This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regexec.c: Fix Debug statement
[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
86451f01
KW
184/* Node whose argument is 'void *', a pointer to void. This needs to be used
185 * very carefully in situations where pointers won't become invalid because of,
186 * say re-mallocs */
187struct regnode_p {
188 U8 flags;
189 U8 type;
190 U16 next_off;
191 void * arg1;
192};
193
6bda09f9
YO
194/* Similar to a regnode_1 but with an extra signed argument */
195struct regnode_2L {
196 U8 flags;
197 U8 type;
198 U16 next_off;
199 U32 arg1;
200 I32 arg2;
201};
202
203/* 'Two field' -- Two 16 bit unsigned args */
c277df42
IZ
204struct regnode_2 {
205 U8 flags;
206 U8 type;
207 U16 next_off;
208 U16 arg1;
209 U16 arg2;
210};
211
dcb20b36 212#define ANYOF_BITMAP_SIZE (NUM_ANYOF_CODE_POINTS / 8) /* 8 bits/Byte */
936ed897 213
1ee208c4
KW
214/* Note that these form structs which are supersets of the next smaller one, by
215 * appending fields. Alignment problems can occur if one of those optional
216 * fields requires stricter alignment than the base struct. And formal
217 * parameters that can really be two or more of the structs should be
218 * declared as the smallest one it could be. See commit message for
219 * 7dcac5f6a5195002b55c935ee1d67f67e1df280b. Regnode allocation is done
220 * without regard to alignment, and changing it to would also require changing
221 * the code that inserts and deletes regnodes. The basic single-argument
222 * regnode has a U32, which is what reganode() allocates as a unit. Therefore
223 * no field can require stricter alignment than U32. */
224
786e8c11 225/* also used by trie */
936ed897
IZ
226struct regnode_charclass {
227 U8 flags;
228 U8 type;
229 U16 next_off;
9da99c5e 230 U32 arg1; /* set by set_ANYOF_arg() */
19103b91 231 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */
936ed897
IZ
232};
233
bf3f2d85 234/* has runtime (locale) \d, \w, ..., [:posix:] classes */
ea461dd6 235struct regnode_charclass_posixl {
93e92956 236 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
936ed897
IZ
237 U8 type;
238 U16 next_off;
7e817188 239 U32 arg1;
03fa83ba 240 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
9cf78fa0 241 U32 classflags; /* and run-time */
936ed897
IZ
242};
243
85c8e306
KW
244/* A synthetic start class (SSC); is a regnode_charclass_posixl_fold, plus an
245 * extra SV*, used only during its construction and which is not used by
246 * regexec.c. Note that the 'next_off' field is unused, as the SSC stands
247 * alone, so there is never a next node. Also, there is no alignment issue,
b8fda935 248 * because these are declared or allocated as a complete unit so the compiler
85c8e306
KW
249 * takes care of alignment. This is unlike the other regnodes which are
250 * allocated in terms of multiples of a single-argument regnode. SSC nodes can
251 * have a pointer field because there is no alignment issue, and because it is
252 * set to NULL after construction, before any cloning of the pattern */
b8f7bb16 253struct regnode_ssc {
93e92956 254 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
b8f7bb16
KW
255 U8 type;
256 U16 next_off;
7e817188 257 U32 arg1;
03fa83ba 258 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
631e9161 259 U32 classflags; /* ... and run-time */
85c8e306
KW
260
261 /* Auxiliary, only used during construction; NULL afterwards: list of code
262 * points matched */
263 SV* invlist;
b8f7bb16
KW
264};
265
71068078
KW
266/* We take advantage of 'next_off' not otherwise being used in the SSC by
267 * actually using it: by setting it to 1. This allows us to test and
268 * distinguish between an SSC and other ANYOF node types, as 'next_off' cannot
269 * otherwise be 1, because it is the offset to the next regnode expressed in
270 * units of regnodes. Since an ANYOF node contains extra fields, it adds up
271 * to 12 regnode units on 32-bit systems, (hence the minimum this can be (if
272 * not 0) is 11 there. Even if things get tightly packed on a 64-bit system,
273 * it still would be more than 1. */
274#define set_ANYOF_SYNTHETIC(n) STMT_START{ OP(n) = ANYOF; \
275 NEXT_OFF(n) = 1; \
276 } STMT_END
c1071334 277#define is_ANYOF_SYNTHETIC(n) (PL_regkind[OP(n)] == ANYOF && NEXT_OFF(n) == 1)
71068078 278
9c7e81e8
AD
279/* XXX fix this description.
280 Impose a limit of REG_INFTY on various pattern matching operations
281 to limit stack growth and to avoid "infinite" recursions.
282*/
4288c5b9
KW
283/* The default size for REG_INFTY is U16_MAX, which is the same as
284 USHORT_MAX (see perl.h). Unfortunately U16 isn't necessarily 16 bits
285 (see handy.h). On the Cray C90, sizeof(short)==4 and hence U16_MAX is
286 ((1<<32)-1), while on the Cray T90, sizeof(short)==8 and U16_MAX is
287 ((1<<64)-1). To limit stack growth to reasonable sizes, supply a
9c7e81e8
AD
288 smaller default.
289 --Andy Dougherty 11 June 1998
290*/
291#if SHORTSIZE > 2
292# ifndef REG_INFTY
4288c5b9 293# define REG_INFTY ((1<<16)-1)
9c7e81e8
AD
294# endif
295#endif
296
297#ifndef REG_INFTY
4288c5b9 298# define REG_INFTY U16_MAX
83cfe48c 299#endif
a687059c 300
e91177ed
IZ
301#define ARG_VALUE(arg) (arg)
302#define ARG__SET(arg,val) ((arg) = (val))
c277df42 303
e8e3b0fb
JH
304#undef ARG
305#undef ARG1
306#undef ARG2
307
c277df42 308#define ARG(p) ARG_VALUE(ARG_LOC(p))
86451f01 309#define ARGp(p) ARG_VALUE(ARGp_LOC(p))
c277df42
IZ
310#define ARG1(p) ARG_VALUE(ARG1_LOC(p))
311#define ARG2(p) ARG_VALUE(ARG2_LOC(p))
6bda09f9 312#define ARG2L(p) ARG_VALUE(ARG2L_LOC(p))
786e8c11 313
c277df42 314#define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val))
86451f01 315#define ARGp_SET(p, val) ARG__SET(ARGp_LOC(p), (val))
c277df42
IZ
316#define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val))
317#define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val))
6bda09f9 318#define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val))
c277df42 319
e8e3b0fb
JH
320#undef NEXT_OFF
321#undef NODE_ALIGN
322
075abff3
AL
323#define NEXT_OFF(p) ((p)->next_off)
324#define NODE_ALIGN(node)
aa48e906
YO
325/* the following define was set to 0xde in 075abff3
326 * as part of some linting logic. I have set it to 0
327 * as otherwise in every place where we /might/ set flags
328 * we have to set it 0 explicitly, which duplicates
329 * assignments and IMO adds an unacceptable level of
330 * surprise to working in the regex engine. If this
331 * is changed from 0 then at the very least make sure
332 * that SBOL for /^/ sets the flags to 0 explicitly.
333 * -- Yves */
334#define NODE_ALIGN_FILL(node) ((node)->flags = 0)
a687059c 335
c277df42
IZ
336#define SIZE_ALIGN NODE_ALIGN
337
e8e3b0fb
JH
338#undef OP
339#undef OPERAND
e8e3b0fb
JH
340#undef STRING
341
e91177ed 342#define OP(p) ((p)->type)
b988e673
KW
343#define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \
344 regnode types. For some, it's the \
345 character set of the regnode */
3f2416ae 346#define STR_LENs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \
5cd61b66 347 ((struct regnode_string *)p)->str_len)
3f2416ae 348#define STRINGs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \
5cd61b66 349 ((struct regnode_string *)p)->string)
ae06e581
KW
350#define OPERANDs(p) STRINGs(p)
351
352/* Long strings. Currently limited to length 18 bits, which handles a 262000
353 * byte string. The limiting factor is the 16 bit 'next_off' field, which
354 * points to the next regnode, so the furthest away it can be is 2**16. On
355 * most architectures, regnodes are 2**2 bytes long, so that yields 2**18
356 * bytes. Should a longer string be desired, we could increase it to 26 bits
357 * fairly easily, by changing this node to have longj type which causes the ARG
358 * field to be used for the link to the next regnode (although code would have
359 * to be changed to account for this), and then use a combination of the flags
360 * and next_off fields for the length. To get 34 bit length, also change the
361 * node to be an ARG2L, using the second 32 bit field for the length, and not
362 * using the flags nor next_off fields at all. One could have an llstring node
363 * and even an lllstring type. */
3f2416ae 364#define STR_LENl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 365 (((struct regnode_lstring *)p)->str_len))
3f2416ae 366#define STRINGl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 367 (((struct regnode_lstring *)p)->string))
ae06e581
KW
368#define OPERANDl(p) STRINGl(p)
369
3f2416ae 370#define STR_LEN(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 371 ? STR_LENl(p) : STR_LENs(p))
3f2416ae 372#define STRING(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 373 ? STRINGl(p) : STRINGs(p))
c5184715
KW
374#define OPERAND(p) STRING(p)
375
c45abc0a
KW
376/* The number of (smallest) regnode equivalents that a string of length l bytes
377 * occupies */
59e54936 378#define STR_SZ(l) (((l) + sizeof(regnode) - 1) / sizeof(regnode))
c45abc0a
KW
379
380/* The number of (smallest) regnode equivalents that the EXACTISH node 'p'
381 * occupies */
ae06e581 382#define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p)) + 1 + regarglen[(p)->type])
c45abc0a 383
c7fd9c67 384#define setSTR_LEN(p,v) \
ae06e581 385 STMT_START{ \
3f2416ae 386 if (OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
ae06e581
KW
387 ((struct regnode_lstring *)(p))->str_len = (v); \
388 else \
389 ((struct regnode_string *)(p))->str_len = (v); \
390 } STMT_END
cd439c50 391
13fcf652
KW
392#define ANYOFR_BASE_BITS 20
393#define ANYOFRbase(p) (ARG(p) & ((1 << ANYOFR_BASE_BITS) - 1))
394#define ANYOFRdelta(p) (ARG(p) >> ANYOFR_BASE_BITS)
395
e8e3b0fb
JH
396#undef NODE_ALIGN
397#undef ARG_LOC
398#undef NEXTOPER
399#undef PREVOPER
400
e91177ed
IZ
401#define NODE_ALIGN(node)
402#define ARG_LOC(p) (((struct regnode_1 *)p)->arg1)
86451f01 403#define ARGp_LOC(p) (((struct regnode_p *)p)->arg1)
e91177ed
IZ
404#define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1)
405#define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2)
6bda09f9 406#define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2)
786e8c11 407
e91177ed
IZ
408#define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */
409#define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2)
410
c277df42
IZ
411#define NEXTOPER(p) ((p) + NODE_STEP_REGNODE)
412#define PREVOPER(p) ((p) - NODE_STEP_REGNODE)
413
f55b7b26
KW
414#define FILL_NODE(offset, op) \
415 STMT_START { \
416 OP(REGNODE_p(offset)) = op; \
417 NEXT_OFF(REGNODE_p(offset)) = 0; \
d2a88850 418 } STMT_END
f55b7b26
KW
419#define FILL_ADVANCE_NODE(offset, op) \
420 STMT_START { \
421 FILL_NODE(offset, op); \
422 (offset)++; \
d2a88850 423 } STMT_END
f55b7b26
KW
424#define FILL_ADVANCE_NODE_ARG(offset, op, arg) \
425 STMT_START { \
426 ARG_SET(REGNODE_p(offset), arg); \
427 FILL_ADVANCE_NODE(offset, op); \
428 /* This is used generically for other operations \
429 * that have a longer argument */ \
430 (offset) += regarglen[op]; \
86451f01
KW
431 } STMT_END
432#define FILL_ADVANCE_NODE_ARGp(offset, op, arg) \
433 STMT_START { \
434 ARGp_SET(REGNODE_p(offset), arg); \
435 FILL_ADVANCE_NODE(offset, op); \
436 (offset) += regarglen[op]; \
93539530 437 } STMT_END
f55b7b26
KW
438#define FILL_ADVANCE_NODE_2L_ARG(offset, op, arg1, arg2) \
439 STMT_START { \
440 ARG_SET(REGNODE_p(offset), arg1); \
441 ARG2L_SET(REGNODE_p(offset), arg2); \
442 FILL_ADVANCE_NODE(offset, op); \
443 (offset) += 2; \
93539530 444 } STMT_END
a687059c 445
22c35a8c 446#define REG_MAGIC 0234
a687059c 447
fe8d1b7c
KW
448/* An ANYOF node is basically a bitmap with the index being a code point. If
449 * the bit for that code point is 1, the code point matches; if 0, it doesn't
450 * match (complemented if inverted). There is an additional mechanism to deal
451 * with cases where the bitmap is insufficient in and of itself. This #define
452 * indicates if the bitmap does fully represent what this ANYOF node can match.
453 * The ARG is set to this special value (since 0, 1, ... are legal, but will
454 * never reach this high). */
93e92956 455#define ANYOF_ONLY_HAS_BITMAP ((U32) -1)
137165a6 456
93539530 457/* When the bitmap isn't completely sufficient for handling the ANYOF node,
fe8d1b7c
KW
458 * flags (in node->flags of the ANYOF node) get set to indicate this. These
459 * are perennially in short supply. Beyond several cases where warnings need
460 * to be raised under certain circumstances, currently, there are six cases
461 * where the bitmap alone isn't sufficient. We could use six flags to
462 * represent the 6 cases, but to save flags bits, we play some games. The
463 * cases are:
d2f465e9 464 *
fe8d1b7c
KW
465 * 1) The bitmap has a compiled-in very finite size. So something else needs
466 * to be used to specify if a code point that is too large for the bitmap
7a94a5fe 467 * actually matches. The mechanism currently is an inversion
fe8d1b7c
KW
468 * list. ANYOF_ONLY_HAS_BITMAP, described above, being TRUE indicates
469 * there are no matches of too-large code points. But if it is FALSE,
470 * then almost certainly there are matches too large for the bitmap. (The
471 * other cases, described below, either imply this one or are extremely
472 * rare in practice.) So we can just assume that a too-large code point
473 * will need something beyond the bitmap if ANYOF_ONLY_HAS_BITMAP is
474 * FALSE, instead of having a separate flag for this.
475 * 2) A subset of item 1) is if all possible code points outside the bitmap
476 * match. This is a common occurrence when the class is complemented,
477 * like /[^ij]/. Therefore a bit is reserved to indicate this,
42d0f81a 478 * rather than having an inversion list created,
d1c40ef5 479 * ANYOF_MATCHES_ALL_ABOVE_BITMAP.
fe8d1b7c
KW
480 * 3) Under /d rules, it can happen that code points that are in the upper
481 * latin1 range (\x80-\xFF or their equivalents on EBCDIC platforms) match
482 * only if the runtime target string being matched against is UTF-8. For
483 * example /[\w[:punct:]]/d. This happens only for posix classes (with a
d1c40ef5
KW
484 * couple of exceptions, like \d where it doesn't happen), and all such
485 * ones also have above-bitmap matches. Thus, 3) implies 1) as well.
486 * Note that /d rules are no longer encouraged; 'use 5.14' or higher
487 * deselects them. But a flag is required so that they can be properly
488 * handled. But it can be a shared flag: see 5) below.
489 * 4) Also under /d rules, something like /[\Wfoo]/ will match everything in
fe8d1b7c 490 * the \x80-\xFF range, unless the string being matched against is UTF-8.
7a94a5fe
KW
491 * An inversion list could be created for this case, but this is
492 * relatively common, and it turns out that it's all or nothing: if any
493 * one of these code points matches, they all do. Hence a single bit
494 * suffices. We use a shared flag that doesn't take up space by itself:
495 * ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER. This
496 * also implies 1), with one exception: [:^cntrl:].
fe8d1b7c
KW
497 * 5) A user-defined \p{} property may not have been defined by the time the
498 * regex is compiled. In this case, we don't know until runtime what it
499 * will match, so we have to assume it could match anything, including
500 * code points that ordinarily would be in the bitmap. A flag bit is
501 * necessary to indicate this , though it can be shared with the item 3)
502 * flag, as that only occurs under /d, and this only occurs under non-d.
503 * This case is quite uncommon in the field, and the /(?[ ...])/ construct
504 * is a better way to accomplish what this feature does. This case also
505 * implies 1).
506 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP
e2f5e63d 507 * is the shared flag.
fe8d1b7c
KW
508 * 6) /[foo]/il may have folds that are only valid if the runtime locale is a
509 * UTF-8 one. These are quite rare, so it would be good to avoid the
510 * expense of looking for them. But /l matching is slow anyway, and we've
e2f5e63d 511 * traditionally not worried too much about its performance. And this
037715a6 512 * condition requires the ANYOFL_FOLD flag to be set, so testing for
fe8d1b7c 513 * that flag would be sufficient to rule out most cases of this. So it is
d1c40ef5
KW
514 * unclear if this should have a flag or not. But, this flag can be
515 * shared with another, so it doesn't occupy extra space.
d2f465e9 516 *
d1c40ef5 517 * At the moment, there is one spare bit, but this could be increased by
93539530 518 * various tricks:
e2f5e63d 519 *
93539530 520 * If just one more bit is needed, as of this writing it seems to khw that the
d1c40ef5
KW
521 * best choice would be to make ANYOF_MATCHES_ALL_ABOVE_BITMAP not a flag, but
522 * something like
d2f465e9 523 *
d1c40ef5 524 * #define ANYOF_MATCHES_ALL_ABOVE_BITMAP ((U32) -2)
fe8d1b7c 525 *
d1c40ef5
KW
526 * and access it through the ARG like ANYOF_ONLY_HAS_BITMAP is. This flag is
527 * used by all ANYOF node types, and it could be used to avoid calling the
528 * handler function, as the macro REGINCLASS in regexec.c does now for other
529 * cases.
530 *
46fc0c43
KW
531 * Another possibility is based on the fact that ANYOF_MATCHES_POSIXL is
532 * redundant with the node type ANYOFPOSIXL. That flag could be removed, but
533 * at the expense of extra code in regexec.c. The flag has been retained
534 * because it allows us to see if we need to call reginsert, or just use the
535 * bitmap in one test.
fe8d1b7c 536 *
46fc0c43
KW
537 * If this is done, an extension would be to make all ANYOFL nodes contain the
538 * extra 32 bits that ANYOFPOSIXL ones do. The posix flags only occupy 30
539 * bits, so the ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD flags
540 * and ANYOFL_FOLD could be moved to that extra space, but it would mean extra
541 * instructions, as there are currently places in the code that assume those
542 * two bits are zero.
d1c40ef5
KW
543 *
544 * All told, 5 bits could be available for other uses if all of the above were
545 * done.
a4525e78 546 *
d1c40ef5 547 * Some flags are not used in synthetic start class (SSC) nodes, so could be
93e92956
KW
548 * shared should new flags be needed for SSCs, like SSC_MATCHES_EMPTY_STRING
549 * now. */
b8c5462f 550
d2f465e9
KW
551/* If this is set, the result of the match should be complemented. regexec.c
552 * is expecting this to be in the low bit. Never in an SSC */
93e92956 553#define ANYOF_INVERT 0x01
1984f6a0 554
a0dd4231 555/* For the SSC node only, which cannot be inverted, so is shared with that bit.
93e92956
KW
556 * This is used only during regex compilation. */
557#define SSC_MATCHES_EMPTY_STRING ANYOF_INVERT
a0dd4231 558
509d4de0
KW
559/* Set if this is a regnode_charclass_posixl vs a regnode_charclass. This
560 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale
561 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved
562 * at compile-time. Only set under /l; can be in SSC */
563#define ANYOF_MATCHES_POSIXL 0x02
b63f73da
KW
564
565/* The fold is calculated and stored in the bitmap where possible at compile
566 * time. However under locale, the actual folding varies depending on
567 * what the locale is at the time of execution, so it has to be deferred until
d2f465e9 568 * then. Only set under /l; never in an SSC */
037715a6 569#define ANYOFL_FOLD 0x04
b63f73da 570
d1c40ef5 571/* Shared bit set only with ANYOFL and SSC nodes:
269bb65b
KW
572 * If ANYOFL_FOLD is set, this flag indicates there are potential matches
573 * valid only if the locale is a UTF-8 one.
574 * If ANYOFL_FOLD is NOT set, this flag means to warn if the runtime locale
d1c40ef5
KW
575 * isn't a UTF-8 one (and the generated node assumes a UTF-8 locale).
576 * None of INVERT, POSIXL,
577 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP
578 * can be set. */
579#define ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD 0x08
580
581/* Convenience macros for teasing apart the meanings when reading the above bit
582 * */
583#define ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags) \
584 ((flags & ( ANYOFL_FOLD /* Both bits are set */ \
585 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \
586 == ( ANYOFL_FOLD \
587 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD))
588
589#define ANYOFL_UTF8_LOCALE_REQD(flags) \
590 ((flags & ( ANYOFL_FOLD /* Only REQD bit is set */ \
591 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \
592 == ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)
593
1744d18d 594/* Spare: Be sure to change ANYOF_FLAGS_ALL if this gets used 0x10 */
c8d3cd88 595
509d4de0
KW
596/* If set, the node matches every code point NUM_ANYOF_CODE_POINTS and above.
597 * Can be in an SSC */
b3b1cf17 598#define ANYOF_MATCHES_ALL_ABOVE_BITMAP 0x20
ef87b810 599
108316fb
KW
600/* Shared bit:
601 * Under /d it means the ANYOFD node matches more things if the target
602 * string is encoded in UTF-8; any such things will be non-ASCII,
7a94a5fe
KW
603 * characters that are < 256, and can be accessed via the inversion
604 * list.
108316fb
KW
605 * When not under /d, it means the ANYOF node contains a user-defined
606 * property that wasn't yet defined at the time the regex was compiled,
7a94a5fe 607 * and so must be looked up at runtime, by creating an inversion list.
108316fb
KW
608 * (These uses are mutually exclusive because a user-defined property is
609 * specified by \p{}, and \p{} implies /u which deselects /d). The long macro
610 * name is to make sure that you are cautioned about its shared nature. Only
611 * the non-/d meaning can be in an SSC */
612#define ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 0x40
f240c685
KW
613
614/* Shared bit:
d2f465e9 615 * Under /d it means the ANYOFD node matches all non-ASCII Latin1
f240c685
KW
616 * characters when the target string is not in utf8.
617 * When not under /d, it means the ANYOF node should raise a warning if
618 * matching against an above-Unicode code point.
619 * (These uses are mutually exclusive because the warning requires a \p{}, and
620 * \p{} implies /u which deselects /d). An SSC node only has this bit set if
621 * what is meant is the warning. The long macro name is to make sure that you
622 * are cautioned about its shared nature */
623#define ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER 0x80
7bc66b18 624
b271ede1 625#define ANYOF_FLAGS_ALL (0xff & ~0x10)
653099ff 626
037715a6 627#define ANYOF_LOCALE_FLAGS (ANYOFL_FOLD | ANYOF_MATCHES_POSIXL)
ace6b0e4 628
eff8b7dc
KW
629/* These are the flags that apply to both regular ANYOF nodes and synthetic
630 * start class nodes during construction of the SSC. During finalization of
d2f465e9 631 * the SSC, other of the flags may get added to it */
d1c40ef5 632#define ANYOF_COMMON_FLAGS 0
eff8b7dc 633
936ed897 634/* Character classes for node->classflags of ANYOF */
653099ff 635/* Should be synchronized with a table in regprop() */
575a8a29 636/* 2n should be the normal one, paired with its complement at 2n+1 */
b8c5462f 637
08ada29e
KW
638#define ANYOF_ALPHA ((_CC_ALPHA) * 2)
639#define ANYOF_NALPHA ((ANYOF_ALPHA) + 1)
e8d596e0
KW
640#define ANYOF_ALPHANUMERIC ((_CC_ALPHANUMERIC) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */
641#define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1)
08ada29e
KW
642#define ANYOF_ASCII ((_CC_ASCII) * 2)
643#define ANYOF_NASCII ((ANYOF_ASCII) + 1)
e8d596e0
KW
644#define ANYOF_BLANK ((_CC_BLANK) * 2) /* GNU extension: space and tab: non-vertical space */
645#define ANYOF_NBLANK ((ANYOF_BLANK) + 1)
b0d691b2
KW
646#define ANYOF_CASED ((_CC_CASED) * 2) /* Pseudo class for [:lower:] or
647 [:upper:] under /i */
648#define ANYOF_NCASED ((ANYOF_CASED) + 1)
08ada29e
KW
649#define ANYOF_CNTRL ((_CC_CNTRL) * 2)
650#define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1)
e8d596e0
KW
651#define ANYOF_DIGIT ((_CC_DIGIT) * 2) /* \d */
652#define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1)
08ada29e
KW
653#define ANYOF_GRAPH ((_CC_GRAPH) * 2)
654#define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1)
655#define ANYOF_LOWER ((_CC_LOWER) * 2)
656#define ANYOF_NLOWER ((ANYOF_LOWER) + 1)
657#define ANYOF_PRINT ((_CC_PRINT) * 2)
658#define ANYOF_NPRINT ((ANYOF_PRINT) + 1)
659#define ANYOF_PUNCT ((_CC_PUNCT) * 2)
660#define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1)
e8d596e0
KW
661#define ANYOF_SPACE ((_CC_SPACE) * 2) /* \s */
662#define ANYOF_NSPACE ((ANYOF_SPACE) + 1)
08ada29e
KW
663#define ANYOF_UPPER ((_CC_UPPER) * 2)
664#define ANYOF_NUPPER ((ANYOF_UPPER) + 1)
e8d596e0
KW
665#define ANYOF_WORDCHAR ((_CC_WORDCHAR) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */
666#define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1)
08ada29e
KW
667#define ANYOF_XDIGIT ((_CC_XDIGIT) * 2)
668#define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1)
b8c5462f 669
87cad4c0 670/* pseudo classes below this, not stored in the class bitmap, but used as flags
e1d1eefb
YO
671 during compilation of char classes */
672
b2ef2e34
KW
673#define ANYOF_VERTWS ((_CC_VERTSPACE) * 2)
674#define ANYOF_NVERTWS ((ANYOF_VERTWS)+1)
e1d1eefb 675
b2ef2e34
KW
676/* It is best if this is the last one, as all above it are stored as bits in a
677 * bitmap, and it isn't part of that bitmap */
678#if _CC_VERTSPACE != _HIGHEST_REGCOMP_DOT_H_SYNC
679# error Problem with handy.h _HIGHEST_REGCOMP_DOT_H_SYNC #define
680#endif
681
8efd3f97
KW
682#define ANYOF_POSIXL_MAX (ANYOF_VERTWS) /* So upper loop limit is written:
683 * '< ANYOF_MAX'
684 * Hence doesn't include VERTWS, as that
685 * is a pseudo class */
686#define ANYOF_MAX ANYOF_POSIXL_MAX
687
688#if (ANYOF_POSIXL_MAX > 32) /* Must fit in 32-bit word */
b2ef2e34 689# error Problem with handy.h _CC_foo #defines
a0947d7b
KW
690#endif
691
8efd3f97
KW
692#define ANYOF_HORIZWS ((ANYOF_POSIXL_MAX)+2) /* = (ANYOF_NVERTWS + 1) */
693#define ANYOF_NHORIZWS ((ANYOF_POSIXL_MAX)+3)
87cad4c0 694
8efd3f97
KW
695#define ANYOF_UNIPROP ((ANYOF_POSIXL_MAX)+4) /* Used to indicate a Unicode
696 property: \p{} or \P{} */
f39a4dc2 697
b8c5462f
JH
698/* Backward source code compatibility. */
699
700#define ANYOF_ALNUML ANYOF_ALNUM
701#define ANYOF_NALNUML ANYOF_NALNUM
702#define ANYOF_SPACEL ANYOF_SPACE
703#define ANYOF_NSPACEL ANYOF_NSPACE
8ebfc29a
KW
704#define ANYOF_ALNUM ANYOF_WORDCHAR
705#define ANYOF_NALNUM ANYOF_NWORDCHAR
b8c5462f
JH
706
707/* Utility macros for the bitmap and classes of ANYOF */
708
936ed897 709#define ANYOF_FLAGS(p) ((p)->flags)
b8c5462f 710
bc51fd78 711#define ANYOF_BIT(c) (1U << ((c) & 7))
b8c5462f 712
14ed74f6
KW
713#define POSIXL_SET(field, c) ((field) |= (1U << (c)))
714#define ANYOF_POSIXL_SET(p, c) POSIXL_SET(((regnode_charclass_posixl*) (p))->classflags, (c))
8efd3f97 715
14ed74f6
KW
716#define POSIXL_CLEAR(field, c) ((field) &= ~ (1U <<(c)))
717#define ANYOF_POSIXL_CLEAR(p, c) POSIXL_CLEAR(((regnode_charclass_posixl*) (p))->classflags, (c))
8efd3f97 718
14ed74f6
KW
719#define POSIXL_TEST(field, c) ((field) & (1U << (c)))
720#define ANYOF_POSIXL_TEST(p, c) POSIXL_TEST(((regnode_charclass_posixl*) (p))->classflags, (c))
b8c5462f 721
14ed74f6
KW
722#define POSIXL_ZERO(field) STMT_START { (field) = 0; } STMT_END
723#define ANYOF_POSIXL_ZERO(ret) POSIXL_ZERO(((regnode_charclass_posixl*) (ret))->classflags)
724
725#define ANYOF_POSIXL_SET_TO_BITMAP(p, bits) \
726 STMT_START { \
727 ((regnode_charclass_posixl*) (p))->classflags = (bits); \
728 } STMT_END
9cf78fa0
KW
729
730/* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */
d1d040e5 731#define ANYOF_POSIXL_SETALL(ret) STMT_START { ((regnode_charclass_posixl*) (ret))->classflags = ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1; } STMT_END
8efd3f97 732#define ANYOF_CLASS_SETALL(ret) ANYOF_POSIXL_SETALL(ret)
9cf78fa0 733
8efd3f97 734#define ANYOF_POSIXL_TEST_ANY_SET(p) \
93e92956 735 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
d1d040e5 736 && (((regnode_charclass_posixl*)(p))->classflags))
8efd3f97 737#define ANYOF_CLASS_TEST_ANY_SET(p) ANYOF_POSIXL_TEST_ANY_SET(p)
e867ce45 738
e0e1be5f
KW
739/* Since an SSC always has this field, we don't have to test for that; nor do
740 * we want to because the bit isn't set for SSC during its construction */
741#define ANYOF_POSIXL_SSC_TEST_ANY_SET(p) \
742 cBOOL(((regnode_ssc*)(p))->classflags)
743#define ANYOF_POSIXL_SSC_TEST_ALL_SET(p) /* Are all bits set? */ \
744 (((regnode_ssc*) (p))->classflags \
745 == ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1)
746
747#define ANYOF_POSIXL_TEST_ALL_SET(p) \
93e92956 748 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
e0e1be5f
KW
749 && ((regnode_charclass_posixl*) (p))->classflags \
750 == ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1)
fb38762f 751
b27fd97a 752#define ANYOF_POSIXL_OR(source, dest) STMT_START { (dest)->classflags |= (source)->classflags ; } STMT_END
8efd3f97 753#define ANYOF_CLASS_OR(source, dest) ANYOF_POSIXL_OR((source), (dest))
936ed897 754
cdd87c1d
KW
755#define ANYOF_POSIXL_AND(source, dest) STMT_START { (dest)->classflags &= (source)->classflags ; } STMT_END
756
428c1aff
KW
757#define ANYOF_BITMAP_ZERO(ret) Zero(((regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char)
758#define ANYOF_BITMAP(p) ((regnode_charclass*)(p))->bitmap
4cbce0a6 759#define ANYOF_BITMAP_BYTE(p, c) BITMAP_BYTE(ANYOF_BITMAP(p), c)
b8c5462f
JH
760#define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c))
761#define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c))
bc51fd78 762#define ANYOF_BITMAP_TEST(p, c) cBOOL(ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c))
b8c5462f 763
f8bef550
NC
764#define ANYOF_BITMAP_SETALL(p) \
765 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE)
766#define ANYOF_BITMAP_CLEARALL(p) \
767 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE)
f8bef550 768
a687059c
LW
769/*
770 * Utility definitions.
771 */
2304df62 772#ifndef CHARMASK
075abff3 773# define UCHARAT(p) ((int)*(const U8*)(p))
a687059c 774#else
075abff3 775# define UCHARAT(p) ((int)*(p)&CHARMASK)
a687059c 776#endif
a687059c 777
c45abc0a
KW
778/* Number of regnode equivalents that 'guy' occupies beyond the size of the
779 * smallest regnode. */
c277df42
IZ
780#define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode))
781
e384d5c1
YO
782#define REG_ZERO_LEN_SEEN 0x00000001
783#define REG_LOOKBEHIND_SEEN 0x00000002
784#define REG_GPOS_SEEN 0x00000004
471f5387 785/* spare */
e384d5c1
YO
786#define REG_RECURSE_SEEN 0x00000020
787#define REG_TOP_LEVEL_BRANCHES_SEEN 0x00000040
788#define REG_VERBARG_SEEN 0x00000080
789#define REG_CUTGROUP_SEEN 0x00000100
790#define REG_RUN_ON_COMMENT_SEEN 0x00000200
791#define REG_UNFOLDED_MULTI_SEEN 0x00000400
d5a00e4a 792/* spare */
ee273784
YO
793#define REG_UNBOUNDED_QUANTIFIER_SEEN 0x00001000
794
d09b2d29 795
73c4f7a1
GS
796START_EXTERN_C
797
be8e71aa
YO
798#ifdef PLUGGABLE_RE_EXTENSION
799#include "re_nodes.h"
800#else
d09b2d29 801#include "regnodes.h"
be8e71aa 802#endif
d09b2d29 803
f9f4320a
YO
804#ifndef PLUGGABLE_RE_EXTENSION
805#ifndef DOINIT
806EXTCONST regexp_engine PL_core_reg_engine;
807#else /* DOINIT */
808EXTCONST regexp_engine PL_core_reg_engine = {
fe578d7f 809 Perl_re_compile,
2fdbfb4d 810 Perl_regexec_flags,
f9f4320a
YO
811 Perl_re_intuit_start,
812 Perl_re_intuit_string,
2fdbfb4d
AB
813 Perl_regfree_internal,
814 Perl_reg_numbered_buff_fetch,
815 Perl_reg_numbered_buff_store,
816 Perl_reg_numbered_buff_length,
192b9cd1
AB
817 Perl_reg_named_buff,
818 Perl_reg_named_buff_iter,
49d7dfbc 819 Perl_reg_qr_package,
f9f4320a 820#if defined(USE_ITHREADS)
3c13cae6 821 Perl_regdupe_internal,
f9f4320a 822#endif
3c13cae6 823 Perl_re_op_compile
f9f4320a
YO
824};
825#endif /* DOINIT */
826#endif /* PLUGGABLE_RE_EXTENSION */
827
828
73c4f7a1 829END_EXTERN_C
cad2e5aa 830
cad2e5aa 831
261faec3
MJD
832/* .what is a character array with one character for each member of .data
833 * The character describes the function of the corresponding .data item:
af534a04 834 * a - AV for paren_name_list under DEBUGGING
0111c4fd 835 * f - start-class data for regstclass optimization
68e2671b 836 * l - start op for literal (?{EVAL}) item
d63c20f2 837 * L - start op for literal (?{EVAL}) item, with separate CV (qr//)
b30fcab9 838 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/
7a94a5fe
KW
839 * s - inversion list for Unicode-style character class, and the
840 * multicharacter strings resulting from casefolding the single-character
841 * entries in the character class
a3621e74 842 * t - trie struct
55eed653 843 * u - trie struct's widecharmap (a HV, so can't share, must dup)
2b8b4781 844 * also used for revcharmap and words under DEBUGGING
07be1b83 845 * T - aho-trie struct
81714fb9 846 * S - sv for named capture lookup
261faec3 847 * 20010712 mjd@plover.com
c3613ed7 848 * (Remember to update re_dup() and pregfree() if you add any items.)
261faec3 849 */
cad2e5aa
JH
850struct reg_data {
851 U32 count;
852 U8 *what;
853 void* data[1];
854};
855
a1cac82e
NC
856/* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses
857 anchored* and float* via array indexes 0 and 1. */
cad2e5aa 858#define anchored_substr substrs->data[0].substr
33b8afdf 859#define anchored_utf8 substrs->data[0].utf8_substr
cad2e5aa 860#define anchored_offset substrs->data[0].min_offset
1de06328
YO
861#define anchored_end_shift substrs->data[0].end_shift
862
cad2e5aa 863#define float_substr substrs->data[1].substr
33b8afdf 864#define float_utf8 substrs->data[1].utf8_substr
cad2e5aa
JH
865#define float_min_offset substrs->data[1].min_offset
866#define float_max_offset substrs->data[1].max_offset
1de06328
YO
867#define float_end_shift substrs->data[1].end_shift
868
cad2e5aa 869#define check_substr substrs->data[2].substr
33b8afdf 870#define check_utf8 substrs->data[2].utf8_substr
cad2e5aa
JH
871#define check_offset_min substrs->data[2].min_offset
872#define check_offset_max substrs->data[2].max_offset
1de06328 873#define check_end_shift substrs->data[2].end_shift
a3621e74 874
8d919b0a
FC
875#define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr)
876#define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8)
877#define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr)
878#define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8)
a3621e74
YO
879
880/* trie related stuff */
5d9a96ca 881
a3621e74
YO
882/* a transition record for the state machine. the
883 check field determines which state "owns" the
884 transition. the char the transition is for is
885 determined by offset from the owning states base
886 field. the next field determines which state
887 is to be transitioned to if any.
888*/
889struct _reg_trie_trans {
890 U32 next;
891 U32 check;
892};
893
894/* a transition list element for the list based representation */
895struct _reg_trie_trans_list_elem {
896 U16 forid;
897 U32 newstate;
898};
899typedef struct _reg_trie_trans_list_elem reg_trie_trans_le;
900
901/* a state for compressed nodes. base is an offset
902 into an array of reg_trie_trans array. If wordnum is
903 nonzero the state is accepting. if base is zero then
904 the state has no children (and will be accepting)
905*/
906struct _reg_trie_state {
907 U16 wordnum;
908 union {
909 U32 base;
910 reg_trie_trans_le* list;
911 } trans;
912};
913
2e64971a
DM
914/* info per word; indexed by wordnum */
915typedef struct {
916 U16 prev; /* previous word in acceptance chain; eg in
917 * zzz|abc|ab/ after matching the chars abc, the
918 * accepted word is #2, and the previous accepted
919 * word is #3 */
920 U32 len; /* how many chars long is this word? */
921 U32 accept; /* accept state for this word */
922} reg_trie_wordinfo;
a3621e74
YO
923
924
a3621e74
YO
925typedef struct _reg_trie_state reg_trie_state;
926typedef struct _reg_trie_trans reg_trie_trans;
927
928
929/* anything in here that needs to be freed later
23eab42c
NC
930 should be dealt with in pregfree.
931 refcount is first in both this and _reg_ac_data to allow a space
932 optimisation in Perl_regdupe. */
a3621e74 933struct _reg_trie_data {
23eab42c 934 U32 refcount; /* number of times this trie is referenced */
786e8c11
YO
935 U32 lasttrans; /* last valid transition element */
936 U16 *charmap; /* byte to charid lookup array */
786e8c11
YO
937 reg_trie_state *states; /* state data */
938 reg_trie_trans *trans; /* array of transition elements */
939 char *bitmap; /* stclass bitmap */
786e8c11
YO
940 U16 *jump; /* optional 1 indexed array of offsets before tail
941 for the node following a given word. */
2e64971a 942 reg_trie_wordinfo *wordinfo; /* array of info per word */
1b5c067c
NC
943 U16 uniquecharcount; /* unique chars in trie (width of trans table) */
944 U32 startstate; /* initial state - used for common prefix optimisation */
945 STRLEN minlen; /* minimum length of words in trie - build/opt only? */
946 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */
2e64971a 947 U32 prefixlen; /* #chars in common prefix */
1e2e3d02
YO
948 U32 statecount; /* Build only - number of states in the states array
949 (including the unused zero state) */
786e8c11 950 U32 wordcount; /* Build only */
a3621e74 951#ifdef DEBUGGING
786e8c11 952 STRLEN charcount; /* Build only */
a3621e74
YO
953#endif
954};
2b8b4781
NC
955/* There is one (3 under DEBUGGING) pointers that logically belong in this
956 structure, but are held outside as they need duplication on thread cloning,
957 whereas the rest of the structure can be read only:
958 HV *widecharmap; code points > 255 to charid
959#ifdef DEBUGGING
960 AV *words; Array of words contained in trie, for dumping
961 AV *revcharmap; Map of each charid back to its character representation
962#endif
963*/
964
965#define TRIE_WORDS_OFFSET 2
966
a3621e74
YO
967typedef struct _reg_trie_data reg_trie_data;
968
23eab42c
NC
969/* refcount is first in both this and _reg_trie_data to allow a space
970 optimisation in Perl_regdupe. */
07be1b83 971struct _reg_ac_data {
23eab42c 972 U32 refcount;
1b5c067c 973 U32 trie;
07be1b83
YO
974 U32 *fail;
975 reg_trie_state *states;
07be1b83
YO
976};
977typedef struct _reg_ac_data reg_ac_data;
978
486ec47a 979/* ANY_BIT doesn't use the structure, so we can borrow it here.
3dab1dad
YO
980 This is simpler than refactoring all of it as wed end up with
981 three different sets... */
982
983#define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap)
4cbce0a6 984#define TRIE_BITMAP_BYTE(p, c) BITMAP_BYTE(TRIE_BITMAP(p), c)
5dfe062f
YO
985#define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c))
986#define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c))
987#define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
3dab1dad 988
1de06328
YO
989#define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC)
990#define IS_TRIE_AC(op) ((op)>=AHOCORASICK)
991
992
0effac60 993#define BITMAP_BYTE(p, c) (( (U8*) p) [ ( ( (UV) (c)) >> 3) ] )
786e8c11 994#define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
3dab1dad 995
a3621e74
YO
996/* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */
997#define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1)
998#define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state))
999#define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state))
1000
3dab1dad 1001#ifdef DEBUGGING
3dab1dad 1002#define TRIE_CHARCOUNT(trie) ((trie)->charcount)
3dab1dad 1003#else
3dab1dad 1004#define TRIE_CHARCOUNT(trie) (trie_charcount)
3dab1dad 1005#endif
a3621e74 1006
0111c4fd
RGS
1007#define RE_TRIE_MAXBUF_INIT 65536
1008#define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF"
a3621e74
YO
1009#define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS"
1010
6ef7fe53
KW
1011#define RE_COMPILE_RECURSION_INIT 1000
1012#define RE_COMPILE_RECURSION_LIMIT "\022E_COMPILE_RECURSION_LIMIT"
1013
be8e71aa
YO
1014/*
1015
1016RE_DEBUG_FLAGS is used to control what debug output is emitted
1017its divided into three groups of options, some of which interact.
1018The three groups are: Compile, Execute, Extra. There is room for a
1019further group, as currently only the low three bytes are used.
1020
1021 Compile Options:
07be1b83 1022
be8e71aa
YO
1023 PARSE
1024 PEEP
1025 TRIE
1026 PROGRAM
1027 OFFSETS
a3621e74 1028
be8e71aa 1029 Execute Options:
a3621e74 1030
be8e71aa
YO
1031 INTUIT
1032 MATCH
1033 TRIE
a3621e74 1034
be8e71aa 1035 Extra Options
a3621e74 1036
be8e71aa
YO
1037 TRIE
1038 OFFSETS
1039
1040If you modify any of these make sure you make corresponding changes to
1041re.pm, especially to the documentation.
a3621e74 1042
be8e71aa
YO
1043*/
1044
1045
1046/* Compile */
1047#define RE_DEBUG_COMPILE_MASK 0x0000FF
1048#define RE_DEBUG_COMPILE_PARSE 0x000001
1049#define RE_DEBUG_COMPILE_OPTIMISE 0x000002
1050#define RE_DEBUG_COMPILE_TRIE 0x000004
1051#define RE_DEBUG_COMPILE_DUMP 0x000008
f7819f85 1052#define RE_DEBUG_COMPILE_FLAGS 0x000010
d9a72fcc 1053#define RE_DEBUG_COMPILE_TEST 0x000020
be8e71aa
YO
1054
1055/* Execute */
1056#define RE_DEBUG_EXECUTE_MASK 0x00FF00
1057#define RE_DEBUG_EXECUTE_INTUIT 0x000100
1058#define RE_DEBUG_EXECUTE_MATCH 0x000200
1059#define RE_DEBUG_EXECUTE_TRIE 0x000400
1060
1061/* Extra */
fd41b2d1
KW
1062#define RE_DEBUG_EXTRA_MASK 0x1FF0000
1063#define RE_DEBUG_EXTRA_TRIE 0x0010000
1064#define RE_DEBUG_EXTRA_OFFSETS 0x0020000
1065#define RE_DEBUG_EXTRA_OFFDEBUG 0x0040000
1066#define RE_DEBUG_EXTRA_STATE 0x0080000
1067#define RE_DEBUG_EXTRA_OPTIMISE 0x0100000
1068#define RE_DEBUG_EXTRA_BUFFERS 0x0400000
1069#define RE_DEBUG_EXTRA_GPOS 0x0800000
1070#define RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE 0x1000000
24b23f37 1071/* combined */
fd41b2d1 1072#define RE_DEBUG_EXTRA_STACK 0x0280000
be8e71aa 1073
e68ec53f 1074#define RE_DEBUG_FLAG(x) (re_debug_flags & x)
be8e71aa
YO
1075/* Compile */
1076#define DEBUG_COMPILE_r(x) DEBUG_r( \
dafb2544 1077 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_MASK)) x )
be8e71aa 1078#define DEBUG_PARSE_r(x) DEBUG_r( \
dafb2544 1079 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_PARSE)) x )
be8e71aa 1080#define DEBUG_OPTIMISE_r(x) DEBUG_r( \
dafb2544 1081 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_OPTIMISE)) x )
be8e71aa 1082#define DEBUG_DUMP_r(x) DEBUG_r( \
dafb2544 1083 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_DUMP)) x )
be8e71aa 1084#define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \
dafb2544 1085 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_TRIE)) x )
f7819f85 1086#define DEBUG_FLAGS_r(x) DEBUG_r( \
dafb2544 1087 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_FLAGS)) x )
d9a72fcc 1088#define DEBUG_TEST_r(x) DEBUG_r( \
dafb2544 1089 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_COMPILE_TEST)) x )
be8e71aa
YO
1090/* Execute */
1091#define DEBUG_EXECUTE_r(x) DEBUG_r( \
dafb2544 1092 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXECUTE_MASK)) x )
be8e71aa 1093#define DEBUG_INTUIT_r(x) DEBUG_r( \
dafb2544 1094 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXECUTE_INTUIT)) x )
be8e71aa 1095#define DEBUG_MATCH_r(x) DEBUG_r( \
dafb2544 1096 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXECUTE_MATCH)) x )
be8e71aa 1097#define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \
dafb2544 1098 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXECUTE_TRIE)) x )
be8e71aa
YO
1099
1100/* Extra */
1101#define DEBUG_EXTRA_r(x) DEBUG_r( \
dafb2544 1102 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_MASK)) x )
a5ca303d 1103#define DEBUG_OFFSETS_r(x) DEBUG_r( \
dafb2544 1104 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_OFFSETS)) x )
ab3bbdeb 1105#define DEBUG_STATE_r(x) DEBUG_r( \
dafb2544 1106 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_STATE)) x )
24b23f37 1107#define DEBUG_STACK_r(x) DEBUG_r( \
dafb2544 1108 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_STACK)) x )
e7707071 1109#define DEBUG_BUFFERS_r(x) DEBUG_r( \
dafb2544 1110 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_BUFFERS)) x )
e7707071 1111
a5ca303d 1112#define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \
dafb2544
KW
1113 if (DEBUG_v_TEST || ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \
1114 (re_debug_flags & (RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE)))) x )
07be1b83 1115#define MJD_OFFSET_DEBUG(x) DEBUG_r( \
dafb2544 1116 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_OFFDEBUG)) \
be8e71aa
YO
1117 Perl_warn_nocontext x )
1118#define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \
dafb2544 1119 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_TRIE)) x )
be8e71aa 1120#define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \
dafb2544 1121 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_TRIE)) x )
be8e71aa
YO
1122
1123#define DEBUG_TRIE_r(x) DEBUG_r( \
dafb2544
KW
1124 if (DEBUG_v_TEST || (re_debug_flags & (RE_DEBUG_COMPILE_TRIE \
1125 | RE_DEBUG_EXECUTE_TRIE ))) x )
2c296965 1126#define DEBUG_GPOS_r(x) DEBUG_r( \
dafb2544 1127 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_GPOS)) x )
be8e71aa 1128
fd41b2d1
KW
1129#define DEBUG_DUMP_PRE_OPTIMIZE_r(x) DEBUG_r( \
1130 if (DEBUG_v_TEST || (re_debug_flags & RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE)) x )
1131
be8e71aa 1132/* initialization */
d2d8c711 1133/* get_sv() can return NULL during global destruction. */
e68ec53f
YO
1134#define GET_RE_DEBUG_FLAGS DEBUG_r({ \
1135 SV * re_debug_flags_sv = NULL; \
243a3b9c 1136 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \
e68ec53f 1137 if (re_debug_flags_sv) { \
8e9a8a48 1138 if (!SvIOK(re_debug_flags_sv)) \
e68ec53f 1139 sv_setuv(re_debug_flags_sv, RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \
8e9a8a48 1140 re_debug_flags=SvIV(re_debug_flags_sv); \
e68ec53f
YO
1141 }\
1142})
a3621e74
YO
1143
1144#ifdef DEBUGGING
ab3bbdeb 1145
8162b70e 1146#define GET_RE_DEBUG_FLAGS_DECL volatile IV re_debug_flags = 0; \
cb65b687 1147 PERL_UNUSED_VAR(re_debug_flags); GET_RE_DEBUG_FLAGS;
ab3bbdeb 1148
eecd4d11
KW
1149#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \
1150 const char * const rpv = \
1151 pv_pretty((dsv), (pv), (l), (m), \
1152 PL_colors[(c1)],PL_colors[(c2)], \
c89df6cf 1153 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \
0df25f3d 1154 const int rlen = SvCUR(dsv)
ab3bbdeb 1155
8ec77c3c 1156/* This is currently unsed in the core */
eecd4d11
KW
1157#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \
1158 const char * const rpv = \
1159 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \
1160 PL_colors[(c1)],PL_colors[(c2)], \
c89df6cf 1161 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) )
ab3bbdeb 1162
eecd4d11
KW
1163#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \
1164 const char * const rpv = \
1165 pv_pretty((dsv), (pv), (l), (m), \
1166 PL_colors[0], PL_colors[1], \
c89df6cf 1167 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \
ab3bbdeb 1168 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \
0a0e9afc 1169 )
ab3bbdeb
YO
1170
1171#define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0))
1172#define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "")
1173
1174#else /* if not DEBUGGING */
1175
a3621e74 1176#define GET_RE_DEBUG_FLAGS_DECL
ab3bbdeb
YO
1177#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2)
1178#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m)
1179#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m)
1180#define RE_SV_DUMPLEN(ItEm)
1181#define RE_SV_TAIL(ItEm)
1182
1183#endif /* DEBUG RELATED DEFINES */
a3621e74 1184
0bc0e44f 1185#define FIRST_NON_ASCII_DECIMAL_DIGIT 0x660 /* ARABIC_INDIC_DIGIT_ZERO */
2ca74f27 1186
64935bc6
KW
1187typedef enum {
1188 TRADITIONAL_BOUND = _CC_WORDCHAR,
ae3bb8ea 1189 GCB_BOUND,
6b659339 1190 LB_BOUND,
06ae2722 1191 SB_BOUND,
ae3bb8ea 1192 WB_BOUND
64935bc6
KW
1193} bound_type;
1194
13fcf652 1195/* This unpacks the FLAGS field of ANYOF[HR]x nodes. The value it contains
3146c00a
KW
1196 * gives the strict lower bound for the UTF-8 start byte of any code point
1197 * matchable by the node, and a loose upper bound as well.
1198 *
1199 * The low bound is stored in the upper 6 bits, plus 0xC0.
1200 * The loose upper bound is determined from the lowest 2 bits and the low bound
1201 * (called x) as follows:
1202 *
1203 * 11 The upper limit of the range can be as much as (EF - x) / 8
1204 * 10 The upper limit of the range can be as much as (EF - x) / 4
1205 * 01 The upper limit of the range can be as much as (EF - x) / 2
1206 * 00 The upper limit of the range can be as much as EF
1207 *
0bf4efd7
KW
1208 * For motivation of this design, see commit message in
1209 * 3146c00a633e9cbed741e10146662fbcedfdb8d3 */
3146c00a
KW
1210#ifdef EBCDIC
1211# define MAX_ANYOF_HRx_BYTE 0xF4
1212#else
1213# define MAX_ANYOF_HRx_BYTE 0xEF
1214#endif
1215#define LOWEST_ANYOF_HRx_BYTE(b) (((b) >> 2) + 0xC0)
1216#define HIGHEST_ANYOF_HRx_BYTE(b) \
1217 (LOWEST_ANYOF_HRx_BYTE(b) \
1218 + ((MAX_ANYOF_HRx_BYTE - LOWEST_ANYOF_HRx_BYTE(b)) >> ((b) & 3)))
1219
3dd7db29
JK
1220#endif /* PERL_REGCOMP_H_ */
1221
e9a8c099 1222/*
14d04a33 1223 * ex: set ts=8 sts=4 sw=4 et:
e9a8c099 1224 */