otherwise being used in the SSC by
257 * actually using it: by setting it to 1. This allows us to test and
258 * distinguish between an SSC and other ANYOF node types, as 'next_off' cannot
259 * otherwise be 1, because it is the offset to the next regnode expressed in
260 * units of regnodes. Since an ANYOF node contains extra fields, it adds up
261 * to 12 regnode units on 32-bit systems, (hence the minimum this can be (if
262 * not 0) is 11 there. Even if things get tightly packed on a 64-bit system,
263 * it still would be more than 1. */
264#define set_ANYOF_SYNTHETIC(n) STMT_START{ OP(n) = ANYOF; \
265 NEXT_OFF(n) = 1; \
266 } STMT_END
c1071334 267#define is_ANYOF_SYNTHETIC(n) (PL_regkind[OP(n)] == ANYOF && NEXT_OFF(n) == 1)
71068078 268
9c7e81e8
AD
269/* XXX fix this description.
270 Impose a limit of REG_INFTY on various pattern matching operations
271 to limit stack growth and to avoid "infinite" recursions.
272*/
4288c5b9
KW
273/* The default size for REG_INFTY is U16_MAX, which is the same as
274 USHORT_MAX (see perl.h). Unfortunately U16 isn't necessarily 16 bits
275 (see handy.h). On the Cray C90, sizeof(short)==4 and hence U16_MAX is
276 ((1<<32)-1), while on the Cray T90, sizeof(short)==8 and U16_MAX is
277 ((1<<64)-1). To limit stack growth to reasonable sizes, supply a
9c7e81e8
AD
278 smaller default.
279 --Andy Dougherty 11 June 1998
280*/
281#if SHORTSIZE > 2
282# ifndef REG_INFTY
4288c5b9 283# define REG_INFTY ((1<<16)-1)
9c7e81e8
AD
284# endif
285#endif
286
287#ifndef REG_INFTY
4288c5b9 288# define REG_INFTY U16_MAX
83cfe48c 289#endif
a687059c 290
e91177ed
IZ
291#define ARG_VALUE(arg) (arg)
292#define ARG__SET(arg,val) ((arg) = (val))
c277df42 293
e8e3b0fb
JH
294#undef ARG
295#undef ARG1
296#undef ARG2
297
c277df42
IZ
298#define ARG(p) ARG_VALUE(ARG_LOC(p))
299#define ARG1(p) ARG_VALUE(ARG1_LOC(p))
300#define ARG2(p) ARG_VALUE(ARG2_LOC(p))
6bda09f9 301#define ARG2L(p) ARG_VALUE(ARG2L_LOC(p))
786e8c11 302
c277df42
IZ
303#define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val))
304#define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val))
305#define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val))
6bda09f9 306#define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val))
c277df42 307
e8e3b0fb
JH
308#undef NEXT_OFF
309#undef NODE_ALIGN
310
075abff3
AL
311#define NEXT_OFF(p) ((p)->next_off)
312#define NODE_ALIGN(node)
aa48e906
YO
313/* the following define was set to 0xde in 075abff3
314 * as part of some linting logic. I have set it to 0
315 * as otherwise in every place where we /might/ set flags
316 * we have to set it 0 explicitly, which duplicates
317 * assignments and IMO adds an unacceptable level of
318 * surprise to working in the regex engine. If this
319 * is changed from 0 then at the very least make sure
320 * that SBOL for /^/ sets the flags to 0 explicitly.
321 * -- Yves */
322#define NODE_ALIGN_FILL(node) ((node)->flags = 0)
a687059c 323
c277df42
IZ
324#define SIZE_ALIGN NODE_ALIGN
325
e8e3b0fb
JH
326#undef OP
327#undef OPERAND
e8e3b0fb
JH
328#undef STRING
329
e91177ed 330#define OP(p) ((p)->type)
b988e673
KW
331#define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \
332 regnode types. For some, it's the \
333 character set of the regnode */
3f2416ae 334#define STR_LENs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \
5cd61b66 335 ((struct regnode_string *)p)->str_len)
3f2416ae 336#define STRINGs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \
5cd61b66 337 ((struct regnode_string *)p)->string)
ae06e581
KW
338#define OPERANDs(p) STRINGs(p)
339
340/* Long strings. Currently limited to length 18 bits, which handles a 262000
341 * byte string. The limiting factor is the 16 bit 'next_off' field, which
342 * points to the next regnode, so the furthest away it can be is 2**16. On
343 * most architectures, regnodes are 2**2 bytes long, so that yields 2**18
344 * bytes. Should a longer string be desired, we could increase it to 26 bits
345 * fairly easily, by changing this node to have longj type which causes the ARG
346 * field to be used for the link to the next regnode (although code would have
347 * to be changed to account for this), and then use a combination of the flags
348 * and next_off fields for the length. To get 34 bit length, also change the
349 * node to be an ARG2L, using the second 32 bit field for the length, and not
350 * using the flags nor next_off fields at all. One could have an llstring node
351 * and even an lllstring type. */
3f2416ae 352#define STR_LENl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 353 (((struct regnode_lstring *)p)->str_len))
3f2416ae 354#define STRINGl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 355 (((struct regnode_lstring *)p)->string))
ae06e581
KW
356#define OPERANDl(p) STRINGl(p)
357
3f2416ae 358#define STR_LEN(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 359 ? STR_LENl(p) : STR_LENs(p))
3f2416ae 360#define STRING(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 361 ? STRINGl(p) : STRINGs(p))
c5184715
KW
362#define OPERAND(p) STRING(p)
363
c45abc0a
KW
364/* The number of (smallest) regnode equivalents that a string of length l bytes
365 * occupies */
59e54936 366#define STR_SZ(l) (((l) + sizeof(regnode) - 1) / sizeof(regnode))
c45abc0a
KW
367
368/* The number of (smallest) regnode equivalents that the EXACTISH node 'p'
369 * occupies */
ae06e581 370#define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p)) + 1 + regarglen[(p)->type])
c45abc0a 371
c7fd9c67 372#define setSTR_LEN(p,v) \
ae06e581 373 STMT_START{ \
3f2416ae 374 if (OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
ae06e581
KW
375 ((struct regnode_lstring *)(p))->str_len = (v); \
376 else \
377 ((struct regnode_string *)(p))->str_len = (v); \
378 } STMT_END
cd439c50 379
13fcf652
KW
380#define ANYOFR_BASE_BITS 20
381#define ANYOFRbase(p) (ARG(p) & ((1 << ANYOFR_BASE_BITS) - 1))
382#define ANYOFRdelta(p) (ARG(p) >> ANYOFR_BASE_BITS)
383
e8e3b0fb
JH
384#undef NODE_ALIGN
385#undef ARG_LOC
386#undef NEXTOPER
387#undef PREVOPER
388
e91177ed
IZ
389#define NODE_ALIGN(node)
390#define ARG_LOC(p) (((struct regnode_1 *)p)->arg1)
391#define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1)
392#define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2)
6bda09f9 393#define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2)
786e8c11 394
e91177ed
IZ
395#define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */
396#define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2)
397
c277df42
IZ
398#define NEXTOPER(p) ((p) + NODE_STEP_REGNODE)
399#define PREVOPER(p) ((p) - NODE_STEP_REGNODE)
400
f55b7b26
KW
401#define FILL_NODE(offset, op) \
402 STMT_START { \
403 OP(REGNODE_p(offset)) = op; \
404 NEXT_OFF(REGNODE_p(offset)) = 0; \
d2a88850 405 } STMT_END
f55b7b26
KW
406#define FILL_ADVANCE_NODE(offset, op) \
407 STMT_START { \
408 FILL_NODE(offset, op); \
409 (offset)++; \
d2a88850 410 } STMT_END
f55b7b26
KW
411#define FILL_ADVANCE_NODE_ARG(offset, op, arg) \
412 STMT_START { \
413 ARG_SET(REGNODE_p(offset), arg); \
414 FILL_ADVANCE_NODE(offset, op); \
415 /* This is used generically for other operations \
416 * that have a longer argument */ \
417 (offset) += regarglen[op]; \
93539530 418 } STMT_END
f55b7b26
KW
419#define FILL_ADVANCE_NODE_2L_ARG(offset, op, arg1, arg2) \
420 STMT_START { \
421 ARG_SET(REGNODE_p(offset), arg1); \
422 ARG2L_SET(REGNODE_p(offset), arg2); \
423 FILL_ADVANCE_NODE(offset, op); \
424 (offset) += 2; \
93539530 425 } STMT_END
a687059c 426
22c35a8c 427#define REG_MAGIC 0234
a687059c 428
fe8d1b7c
KW
429/* An ANYOF node is basically a bitmap with the index being a code point. If
430 * the bit for that code point is 1, the code point matches; if 0, it doesn't
431 * match (complemented if inverted). There is an additional mechanism to deal
432 * with cases where the bitmap is insufficient in and of itself. This #define
433 * indicates if the bitmap does fully represent what this ANYOF node can match.
434 * The ARG is set to this special value (since 0, 1, ... are legal, but will
435 * never reach this high). */
93e92956 436#define ANYOF_ONLY_HAS_BITMAP ((U32) -1)
137165a6 437
93539530 438/* When the bitmap isn't completely sufficient for handling the ANYOF node,
fe8d1b7c
KW
439 * flags (in node->flags of the ANYOF node) get set to indicate this. These
440 * are perennially in short supply. Beyond several cases where warnings need
441 * to be raised under certain circumstances, currently, there are six cases
442 * where the bitmap alone isn't sufficient. We could use six flags to
443 * represent the 6 cases, but to save flags bits, we play some games. The
444 * cases are:
d2f465e9 445 *
fe8d1b7c
KW
446 * 1) The bitmap has a compiled-in very finite size. So something else needs
447 * to be used to specify if a code point that is too large for the bitmap
7a94a5fe 448 * actually matches. The mechanism currently is an inversion
fe8d1b7c
KW
449 * list. ANYOF_ONLY_HAS_BITMAP, described above, being TRUE indicates
450 * there are no matches of too-large code points. But if it is FALSE,
451 * then almost certainly there are matches too large for the bitmap. (The
452 * other cases, described below, either imply this one or are extremely
453 * rare in practice.) So we can just assume that a too-large code point
454 * will need something beyond the bitmap if ANYOF_ONLY_HAS_BITMAP is
455 * FALSE, instead of having a separate flag for this.
456 * 2) A subset of item 1) is if all possible code points outside the bitmap
457 * match. This is a common occurrence when the class is complemented,
458 * like /[^ij]/. Therefore a bit is reserved to indicate this,
42d0f81a 459 * rather than having an inversion list created,
d1c40ef5 460 * ANYOF_MATCHES_ALL_ABOVE_BITMAP.
fe8d1b7c
KW
461 * 3) Under /d rules, it can happen that code points that are in the upper
462 * latin1 range (\x80-\xFF or their equivalents on EBCDIC platforms) match
463 * only if the runtime target string being matched against is UTF-8. For
464 * example /[\w[:punct:]]/d. This happens only for posix classes (with a
d1c40ef5
KW
465 * couple of exceptions, like \d where it doesn't happen), and all such
466 * ones also have above-bitmap matches. Thus, 3) implies 1) as well.
467 * Note that /d rules are no longer encouraged; 'use 5.14' or higher
468 * deselects them. But a flag is required so that they can be properly
469 * handled. But it can be a shared flag: see 5) below.
470 * 4) Also under /d rules, something like /[\Wfoo]/ will match everything in
fe8d1b7c 471 * the \x80-\xFF range, unless the string being matched against is UTF-8.
7a94a5fe
KW
472 * An inversion list could be created for this case, but this is
473 * relatively common, and it turns out that it's all or nothing: if any
474 * one of these code points matches, they all do. Hence a single bit
475 * suffices. We use a shared flag that doesn't take up space by itself:
476 * ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER. This
477 * also implies 1), with one exception: [:^cntrl:].
fe8d1b7c
KW
478 * 5) A user-defined \p{} property may not have been defined by the time the
479 * regex is compiled. In this case, we don't know until runtime what it
480 * will match, so we have to assume it could match anything, including
481 * code points that ordinarily would be in the bitmap. A flag bit is
482 * necessary to indicate this , though it can be shared with the item 3)
483 * flag, as that only occurs under /d, and this only occurs under non-d.
484 * This case is quite uncommon in the field, and the /(?[ ...])/ construct
485 * is a better way to accomplish what this feature does. This case also
486 * implies 1).
487 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP
e2f5e63d 488 * is the shared flag.
fe8d1b7c
KW
489 * 6) /[foo]/il may have folds that are only valid if the runtime locale is a
490 * UTF-8 one. These are quite rare, so it would be good to avoid the
491 * expense of looking for them. But /l matching is slow anyway, and we've
e2f5e63d 492 * traditionally not worried too much about its performance. And this
037715a6 493 * condition requires the ANYOFL_FOLD flag to be set, so testing for
fe8d1b7c 494 * that flag would be sufficient to rule out most cases of this. So it is
d1c40ef5
KW
495 * unclear if this should have a flag or not. But, this flag can be
496 * shared with another, so it doesn't occupy extra space.
d2f465e9 497 *
d1c40ef5 498 * At the moment, there is one spare bit, but this could be increased by
93539530 499 * various tricks:
e2f5e63d 500 *
93539530 501 * If just one more bit is needed, as of this writing it seems to khw that the
d1c40ef5
KW
502 * best choice would be to make ANYOF_MATCHES_ALL_ABOVE_BITMAP not a flag, but
503 * something like
d2f465e9 504 *
d1c40ef5 505 * #define ANYOF_MATCHES_ALL_ABOVE_BITMAP ((U32) -2)
fe8d1b7c 506 *
d1c40ef5
KW
507 * and access it through the ARG like ANYOF_ONLY_HAS_BITMAP is. This flag is
508 * used by all ANYOF node types, and it could be used to avoid calling the
509 * handler function, as the macro REGINCLASS in regexec.c does now for other
510 * cases.
511 *
46fc0c43
KW
512 * Another possibility is based on the fact that ANYOF_MATCHES_POSIXL is
513 * redundant with the node type ANYOFPOSIXL. That flag could be removed, but
514 * at the expense of extra code in regexec.c. The flag has been retained
515 * because it allows us to see if we need to call reginsert, or just use the
516 * bitmap in one test.
fe8d1b7c 517 *
46fc0c43
KW
518 * If this is done, an extension would be to make all ANYOFL nodes contain the
519 * extra 32 bits that ANYOFPOSIXL ones do. The posix flags only occupy 30
520 * bits, so the ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD flags
521 * and ANYOFL_FOLD could be moved to that extra space, but it would mean extra
522 * instructions, as there are currently places in the code that assume those
523 * two bits are zero.
d1c40ef5
KW
524 *
525 * All told, 5 bits could be available for other uses if all of the above were
526 * done.
a4525e78 527 *
d1c40ef5 528 * Some flags are not used in synthetic start class (SSC) nodes, so could be
93e92956
KW
529 * shared should new flags be needed for SSCs, like SSC_MATCHES_EMPTY_STRING
530 * now. */
b8c5462f 531
d2f465e9
KW
532/* If this is set, the result of the match should be complemented. regexec.c
533 * is expecting this to be in the low bit. Never in an SSC */
93e92956 534#define ANYOF_INVERT 0x01
1984f6a0 535
a0dd4231 536/* For the SSC node only, which cannot be inverted, so is shared with that bit.
93e92956
KW
537 * This is used only during regex compilation. */
538#define SSC_MATCHES_EMPTY_STRING ANYOF_INVERT
a0dd4231 539
509d4de0
KW
540/* Set if this is a regnode_charclass_posixl vs a regnode_charclass. This
541 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale
542 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved
543 * at compile-time. Only set under /l; can be in SSC */
544#define ANYOF_MATCHES_POSIXL 0x02
b63f73da
KW
545
546/* The fold is calculated and stored in the bitmap where possible at compile
547 * time. However under locale, the actual folding varies depending on
548 * what the locale is at the time of execution, so it has to be deferred until
d2f465e9 549 * then. Only set under /l; never in an SSC */
037715a6 550#define ANYOFL_FOLD 0x04
b63f73da 551
d1c40ef5 552/* Shared bit set only with ANYOFL and SSC nodes:
269bb65b
KW
553 * If ANYOFL_FOLD is set, this flag indicates there are potential matches
554 * valid only if the locale is a UTF-8 one.
555 * If ANYOFL_FOLD is NOT set, this flag means to warn if the runtime locale
d1c40ef5
KW
556 * isn't a UTF-8 one (and the generated node assumes a UTF-8 locale).
557 * None of INVERT, POSIXL,
558 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP
559 * can be set. */
560#define ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD 0x08
561
562/* Convenience macros for teasing apart the meanings when reading the above bit
563 * */
564#define ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags) \
565 ((flags & ( ANYOFL_FOLD /* Both bits are set */ \
566 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \
567 == ( ANYOFL_FOLD \
568 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD))
569
570#define ANYOFL_UTF8_LOCALE_REQD(flags) \
571 ((flags & ( ANYOFL_FOLD /* Only REQD bit is set */ \
572 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \
573 == ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)
574
1744d18d 575/* Spare: Be sure to change ANYOF_FLAGS_ALL if this gets used 0x10 */
c8d3cd88 576
509d4de0
KW
577/* If set, the node matches every code point NUM_ANYOF_CODE_POINTS and above.
578 * Can be in an SSC */
b3b1cf17 579#define ANYOF_MATCHES_ALL_ABOVE_BITMAP 0x20
ef87b810 580
108316fb
KW
581/* Shared bit:
582 * Under /d it means the ANYOFD node matches more things if the target
583 * string is encoded in UTF-8; any such things will be non-ASCII,
7a94a5fe
KW
584 * characters that are < 256, and can be accessed via the inversion
585 * list.
108316fb
KW
586 * When not under /d, it means the ANYOF node contains a user-defined
587 * property that wasn't yet defined at the time the regex was compiled,
7a94a5fe 588 * and so must be looked up at runtime, by creating an inversion list.
108316fb
KW
589 * (These uses are mutually exclusive because a user-defined property is
590 * specified by \p{}, and \p{} implies /u which deselects /d). The long macro
591 * name is to make sure that you are cautioned about its shared nature. Only
592 * the non-/d meaning can be in an SSC */
593#define ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 0x40
f240c685
KW
594
595/* Shared bit:
d2f465e9 596 * Under /d it means the ANYOFD node matches all non-ASCII Latin1
f240c685
KW
597 * characters when the target string is not in utf8.
598 * When not under /d, it means the ANYOF node should raise a warning if
599 * matching against an above-Unicode code point.
600 * (These uses are mutually exclusive because the warning requires a \p{}, and
601 * \p{} implies /u which deselects /d). An SSC node only has this bit set if
602 * what is meant is the warning. The long macro name is to make sure that you
603 * are cautioned about its shared nature */
604#define ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER 0x80
7bc66b18 605
b271ede1 606#define ANYOF_FLAGS_ALL (0xff & ~0x10)
653099ff 607
037715a6 608#define ANYOF_LOCALE_FLAGS (ANYOFL_FOLD | ANYOF_MATCHES_POSIXL)
ace6b0e4 609
eff8b7dc
KW
610/* These are the flags that apply to both regular ANYOF nodes and synthetic
611 * start class nodes during construction of the SSC. During finalization of
d2f465e9 612 * the SSC, other of the flags may get added to it */
d1c40ef5 613#define ANYOF_COMMON_FLAGS 0
eff8b7dc 614
936ed897 615/* Character classes for node->classflags of ANYOF */
653099ff 616/* Should be synchronized with a table in regprop() */
575a8a29 617/* 2n should be the normal one, paired with its complement at 2n+1 */
b8c5462f 618
08ada29e
KW
619#define ANYOF_ALPHA ((_CC_ALPHA) * 2)
620#define ANYOF_NALPHA ((ANYOF_ALPHA) + 1)
e8d596e0
KW
621#define ANYOF_ALPHANUMERIC ((_CC_ALPHANUMERIC) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */
622#define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1)
08ada29e
KW
623#define ANYOF_ASCII ((_CC_ASCII) * 2)
624#define ANYOF_NASCII ((ANYOF_ASCII) + 1)
e8d596e0
KW
625#define ANYOF_BLANK ((_CC_BLANK) * 2) /* GNU extension: space and tab: non-vertical space */
626#define ANYOF_NBLANK ((ANYOF_BLANK) + 1)
b0d691b2
KW
627#define ANYOF_CASED ((_CC_CASED) * 2) /* Pseudo class for [:lower:] or
628 [:upper:] under /i */
629#define ANYOF_NCASED ((ANYOF_CASED) + 1)
08ada29e
KW
630#define ANYOF_CNTRL ((_CC_CNTRL) * 2)
631#define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1)
e8d596e0
KW
632#define ANYOF_DIGIT ((_CC_DIGIT) * 2) /* \d */
633#define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1)
08ada29e
KW
634#define ANYOF_GRAPH ((_CC_GRAPH) * 2)
635#define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1)
636#define ANYOF_LOWER ((_CC_LOWER) * 2)
637#define ANYOF_NLOWER ((ANYOF_LOWER) + 1)
638#define ANYOF_PRINT ((_CC_PRINT) * 2)
639#define ANYOF_NPRINT ((ANYOF_PRINT) + 1)
640#define ANYOF_PUNCT ((_CC_PUNCT) * 2)
641#define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1)
e8d596e0
KW
642#define ANYOF_SPACE ((_CC_SPACE) * 2) /* \s */
643#define ANYOF_NSPACE ((ANYOF_SPACE) + 1)
08ada29e
KW
644#define ANYOF_UPPER ((_CC_UPPER) * 2)
645#define ANYOF_NUPPER ((ANYOF_UPPER) + 1)
e8d596e0
KW
646#define ANYOF_WORDCHAR ((_CC_WORDCHAR) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */
647#define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1)
08ada29e
KW
648#define ANYOF_XDIGIT ((_CC_XDIGIT) * 2)
649#define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1)
b8c5462f 650
87cad4c0 651/* pseudo classes below this, not stored in the class bitmap, but used as flags
e1d1eefb
YO
652 during compilation of char classes */
653
b2ef2e34
KW
654#define ANYOF_VERTWS ((_CC_VERTSPACE) * 2)
655#define ANYOF_NVERTWS ((ANYOF_VERTWS)+1)
e1d1eefb 656
b2ef2e34
KW
657/* It is best if this is the last one, as all above it are stored as bits in a
658 * bitmap, and it isn't part of that bitmap */
659#if _CC_VERTSPACE != _HIGHEST_REGCOMP_DOT_H_SYNC
660# error Problem with handy.h _HIGHEST_REGCOMP_DOT_H_SYNC #define
661#endif
662
8efd3f97
KW
663#define ANYOF_POSIXL_MAX (ANYOF_VERTWS) /* So upper loop limit is written:
664 * '< ANYOF_MAX'
665 * Hence doesn't include VERTWS, as that
666 * is a pseudo class */
667#define ANYOF_MAX ANYOF_POSIXL_MAX
668
669#if (ANYOF_POSIXL_MAX > 32) /* Must fit in 32-bit word */
b2ef2e34 670# error Problem with handy.h _CC_foo #defines
a0947d7b
KW
671#endif
672
8efd3f97
KW
673#define ANYOF_HORIZWS ((ANYOF_POSIXL_MAX)+2) /* = (ANYOF_NVERTWS + 1) */
674#define ANYOF_NHORIZWS ((ANYOF_POSIXL_MAX)+3)
87cad4c0 675
8efd3f97
KW
676#define ANYOF_UNIPROP ((ANYOF_POSIXL_MAX)+4) /* Used to indicate a Unicode
677 property: \p{} or \P{} */
f39a4dc2 678
b8c5462f
JH
679/* Backward source code compatibility. */
680
681#define ANYOF_ALNUML ANYOF_ALNUM
682#define ANYOF_NALNUML ANYOF_NALNUM
683#define ANYOF_SPACEL ANYOF_SPACE
684#define ANYOF_NSPACEL ANYOF_NSPACE
8ebfc29a
KW
685#define ANYOF_ALNUM ANYOF_WORDCHAR
686#define ANYOF_NALNUM ANYOF_NWORDCHAR
b8c5462f
JH
687
688/* Utility macros for the bitmap and classes of ANYOF */
689
936ed897 690#define ANYOF_FLAGS(p) ((p)->flags)
b8c5462f 691
bc51fd78 692#define ANYOF_BIT(c) (1U << ((c) & 7))
b8c5462f 693
14ed74f6
KW
694#define POSIXL_SET(field, c) ((field) |= (1U << (c)))
695#define ANYOF_POSIXL_SET(p, c) POSIXL_SET(((regnode_charclass_posixl*) (p))->classflags, (c))
8efd3f97 696
14ed74f6
KW
697#define POSIXL_CLEAR(field, c) ((field) &= ~ (1U <<(c)))
698#define ANYOF_POSIXL_CLEAR(p, c) POSIXL_CLEAR(((regnode_charclass_posixl*) (p))->classflags, (c))
8efd3f97 699
14ed74f6
KW
700#define POSIXL_TEST(field, c) ((field) & (1U << (c)))
701#define ANYOF_POSIXL_TEST(p, c) POSIXL_TEST(((regnode_charclass_posixl*) (p))->classflags, (c))
b8c5462f 702
14ed74f6
KW
703#define POSIXL_ZERO(field) STMT_START { (field) = 0; } STMT_END
704#define ANYOF_POSIXL_ZERO(ret) POSIXL_ZERO(((regnode_charclass_posixl*) (ret))->classflags)
705
706#define ANYOF_POSIXL_SET_TO_BITMAP(p, bits) \
707 STMT_START { \
708 ((regnode_charclass_posixl*) (p))->classflags = (bits); \
709 } STMT_END
9cf78fa0
KW
710
711/* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */
d1d040e5 712#define ANYOF_POSIXL_SETALL(ret) STMT_START { ((regnode_charclass_posixl*) (ret))->classflags = ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1; } STMT_END
8efd3f97 713#define ANYOF_CLASS_SETALL(ret) ANYOF_POSIXL_SETALL(ret)
9cf78fa0 714
8efd3f97 715#define ANYOF_POSIXL_TEST_ANY_SET(p) \
93e92956 716 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
d1d040e5 717 && (((regnode_charclass_posixl*)(p))->classflags))
8efd3f97 718#define ANYOF_CLASS_TEST_ANY_SET(p) ANYOF_POSIXL_TEST_ANY_SET(p)
e867ce45 719
e0e1be5f
KW
720/* Since an SSC always has this field, we don't have to test for that; nor do
721 * we want to because the bit isn't set for SSC during its construction */
722#define ANYOF_POSIXL_SSC_TEST_ANY_SET(p) \
723 cBOOL(((regnode_ssc*)(p))->classflags)
724#define ANYOF_POSIXL_SSC_TEST_ALL_SET(p) /* Are all bits set? */ \
725 (((regnode_ssc*) (p))->classflags \
726 == ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1)
727
728#define ANYOF_POSIXL_TEST_ALL_SET(p) \
93e92956 729 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
e0e1be5f
KW
730 && ((regnode_charclass_posixl*) (p))->classflags \
731 == ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1)
fb38762f 732
b27fd97a 733#define ANYOF_POSIXL_OR(source, dest) STMT_START { (dest)->classflags |= (source)->classflags ; } STMT_END
8efd3f97 734#define ANYOF_CLASS_OR(source, dest) ANYOF_POSIXL_OR((source), (dest))
936ed897 735
cdd87c1d
KW
736#define ANYOF_POSIXL_AND(source, dest) STMT_START { (dest)->classflags &= (source)->classflags ; } STMT_END
737
428c1aff
KW
738#define ANYOF_BITMAP_ZERO(ret) Zero(((regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char)
739#define ANYOF_BITMAP(p) ((regnode_charclass*)(p))->bitmap
4cbce0a6 740#define ANYOF_BITMAP_BYTE(p, c) BITMAP_BYTE(ANYOF_BITMAP(p), c)
b8c5462f
JH
741#define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c))
742#define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c))
bc51fd78 743#define ANYOF_BITMAP_TEST(p, c) cBOOL(ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c))
b8c5462f 744
f8bef550
NC
745#define ANYOF_BITMAP_SETALL(p) \
746 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE)
747#define ANYOF_BITMAP_CLEARALL(p) \
748 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE)
f8bef550 749
a687059c
LW
750/*
751 * Utility definitions.
752 */
2304df62 753#ifndef CHARMASK
075abff3 754# define UCHARAT(p) ((int)*(const U8*)(p))
a687059c 755#else
075abff3 756# define UCHARAT(p) ((int)*(p)&CHARMASK)
a687059c 757#endif
a687059c 758
c45abc0a
KW
759/* Number of regnode equivalents that 'guy' occupies beyond the size of the
760 * smallest regnode. */
c277df42
IZ
761#define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode))
762
e384d5c1
YO
763#define REG_ZERO_LEN_SEEN 0x00000001
764#define REG_LOOKBEHIND_SEEN 0x00000002
765#define REG_GPOS_SEEN 0x00000004
471f5387 766/* spare */
e384d5c1
YO
767#define REG_RECURSE_SEEN 0x00000020
768#define REG_TOP_LEVEL_BRANCHES_SEEN 0x00000040
769#define REG_VERBARG_SEEN 0x00000080
770#define REG_CUTGROUP_SEEN 0x00000100
771#define REG_RUN_ON_COMMENT_SEEN 0x00000200
772#define REG_UNFOLDED_MULTI_SEEN 0x00000400
d5a00e4a 773/* spare */
ee273784
YO
774#define REG_UNBOUNDED_QUANTIFIER_SEEN 0x00001000
775
d09b2d29 776
73c4f7a1
GS
777START_EXTERN_C
778
be8e71aa
YO
779#ifdef PLUGGABLE_RE_EXTENSION
780#include "re_nodes.h"
781#else
d09b2d29 782#include "regnodes.h"
be8e71aa 783#endif
d09b2d29 784
f9f4320a
YO
785#ifndef PLUGGABLE_RE_EXTENSION
786#ifndef DOINIT
787EXTCONST regexp_engine PL_core_reg_engine;
788#else /* DOINIT */
789EXTCONST regexp_engine PL_core_reg_engine = {
fe578d7f 790 Perl_re_compile,
2fdbfb4d 791 Perl_regexec_flags,
f9f4320a
YO
792 Perl_re_intuit_start,
793 Perl_re_intuit_string,
2fdbfb4d
AB
794 Perl_regfree_internal,
795 Perl_reg_numbered_buff_fetch,
796 Perl_reg_numbered_buff_store,
797 Perl_reg_numbered_buff_length,
192b9cd1
AB
798 Perl_reg_named_buff,
799 Perl_reg_named_buff_iter,
49d7dfbc 800 Perl_reg_qr_package,
f9f4320a 801#if defined(USE_ITHREADS)
3c13cae6 802 Perl_regdupe_internal,
f9f4320a 803#endif
3c13cae6 804 Perl_re_op_compile
f9f4320a
YO
805};
806#endif /* DOINIT */
807#endif /* PLUGGABLE_RE_EXTENSION */
808
809
73c4f7a1 810END_EXTERN_C
cad2e5aa 811
cad2e5aa 812
261faec3
MJD
813/* .what is a character array with one character for each member of .data
814 * The character describes the function of the corresponding .data item:
af534a04 815 * a - AV for paren_name_list under DEBUGGING
0111c4fd 816 * f - start-class data for regstclass optimization
68e2671b 817 * l - start op for literal (?{EVAL}) item
d63c20f2 818 * L - start op for literal (?{EVAL}) item, with separate CV (qr//)
b30fcab9 819 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/
7a94a5fe
KW
820 * s - inversion list for Unicode-style character class, and the
821 * multicharacter strings resulting from casefolding the single-character
822 * entries in the character class
a3621e74 823 * t - trie struct
55eed653 824 * u - trie struct's widecharmap (a HV, so can't share, must dup)
2b8b4781 825 * also used for revcharmap and words under DEBUGGING
07be1b83 826 * T - aho-trie struct
81714fb9 827 * S - sv for named capture lookup
261faec3 828 * 20010712 mjd@plover.com
c3613ed7 829 * (Remember to update re_dup() and pregfree() if you add any items.)
261faec3 830 */
cad2e5aa
JH
831struct reg_data {
832 U32 count;
833 U8 *what;
834 void* data[1];
835};
836
a1cac82e
NC
837/* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses
838 anchored* and float* via array indexes 0 and 1. */
cad2e5aa 839#define anchored_substr substrs->data[0].substr
33b8afdf 840#define anchored_utf8 substrs->data[0].utf8_substr
cad2e5aa 841#define anchored_offset substrs->data[0].min_offset
1de06328
YO
842#define anchored_end_shift substrs->data[0].end_shift
843
cad2e5aa 844#define float_substr substrs->data[1].substr
33b8afdf 845#define float_utf8 substrs->data[1].utf8_substr
cad2e5aa
JH
846#define float_min_offset substrs->data[1].min_offset
847#define float_max_offset substrs->data[1].max_offset
1de06328
YO
848#define float_end_shift substrs->data[1].end_shift
849
cad2e5aa 850#define check_substr substrs->data[2].substr
33b8afdf 851#define check_utf8 substrs->data[2].utf8_substr
cad2e5aa
JH
852#define check_offset_min substrs->data[2].min_offset
853#define check_offset_max substrs->data[2].max_offset
1de06328 854#define check_end_shift substrs->data[2].end_shift
a3621e74 855
8d919b0a
FC
856#define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr)
857#define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8)
858#define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr)
859#define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8)
a3621e74
YO
860
861/* trie related stuff */
5d9a96ca 862
a3621e74
YO
863/* a transition record for the state machine. the
864 check field determines which state "owns" the
865 transition. the char the transition is for is
866 determined by offset from the owning states base
867 field. the next field determines which state
868 is to be transitioned to if any.
869*/
870struct _reg_trie_trans {
871 U32 next;
872 U32 check;
873};
874
875/* a transition list element for the list based representation */
876struct _reg_trie_trans_list_elem {
877 U16 forid;
878 U32 newstate;
879};
880typedef struct _reg_trie_trans_list_elem reg_trie_trans_le;
881
882/* a state for compressed nodes. base is an offset
883 into an array of reg_trie_trans array. If wordnum is
884 nonzero the state is accepting. if base is zero then
885 the state has no children (and will be accepting)
886*/
887struct _reg_trie_state {
888 U16 wordnum;
889 union {
890 U32 base;
891 reg_trie_trans_le* list;
892 } trans;
893};
894
2e64971a
DM
895/* info per word; indexed by wordnum */
896typedef struct {
897 U16 prev; /* previous word in acceptance chain; eg in
898 * zzz|abc|ab/ after matching the chars abc, the
899 * accepted word is #2, and the previous accepted
900 * word is #3 */
901 U32 len; /* how many chars long is this word? */
902 U32 accept; /* accept state for this word */
903} reg_trie_wordinfo;
a3621e74
YO
904
905
a3621e74
YO
906typedef struct _reg_trie_state reg_trie_state;
907typedef struct _reg_trie_trans reg_trie_trans;
908
909
910/* anything in here that needs to be freed later
23eab42c
NC
911 should be dealt with in pregfree.
912 refcount is first in both this and _reg_ac_data to allow a space
913 optimisation in Perl_regdupe. */
a3621e74 914struct _reg_trie_data {
23eab42c 915 U32 refcount; /* number of times this trie is referenced */
786e8c11
YO
916 U32 lasttrans; /* last valid transition element */
917 U16 *charmap; /* byte to charid lookup array */
786e8c11
YO
918 reg_trie_state *states; /* state data */
919 reg_trie_trans *trans; /* array of transition elements */
920 char *bitmap; /* stclass bitmap */
786e8c11
YO
921 U16 *jump; /* optional 1 indexed array of offsets before tail
922 for the node following a given word. */
2e64971a 923 reg_trie_wordinfo *wordinfo; /* array of info per word */
1b5c067c
NC
924 U16 uniquecharcount; /* unique chars in trie (width of trans table) */
925 U32 startstate; /* initial state - used for common prefix optimisation */
926 STRLEN minlen; /* minimum length of words in trie - build/opt only? */
927 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */
2e64971a 928 U32 prefixlen; /* #chars in common prefix */
1e2e3d02
YO
929 U32 statecount; /* Build only - number of states in the states array
930 (including the unused zero state) */
786e8c11 931 U32 wordcount; /* Build only */
a3621e74 932#ifdef DEBUGGING
786e8c11 933 STRLEN charcount; /* Build only */
a3621e74
YO
934#endif
935};
2b8b4781
NC
936/* There is one (3 under DEBUGGING) pointers that logically belong in this
937 structure, but are held outside as they need duplication on thread cloning,
938 whereas the rest of the structure can be read only:
939 HV *widecharmap; code points > 255 to charid
940#ifdef DEBUGGING
941 AV *words; Array of words contained in trie, for dumping
942 AV *revcharmap; Map of each charid back to its character representation
943#endif
944*/
945
946#define TRIE_WORDS_OFFSET 2
947
a3621e74
YO
948typedef struct _reg_trie_data reg_trie_data;
949
23eab42c
NC
950/* refcount is first in both this and _reg_trie_data to allow a space
951 optimisation in Perl_regdupe. */
07be1b83 952struct _reg_ac_data {
23eab42c 953 U32 refcount;
1b5c067c 954 U32 trie;
07be1b83
YO
955 U32 *fail;
956 reg_trie_state *states;
07be1b83
YO
957};
958typedef struct _reg_ac_data reg_ac_data;
959
486ec47a 960/* ANY_BIT doesn't use the structure, so we can borrow it here.
3dab1dad
YO
961 This is simpler than refactoring all of it as wed end up with
962 three different sets... */
963
964#define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap)
4cbce0a6 965#define TRIE_BITMAP_BYTE(p, c) BITMAP_BYTE(TRIE_BITMAP(p), c)
5dfe062f
YO
966#define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c))
967#define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c))
968#define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
3dab1dad 969
1de06328
YO
970#define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC)
971#define IS_TRIE_AC(op) ((op)>=AHOCORASICK)
972
973
0effac60 974#define BITMAP_BYTE(p, c) (( (U8*) p) [ ( ( (UV) (c)) >> 3) ] )
786e8c11 975#define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
3dab1dad 976
a3621e74
YO
977/* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */
978#define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1)
979#define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state))
980#define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state))
981
3dab1dad 982#ifdef DEBUGGING
3dab1dad 983#define TRIE_CHARCOUNT(trie) ((trie)->charcount)
3dab1dad 984#else
3dab1dad 985#define TRIE_CHARCOUNT(trie) (trie_charcount)
3dab1dad 986#endif
a3621e74 987
0111c4fd
RGS
988#define RE_TRIE_MAXBUF_INIT 65536
989#define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF"
a3621e74
YO
990#define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS"
991
6ef7fe53
KW
992#define RE_COMPILE_RECURSION_INIT 1000
993#define RE_COMPILE_RECURSION_LIMIT "\022E_COMPILE_RECURSION_LIMIT"
994
be8e71aa
YO
995/*
996
997RE_DEBUG_FLAGS is used to control what debug output is emitted
998its divided into three groups of options, some of which interact.
999The three groups are: Compile, Execute, Extra. There is room for a
1000further group, as currently only the low three bytes are used.
1001
1002 Compile Options:
07be1b83 1003
be8e71aa
YO
1004 PARSE
1005 PEEP
1006 TRIE
1007 PROGRAM
1008 OFFSETS
a3621e74 1009
be8e71aa 1010 Execute Options:
a3621e74 1011
be8e71aa
YO
1012 INTUIT
1013 MATCH
1014 TRIE
a3621e74 1015
be8e71aa 1016 Extra Options
a3621e74 1017
be8e71aa
YO
1018 TRIE
1019 OFFSETS
1020
1021If you modify any of these make sure you make corresponding changes to
1022re.pm, especially to the documentation.
a3621e74 1023
be8e71aa
YO
1024*/
1025
1026
1027/* Compile */
1028#define RE_DEBUG_COMPILE_MASK 0x0000FF
1029#define RE_DEBUG_COMPILE_PARSE 0x000001
1030#define RE_DEBUG_COMPILE_OPTIMISE 0x000002
1031#define RE_DEBUG_COMPILE_TRIE 0x000004
1032#define RE_DEBUG_COMPILE_DUMP 0x000008
f7819f85 1033#define RE_DEBUG_COMPILE_FLAGS 0x000010
d9a72fcc 1034#define RE_DEBUG_COMPILE_TEST 0x000020
be8e71aa
YO
1035
1036/* Execute */
1037#define RE_DEBUG_EXECUTE_MASK 0x00FF00
1038#define RE_DEBUG_EXECUTE_INTUIT 0x000100
1039#define RE_DEBUG_EXECUTE_MATCH 0x000200
1040#define RE_DEBUG_EXECUTE_TRIE 0x000400
1041
1042/* Extra */
fd41b2d1
KW
1043#define RE_DEBUG_EXTRA_MASK 0x1FF0000
1044#define RE_DEBUG_EXTRA_TRIE 0x0010000
1045#define RE_DEBUG_EXTRA_OFFSETS 0x0020000
1046#define RE_DEBUG_EXTRA_OFFDEBUG 0x0040000
1047#define RE_DEBUG_EXTRA_STATE 0x0080000
1048#define RE_DEBUG_EXTRA_OPTIMISE 0x0100000
1049#define RE_DEBUG_EXTRA_BUFFERS 0x0400000
1050#define RE_DEBUG_EXTRA_GPOS 0x0800000
1051#define RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE 0x1000000
24b23f37 1052/* combined */
fd41b2d1 1053#define RE_DEBUG_EXTRA_STACK 0x0280000
be8e71aa 1054
e68ec53f 1055#define RE_DEBUG_FLAG(x) (re_debug_flags & x)
be8e71aa
YO
1056/* Compile */
1057#define DEBUG_COMPILE_r(x) DEBUG_r( \
dafb2544 1058 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_MASK)) x )
be8e71aa 1059#define DEBUG_PARSE_r(x) DEBUG_r( \
dafb2544 1060 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_PARSE)) x )
be8e71aa 1061#define DEBUG_OPTIMISE_r(x) DEBUG_r( \
dafb2544 1062 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_OPTIMISE)) x )
be8e71aa 1063#define DEBUG_DUMP_r(x) DEBUG_r( \
dafb2544 1064 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_DUMP)) x )
be8e71aa 1065#define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \
dafb2544 1066 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_TRIE)) x )
f7819f85 1067#define DEBUG_FLAGS_r(x) DEBUG_r( \
dafb2544 1068 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_FLAGS)) x )
d9a72fcc 1069#define DEBUG_TEST_r(x) DEBUG_r( \
dafb2544 1070 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_TEST)) x )
be8e71aa
YO
1071/* Execute */
1072#define DEBUG_EXECUTE_r(x) DEBUG_r( \
dafb2544 1073 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXECUTE_MASK)) x )
be8e71aa 1074#define DEBUG_INTUIT_r(x) DEBUG_r( \
dafb2544 1075 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXECUTE_INTUIT)) x )
be8e71aa 1076#define DEBUG_MATCH_r(x) DEBUG_r( \
dafb2544 1077 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXECUTE_MATCH)) x )
be8e71aa 1078#define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \
dafb2544 1079 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXECUTE_TRIE)) x )
be8e71aa
YO
1080
1081/* Extra */
1082#define DEBUG_EXTRA_r(x) DEBUG_r( \
dafb2544 1083 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_MASK)) x )
a5ca303d 1084#define DEBUG_OFFSETS_r(x) DEBUG_r( \
dafb2544 1085 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_OFFSETS)) x )
ab3bbdeb 1086#define DEBUG_STATE_r(x) DEBUG_r( \
dafb2544 1087 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_STATE)) x )
24b23f37 1088#define DEBUG_STACK_r(x) DEBUG_r( \
dafb2544 1089 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_STACK)) x )
e7707071 1090#define DEBUG_BUFFERS_r(x) DEBUG_r( \
dafb2544 1091 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_BUFFERS)) x )
e7707071 1092
a5ca303d 1093#define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \
dafb2544
KW
1094 if (DEBUG_v_TEST || ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \
1095 (re_debug_flags & (RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE)))) x )
07be1b83 1096#define MJD_OFFSET_DEBUG(x) DEBUG_r( \
dafb2544 1097 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_OFFDEBUG)) \
be8e71aa
YO
1098 Perl_warn_nocontext x )
1099#define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \
dafb2544 1100 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_TRIE)) x )
be8e71aa 1101#define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \
dafb2544 1102 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_TRIE)) x )
be8e71aa
YO
1103
1104#define DEBUG_TRIE_r(x) DEBUG_r( \
dafb2544
KW
1105 if (DEBUG_v_TEST || (re_debug_flags & (RE_DEBUG_COMPILE_TRIE \
1106 | RE_DEBUG_EXECUTE_TRIE ))) x )
2c296965 1107#define DEBUG_GPOS_r(x) DEBUG_r( \
dafb2544 1108 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_GPOS)) x )
be8e71aa 1109
fd41b2d1
KW
1110#define DEBUG_DUMP_PRE_OPTIMIZE_r(x) DEBUG_r( \
1111 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE)) x )
1112
be8e71aa 1113/* initialization */
d2d8c711 1114/* get_sv() can return NULL during global destruction. */
e68ec53f
YO
1115#define GET_RE_DEBUG_FLAGS DEBUG_r({ \
1116 SV * re_debug_flags_sv = NULL; \
243a3b9c 1117 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \
e68ec53f 1118 if (re_debug_flags_sv) { \
8e9a8a48 1119 if (!SvIOK(re_debug_flags_sv)) \
e68ec53f 1120 sv_setuv(re_debug_flags_sv, RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \
8e9a8a48 1121 re_debug_flags=SvIV(re_debug_flags_sv); \
e68ec53f
YO
1122 }\
1123})
a3621e74
YO
1124
1125#ifdef DEBUGGING
ab3bbdeb 1126
8162b70e 1127#define GET_RE_DEBUG_FLAGS_DECL volatile IV re_debug_flags = 0; \
cb65b687 1128 PERL_UNUSED_VAR(re_debug_flags); GET_RE_DEBUG_FLAGS;
ab3bbdeb 1129
eecd4d11
KW
1130#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \
1131 const char * const rpv = \
1132 pv_pretty((dsv), (pv), (l), (m), \
1133 PL_colors[(c1)],PL_colors[(c2)], \
c89df6cf 1134 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \
0df25f3d 1135 const int rlen = SvCUR(dsv)
ab3bbdeb 1136
8ec77c3c 1137/* This is currently unsed in the core */
eecd4d11
KW
1138#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \
1139 const char * const rpv = \
1140 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \
1141 PL_colors[(c1)],PL_colors[(c2)], \
c89df6cf 1142 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) )
ab3bbdeb 1143
eecd4d11
KW
1144#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \
1145 const char * const rpv = \
1146 pv_pretty((dsv), (pv), (l), (m), \
1147 PL_colors[0], PL_colors[1], \
c89df6cf 1148 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \
ab3bbdeb 1149 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \
0a0e9afc 1150 )
ab3bbdeb
YO
1151
1152#define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0))
1153#define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "")
1154
1155#else /* if not DEBUGGING */
1156
a3621e74 1157#define GET_RE_DEBUG_FLAGS_DECL
ab3bbdeb
YO
1158#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2)
1159#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m)
1160#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m)
1161#define RE_SV_DUMPLEN(ItEm)
1162#define RE_SV_TAIL(ItEm)
1163
1164#endif /* DEBUG RELATED DEFINES */
a3621e74 1165
0bc0e44f 1166#define FIRST_NON_ASCII_DECIMAL_DIGIT 0x660 /* ARABIC_INDIC_DIGIT_ZERO */
2ca74f27 1167
64935bc6
KW
1168typedef enum {
1169 TRADITIONAL_BOUND = _CC_WORDCHAR,
ae3bb8ea 1170 GCB_BOUND,
6b659339 1171 LB_BOUND,
06ae2722 1172 SB_BOUND,
ae3bb8ea 1173 WB_BOUND
64935bc6
KW
1174} bound_type;
1175
13fcf652 1176/* This unpacks the FLAGS field of ANYOF[HR]x nodes. The value it contains
3146c00a
KW
1177 * gives the strict lower bound for the UTF-8 start byte of any code point
1178 * matchable by the node, and a loose upper bound as well.
1179 *
1180 * The low bound is stored in the upper 6 bits, plus 0xC0.
1181 * The loose upper bound is determined from the lowest 2 bits and the low bound
1182 * (called x) as follows:
1183 *
1184 * 11 The upper limit of the range can be as much as (EF - x) / 8
1185 * 10 The upper limit of the range can be as much as (EF - x) / 4
1186 * 01 The upper limit of the range can be as much as (EF - x) / 2
1187 * 00 The upper limit of the range can be as much as EF
1188 *
0bf4efd7
KW
1189 * For motivation of this design, see commit message in
1190 * 3146c00a633e9cbed741e10146662fbcedfdb8d3 */
3146c00a
KW
1191#ifdef EBCDIC
1192# define MAX_ANYOF_HRx_BYTE 0xF4
1193#else
1194# define MAX_ANYOF_HRx_BYTE 0xEF
1195#endif
1196#define LOWEST_ANYOF_HRx_BYTE(b) (((b) >> 2) + 0xC0)
1197#define HIGHEST_ANYOF_HRx_BYTE(b) \
1198 (LOWEST_ANYOF_HRx_BYTE(b) \
1199 + ((MAX_ANYOF_HRx_BYTE - LOWEST_ANYOF_HRx_BYTE(b)) >> ((b) & 3)))
1200
3dd7db29
JK
1201#endif /* PERL_REGCOMP_H_ */
1202
e9a8c099 1203/*
14d04a33 1204 * ex: set ts=8 sts=4 sw=4 et:
e9a8c099 1205 */