This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
handy.h: Create nBIT_UMAX() macro
[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 33/*
a687059c
LW
34 * Structure for regexp "program". This is essentially a linear encoding
35 * of a nondeterministic finite-state machine (aka syntax charts or
36 * "railroad normal form" in parsing technology). Each node is an opcode
37 * plus a "next" pointer, possibly plus an operand. "Next" pointers of
38 * all nodes except BRANCH implement concatenation; a "next" pointer with
39 * a BRANCH on both ends of it is connecting two alternatives. (Here we
40 * have one of the subtle syntax dependencies: an individual BRANCH (as
41 * opposed to a collection of them) is never concatenated with anything
42 * because of operator precedence.) The operand of some types of node is
43 * a literal string; for others, it is a node leading into a sub-FSM. In
44 * particular, the operand of a BRANCH node is the first node of the branch.
45 * (NB this is *not* a tree structure: the tail of the branch connects
a3621e74
YO
46 * to the thing following the set of BRANCHes.) The opcodes are defined
47 * in regnodes.h which is generated from regcomp.sym by regcomp.pl.
a687059c
LW
48 */
49
a687059c
LW
50/*
51 * A node is one char of opcode followed by two chars of "next" pointer.
52 * "Next" pointers are stored as two 8-bit pieces, high order first. The
53 * value is a positive offset from the opcode of the node containing it.
54 * An operand, if any, simply follows the node. (Note that much of the
55 * code generation knows about this implicit relationship.)
56 *
57 * Using two bytes for the "next" pointer is vast overkill for most things,
58 * but allows patterns to get big without disasters.
59 *
e91177ed 60 * [The "next" pointer is always aligned on an even
62e6ef33 61 * boundary, and reads the offset directly as a short.]
a687059c 62 */
785a26d5
YO
63
64/* This is the stuff that used to live in regexp.h that was truly
65 private to the engine itself. It now lives here. */
66
785a26d5 67 typedef struct regexp_internal {
7122b237
YO
68 union {
69 U32 *offsets; /* offset annotations 20001228 MJD
70 data about mapping the program to the
71 string -
72 offsets[0] is proglen when this is used
73 */
74 U32 proglen;
75 } u;
76
6d5c990f
RGS
77 regnode *regstclass; /* Optional startclass as identified or constructed
78 by the optimiser */
79 struct reg_data *data; /* Additional miscellaneous data used by the program.
80 Used to make it easier to clone and free arbitrary
81 data that the regops need. Often the ARG field of
82 a regop is an index into this structure */
1acab4c5 83 struct reg_code_blocks *code_blocks;/* positions of literal (?{}) */
bba699a8 84 int name_list_idx; /* Optional data index of an array of paren names */
6d5c990f
RGS
85 regnode program[1]; /* Unwarranted chumminess with compiler. */
86} regexp_internal;
87
88#define RXi_SET(x,y) (x)->pprivate = (void*)(y)
89#define RXi_GET(x) ((regexp_internal *)((x)->pprivate))
90#define RXi_GET_DECL(r,ri) regexp_internal *ri = RXi_GET(r)
785a26d5
YO
91/*
92 * Flags stored in regexp->intflags
93 * These are used only internally to the regexp engine
94 *
95 * See regexp.h for flags used externally to the regexp engine
96 */
e3e400ec
YO
97#define RXp_INTFLAGS(rx) ((rx)->intflags)
98#define RX_INTFLAGS(prog) RXp_INTFLAGS(ReANY(prog))
99
785a26d5
YO
100#define PREGf_SKIP 0x00000001
101#define PREGf_IMPLICIT 0x00000002 /* Converted .* to ^.* */
102#define PREGf_NAUGHTY 0x00000004 /* how exponential is this pattern? */
103#define PREGf_VERBARG_SEEN 0x00000008
104#define PREGf_CUTGROUP_SEEN 0x00000010
732caac7 105#define PREGf_USE_RE_EVAL 0x00000020 /* compiled with "use re 'eval'" */
e3e400ec
YO
106/* these used to be extflags, but are now intflags */
107#define PREGf_NOSCAN 0x00000040
33c28ab2 108 /* spare */
8e1490ee
YO
109#define PREGf_GPOS_SEEN 0x00000100
110#define PREGf_GPOS_FLOAT 0x00000200
785a26d5 111
d3d47aac
YO
112#define PREGf_ANCH_MBOL 0x00000400
113#define PREGf_ANCH_SBOL 0x00000800
114#define PREGf_ANCH_GPOS 0x00001000
d5a00e4a 115#define PREGf_RECURSE_SEEN 0x00002000
8e1490ee 116
d3d47aac
YO
117#define PREGf_ANCH \
118 ( PREGf_ANCH_SBOL | PREGf_ANCH_GPOS | PREGf_ANCH_MBOL )
785a26d5
YO
119
120/* this is where the old regcomp.h started */
a687059c 121
c277df42 122struct regnode_string {
cd439c50 123 U8 str_len;
c277df42
IZ
124 U8 type;
125 U16 next_off;
cd439c50 126 char string[1];
c277df42 127};
a687059c 128
ae06e581
KW
129struct regnode_lstring { /* Constructed this way to keep the string aligned. */
130 U8 flags;
131 U8 type;
132 U16 next_off;
1d425dad 133 U32 str_len; /* Only 18 bits allowed before would overflow 'next_off' */
ae06e581
KW
134 char string[1];
135};
136
34924db0 137struct regnode_anyofhs { /* Constructed this way to keep the string aligned. */
53d42e43 138 U8 str_len;
34924db0
KW
139 U8 type;
140 U16 next_off;
141 U32 arg1; /* set by set_ANYOF_arg() */
142 char string[1];
143};
144
6bda09f9
YO
145/* Argument bearing node - workhorse,
146 arg1 is often for the data field */
c277df42
IZ
147struct regnode_1 {
148 U8 flags;
149 U8 type;
150 U16 next_off;
151 U32 arg1;
152};
153
d3e3ece8
KW
154/* Node whose argument is 'SV *'. This needs to be used very carefully in
155 * situations where pointers won't become invalid because of, say re-mallocs */
86451f01
KW
156struct regnode_p {
157 U8 flags;
158 U8 type;
159 U16 next_off;
d3e3ece8 160 SV * arg1;
86451f01
KW
161};
162
6bda09f9
YO
163/* Similar to a regnode_1 but with an extra signed argument */
164struct regnode_2L {
165 U8 flags;
166 U8 type;
167 U16 next_off;
168 U32 arg1;
169 I32 arg2;
170};
171
172/* 'Two field' -- Two 16 bit unsigned args */
c277df42
IZ
173struct regnode_2 {
174 U8 flags;
175 U8 type;
176 U16 next_off;
177 U16 arg1;
178 U16 arg2;
179};
180
dcb20b36 181#define ANYOF_BITMAP_SIZE (NUM_ANYOF_CODE_POINTS / 8) /* 8 bits/Byte */
936ed897 182
1ee208c4
KW
183/* Note that these form structs which are supersets of the next smaller one, by
184 * appending fields. Alignment problems can occur if one of those optional
185 * fields requires stricter alignment than the base struct. And formal
186 * parameters that can really be two or more of the structs should be
187 * declared as the smallest one it could be. See commit message for
188 * 7dcac5f6a5195002b55c935ee1d67f67e1df280b. Regnode allocation is done
189 * without regard to alignment, and changing it to would also require changing
190 * the code that inserts and deletes regnodes. The basic single-argument
191 * regnode has a U32, which is what reganode() allocates as a unit. Therefore
192 * no field can require stricter alignment than U32. */
193
786e8c11 194/* also used by trie */
936ed897
IZ
195struct regnode_charclass {
196 U8 flags;
197 U8 type;
198 U16 next_off;
9da99c5e 199 U32 arg1; /* set by set_ANYOF_arg() */
19103b91 200 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */
936ed897
IZ
201};
202
bf3f2d85 203/* has runtime (locale) \d, \w, ..., [:posix:] classes */
ea461dd6 204struct regnode_charclass_posixl {
93e92956 205 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
936ed897
IZ
206 U8 type;
207 U16 next_off;
7e817188 208 U32 arg1;
03fa83ba 209 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
9cf78fa0 210 U32 classflags; /* and run-time */
936ed897
IZ
211};
212
85c8e306
KW
213/* A synthetic start class (SSC); is a regnode_charclass_posixl_fold, plus an
214 * extra SV*, used only during its construction and which is not used by
215 * regexec.c. Note that the 'next_off' field is unused, as the SSC stands
216 * alone, so there is never a next node. Also, there is no alignment issue,
b8fda935 217 * because these are declared or allocated as a complete unit so the compiler
85c8e306
KW
218 * takes care of alignment. This is unlike the other regnodes which are
219 * allocated in terms of multiples of a single-argument regnode. SSC nodes can
220 * have a pointer field because there is no alignment issue, and because it is
221 * set to NULL after construction, before any cloning of the pattern */
b8f7bb16 222struct regnode_ssc {
93e92956 223 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
b8f7bb16
KW
224 U8 type;
225 U16 next_off;
7e817188 226 U32 arg1;
03fa83ba 227 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
631e9161 228 U32 classflags; /* ... and run-time */
85c8e306
KW
229
230 /* Auxiliary, only used during construction; NULL afterwards: list of code
231 * points matched */
232 SV* invlist;
b8f7bb16
KW
233};
234
71068078
KW
235/* We take advantage of 'next_off' not otherwise being used in the SSC by
236 * actually using it: by setting it to 1. This allows us to test and
237 * distinguish between an SSC and other ANYOF node types, as 'next_off' cannot
238 * otherwise be 1, because it is the offset to the next regnode expressed in
239 * units of regnodes. Since an ANYOF node contains extra fields, it adds up
240 * to 12 regnode units on 32-bit systems, (hence the minimum this can be (if
241 * not 0) is 11 there. Even if things get tightly packed on a 64-bit system,
242 * it still would be more than 1. */
243#define set_ANYOF_SYNTHETIC(n) STMT_START{ OP(n) = ANYOF; \
244 NEXT_OFF(n) = 1; \
245 } STMT_END
c1071334 246#define is_ANYOF_SYNTHETIC(n) (PL_regkind[OP(n)] == ANYOF && NEXT_OFF(n) == 1)
71068078 247
9c7e81e8
AD
248/* XXX fix this description.
249 Impose a limit of REG_INFTY on various pattern matching operations
250 to limit stack growth and to avoid "infinite" recursions.
251*/
4288c5b9
KW
252/* The default size for REG_INFTY is U16_MAX, which is the same as
253 USHORT_MAX (see perl.h). Unfortunately U16 isn't necessarily 16 bits
254 (see handy.h). On the Cray C90, sizeof(short)==4 and hence U16_MAX is
255 ((1<<32)-1), while on the Cray T90, sizeof(short)==8 and U16_MAX is
256 ((1<<64)-1). To limit stack growth to reasonable sizes, supply a
9c7e81e8
AD
257 smaller default.
258 --Andy Dougherty 11 June 1998
259*/
260#if SHORTSIZE > 2
261# ifndef REG_INFTY
1381ccb1 262# define REG_INFTY nBIT_UMAX(16)
9c7e81e8
AD
263# endif
264#endif
265
266#ifndef REG_INFTY
4288c5b9 267# define REG_INFTY U16_MAX
83cfe48c 268#endif
a687059c 269
e91177ed
IZ
270#define ARG_VALUE(arg) (arg)
271#define ARG__SET(arg,val) ((arg) = (val))
c277df42 272
e8e3b0fb
JH
273#undef ARG
274#undef ARG1
275#undef ARG2
276
c277df42 277#define ARG(p) ARG_VALUE(ARG_LOC(p))
86451f01 278#define ARGp(p) ARG_VALUE(ARGp_LOC(p))
c277df42
IZ
279#define ARG1(p) ARG_VALUE(ARG1_LOC(p))
280#define ARG2(p) ARG_VALUE(ARG2_LOC(p))
6bda09f9 281#define ARG2L(p) ARG_VALUE(ARG2L_LOC(p))
786e8c11 282
c277df42 283#define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val))
86451f01 284#define ARGp_SET(p, val) ARG__SET(ARGp_LOC(p), (val))
c277df42
IZ
285#define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val))
286#define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val))
6bda09f9 287#define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val))
c277df42 288
e8e3b0fb
JH
289#undef NEXT_OFF
290#undef NODE_ALIGN
291
075abff3
AL
292#define NEXT_OFF(p) ((p)->next_off)
293#define NODE_ALIGN(node)
aa48e906
YO
294/* the following define was set to 0xde in 075abff3
295 * as part of some linting logic. I have set it to 0
296 * as otherwise in every place where we /might/ set flags
297 * we have to set it 0 explicitly, which duplicates
298 * assignments and IMO adds an unacceptable level of
299 * surprise to working in the regex engine. If this
300 * is changed from 0 then at the very least make sure
301 * that SBOL for /^/ sets the flags to 0 explicitly.
302 * -- Yves */
303#define NODE_ALIGN_FILL(node) ((node)->flags = 0)
a687059c 304
c277df42
IZ
305#define SIZE_ALIGN NODE_ALIGN
306
e8e3b0fb
JH
307#undef OP
308#undef OPERAND
e8e3b0fb
JH
309#undef STRING
310
e91177ed 311#define OP(p) ((p)->type)
b988e673
KW
312#define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \
313 regnode types. For some, it's the \
314 character set of the regnode */
3f2416ae 315#define STR_LENs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \
5cd61b66 316 ((struct regnode_string *)p)->str_len)
3f2416ae 317#define STRINGs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \
5cd61b66 318 ((struct regnode_string *)p)->string)
ae06e581
KW
319#define OPERANDs(p) STRINGs(p)
320
321/* Long strings. Currently limited to length 18 bits, which handles a 262000
322 * byte string. The limiting factor is the 16 bit 'next_off' field, which
323 * points to the next regnode, so the furthest away it can be is 2**16. On
324 * most architectures, regnodes are 2**2 bytes long, so that yields 2**18
325 * bytes. Should a longer string be desired, we could increase it to 26 bits
326 * fairly easily, by changing this node to have longj type which causes the ARG
327 * field to be used for the link to the next regnode (although code would have
328 * to be changed to account for this), and then use a combination of the flags
329 * and next_off fields for the length. To get 34 bit length, also change the
330 * node to be an ARG2L, using the second 32 bit field for the length, and not
331 * using the flags nor next_off fields at all. One could have an llstring node
332 * and even an lllstring type. */
3f2416ae 333#define STR_LENl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 334 (((struct regnode_lstring *)p)->str_len))
3f2416ae 335#define STRINGl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 336 (((struct regnode_lstring *)p)->string))
ae06e581
KW
337#define OPERANDl(p) STRINGl(p)
338
3f2416ae 339#define STR_LEN(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 340 ? STR_LENl(p) : STR_LENs(p))
3f2416ae 341#define STRING(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 342 ? STRINGl(p) : STRINGs(p))
c5184715
KW
343#define OPERAND(p) STRING(p)
344
c45abc0a
KW
345/* The number of (smallest) regnode equivalents that a string of length l bytes
346 * occupies */
59e54936 347#define STR_SZ(l) (((l) + sizeof(regnode) - 1) / sizeof(regnode))
c45abc0a
KW
348
349/* The number of (smallest) regnode equivalents that the EXACTISH node 'p'
350 * occupies */
ae06e581 351#define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p)) + 1 + regarglen[(p)->type])
c45abc0a 352
c7fd9c67 353#define setSTR_LEN(p,v) \
ae06e581 354 STMT_START{ \
3f2416ae 355 if (OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
ae06e581
KW
356 ((struct regnode_lstring *)(p))->str_len = (v); \
357 else \
358 ((struct regnode_string *)(p))->str_len = (v); \
359 } STMT_END
cd439c50 360
13fcf652 361#define ANYOFR_BASE_BITS 20
d223e1ea 362#define ANYOFRbase(p) (ARG(p) & nBIT_MASK(ANYOFR_BASE_BITS))
13fcf652
KW
363#define ANYOFRdelta(p) (ARG(p) >> ANYOFR_BASE_BITS)
364
e8e3b0fb
JH
365#undef NODE_ALIGN
366#undef ARG_LOC
367#undef NEXTOPER
368#undef PREVOPER
369
e91177ed
IZ
370#define NODE_ALIGN(node)
371#define ARG_LOC(p) (((struct regnode_1 *)p)->arg1)
86451f01 372#define ARGp_LOC(p) (((struct regnode_p *)p)->arg1)
e91177ed
IZ
373#define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1)
374#define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2)
6bda09f9 375#define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2)
786e8c11 376
e91177ed
IZ
377#define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */
378#define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2)
379
c277df42
IZ
380#define NEXTOPER(p) ((p) + NODE_STEP_REGNODE)
381#define PREVOPER(p) ((p) - NODE_STEP_REGNODE)
382
f55b7b26
KW
383#define FILL_NODE(offset, op) \
384 STMT_START { \
385 OP(REGNODE_p(offset)) = op; \
386 NEXT_OFF(REGNODE_p(offset)) = 0; \
d2a88850 387 } STMT_END
f55b7b26
KW
388#define FILL_ADVANCE_NODE(offset, op) \
389 STMT_START { \
390 FILL_NODE(offset, op); \
391 (offset)++; \
d2a88850 392 } STMT_END
f55b7b26
KW
393#define FILL_ADVANCE_NODE_ARG(offset, op, arg) \
394 STMT_START { \
395 ARG_SET(REGNODE_p(offset), arg); \
396 FILL_ADVANCE_NODE(offset, op); \
397 /* This is used generically for other operations \
398 * that have a longer argument */ \
399 (offset) += regarglen[op]; \
86451f01
KW
400 } STMT_END
401#define FILL_ADVANCE_NODE_ARGp(offset, op, arg) \
402 STMT_START { \
403 ARGp_SET(REGNODE_p(offset), arg); \
404 FILL_ADVANCE_NODE(offset, op); \
405 (offset) += regarglen[op]; \
93539530 406 } STMT_END
f55b7b26
KW
407#define FILL_ADVANCE_NODE_2L_ARG(offset, op, arg1, arg2) \
408 STMT_START { \
409 ARG_SET(REGNODE_p(offset), arg1); \
410 ARG2L_SET(REGNODE_p(offset), arg2); \
411 FILL_ADVANCE_NODE(offset, op); \
412 (offset) += 2; \
93539530 413 } STMT_END
a687059c 414
22c35a8c 415#define REG_MAGIC 0234
a687059c 416
fe8d1b7c
KW
417/* An ANYOF node is basically a bitmap with the index being a code point. If
418 * the bit for that code point is 1, the code point matches; if 0, it doesn't
419 * match (complemented if inverted). There is an additional mechanism to deal
420 * with cases where the bitmap is insufficient in and of itself. This #define
421 * indicates if the bitmap does fully represent what this ANYOF node can match.
422 * The ARG is set to this special value (since 0, 1, ... are legal, but will
423 * never reach this high). */
93e92956 424#define ANYOF_ONLY_HAS_BITMAP ((U32) -1)
137165a6 425
93539530 426/* When the bitmap isn't completely sufficient for handling the ANYOF node,
fe8d1b7c
KW
427 * flags (in node->flags of the ANYOF node) get set to indicate this. These
428 * are perennially in short supply. Beyond several cases where warnings need
429 * to be raised under certain circumstances, currently, there are six cases
430 * where the bitmap alone isn't sufficient. We could use six flags to
431 * represent the 6 cases, but to save flags bits, we play some games. The
432 * cases are:
d2f465e9 433 *
fe8d1b7c
KW
434 * 1) The bitmap has a compiled-in very finite size. So something else needs
435 * to be used to specify if a code point that is too large for the bitmap
7a94a5fe 436 * actually matches. The mechanism currently is an inversion
fe8d1b7c
KW
437 * list. ANYOF_ONLY_HAS_BITMAP, described above, being TRUE indicates
438 * there are no matches of too-large code points. But if it is FALSE,
439 * then almost certainly there are matches too large for the bitmap. (The
440 * other cases, described below, either imply this one or are extremely
441 * rare in practice.) So we can just assume that a too-large code point
442 * will need something beyond the bitmap if ANYOF_ONLY_HAS_BITMAP is
443 * FALSE, instead of having a separate flag for this.
444 * 2) A subset of item 1) is if all possible code points outside the bitmap
445 * match. This is a common occurrence when the class is complemented,
446 * like /[^ij]/. Therefore a bit is reserved to indicate this,
42d0f81a 447 * rather than having an inversion list created,
d1c40ef5 448 * ANYOF_MATCHES_ALL_ABOVE_BITMAP.
fe8d1b7c
KW
449 * 3) Under /d rules, it can happen that code points that are in the upper
450 * latin1 range (\x80-\xFF or their equivalents on EBCDIC platforms) match
451 * only if the runtime target string being matched against is UTF-8. For
452 * example /[\w[:punct:]]/d. This happens only for posix classes (with a
d1c40ef5
KW
453 * couple of exceptions, like \d where it doesn't happen), and all such
454 * ones also have above-bitmap matches. Thus, 3) implies 1) as well.
455 * Note that /d rules are no longer encouraged; 'use 5.14' or higher
456 * deselects them. But a flag is required so that they can be properly
457 * handled. But it can be a shared flag: see 5) below.
458 * 4) Also under /d rules, something like /[\Wfoo]/ will match everything in
fe8d1b7c 459 * the \x80-\xFF range, unless the string being matched against is UTF-8.
7a94a5fe
KW
460 * An inversion list could be created for this case, but this is
461 * relatively common, and it turns out that it's all or nothing: if any
462 * one of these code points matches, they all do. Hence a single bit
463 * suffices. We use a shared flag that doesn't take up space by itself:
464 * ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER. This
465 * also implies 1), with one exception: [:^cntrl:].
fe8d1b7c
KW
466 * 5) A user-defined \p{} property may not have been defined by the time the
467 * regex is compiled. In this case, we don't know until runtime what it
468 * will match, so we have to assume it could match anything, including
469 * code points that ordinarily would be in the bitmap. A flag bit is
470 * necessary to indicate this , though it can be shared with the item 3)
471 * flag, as that only occurs under /d, and this only occurs under non-d.
472 * This case is quite uncommon in the field, and the /(?[ ...])/ construct
473 * is a better way to accomplish what this feature does. This case also
474 * implies 1).
475 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP
e2f5e63d 476 * is the shared flag.
fe8d1b7c
KW
477 * 6) /[foo]/il may have folds that are only valid if the runtime locale is a
478 * UTF-8 one. These are quite rare, so it would be good to avoid the
479 * expense of looking for them. But /l matching is slow anyway, and we've
e2f5e63d 480 * traditionally not worried too much about its performance. And this
037715a6 481 * condition requires the ANYOFL_FOLD flag to be set, so testing for
fe8d1b7c 482 * that flag would be sufficient to rule out most cases of this. So it is
d1c40ef5
KW
483 * unclear if this should have a flag or not. But, this flag can be
484 * shared with another, so it doesn't occupy extra space.
d2f465e9 485 *
d1c40ef5 486 * At the moment, there is one spare bit, but this could be increased by
93539530 487 * various tricks:
e2f5e63d 488 *
93539530 489 * If just one more bit is needed, as of this writing it seems to khw that the
d1c40ef5
KW
490 * best choice would be to make ANYOF_MATCHES_ALL_ABOVE_BITMAP not a flag, but
491 * something like
d2f465e9 492 *
d1c40ef5 493 * #define ANYOF_MATCHES_ALL_ABOVE_BITMAP ((U32) -2)
fe8d1b7c 494 *
d1c40ef5
KW
495 * and access it through the ARG like ANYOF_ONLY_HAS_BITMAP is. This flag is
496 * used by all ANYOF node types, and it could be used to avoid calling the
497 * handler function, as the macro REGINCLASS in regexec.c does now for other
498 * cases.
499 *
46fc0c43
KW
500 * Another possibility is based on the fact that ANYOF_MATCHES_POSIXL is
501 * redundant with the node type ANYOFPOSIXL. That flag could be removed, but
502 * at the expense of extra code in regexec.c. The flag has been retained
503 * because it allows us to see if we need to call reginsert, or just use the
504 * bitmap in one test.
fe8d1b7c 505 *
46fc0c43
KW
506 * If this is done, an extension would be to make all ANYOFL nodes contain the
507 * extra 32 bits that ANYOFPOSIXL ones do. The posix flags only occupy 30
508 * bits, so the ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD flags
509 * and ANYOFL_FOLD could be moved to that extra space, but it would mean extra
510 * instructions, as there are currently places in the code that assume those
511 * two bits are zero.
d1c40ef5
KW
512 *
513 * All told, 5 bits could be available for other uses if all of the above were
514 * done.
a4525e78 515 *
d1c40ef5 516 * Some flags are not used in synthetic start class (SSC) nodes, so could be
93e92956
KW
517 * shared should new flags be needed for SSCs, like SSC_MATCHES_EMPTY_STRING
518 * now. */
b8c5462f 519
d2f465e9
KW
520/* If this is set, the result of the match should be complemented. regexec.c
521 * is expecting this to be in the low bit. Never in an SSC */
93e92956 522#define ANYOF_INVERT 0x01
1984f6a0 523
a0dd4231 524/* For the SSC node only, which cannot be inverted, so is shared with that bit.
93e92956
KW
525 * This is used only during regex compilation. */
526#define SSC_MATCHES_EMPTY_STRING ANYOF_INVERT
a0dd4231 527
509d4de0
KW
528/* Set if this is a regnode_charclass_posixl vs a regnode_charclass. This
529 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale
530 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved
531 * at compile-time. Only set under /l; can be in SSC */
532#define ANYOF_MATCHES_POSIXL 0x02
b63f73da
KW
533
534/* The fold is calculated and stored in the bitmap where possible at compile
535 * time. However under locale, the actual folding varies depending on
536 * what the locale is at the time of execution, so it has to be deferred until
d2f465e9 537 * then. Only set under /l; never in an SSC */
037715a6 538#define ANYOFL_FOLD 0x04
b63f73da 539
d1c40ef5 540/* Shared bit set only with ANYOFL and SSC nodes:
269bb65b
KW
541 * If ANYOFL_FOLD is set, this flag indicates there are potential matches
542 * valid only if the locale is a UTF-8 one.
543 * If ANYOFL_FOLD is NOT set, this flag means to warn if the runtime locale
d1c40ef5
KW
544 * isn't a UTF-8 one (and the generated node assumes a UTF-8 locale).
545 * None of INVERT, POSIXL,
546 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP
547 * can be set. */
548#define ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD 0x08
549
550/* Convenience macros for teasing apart the meanings when reading the above bit
551 * */
552#define ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags) \
553 ((flags & ( ANYOFL_FOLD /* Both bits are set */ \
554 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \
555 == ( ANYOFL_FOLD \
556 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD))
557
558#define ANYOFL_UTF8_LOCALE_REQD(flags) \
559 ((flags & ( ANYOFL_FOLD /* Only REQD bit is set */ \
560 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \
561 == ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)
562
1744d18d 563/* Spare: Be sure to change ANYOF_FLAGS_ALL if this gets used 0x10 */
c8d3cd88 564
509d4de0
KW
565/* If set, the node matches every code point NUM_ANYOF_CODE_POINTS and above.
566 * Can be in an SSC */
b3b1cf17 567#define ANYOF_MATCHES_ALL_ABOVE_BITMAP 0x20
ef87b810 568
108316fb
KW
569/* Shared bit:
570 * Under /d it means the ANYOFD node matches more things if the target
571 * string is encoded in UTF-8; any such things will be non-ASCII,
7a94a5fe
KW
572 * characters that are < 256, and can be accessed via the inversion
573 * list.
108316fb
KW
574 * When not under /d, it means the ANYOF node contains a user-defined
575 * property that wasn't yet defined at the time the regex was compiled,
7a94a5fe 576 * and so must be looked up at runtime, by creating an inversion list.
108316fb
KW
577 * (These uses are mutually exclusive because a user-defined property is
578 * specified by \p{}, and \p{} implies /u which deselects /d). The long macro
579 * name is to make sure that you are cautioned about its shared nature. Only
580 * the non-/d meaning can be in an SSC */
581#define ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 0x40
f240c685
KW
582
583/* Shared bit:
d2f465e9 584 * Under /d it means the ANYOFD node matches all non-ASCII Latin1
f240c685
KW
585 * characters when the target string is not in utf8.
586 * When not under /d, it means the ANYOF node should raise a warning if
587 * matching against an above-Unicode code point.
588 * (These uses are mutually exclusive because the warning requires a \p{}, and
589 * \p{} implies /u which deselects /d). An SSC node only has this bit set if
590 * what is meant is the warning. The long macro name is to make sure that you
591 * are cautioned about its shared nature */
592#define ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER 0x80
7bc66b18 593
b271ede1 594#define ANYOF_FLAGS_ALL (0xff & ~0x10)
653099ff 595
037715a6 596#define ANYOF_LOCALE_FLAGS (ANYOFL_FOLD | ANYOF_MATCHES_POSIXL)
ace6b0e4 597
eff8b7dc
KW
598/* These are the flags that apply to both regular ANYOF nodes and synthetic
599 * start class nodes during construction of the SSC. During finalization of
d2f465e9 600 * the SSC, other of the flags may get added to it */
d1c40ef5 601#define ANYOF_COMMON_FLAGS 0
eff8b7dc 602
936ed897 603/* Character classes for node->classflags of ANYOF */
653099ff 604/* Should be synchronized with a table in regprop() */
575a8a29 605/* 2n should be the normal one, paired with its complement at 2n+1 */
b8c5462f 606
08ada29e
KW
607#define ANYOF_ALPHA ((_CC_ALPHA) * 2)
608#define ANYOF_NALPHA ((ANYOF_ALPHA) + 1)
e8d596e0
KW
609#define ANYOF_ALPHANUMERIC ((_CC_ALPHANUMERIC) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */
610#define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1)
08ada29e
KW
611#define ANYOF_ASCII ((_CC_ASCII) * 2)
612#define ANYOF_NASCII ((ANYOF_ASCII) + 1)
e8d596e0
KW
613#define ANYOF_BLANK ((_CC_BLANK) * 2) /* GNU extension: space and tab: non-vertical space */
614#define ANYOF_NBLANK ((ANYOF_BLANK) + 1)
b0d691b2
KW
615#define ANYOF_CASED ((_CC_CASED) * 2) /* Pseudo class for [:lower:] or
616 [:upper:] under /i */
617#define ANYOF_NCASED ((ANYOF_CASED) + 1)
08ada29e
KW
618#define ANYOF_CNTRL ((_CC_CNTRL) * 2)
619#define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1)
e8d596e0
KW
620#define ANYOF_DIGIT ((_CC_DIGIT) * 2) /* \d */
621#define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1)
08ada29e
KW
622#define ANYOF_GRAPH ((_CC_GRAPH) * 2)
623#define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1)
624#define ANYOF_LOWER ((_CC_LOWER) * 2)
625#define ANYOF_NLOWER ((ANYOF_LOWER) + 1)
626#define ANYOF_PRINT ((_CC_PRINT) * 2)
627#define ANYOF_NPRINT ((ANYOF_PRINT) + 1)
628#define ANYOF_PUNCT ((_CC_PUNCT) * 2)
629#define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1)
e8d596e0
KW
630#define ANYOF_SPACE ((_CC_SPACE) * 2) /* \s */
631#define ANYOF_NSPACE ((ANYOF_SPACE) + 1)
08ada29e
KW
632#define ANYOF_UPPER ((_CC_UPPER) * 2)
633#define ANYOF_NUPPER ((ANYOF_UPPER) + 1)
e8d596e0
KW
634#define ANYOF_WORDCHAR ((_CC_WORDCHAR) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */
635#define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1)
08ada29e
KW
636#define ANYOF_XDIGIT ((_CC_XDIGIT) * 2)
637#define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1)
b8c5462f 638
87cad4c0 639/* pseudo classes below this, not stored in the class bitmap, but used as flags
e1d1eefb
YO
640 during compilation of char classes */
641
b2ef2e34
KW
642#define ANYOF_VERTWS ((_CC_VERTSPACE) * 2)
643#define ANYOF_NVERTWS ((ANYOF_VERTWS)+1)
e1d1eefb 644
b2ef2e34
KW
645/* It is best if this is the last one, as all above it are stored as bits in a
646 * bitmap, and it isn't part of that bitmap */
647#if _CC_VERTSPACE != _HIGHEST_REGCOMP_DOT_H_SYNC
648# error Problem with handy.h _HIGHEST_REGCOMP_DOT_H_SYNC #define
649#endif
650
8efd3f97
KW
651#define ANYOF_POSIXL_MAX (ANYOF_VERTWS) /* So upper loop limit is written:
652 * '< ANYOF_MAX'
653 * Hence doesn't include VERTWS, as that
654 * is a pseudo class */
655#define ANYOF_MAX ANYOF_POSIXL_MAX
656
657#if (ANYOF_POSIXL_MAX > 32) /* Must fit in 32-bit word */
b2ef2e34 658# error Problem with handy.h _CC_foo #defines
a0947d7b
KW
659#endif
660
8efd3f97
KW
661#define ANYOF_HORIZWS ((ANYOF_POSIXL_MAX)+2) /* = (ANYOF_NVERTWS + 1) */
662#define ANYOF_NHORIZWS ((ANYOF_POSIXL_MAX)+3)
87cad4c0 663
8efd3f97
KW
664#define ANYOF_UNIPROP ((ANYOF_POSIXL_MAX)+4) /* Used to indicate a Unicode
665 property: \p{} or \P{} */
f39a4dc2 666
b8c5462f
JH
667/* Backward source code compatibility. */
668
669#define ANYOF_ALNUML ANYOF_ALNUM
670#define ANYOF_NALNUML ANYOF_NALNUM
671#define ANYOF_SPACEL ANYOF_SPACE
672#define ANYOF_NSPACEL ANYOF_NSPACE
8ebfc29a
KW
673#define ANYOF_ALNUM ANYOF_WORDCHAR
674#define ANYOF_NALNUM ANYOF_NWORDCHAR
b8c5462f
JH
675
676/* Utility macros for the bitmap and classes of ANYOF */
677
936ed897 678#define ANYOF_FLAGS(p) ((p)->flags)
b8c5462f 679
bc51fd78 680#define ANYOF_BIT(c) (1U << ((c) & 7))
b8c5462f 681
14ed74f6
KW
682#define POSIXL_SET(field, c) ((field) |= (1U << (c)))
683#define ANYOF_POSIXL_SET(p, c) POSIXL_SET(((regnode_charclass_posixl*) (p))->classflags, (c))
8efd3f97 684
14ed74f6
KW
685#define POSIXL_CLEAR(field, c) ((field) &= ~ (1U <<(c)))
686#define ANYOF_POSIXL_CLEAR(p, c) POSIXL_CLEAR(((regnode_charclass_posixl*) (p))->classflags, (c))
8efd3f97 687
14ed74f6
KW
688#define POSIXL_TEST(field, c) ((field) & (1U << (c)))
689#define ANYOF_POSIXL_TEST(p, c) POSIXL_TEST(((regnode_charclass_posixl*) (p))->classflags, (c))
b8c5462f 690
14ed74f6
KW
691#define POSIXL_ZERO(field) STMT_START { (field) = 0; } STMT_END
692#define ANYOF_POSIXL_ZERO(ret) POSIXL_ZERO(((regnode_charclass_posixl*) (ret))->classflags)
693
694#define ANYOF_POSIXL_SET_TO_BITMAP(p, bits) \
695 STMT_START { \
696 ((regnode_charclass_posixl*) (p))->classflags = (bits); \
697 } STMT_END
9cf78fa0
KW
698
699/* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */
d223e1ea 700#define ANYOF_POSIXL_SETALL(ret) STMT_START { ((regnode_charclass_posixl*) (ret))->classflags = nBIT_MASK(ANYOF_POSIXL_MAX); } STMT_END
8efd3f97 701#define ANYOF_CLASS_SETALL(ret) ANYOF_POSIXL_SETALL(ret)
9cf78fa0 702
8efd3f97 703#define ANYOF_POSIXL_TEST_ANY_SET(p) \
93e92956 704 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
d1d040e5 705 && (((regnode_charclass_posixl*)(p))->classflags))
8efd3f97 706#define ANYOF_CLASS_TEST_ANY_SET(p) ANYOF_POSIXL_TEST_ANY_SET(p)
e867ce45 707
e0e1be5f
KW
708/* Since an SSC always has this field, we don't have to test for that; nor do
709 * we want to because the bit isn't set for SSC during its construction */
710#define ANYOF_POSIXL_SSC_TEST_ANY_SET(p) \
711 cBOOL(((regnode_ssc*)(p))->classflags)
712#define ANYOF_POSIXL_SSC_TEST_ALL_SET(p) /* Are all bits set? */ \
713 (((regnode_ssc*) (p))->classflags \
d223e1ea 714 == nBIT_MASK(ANYOF_POSIXL_MAX))
e0e1be5f
KW
715
716#define ANYOF_POSIXL_TEST_ALL_SET(p) \
d223e1ea 717 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
e0e1be5f 718 && ((regnode_charclass_posixl*) (p))->classflags \
d223e1ea 719 == nBIT_MASK(ANYOF_POSIXL_MAX))
fb38762f 720
b27fd97a 721#define ANYOF_POSIXL_OR(source, dest) STMT_START { (dest)->classflags |= (source)->classflags ; } STMT_END
8efd3f97 722#define ANYOF_CLASS_OR(source, dest) ANYOF_POSIXL_OR((source), (dest))
936ed897 723
cdd87c1d
KW
724#define ANYOF_POSIXL_AND(source, dest) STMT_START { (dest)->classflags &= (source)->classflags ; } STMT_END
725
428c1aff
KW
726#define ANYOF_BITMAP_ZERO(ret) Zero(((regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char)
727#define ANYOF_BITMAP(p) ((regnode_charclass*)(p))->bitmap
4cbce0a6 728#define ANYOF_BITMAP_BYTE(p, c) BITMAP_BYTE(ANYOF_BITMAP(p), c)
b8c5462f
JH
729#define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c))
730#define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c))
bc51fd78 731#define ANYOF_BITMAP_TEST(p, c) cBOOL(ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c))
b8c5462f 732
f8bef550
NC
733#define ANYOF_BITMAP_SETALL(p) \
734 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE)
735#define ANYOF_BITMAP_CLEARALL(p) \
736 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE)
f8bef550 737
a687059c
LW
738/*
739 * Utility definitions.
740 */
2304df62 741#ifndef CHARMASK
075abff3 742# define UCHARAT(p) ((int)*(const U8*)(p))
a687059c 743#else
075abff3 744# define UCHARAT(p) ((int)*(p)&CHARMASK)
a687059c 745#endif
a687059c 746
c45abc0a
KW
747/* Number of regnode equivalents that 'guy' occupies beyond the size of the
748 * smallest regnode. */
c277df42
IZ
749#define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode))
750
e384d5c1
YO
751#define REG_ZERO_LEN_SEEN 0x00000001
752#define REG_LOOKBEHIND_SEEN 0x00000002
753#define REG_GPOS_SEEN 0x00000004
471f5387 754/* spare */
e384d5c1
YO
755#define REG_RECURSE_SEEN 0x00000020
756#define REG_TOP_LEVEL_BRANCHES_SEEN 0x00000040
757#define REG_VERBARG_SEEN 0x00000080
758#define REG_CUTGROUP_SEEN 0x00000100
759#define REG_RUN_ON_COMMENT_SEEN 0x00000200
760#define REG_UNFOLDED_MULTI_SEEN 0x00000400
d5a00e4a 761/* spare */
ee273784
YO
762#define REG_UNBOUNDED_QUANTIFIER_SEEN 0x00001000
763
d09b2d29 764
73c4f7a1
GS
765START_EXTERN_C
766
be8e71aa
YO
767#ifdef PLUGGABLE_RE_EXTENSION
768#include "re_nodes.h"
769#else
d09b2d29 770#include "regnodes.h"
be8e71aa 771#endif
d09b2d29 772
f9f4320a
YO
773#ifndef PLUGGABLE_RE_EXTENSION
774#ifndef DOINIT
775EXTCONST regexp_engine PL_core_reg_engine;
776#else /* DOINIT */
777EXTCONST regexp_engine PL_core_reg_engine = {
fe578d7f 778 Perl_re_compile,
2fdbfb4d 779 Perl_regexec_flags,
f9f4320a
YO
780 Perl_re_intuit_start,
781 Perl_re_intuit_string,
2fdbfb4d
AB
782 Perl_regfree_internal,
783 Perl_reg_numbered_buff_fetch,
784 Perl_reg_numbered_buff_store,
785 Perl_reg_numbered_buff_length,
192b9cd1
AB
786 Perl_reg_named_buff,
787 Perl_reg_named_buff_iter,
49d7dfbc 788 Perl_reg_qr_package,
f9f4320a 789#if defined(USE_ITHREADS)
3c13cae6 790 Perl_regdupe_internal,
f9f4320a 791#endif
3c13cae6 792 Perl_re_op_compile
f9f4320a
YO
793};
794#endif /* DOINIT */
795#endif /* PLUGGABLE_RE_EXTENSION */
796
797
73c4f7a1 798END_EXTERN_C
cad2e5aa 799
cad2e5aa 800
261faec3
MJD
801/* .what is a character array with one character for each member of .data
802 * The character describes the function of the corresponding .data item:
af534a04 803 * a - AV for paren_name_list under DEBUGGING
0111c4fd 804 * f - start-class data for regstclass optimization
68e2671b 805 * l - start op for literal (?{EVAL}) item
d63c20f2 806 * L - start op for literal (?{EVAL}) item, with separate CV (qr//)
b30fcab9 807 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/
7a94a5fe
KW
808 * s - inversion list for Unicode-style character class, and the
809 * multicharacter strings resulting from casefolding the single-character
810 * entries in the character class
a3621e74 811 * t - trie struct
55eed653 812 * u - trie struct's widecharmap (a HV, so can't share, must dup)
2b8b4781 813 * also used for revcharmap and words under DEBUGGING
07be1b83 814 * T - aho-trie struct
81714fb9 815 * S - sv for named capture lookup
261faec3 816 * 20010712 mjd@plover.com
c3613ed7 817 * (Remember to update re_dup() and pregfree() if you add any items.)
261faec3 818 */
cad2e5aa
JH
819struct reg_data {
820 U32 count;
821 U8 *what;
822 void* data[1];
823};
824
a1cac82e
NC
825/* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses
826 anchored* and float* via array indexes 0 and 1. */
cad2e5aa 827#define anchored_substr substrs->data[0].substr
33b8afdf 828#define anchored_utf8 substrs->data[0].utf8_substr
cad2e5aa 829#define anchored_offset substrs->data[0].min_offset
1de06328
YO
830#define anchored_end_shift substrs->data[0].end_shift
831
cad2e5aa 832#define float_substr substrs->data[1].substr
33b8afdf 833#define float_utf8 substrs->data[1].utf8_substr
cad2e5aa
JH
834#define float_min_offset substrs->data[1].min_offset
835#define float_max_offset substrs->data[1].max_offset
1de06328
YO
836#define float_end_shift substrs->data[1].end_shift
837
cad2e5aa 838#define check_substr substrs->data[2].substr
33b8afdf 839#define check_utf8 substrs->data[2].utf8_substr
cad2e5aa
JH
840#define check_offset_min substrs->data[2].min_offset
841#define check_offset_max substrs->data[2].max_offset
1de06328 842#define check_end_shift substrs->data[2].end_shift
a3621e74 843
8d919b0a
FC
844#define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr)
845#define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8)
846#define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr)
847#define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8)
a3621e74
YO
848
849/* trie related stuff */
5d9a96ca 850
a3621e74
YO
851/* a transition record for the state machine. the
852 check field determines which state "owns" the
853 transition. the char the transition is for is
854 determined by offset from the owning states base
855 field. the next field determines which state
856 is to be transitioned to if any.
857*/
858struct _reg_trie_trans {
859 U32 next;
860 U32 check;
861};
862
863/* a transition list element for the list based representation */
864struct _reg_trie_trans_list_elem {
865 U16 forid;
866 U32 newstate;
867};
868typedef struct _reg_trie_trans_list_elem reg_trie_trans_le;
869
870/* a state for compressed nodes. base is an offset
871 into an array of reg_trie_trans array. If wordnum is
872 nonzero the state is accepting. if base is zero then
873 the state has no children (and will be accepting)
874*/
875struct _reg_trie_state {
876 U16 wordnum;
877 union {
878 U32 base;
879 reg_trie_trans_le* list;
880 } trans;
881};
882
2e64971a
DM
883/* info per word; indexed by wordnum */
884typedef struct {
885 U16 prev; /* previous word in acceptance chain; eg in
886 * zzz|abc|ab/ after matching the chars abc, the
887 * accepted word is #2, and the previous accepted
888 * word is #3 */
889 U32 len; /* how many chars long is this word? */
890 U32 accept; /* accept state for this word */
891} reg_trie_wordinfo;
a3621e74
YO
892
893
a3621e74
YO
894typedef struct _reg_trie_state reg_trie_state;
895typedef struct _reg_trie_trans reg_trie_trans;
896
897
898/* anything in here that needs to be freed later
23eab42c
NC
899 should be dealt with in pregfree.
900 refcount is first in both this and _reg_ac_data to allow a space
901 optimisation in Perl_regdupe. */
a3621e74 902struct _reg_trie_data {
23eab42c 903 U32 refcount; /* number of times this trie is referenced */
786e8c11
YO
904 U32 lasttrans; /* last valid transition element */
905 U16 *charmap; /* byte to charid lookup array */
786e8c11
YO
906 reg_trie_state *states; /* state data */
907 reg_trie_trans *trans; /* array of transition elements */
908 char *bitmap; /* stclass bitmap */
786e8c11
YO
909 U16 *jump; /* optional 1 indexed array of offsets before tail
910 for the node following a given word. */
2e64971a 911 reg_trie_wordinfo *wordinfo; /* array of info per word */
1b5c067c
NC
912 U16 uniquecharcount; /* unique chars in trie (width of trans table) */
913 U32 startstate; /* initial state - used for common prefix optimisation */
914 STRLEN minlen; /* minimum length of words in trie - build/opt only? */
915 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */
2e64971a 916 U32 prefixlen; /* #chars in common prefix */
1e2e3d02
YO
917 U32 statecount; /* Build only - number of states in the states array
918 (including the unused zero state) */
786e8c11 919 U32 wordcount; /* Build only */
a3621e74 920#ifdef DEBUGGING
786e8c11 921 STRLEN charcount; /* Build only */
a3621e74
YO
922#endif
923};
2b8b4781
NC
924/* There is one (3 under DEBUGGING) pointers that logically belong in this
925 structure, but are held outside as they need duplication on thread cloning,
926 whereas the rest of the structure can be read only:
927 HV *widecharmap; code points > 255 to charid
928#ifdef DEBUGGING
929 AV *words; Array of words contained in trie, for dumping
930 AV *revcharmap; Map of each charid back to its character representation
931#endif
932*/
933
934#define TRIE_WORDS_OFFSET 2
935
a3621e74
YO
936typedef struct _reg_trie_data reg_trie_data;
937
23eab42c
NC
938/* refcount is first in both this and _reg_trie_data to allow a space
939 optimisation in Perl_regdupe. */
07be1b83 940struct _reg_ac_data {
23eab42c 941 U32 refcount;
1b5c067c 942 U32 trie;
07be1b83
YO
943 U32 *fail;
944 reg_trie_state *states;
07be1b83
YO
945};
946typedef struct _reg_ac_data reg_ac_data;
947
486ec47a 948/* ANY_BIT doesn't use the structure, so we can borrow it here.
3dab1dad
YO
949 This is simpler than refactoring all of it as wed end up with
950 three different sets... */
951
952#define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap)
4cbce0a6 953#define TRIE_BITMAP_BYTE(p, c) BITMAP_BYTE(TRIE_BITMAP(p), c)
5dfe062f
YO
954#define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c))
955#define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c))
956#define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
3dab1dad 957
1de06328
YO
958#define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC)
959#define IS_TRIE_AC(op) ((op)>=AHOCORASICK)
960
961
0effac60 962#define BITMAP_BYTE(p, c) (( (U8*) p) [ ( ( (UV) (c)) >> 3) ] )
786e8c11 963#define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
3dab1dad 964
a3621e74
YO
965/* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */
966#define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1)
967#define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state))
968#define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state))
969
3dab1dad 970#ifdef DEBUGGING
3dab1dad 971#define TRIE_CHARCOUNT(trie) ((trie)->charcount)
3dab1dad 972#else
3dab1dad 973#define TRIE_CHARCOUNT(trie) (trie_charcount)
3dab1dad 974#endif
a3621e74 975
0111c4fd
RGS
976#define RE_TRIE_MAXBUF_INIT 65536
977#define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF"
a3621e74
YO
978#define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS"
979
6ef7fe53
KW
980#define RE_COMPILE_RECURSION_INIT 1000
981#define RE_COMPILE_RECURSION_LIMIT "\022E_COMPILE_RECURSION_LIMIT"
982
be8e71aa
YO
983/*
984
985RE_DEBUG_FLAGS is used to control what debug output is emitted
986its divided into three groups of options, some of which interact.
987The three groups are: Compile, Execute, Extra. There is room for a
988further group, as currently only the low three bytes are used.
989
990 Compile Options:
07be1b83 991
be8e71aa
YO
992 PARSE
993 PEEP
994 TRIE
995 PROGRAM
996 OFFSETS
a3621e74 997
be8e71aa 998 Execute Options:
a3621e74 999
be8e71aa
YO
1000 INTUIT
1001 MATCH
1002 TRIE
a3621e74 1003
be8e71aa 1004 Extra Options
a3621e74 1005
be8e71aa
YO
1006 TRIE
1007 OFFSETS
1008
1009If you modify any of these make sure you make corresponding changes to
1010re.pm, especially to the documentation.
a3621e74 1011
be8e71aa
YO
1012*/
1013
1014
1015/* Compile */
1016#define RE_DEBUG_COMPILE_MASK 0x0000FF
1017#define RE_DEBUG_COMPILE_PARSE 0x000001
1018#define RE_DEBUG_COMPILE_OPTIMISE 0x000002
1019#define RE_DEBUG_COMPILE_TRIE 0x000004
1020#define RE_DEBUG_COMPILE_DUMP 0x000008
f7819f85 1021#define RE_DEBUG_COMPILE_FLAGS 0x000010
d9a72fcc 1022#define RE_DEBUG_COMPILE_TEST 0x000020
be8e71aa
YO
1023
1024/* Execute */
1025#define RE_DEBUG_EXECUTE_MASK 0x00FF00
1026#define RE_DEBUG_EXECUTE_INTUIT 0x000100
1027#define RE_DEBUG_EXECUTE_MATCH 0x000200
1028#define RE_DEBUG_EXECUTE_TRIE 0x000400
1029
1030/* Extra */
51bd1941 1031#define RE_DEBUG_EXTRA_MASK 0x3FF0000
fd41b2d1
KW
1032#define RE_DEBUG_EXTRA_TRIE 0x0010000
1033#define RE_DEBUG_EXTRA_OFFSETS 0x0020000
1034#define RE_DEBUG_EXTRA_OFFDEBUG 0x0040000
1035#define RE_DEBUG_EXTRA_STATE 0x0080000
1036#define RE_DEBUG_EXTRA_OPTIMISE 0x0100000
1037#define RE_DEBUG_EXTRA_BUFFERS 0x0400000
1038#define RE_DEBUG_EXTRA_GPOS 0x0800000
1039#define RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE 0x1000000
51bd1941 1040#define RE_DEBUG_EXTRA_WILDCARD 0x2000000
24b23f37 1041/* combined */
fd41b2d1 1042#define RE_DEBUG_EXTRA_STACK 0x0280000
be8e71aa 1043
55a0b497 1044#define RE_DEBUG_FLAG(x) (re_debug_flags & (x))
be8e71aa
YO
1045/* Compile */
1046#define DEBUG_COMPILE_r(x) DEBUG_r( \
55a0b497 1047 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_MASK)) x )
be8e71aa 1048#define DEBUG_PARSE_r(x) DEBUG_r( \
55a0b497 1049 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_PARSE)) x )
be8e71aa 1050#define DEBUG_OPTIMISE_r(x) DEBUG_r( \
55a0b497 1051 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE)) x )
be8e71aa 1052#define DEBUG_DUMP_r(x) DEBUG_r( \
55a0b497 1053 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_DUMP)) x )
be8e71aa 1054#define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \
55a0b497 1055 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TRIE)) x )
f7819f85 1056#define DEBUG_FLAGS_r(x) DEBUG_r( \
55a0b497 1057 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_FLAGS)) x )
d9a72fcc 1058#define DEBUG_TEST_r(x) DEBUG_r( \
55a0b497 1059 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TEST)) x )
be8e71aa
YO
1060/* Execute */
1061#define DEBUG_EXECUTE_r(x) DEBUG_r( \
55a0b497 1062 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_MASK)) x )
be8e71aa 1063#define DEBUG_INTUIT_r(x) DEBUG_r( \
55a0b497 1064 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_INTUIT)) x )
be8e71aa 1065#define DEBUG_MATCH_r(x) DEBUG_r( \
55a0b497 1066 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_MATCH)) x )
be8e71aa 1067#define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \
55a0b497 1068 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_TRIE)) x )
be8e71aa
YO
1069
1070/* Extra */
1071#define DEBUG_EXTRA_r(x) DEBUG_r( \
55a0b497 1072 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_MASK)) x )
a5ca303d 1073#define DEBUG_OFFSETS_r(x) DEBUG_r( \
55a0b497 1074 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_OFFSETS)) x )
ab3bbdeb 1075#define DEBUG_STATE_r(x) DEBUG_r( \
55a0b497 1076 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_STATE)) x )
24b23f37 1077#define DEBUG_STACK_r(x) DEBUG_r( \
55a0b497 1078 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_STACK)) x )
e7707071 1079#define DEBUG_BUFFERS_r(x) DEBUG_r( \
55a0b497 1080 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_BUFFERS)) x )
e7707071 1081
a5ca303d 1082#define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \
dafb2544 1083 if (DEBUG_v_TEST || ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \
55a0b497 1084 RE_DEBUG_FLAG(RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE))) x )
07be1b83 1085#define MJD_OFFSET_DEBUG(x) DEBUG_r( \
55a0b497 1086 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_OFFDEBUG)) \
be8e71aa
YO
1087 Perl_warn_nocontext x )
1088#define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \
55a0b497 1089 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_TRIE)) x )
be8e71aa 1090#define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \
55a0b497 1091 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_TRIE)) x )
be8e71aa
YO
1092
1093#define DEBUG_TRIE_r(x) DEBUG_r( \
55a0b497
KW
1094 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TRIE \
1095 | RE_DEBUG_EXECUTE_TRIE )) x )
2c296965 1096#define DEBUG_GPOS_r(x) DEBUG_r( \
55a0b497 1097 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_GPOS)) x )
be8e71aa 1098
fd41b2d1 1099#define DEBUG_DUMP_PRE_OPTIMIZE_r(x) DEBUG_r( \
55a0b497 1100 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE)) x )
fd41b2d1 1101
be8e71aa 1102/* initialization */
1fe1d354
KW
1103/* Get the debug flags for code not in regcomp.c nor regexec.c. This doesn't
1104 * initialize the variable if it isn't already there, instead it just assumes
1105 * the flags are 0 */
1106#define DECLARE_AND_GET_RE_DEBUG_FLAGS_NON_REGEX \
1107 volatile IV re_debug_flags = 0; PERL_UNUSED_VAR(re_debug_flags); \
1108 STMT_START { \
1109 SV * re_debug_flags_sv = NULL; \
1110 /* get_sv() can return NULL during global destruction. */ \
1111 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \
1112 if (re_debug_flags_sv && SvIOK(re_debug_flags_sv)) \
1113 re_debug_flags=SvIV(re_debug_flags_sv); \
1114 } STMT_END
1115
1116
a3621e74 1117#ifdef DEBUGGING
ab3bbdeb 1118
1fe1d354
KW
1119/* For use in regcomp.c and regexec.c, Get the debug flags, and initialize to
1120 * the defaults if not done already */
619d7873
KW
1121#define DECLARE_AND_GET_RE_DEBUG_FLAGS \
1122 volatile IV re_debug_flags = 0; PERL_UNUSED_VAR(re_debug_flags); \
1123 STMT_START { \
1124 SV * re_debug_flags_sv = NULL; \
1125 /* get_sv() can return NULL during global destruction. */ \
1126 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \
1127 if (re_debug_flags_sv) { \
1128 if (!SvIOK(re_debug_flags_sv)) /* If doesnt exist set to default */\
1129 sv_setuv(re_debug_flags_sv, \
1130 /* These defaults should be kept in sync with re.pm */ \
1131 RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \
1132 re_debug_flags=SvIV(re_debug_flags_sv); \
1133 } \
1134 } STMT_END
ab3bbdeb 1135
51bd1941
KW
1136#define isDEBUG_WILDCARD (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_WILDCARD))
1137
eecd4d11
KW
1138#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \
1139 const char * const rpv = \
1140 pv_pretty((dsv), (pv), (l), (m), \
1141 PL_colors[(c1)],PL_colors[(c2)], \
c89df6cf 1142 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \
0df25f3d 1143 const int rlen = SvCUR(dsv)
ab3bbdeb 1144
8ec77c3c 1145/* This is currently unsed in the core */
eecd4d11
KW
1146#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \
1147 const char * const rpv = \
1148 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \
1149 PL_colors[(c1)],PL_colors[(c2)], \
c89df6cf 1150 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) )
ab3bbdeb 1151
eecd4d11
KW
1152#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \
1153 const char * const rpv = \
1154 pv_pretty((dsv), (pv), (l), (m), \
1155 PL_colors[0], PL_colors[1], \
c89df6cf 1156 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \
ab3bbdeb 1157 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \
0a0e9afc 1158 )
ab3bbdeb
YO
1159
1160#define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0))
1161#define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "")
1162
1163#else /* if not DEBUGGING */
1164
82dc0782
KW
1165#define DECLARE_AND_GET_RE_DEBUG_FLAGS dNOOP
1166#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) dNOOP
ab3bbdeb 1167#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m)
82dc0782 1168#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) dNOOP
ab3bbdeb
YO
1169#define RE_SV_DUMPLEN(ItEm)
1170#define RE_SV_TAIL(ItEm)
51bd1941 1171#define isDEBUG_WILDCARD 0
ab3bbdeb
YO
1172
1173#endif /* DEBUG RELATED DEFINES */
a3621e74 1174
0bc0e44f 1175#define FIRST_NON_ASCII_DECIMAL_DIGIT 0x660 /* ARABIC_INDIC_DIGIT_ZERO */
2ca74f27 1176
64935bc6
KW
1177typedef enum {
1178 TRADITIONAL_BOUND = _CC_WORDCHAR,
ae3bb8ea 1179 GCB_BOUND,
6b659339 1180 LB_BOUND,
06ae2722 1181 SB_BOUND,
ae3bb8ea 1182 WB_BOUND
64935bc6
KW
1183} bound_type;
1184
13fcf652 1185/* This unpacks the FLAGS field of ANYOF[HR]x nodes. The value it contains
3146c00a
KW
1186 * gives the strict lower bound for the UTF-8 start byte of any code point
1187 * matchable by the node, and a loose upper bound as well.
1188 *
1189 * The low bound is stored in the upper 6 bits, plus 0xC0.
1190 * The loose upper bound is determined from the lowest 2 bits and the low bound
1191 * (called x) as follows:
1192 *
1193 * 11 The upper limit of the range can be as much as (EF - x) / 8
1194 * 10 The upper limit of the range can be as much as (EF - x) / 4
1195 * 01 The upper limit of the range can be as much as (EF - x) / 2
1196 * 00 The upper limit of the range can be as much as EF
1197 *
0bf4efd7
KW
1198 * For motivation of this design, see commit message in
1199 * 3146c00a633e9cbed741e10146662fbcedfdb8d3 */
3146c00a
KW
1200#ifdef EBCDIC
1201# define MAX_ANYOF_HRx_BYTE 0xF4
1202#else
1203# define MAX_ANYOF_HRx_BYTE 0xEF
1204#endif
1205#define LOWEST_ANYOF_HRx_BYTE(b) (((b) >> 2) + 0xC0)
1206#define HIGHEST_ANYOF_HRx_BYTE(b) \
1207 (LOWEST_ANYOF_HRx_BYTE(b) \
1208 + ((MAX_ANYOF_HRx_BYTE - LOWEST_ANYOF_HRx_BYTE(b)) >> ((b) & 3)))
1209
3dd7db29
JK
1210#endif /* PERL_REGCOMP_H_ */
1211
e9a8c099 1212/*
14d04a33 1213 * ex: set ts=8 sts=4 sw=4 et:
e9a8c099 1214 */