This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
msgrcv: properly downgrade the receive buffer
[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 10
21d0beeb
KW
11#if ! defined(PERL_REGCOMP_H_) && ( defined(PERL_CORE) \
12 || defined(PERL_EXT_RE_BUILD))
3dd7db29
JK
13#define PERL_REGCOMP_H_
14
e1d1eefb 15#include "regcharclass.h"
a687059c 16
58e23c8d 17/* Convert branch sequences to more efficient trie ops? */
07be1b83 18#define PERL_ENABLE_TRIE_OPTIMISATION 1
58e23c8d 19
486ec47a 20/* Be really aggressive about optimising patterns with trie sequences? */
07be1b83 21#define PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION 1
58e23c8d
YO
22
23/* Should the optimiser take positive assertions into account? */
68ba3a3f 24#define PERL_ENABLE_POSITIVE_ASSERTION_STUDY 0
58e23c8d
YO
25
26/* Not for production use: */
07be1b83 27#define PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS 0
58e23c8d 28
7122b237
YO
29/* Activate offsets code - set to if 1 to enable */
30#ifdef DEBUGGING
31#define RE_TRACK_PATTERN_OFFSETS
32#endif
33
a687059c 34/*
a687059c
LW
35 * Structure for regexp "program". This is essentially a linear encoding
36 * of a nondeterministic finite-state machine (aka syntax charts or
37 * "railroad normal form" in parsing technology). Each node is an opcode
38 * plus a "next" pointer, possibly plus an operand. "Next" pointers of
39 * all nodes except BRANCH implement concatenation; a "next" pointer with
40 * a BRANCH on both ends of it is connecting two alternatives. (Here we
41 * have one of the subtle syntax dependencies: an individual BRANCH (as
42 * opposed to a collection of them) is never concatenated with anything
43 * because of operator precedence.) The operand of some types of node is
44 * a literal string; for others, it is a node leading into a sub-FSM. In
45 * particular, the operand of a BRANCH node is the first node of the branch.
46 * (NB this is *not* a tree structure: the tail of the branch connects
a3621e74
YO
47 * to the thing following the set of BRANCHes.) The opcodes are defined
48 * in regnodes.h which is generated from regcomp.sym by regcomp.pl.
a687059c
LW
49 */
50
a687059c
LW
51/*
52 * A node is one char of opcode followed by two chars of "next" pointer.
53 * "Next" pointers are stored as two 8-bit pieces, high order first. The
54 * value is a positive offset from the opcode of the node containing it.
55 * An operand, if any, simply follows the node. (Note that much of the
56 * code generation knows about this implicit relationship.)
57 *
58 * Using two bytes for the "next" pointer is vast overkill for most things,
59 * but allows patterns to get big without disasters.
60 *
e91177ed 61 * [The "next" pointer is always aligned on an even
62e6ef33 62 * boundary, and reads the offset directly as a short.]
a687059c 63 */
785a26d5
YO
64
65/* This is the stuff that used to live in regexp.h that was truly
66 private to the engine itself. It now lives here. */
67
785a26d5 68 typedef struct regexp_internal {
7122b237
YO
69 union {
70 U32 *offsets; /* offset annotations 20001228 MJD
71 data about mapping the program to the
72 string -
73 offsets[0] is proglen when this is used
74 */
75 U32 proglen;
76 } u;
77
6d5c990f
RGS
78 regnode *regstclass; /* Optional startclass as identified or constructed
79 by the optimiser */
80 struct reg_data *data; /* Additional miscellaneous data used by the program.
81 Used to make it easier to clone and free arbitrary
82 data that the regops need. Often the ARG field of
83 a regop is an index into this structure */
1acab4c5 84 struct reg_code_blocks *code_blocks;/* positions of literal (?{}) */
bba699a8 85 int name_list_idx; /* Optional data index of an array of paren names */
6d5c990f
RGS
86 regnode program[1]; /* Unwarranted chumminess with compiler. */
87} regexp_internal;
88
89#define RXi_SET(x,y) (x)->pprivate = (void*)(y)
90#define RXi_GET(x) ((regexp_internal *)((x)->pprivate))
91#define RXi_GET_DECL(r,ri) regexp_internal *ri = RXi_GET(r)
785a26d5
YO
92/*
93 * Flags stored in regexp->intflags
94 * These are used only internally to the regexp engine
95 *
96 * See regexp.h for flags used externally to the regexp engine
97 */
e3e400ec
YO
98#define RXp_INTFLAGS(rx) ((rx)->intflags)
99#define RX_INTFLAGS(prog) RXp_INTFLAGS(ReANY(prog))
100
785a26d5
YO
101#define PREGf_SKIP 0x00000001
102#define PREGf_IMPLICIT 0x00000002 /* Converted .* to ^.* */
103#define PREGf_NAUGHTY 0x00000004 /* how exponential is this pattern? */
104#define PREGf_VERBARG_SEEN 0x00000008
105#define PREGf_CUTGROUP_SEEN 0x00000010
732caac7 106#define PREGf_USE_RE_EVAL 0x00000020 /* compiled with "use re 'eval'" */
e3e400ec
YO
107/* these used to be extflags, but are now intflags */
108#define PREGf_NOSCAN 0x00000040
33c28ab2 109 /* spare */
8e1490ee
YO
110#define PREGf_GPOS_SEEN 0x00000100
111#define PREGf_GPOS_FLOAT 0x00000200
785a26d5 112
d3d47aac
YO
113#define PREGf_ANCH_MBOL 0x00000400
114#define PREGf_ANCH_SBOL 0x00000800
115#define PREGf_ANCH_GPOS 0x00001000
d5a00e4a 116#define PREGf_RECURSE_SEEN 0x00002000
8e1490ee 117
d3d47aac
YO
118#define PREGf_ANCH \
119 ( PREGf_ANCH_SBOL | PREGf_ANCH_GPOS | PREGf_ANCH_MBOL )
785a26d5
YO
120
121/* this is where the old regcomp.h started */
a687059c 122
c277df42 123struct regnode_string {
cd439c50 124 U8 str_len;
c277df42
IZ
125 U8 type;
126 U16 next_off;
cd439c50 127 char string[1];
c277df42 128};
a687059c 129
ae06e581
KW
130struct regnode_lstring { /* Constructed this way to keep the string aligned. */
131 U8 flags;
132 U8 type;
133 U16 next_off;
1d425dad 134 U32 str_len; /* Only 18 bits allowed before would overflow 'next_off' */
ae06e581
KW
135 char string[1];
136};
137
34924db0 138struct regnode_anyofhs { /* Constructed this way to keep the string aligned. */
53d42e43 139 U8 str_len;
34924db0
KW
140 U8 type;
141 U16 next_off;
142 U32 arg1; /* set by set_ANYOF_arg() */
143 char string[1];
144};
145
6bda09f9
YO
146/* Argument bearing node - workhorse,
147 arg1 is often for the data field */
c277df42
IZ
148struct regnode_1 {
149 U8 flags;
150 U8 type;
151 U16 next_off;
152 U32 arg1;
153};
154
d3e3ece8
KW
155/* Node whose argument is 'SV *'. This needs to be used very carefully in
156 * situations where pointers won't become invalid because of, say re-mallocs */
86451f01
KW
157struct regnode_p {
158 U8 flags;
159 U8 type;
160 U16 next_off;
d3e3ece8 161 SV * arg1;
86451f01
KW
162};
163
6bda09f9
YO
164/* Similar to a regnode_1 but with an extra signed argument */
165struct regnode_2L {
166 U8 flags;
167 U8 type;
168 U16 next_off;
169 U32 arg1;
170 I32 arg2;
171};
172
173/* 'Two field' -- Two 16 bit unsigned args */
c277df42
IZ
174struct regnode_2 {
175 U8 flags;
176 U8 type;
177 U16 next_off;
178 U16 arg1;
179 U16 arg2;
180};
181
dcb20b36 182#define ANYOF_BITMAP_SIZE (NUM_ANYOF_CODE_POINTS / 8) /* 8 bits/Byte */
936ed897 183
1ee208c4
KW
184/* Note that these form structs which are supersets of the next smaller one, by
185 * appending fields. Alignment problems can occur if one of those optional
186 * fields requires stricter alignment than the base struct. And formal
187 * parameters that can really be two or more of the structs should be
188 * declared as the smallest one it could be. See commit message for
189 * 7dcac5f6a5195002b55c935ee1d67f67e1df280b. Regnode allocation is done
190 * without regard to alignment, and changing it to would also require changing
191 * the code that inserts and deletes regnodes. The basic single-argument
192 * regnode has a U32, which is what reganode() allocates as a unit. Therefore
193 * no field can require stricter alignment than U32. */
194
786e8c11 195/* also used by trie */
936ed897
IZ
196struct regnode_charclass {
197 U8 flags;
198 U8 type;
199 U16 next_off;
9da99c5e 200 U32 arg1; /* set by set_ANYOF_arg() */
19103b91 201 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */
936ed897
IZ
202};
203
bf3f2d85 204/* has runtime (locale) \d, \w, ..., [:posix:] classes */
ea461dd6 205struct regnode_charclass_posixl {
93e92956 206 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
936ed897
IZ
207 U8 type;
208 U16 next_off;
7e817188 209 U32 arg1;
03fa83ba 210 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
9cf78fa0 211 U32 classflags; /* and run-time */
936ed897
IZ
212};
213
85c8e306
KW
214/* A synthetic start class (SSC); is a regnode_charclass_posixl_fold, plus an
215 * extra SV*, used only during its construction and which is not used by
216 * regexec.c. Note that the 'next_off' field is unused, as the SSC stands
217 * alone, so there is never a next node. Also, there is no alignment issue,
b8fda935 218 * because these are declared or allocated as a complete unit so the compiler
85c8e306
KW
219 * takes care of alignment. This is unlike the other regnodes which are
220 * allocated in terms of multiples of a single-argument regnode. SSC nodes can
221 * have a pointer field because there is no alignment issue, and because it is
222 * set to NULL after construction, before any cloning of the pattern */
b8f7bb16 223struct regnode_ssc {
93e92956 224 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
b8f7bb16
KW
225 U8 type;
226 U16 next_off;
7e817188 227 U32 arg1;
03fa83ba 228 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
631e9161 229 U32 classflags; /* ... and run-time */
85c8e306
KW
230
231 /* Auxiliary, only used during construction; NULL afterwards: list of code
232 * points matched */
233 SV* invlist;
b8f7bb16
KW
234};
235
71068078
KW
236/* We take advantage of 'next_off' not otherwise being used in the SSC by
237 * actually using it: by setting it to 1. This allows us to test and
238 * distinguish between an SSC and other ANYOF node types, as 'next_off' cannot
239 * otherwise be 1, because it is the offset to the next regnode expressed in
240 * units of regnodes. Since an ANYOF node contains extra fields, it adds up
241 * to 12 regnode units on 32-bit systems, (hence the minimum this can be (if
242 * not 0) is 11 there. Even if things get tightly packed on a 64-bit system,
243 * it still would be more than 1. */
244#define set_ANYOF_SYNTHETIC(n) STMT_START{ OP(n) = ANYOF; \
245 NEXT_OFF(n) = 1; \
246 } STMT_END
c1071334 247#define is_ANYOF_SYNTHETIC(n) (PL_regkind[OP(n)] == ANYOF && NEXT_OFF(n) == 1)
71068078 248
9c7e81e8
AD
249/* XXX fix this description.
250 Impose a limit of REG_INFTY on various pattern matching operations
251 to limit stack growth and to avoid "infinite" recursions.
252*/
4288c5b9
KW
253/* The default size for REG_INFTY is U16_MAX, which is the same as
254 USHORT_MAX (see perl.h). Unfortunately U16 isn't necessarily 16 bits
255 (see handy.h). On the Cray C90, sizeof(short)==4 and hence U16_MAX is
256 ((1<<32)-1), while on the Cray T90, sizeof(short)==8 and U16_MAX is
257 ((1<<64)-1). To limit stack growth to reasonable sizes, supply a
9c7e81e8
AD
258 smaller default.
259 --Andy Dougherty 11 June 1998
260*/
261#if SHORTSIZE > 2
262# ifndef REG_INFTY
1381ccb1 263# define REG_INFTY nBIT_UMAX(16)
9c7e81e8
AD
264# endif
265#endif
266
267#ifndef REG_INFTY
4288c5b9 268# define REG_INFTY U16_MAX
83cfe48c 269#endif
a687059c 270
e91177ed
IZ
271#define ARG_VALUE(arg) (arg)
272#define ARG__SET(arg,val) ((arg) = (val))
c277df42 273
e8e3b0fb
JH
274#undef ARG
275#undef ARG1
276#undef ARG2
277
c277df42 278#define ARG(p) ARG_VALUE(ARG_LOC(p))
86451f01 279#define ARGp(p) ARG_VALUE(ARGp_LOC(p))
c277df42
IZ
280#define ARG1(p) ARG_VALUE(ARG1_LOC(p))
281#define ARG2(p) ARG_VALUE(ARG2_LOC(p))
6bda09f9 282#define ARG2L(p) ARG_VALUE(ARG2L_LOC(p))
786e8c11 283
c277df42 284#define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val))
86451f01 285#define ARGp_SET(p, val) ARG__SET(ARGp_LOC(p), (val))
c277df42
IZ
286#define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val))
287#define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val))
6bda09f9 288#define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val))
c277df42 289
e8e3b0fb
JH
290#undef NEXT_OFF
291#undef NODE_ALIGN
292
075abff3
AL
293#define NEXT_OFF(p) ((p)->next_off)
294#define NODE_ALIGN(node)
aa48e906
YO
295/* the following define was set to 0xde in 075abff3
296 * as part of some linting logic. I have set it to 0
297 * as otherwise in every place where we /might/ set flags
298 * we have to set it 0 explicitly, which duplicates
299 * assignments and IMO adds an unacceptable level of
300 * surprise to working in the regex engine. If this
301 * is changed from 0 then at the very least make sure
302 * that SBOL for /^/ sets the flags to 0 explicitly.
303 * -- Yves */
304#define NODE_ALIGN_FILL(node) ((node)->flags = 0)
a687059c 305
c277df42
IZ
306#define SIZE_ALIGN NODE_ALIGN
307
e8e3b0fb
JH
308#undef OP
309#undef OPERAND
e8e3b0fb
JH
310#undef STRING
311
e91177ed 312#define OP(p) ((p)->type)
b988e673
KW
313#define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \
314 regnode types. For some, it's the \
315 character set of the regnode */
3f2416ae 316#define STR_LENs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \
5cd61b66 317 ((struct regnode_string *)p)->str_len)
3f2416ae 318#define STRINGs(p) (__ASSERT_(OP(p) != LEXACT && OP(p) != LEXACT_REQ8) \
5cd61b66 319 ((struct regnode_string *)p)->string)
ae06e581
KW
320#define OPERANDs(p) STRINGs(p)
321
322/* Long strings. Currently limited to length 18 bits, which handles a 262000
323 * byte string. The limiting factor is the 16 bit 'next_off' field, which
324 * points to the next regnode, so the furthest away it can be is 2**16. On
325 * most architectures, regnodes are 2**2 bytes long, so that yields 2**18
326 * bytes. Should a longer string be desired, we could increase it to 26 bits
327 * fairly easily, by changing this node to have longj type which causes the ARG
328 * field to be used for the link to the next regnode (although code would have
329 * to be changed to account for this), and then use a combination of the flags
330 * and next_off fields for the length. To get 34 bit length, also change the
331 * node to be an ARG2L, using the second 32 bit field for the length, and not
332 * using the flags nor next_off fields at all. One could have an llstring node
333 * and even an lllstring type. */
3f2416ae 334#define STR_LENl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 335 (((struct regnode_lstring *)p)->str_len))
3f2416ae 336#define STRINGl(p) (__ASSERT_(OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 337 (((struct regnode_lstring *)p)->string))
ae06e581
KW
338#define OPERANDl(p) STRINGl(p)
339
3f2416ae 340#define STR_LEN(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 341 ? STR_LENl(p) : STR_LENs(p))
3f2416ae 342#define STRING(p) ((OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
5cd61b66 343 ? STRINGl(p) : STRINGs(p))
c5184715
KW
344#define OPERAND(p) STRING(p)
345
c45abc0a
KW
346/* The number of (smallest) regnode equivalents that a string of length l bytes
347 * occupies */
59e54936 348#define STR_SZ(l) (((l) + sizeof(regnode) - 1) / sizeof(regnode))
c45abc0a
KW
349
350/* The number of (smallest) regnode equivalents that the EXACTISH node 'p'
351 * occupies */
ae06e581 352#define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p)) + 1 + regarglen[(p)->type])
c45abc0a 353
c7fd9c67 354#define setSTR_LEN(p,v) \
ae06e581 355 STMT_START{ \
3f2416ae 356 if (OP(p) == LEXACT || OP(p) == LEXACT_REQ8) \
ae06e581
KW
357 ((struct regnode_lstring *)(p))->str_len = (v); \
358 else \
359 ((struct regnode_string *)(p))->str_len = (v); \
360 } STMT_END
cd439c50 361
13fcf652 362#define ANYOFR_BASE_BITS 20
d223e1ea 363#define ANYOFRbase(p) (ARG(p) & nBIT_MASK(ANYOFR_BASE_BITS))
13fcf652
KW
364#define ANYOFRdelta(p) (ARG(p) >> ANYOFR_BASE_BITS)
365
e8e3b0fb
JH
366#undef NODE_ALIGN
367#undef ARG_LOC
368#undef NEXTOPER
369#undef PREVOPER
370
e91177ed
IZ
371#define NODE_ALIGN(node)
372#define ARG_LOC(p) (((struct regnode_1 *)p)->arg1)
86451f01 373#define ARGp_LOC(p) (((struct regnode_p *)p)->arg1)
e91177ed
IZ
374#define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1)
375#define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2)
6bda09f9 376#define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2)
786e8c11 377
e91177ed
IZ
378#define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */
379#define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2)
380
c277df42
IZ
381#define NEXTOPER(p) ((p) + NODE_STEP_REGNODE)
382#define PREVOPER(p) ((p) - NODE_STEP_REGNODE)
383
f55b7b26
KW
384#define FILL_NODE(offset, op) \
385 STMT_START { \
386 OP(REGNODE_p(offset)) = op; \
387 NEXT_OFF(REGNODE_p(offset)) = 0; \
d2a88850 388 } STMT_END
f55b7b26
KW
389#define FILL_ADVANCE_NODE(offset, op) \
390 STMT_START { \
391 FILL_NODE(offset, op); \
392 (offset)++; \
d2a88850 393 } STMT_END
f55b7b26
KW
394#define FILL_ADVANCE_NODE_ARG(offset, op, arg) \
395 STMT_START { \
396 ARG_SET(REGNODE_p(offset), arg); \
397 FILL_ADVANCE_NODE(offset, op); \
398 /* This is used generically for other operations \
399 * that have a longer argument */ \
400 (offset) += regarglen[op]; \
86451f01
KW
401 } STMT_END
402#define FILL_ADVANCE_NODE_ARGp(offset, op, arg) \
403 STMT_START { \
404 ARGp_SET(REGNODE_p(offset), arg); \
405 FILL_ADVANCE_NODE(offset, op); \
406 (offset) += regarglen[op]; \
93539530 407 } STMT_END
f55b7b26
KW
408#define FILL_ADVANCE_NODE_2L_ARG(offset, op, arg1, arg2) \
409 STMT_START { \
410 ARG_SET(REGNODE_p(offset), arg1); \
411 ARG2L_SET(REGNODE_p(offset), arg2); \
412 FILL_ADVANCE_NODE(offset, op); \
413 (offset) += 2; \
93539530 414 } STMT_END
a687059c 415
22c35a8c 416#define REG_MAGIC 0234
a687059c 417
fe8d1b7c
KW
418/* An ANYOF node is basically a bitmap with the index being a code point. If
419 * the bit for that code point is 1, the code point matches; if 0, it doesn't
420 * match (complemented if inverted). There is an additional mechanism to deal
421 * with cases where the bitmap is insufficient in and of itself. This #define
422 * indicates if the bitmap does fully represent what this ANYOF node can match.
423 * The ARG is set to this special value (since 0, 1, ... are legal, but will
424 * never reach this high). */
93e92956 425#define ANYOF_ONLY_HAS_BITMAP ((U32) -1)
137165a6 426
93539530 427/* When the bitmap isn't completely sufficient for handling the ANYOF node,
fe8d1b7c
KW
428 * flags (in node->flags of the ANYOF node) get set to indicate this. These
429 * are perennially in short supply. Beyond several cases where warnings need
430 * to be raised under certain circumstances, currently, there are six cases
431 * where the bitmap alone isn't sufficient. We could use six flags to
432 * represent the 6 cases, but to save flags bits, we play some games. The
433 * cases are:
d2f465e9 434 *
fe8d1b7c
KW
435 * 1) The bitmap has a compiled-in very finite size. So something else needs
436 * to be used to specify if a code point that is too large for the bitmap
7a94a5fe 437 * actually matches. The mechanism currently is an inversion
fe8d1b7c
KW
438 * list. ANYOF_ONLY_HAS_BITMAP, described above, being TRUE indicates
439 * there are no matches of too-large code points. But if it is FALSE,
440 * then almost certainly there are matches too large for the bitmap. (The
441 * other cases, described below, either imply this one or are extremely
442 * rare in practice.) So we can just assume that a too-large code point
443 * will need something beyond the bitmap if ANYOF_ONLY_HAS_BITMAP is
444 * FALSE, instead of having a separate flag for this.
445 * 2) A subset of item 1) is if all possible code points outside the bitmap
446 * match. This is a common occurrence when the class is complemented,
447 * like /[^ij]/. Therefore a bit is reserved to indicate this,
42d0f81a 448 * rather than having an inversion list created,
d1c40ef5 449 * ANYOF_MATCHES_ALL_ABOVE_BITMAP.
fe8d1b7c
KW
450 * 3) Under /d rules, it can happen that code points that are in the upper
451 * latin1 range (\x80-\xFF or their equivalents on EBCDIC platforms) match
452 * only if the runtime target string being matched against is UTF-8. For
453 * example /[\w[:punct:]]/d. This happens only for posix classes (with a
d1c40ef5
KW
454 * couple of exceptions, like \d where it doesn't happen), and all such
455 * ones also have above-bitmap matches. Thus, 3) implies 1) as well.
456 * Note that /d rules are no longer encouraged; 'use 5.14' or higher
457 * deselects them. But a flag is required so that they can be properly
458 * handled. But it can be a shared flag: see 5) below.
459 * 4) Also under /d rules, something like /[\Wfoo]/ will match everything in
fe8d1b7c 460 * the \x80-\xFF range, unless the string being matched against is UTF-8.
7a94a5fe
KW
461 * An inversion list could be created for this case, but this is
462 * relatively common, and it turns out that it's all or nothing: if any
463 * one of these code points matches, they all do. Hence a single bit
464 * suffices. We use a shared flag that doesn't take up space by itself:
465 * ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER. This
466 * also implies 1), with one exception: [:^cntrl:].
fe8d1b7c
KW
467 * 5) A user-defined \p{} property may not have been defined by the time the
468 * regex is compiled. In this case, we don't know until runtime what it
469 * will match, so we have to assume it could match anything, including
470 * code points that ordinarily would be in the bitmap. A flag bit is
471 * necessary to indicate this , though it can be shared with the item 3)
472 * flag, as that only occurs under /d, and this only occurs under non-d.
473 * This case is quite uncommon in the field, and the /(?[ ...])/ construct
474 * is a better way to accomplish what this feature does. This case also
475 * implies 1).
476 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP
e2f5e63d 477 * is the shared flag.
fe8d1b7c
KW
478 * 6) /[foo]/il may have folds that are only valid if the runtime locale is a
479 * UTF-8 one. These are quite rare, so it would be good to avoid the
480 * expense of looking for them. But /l matching is slow anyway, and we've
e2f5e63d 481 * traditionally not worried too much about its performance. And this
037715a6 482 * condition requires the ANYOFL_FOLD flag to be set, so testing for
fe8d1b7c 483 * that flag would be sufficient to rule out most cases of this. So it is
d1c40ef5
KW
484 * unclear if this should have a flag or not. But, this flag can be
485 * shared with another, so it doesn't occupy extra space.
d2f465e9 486 *
d1c40ef5 487 * At the moment, there is one spare bit, but this could be increased by
93539530 488 * various tricks:
e2f5e63d 489 *
93539530 490 * If just one more bit is needed, as of this writing it seems to khw that the
d1c40ef5
KW
491 * best choice would be to make ANYOF_MATCHES_ALL_ABOVE_BITMAP not a flag, but
492 * something like
d2f465e9 493 *
d1c40ef5 494 * #define ANYOF_MATCHES_ALL_ABOVE_BITMAP ((U32) -2)
fe8d1b7c 495 *
d1c40ef5
KW
496 * and access it through the ARG like ANYOF_ONLY_HAS_BITMAP is. This flag is
497 * used by all ANYOF node types, and it could be used to avoid calling the
498 * handler function, as the macro REGINCLASS in regexec.c does now for other
499 * cases.
500 *
46fc0c43
KW
501 * Another possibility is based on the fact that ANYOF_MATCHES_POSIXL is
502 * redundant with the node type ANYOFPOSIXL. That flag could be removed, but
503 * at the expense of extra code in regexec.c. The flag has been retained
504 * because it allows us to see if we need to call reginsert, or just use the
505 * bitmap in one test.
fe8d1b7c 506 *
46fc0c43
KW
507 * If this is done, an extension would be to make all ANYOFL nodes contain the
508 * extra 32 bits that ANYOFPOSIXL ones do. The posix flags only occupy 30
509 * bits, so the ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD flags
510 * and ANYOFL_FOLD could be moved to that extra space, but it would mean extra
511 * instructions, as there are currently places in the code that assume those
512 * two bits are zero.
d1c40ef5
KW
513 *
514 * All told, 5 bits could be available for other uses if all of the above were
515 * done.
a4525e78 516 *
d1c40ef5 517 * Some flags are not used in synthetic start class (SSC) nodes, so could be
93e92956
KW
518 * shared should new flags be needed for SSCs, like SSC_MATCHES_EMPTY_STRING
519 * now. */
b8c5462f 520
d2f465e9
KW
521/* If this is set, the result of the match should be complemented. regexec.c
522 * is expecting this to be in the low bit. Never in an SSC */
93e92956 523#define ANYOF_INVERT 0x01
1984f6a0 524
a0dd4231 525/* For the SSC node only, which cannot be inverted, so is shared with that bit.
93e92956
KW
526 * This is used only during regex compilation. */
527#define SSC_MATCHES_EMPTY_STRING ANYOF_INVERT
a0dd4231 528
509d4de0
KW
529/* Set if this is a regnode_charclass_posixl vs a regnode_charclass. This
530 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale
531 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved
532 * at compile-time. Only set under /l; can be in SSC */
533#define ANYOF_MATCHES_POSIXL 0x02
b63f73da
KW
534
535/* The fold is calculated and stored in the bitmap where possible at compile
536 * time. However under locale, the actual folding varies depending on
537 * what the locale is at the time of execution, so it has to be deferred until
d2f465e9 538 * then. Only set under /l; never in an SSC */
037715a6 539#define ANYOFL_FOLD 0x04
b63f73da 540
d1c40ef5 541/* Shared bit set only with ANYOFL and SSC nodes:
269bb65b
KW
542 * If ANYOFL_FOLD is set, this flag indicates there are potential matches
543 * valid only if the locale is a UTF-8 one.
544 * If ANYOFL_FOLD is NOT set, this flag means to warn if the runtime locale
d1c40ef5
KW
545 * isn't a UTF-8 one (and the generated node assumes a UTF-8 locale).
546 * None of INVERT, POSIXL,
547 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP
548 * can be set. */
549#define ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD 0x08
550
551/* Convenience macros for teasing apart the meanings when reading the above bit
552 * */
553#define ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags) \
554 ((flags & ( ANYOFL_FOLD /* Both bits are set */ \
555 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \
556 == ( ANYOFL_FOLD \
557 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD))
558
559#define ANYOFL_UTF8_LOCALE_REQD(flags) \
560 ((flags & ( ANYOFL_FOLD /* Only REQD bit is set */ \
561 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \
562 == ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)
563
1744d18d 564/* Spare: Be sure to change ANYOF_FLAGS_ALL if this gets used 0x10 */
c8d3cd88 565
509d4de0
KW
566/* If set, the node matches every code point NUM_ANYOF_CODE_POINTS and above.
567 * Can be in an SSC */
b3b1cf17 568#define ANYOF_MATCHES_ALL_ABOVE_BITMAP 0x20
ef87b810 569
108316fb
KW
570/* Shared bit:
571 * Under /d it means the ANYOFD node matches more things if the target
572 * string is encoded in UTF-8; any such things will be non-ASCII,
7a94a5fe
KW
573 * characters that are < 256, and can be accessed via the inversion
574 * list.
108316fb
KW
575 * When not under /d, it means the ANYOF node contains a user-defined
576 * property that wasn't yet defined at the time the regex was compiled,
7a94a5fe 577 * and so must be looked up at runtime, by creating an inversion list.
108316fb
KW
578 * (These uses are mutually exclusive because a user-defined property is
579 * specified by \p{}, and \p{} implies /u which deselects /d). The long macro
580 * name is to make sure that you are cautioned about its shared nature. Only
581 * the non-/d meaning can be in an SSC */
582#define ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 0x40
f240c685
KW
583
584/* Shared bit:
d2f465e9 585 * Under /d it means the ANYOFD node matches all non-ASCII Latin1
f240c685
KW
586 * characters when the target string is not in utf8.
587 * When not under /d, it means the ANYOF node should raise a warning if
588 * matching against an above-Unicode code point.
589 * (These uses are mutually exclusive because the warning requires a \p{}, and
590 * \p{} implies /u which deselects /d). An SSC node only has this bit set if
591 * what is meant is the warning. The long macro name is to make sure that you
592 * are cautioned about its shared nature */
593#define ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER 0x80
7bc66b18 594
b271ede1 595#define ANYOF_FLAGS_ALL (0xff & ~0x10)
653099ff 596
037715a6 597#define ANYOF_LOCALE_FLAGS (ANYOFL_FOLD | ANYOF_MATCHES_POSIXL)
ace6b0e4 598
eff8b7dc
KW
599/* These are the flags that apply to both regular ANYOF nodes and synthetic
600 * start class nodes during construction of the SSC. During finalization of
d2f465e9 601 * the SSC, other of the flags may get added to it */
d1c40ef5 602#define ANYOF_COMMON_FLAGS 0
eff8b7dc 603
936ed897 604/* Character classes for node->classflags of ANYOF */
653099ff 605/* Should be synchronized with a table in regprop() */
575a8a29 606/* 2n should be the normal one, paired with its complement at 2n+1 */
b8c5462f 607
08ada29e
KW
608#define ANYOF_ALPHA ((_CC_ALPHA) * 2)
609#define ANYOF_NALPHA ((ANYOF_ALPHA) + 1)
e8d596e0
KW
610#define ANYOF_ALPHANUMERIC ((_CC_ALPHANUMERIC) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */
611#define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1)
08ada29e
KW
612#define ANYOF_ASCII ((_CC_ASCII) * 2)
613#define ANYOF_NASCII ((ANYOF_ASCII) + 1)
e8d596e0
KW
614#define ANYOF_BLANK ((_CC_BLANK) * 2) /* GNU extension: space and tab: non-vertical space */
615#define ANYOF_NBLANK ((ANYOF_BLANK) + 1)
b0d691b2
KW
616#define ANYOF_CASED ((_CC_CASED) * 2) /* Pseudo class for [:lower:] or
617 [:upper:] under /i */
618#define ANYOF_NCASED ((ANYOF_CASED) + 1)
08ada29e
KW
619#define ANYOF_CNTRL ((_CC_CNTRL) * 2)
620#define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1)
e8d596e0
KW
621#define ANYOF_DIGIT ((_CC_DIGIT) * 2) /* \d */
622#define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1)
08ada29e
KW
623#define ANYOF_GRAPH ((_CC_GRAPH) * 2)
624#define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1)
625#define ANYOF_LOWER ((_CC_LOWER) * 2)
626#define ANYOF_NLOWER ((ANYOF_LOWER) + 1)
627#define ANYOF_PRINT ((_CC_PRINT) * 2)
628#define ANYOF_NPRINT ((ANYOF_PRINT) + 1)
629#define ANYOF_PUNCT ((_CC_PUNCT) * 2)
630#define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1)
e8d596e0
KW
631#define ANYOF_SPACE ((_CC_SPACE) * 2) /* \s */
632#define ANYOF_NSPACE ((ANYOF_SPACE) + 1)
08ada29e
KW
633#define ANYOF_UPPER ((_CC_UPPER) * 2)
634#define ANYOF_NUPPER ((ANYOF_UPPER) + 1)
e8d596e0
KW
635#define ANYOF_WORDCHAR ((_CC_WORDCHAR) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */
636#define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1)
08ada29e
KW
637#define ANYOF_XDIGIT ((_CC_XDIGIT) * 2)
638#define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1)
b8c5462f 639
87cad4c0 640/* pseudo classes below this, not stored in the class bitmap, but used as flags
e1d1eefb
YO
641 during compilation of char classes */
642
b2ef2e34
KW
643#define ANYOF_VERTWS ((_CC_VERTSPACE) * 2)
644#define ANYOF_NVERTWS ((ANYOF_VERTWS)+1)
e1d1eefb 645
b2ef2e34
KW
646/* It is best if this is the last one, as all above it are stored as bits in a
647 * bitmap, and it isn't part of that bitmap */
648#if _CC_VERTSPACE != _HIGHEST_REGCOMP_DOT_H_SYNC
649# error Problem with handy.h _HIGHEST_REGCOMP_DOT_H_SYNC #define
650#endif
651
8efd3f97
KW
652#define ANYOF_POSIXL_MAX (ANYOF_VERTWS) /* So upper loop limit is written:
653 * '< ANYOF_MAX'
654 * Hence doesn't include VERTWS, as that
655 * is a pseudo class */
656#define ANYOF_MAX ANYOF_POSIXL_MAX
657
658#if (ANYOF_POSIXL_MAX > 32) /* Must fit in 32-bit word */
b2ef2e34 659# error Problem with handy.h _CC_foo #defines
a0947d7b
KW
660#endif
661
8efd3f97
KW
662#define ANYOF_HORIZWS ((ANYOF_POSIXL_MAX)+2) /* = (ANYOF_NVERTWS + 1) */
663#define ANYOF_NHORIZWS ((ANYOF_POSIXL_MAX)+3)
87cad4c0 664
8efd3f97
KW
665#define ANYOF_UNIPROP ((ANYOF_POSIXL_MAX)+4) /* Used to indicate a Unicode
666 property: \p{} or \P{} */
f39a4dc2 667
b8c5462f
JH
668/* Backward source code compatibility. */
669
670#define ANYOF_ALNUML ANYOF_ALNUM
671#define ANYOF_NALNUML ANYOF_NALNUM
672#define ANYOF_SPACEL ANYOF_SPACE
673#define ANYOF_NSPACEL ANYOF_NSPACE
8ebfc29a
KW
674#define ANYOF_ALNUM ANYOF_WORDCHAR
675#define ANYOF_NALNUM ANYOF_NWORDCHAR
b8c5462f
JH
676
677/* Utility macros for the bitmap and classes of ANYOF */
678
936ed897 679#define ANYOF_FLAGS(p) ((p)->flags)
b8c5462f 680
bc51fd78 681#define ANYOF_BIT(c) (1U << ((c) & 7))
b8c5462f 682
14ed74f6
KW
683#define POSIXL_SET(field, c) ((field) |= (1U << (c)))
684#define ANYOF_POSIXL_SET(p, c) POSIXL_SET(((regnode_charclass_posixl*) (p))->classflags, (c))
8efd3f97 685
14ed74f6
KW
686#define POSIXL_CLEAR(field, c) ((field) &= ~ (1U <<(c)))
687#define ANYOF_POSIXL_CLEAR(p, c) POSIXL_CLEAR(((regnode_charclass_posixl*) (p))->classflags, (c))
8efd3f97 688
14ed74f6
KW
689#define POSIXL_TEST(field, c) ((field) & (1U << (c)))
690#define ANYOF_POSIXL_TEST(p, c) POSIXL_TEST(((regnode_charclass_posixl*) (p))->classflags, (c))
b8c5462f 691
14ed74f6
KW
692#define POSIXL_ZERO(field) STMT_START { (field) = 0; } STMT_END
693#define ANYOF_POSIXL_ZERO(ret) POSIXL_ZERO(((regnode_charclass_posixl*) (ret))->classflags)
694
695#define ANYOF_POSIXL_SET_TO_BITMAP(p, bits) \
696 STMT_START { \
697 ((regnode_charclass_posixl*) (p))->classflags = (bits); \
698 } STMT_END
9cf78fa0
KW
699
700/* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */
d223e1ea 701#define ANYOF_POSIXL_SETALL(ret) STMT_START { ((regnode_charclass_posixl*) (ret))->classflags = nBIT_MASK(ANYOF_POSIXL_MAX); } STMT_END
8efd3f97 702#define ANYOF_CLASS_SETALL(ret) ANYOF_POSIXL_SETALL(ret)
9cf78fa0 703
8efd3f97 704#define ANYOF_POSIXL_TEST_ANY_SET(p) \
93e92956 705 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
d1d040e5 706 && (((regnode_charclass_posixl*)(p))->classflags))
8efd3f97 707#define ANYOF_CLASS_TEST_ANY_SET(p) ANYOF_POSIXL_TEST_ANY_SET(p)
e867ce45 708
e0e1be5f
KW
709/* Since an SSC always has this field, we don't have to test for that; nor do
710 * we want to because the bit isn't set for SSC during its construction */
711#define ANYOF_POSIXL_SSC_TEST_ANY_SET(p) \
712 cBOOL(((regnode_ssc*)(p))->classflags)
713#define ANYOF_POSIXL_SSC_TEST_ALL_SET(p) /* Are all bits set? */ \
714 (((regnode_ssc*) (p))->classflags \
d223e1ea 715 == nBIT_MASK(ANYOF_POSIXL_MAX))
e0e1be5f
KW
716
717#define ANYOF_POSIXL_TEST_ALL_SET(p) \
d223e1ea 718 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
e0e1be5f 719 && ((regnode_charclass_posixl*) (p))->classflags \
d223e1ea 720 == nBIT_MASK(ANYOF_POSIXL_MAX))
fb38762f 721
b27fd97a 722#define ANYOF_POSIXL_OR(source, dest) STMT_START { (dest)->classflags |= (source)->classflags ; } STMT_END
8efd3f97 723#define ANYOF_CLASS_OR(source, dest) ANYOF_POSIXL_OR((source), (dest))
936ed897 724
cdd87c1d
KW
725#define ANYOF_POSIXL_AND(source, dest) STMT_START { (dest)->classflags &= (source)->classflags ; } STMT_END
726
428c1aff
KW
727#define ANYOF_BITMAP_ZERO(ret) Zero(((regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char)
728#define ANYOF_BITMAP(p) ((regnode_charclass*)(p))->bitmap
4cbce0a6 729#define ANYOF_BITMAP_BYTE(p, c) BITMAP_BYTE(ANYOF_BITMAP(p), c)
b8c5462f
JH
730#define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c))
731#define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c))
bc51fd78 732#define ANYOF_BITMAP_TEST(p, c) cBOOL(ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c))
b8c5462f 733
f8bef550
NC
734#define ANYOF_BITMAP_SETALL(p) \
735 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE)
736#define ANYOF_BITMAP_CLEARALL(p) \
737 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE)
f8bef550 738
a687059c
LW
739/*
740 * Utility definitions.
741 */
2304df62 742#ifndef CHARMASK
075abff3 743# define UCHARAT(p) ((int)*(const U8*)(p))
a687059c 744#else
075abff3 745# define UCHARAT(p) ((int)*(p)&CHARMASK)
a687059c 746#endif
a687059c 747
c45abc0a
KW
748/* Number of regnode equivalents that 'guy' occupies beyond the size of the
749 * smallest regnode. */
c277df42
IZ
750#define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode))
751
e384d5c1
YO
752#define REG_ZERO_LEN_SEEN 0x00000001
753#define REG_LOOKBEHIND_SEEN 0x00000002
754#define REG_GPOS_SEEN 0x00000004
471f5387 755/* spare */
e384d5c1
YO
756#define REG_RECURSE_SEEN 0x00000020
757#define REG_TOP_LEVEL_BRANCHES_SEEN 0x00000040
758#define REG_VERBARG_SEEN 0x00000080
759#define REG_CUTGROUP_SEEN 0x00000100
760#define REG_RUN_ON_COMMENT_SEEN 0x00000200
761#define REG_UNFOLDED_MULTI_SEEN 0x00000400
d5a00e4a 762/* spare */
ee273784
YO
763#define REG_UNBOUNDED_QUANTIFIER_SEEN 0x00001000
764
d09b2d29 765
73c4f7a1
GS
766START_EXTERN_C
767
be8e71aa
YO
768#ifdef PLUGGABLE_RE_EXTENSION
769#include "re_nodes.h"
770#else
d09b2d29 771#include "regnodes.h"
be8e71aa 772#endif
d09b2d29 773
f9f4320a
YO
774#ifndef PLUGGABLE_RE_EXTENSION
775#ifndef DOINIT
776EXTCONST regexp_engine PL_core_reg_engine;
777#else /* DOINIT */
778EXTCONST regexp_engine PL_core_reg_engine = {
fe578d7f 779 Perl_re_compile,
2fdbfb4d 780 Perl_regexec_flags,
f9f4320a
YO
781 Perl_re_intuit_start,
782 Perl_re_intuit_string,
2fdbfb4d
AB
783 Perl_regfree_internal,
784 Perl_reg_numbered_buff_fetch,
785 Perl_reg_numbered_buff_store,
786 Perl_reg_numbered_buff_length,
192b9cd1
AB
787 Perl_reg_named_buff,
788 Perl_reg_named_buff_iter,
49d7dfbc 789 Perl_reg_qr_package,
f9f4320a 790#if defined(USE_ITHREADS)
3c13cae6 791 Perl_regdupe_internal,
f9f4320a 792#endif
3c13cae6 793 Perl_re_op_compile
f9f4320a
YO
794};
795#endif /* DOINIT */
796#endif /* PLUGGABLE_RE_EXTENSION */
797
798
73c4f7a1 799END_EXTERN_C
cad2e5aa 800
cad2e5aa 801
261faec3
MJD
802/* .what is a character array with one character for each member of .data
803 * The character describes the function of the corresponding .data item:
af534a04 804 * a - AV for paren_name_list under DEBUGGING
0111c4fd 805 * f - start-class data for regstclass optimization
68e2671b 806 * l - start op for literal (?{EVAL}) item
d63c20f2 807 * L - start op for literal (?{EVAL}) item, with separate CV (qr//)
b30fcab9 808 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/
7a94a5fe
KW
809 * s - inversion list for Unicode-style character class, and the
810 * multicharacter strings resulting from casefolding the single-character
811 * entries in the character class
a3621e74 812 * t - trie struct
55eed653 813 * u - trie struct's widecharmap (a HV, so can't share, must dup)
2b8b4781 814 * also used for revcharmap and words under DEBUGGING
07be1b83 815 * T - aho-trie struct
81714fb9 816 * S - sv for named capture lookup
261faec3 817 * 20010712 mjd@plover.com
c3613ed7 818 * (Remember to update re_dup() and pregfree() if you add any items.)
261faec3 819 */
cad2e5aa
JH
820struct reg_data {
821 U32 count;
822 U8 *what;
823 void* data[1];
824};
825
a1cac82e
NC
826/* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses
827 anchored* and float* via array indexes 0 and 1. */
cad2e5aa 828#define anchored_substr substrs->data[0].substr
33b8afdf 829#define anchored_utf8 substrs->data[0].utf8_substr
cad2e5aa 830#define anchored_offset substrs->data[0].min_offset
1de06328
YO
831#define anchored_end_shift substrs->data[0].end_shift
832
cad2e5aa 833#define float_substr substrs->data[1].substr
33b8afdf 834#define float_utf8 substrs->data[1].utf8_substr
cad2e5aa
JH
835#define float_min_offset substrs->data[1].min_offset
836#define float_max_offset substrs->data[1].max_offset
1de06328
YO
837#define float_end_shift substrs->data[1].end_shift
838
cad2e5aa 839#define check_substr substrs->data[2].substr
33b8afdf 840#define check_utf8 substrs->data[2].utf8_substr
cad2e5aa
JH
841#define check_offset_min substrs->data[2].min_offset
842#define check_offset_max substrs->data[2].max_offset
1de06328 843#define check_end_shift substrs->data[2].end_shift
a3621e74 844
8d919b0a
FC
845#define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr)
846#define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8)
847#define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr)
848#define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8)
a3621e74
YO
849
850/* trie related stuff */
5d9a96ca 851
a3621e74
YO
852/* a transition record for the state machine. the
853 check field determines which state "owns" the
854 transition. the char the transition is for is
855 determined by offset from the owning states base
856 field. the next field determines which state
857 is to be transitioned to if any.
858*/
859struct _reg_trie_trans {
860 U32 next;
861 U32 check;
862};
863
864/* a transition list element for the list based representation */
865struct _reg_trie_trans_list_elem {
866 U16 forid;
867 U32 newstate;
868};
869typedef struct _reg_trie_trans_list_elem reg_trie_trans_le;
870
871/* a state for compressed nodes. base is an offset
872 into an array of reg_trie_trans array. If wordnum is
873 nonzero the state is accepting. if base is zero then
874 the state has no children (and will be accepting)
875*/
876struct _reg_trie_state {
877 U16 wordnum;
878 union {
879 U32 base;
880 reg_trie_trans_le* list;
881 } trans;
882};
883
2e64971a
DM
884/* info per word; indexed by wordnum */
885typedef struct {
886 U16 prev; /* previous word in acceptance chain; eg in
887 * zzz|abc|ab/ after matching the chars abc, the
888 * accepted word is #2, and the previous accepted
889 * word is #3 */
890 U32 len; /* how many chars long is this word? */
891 U32 accept; /* accept state for this word */
892} reg_trie_wordinfo;
a3621e74
YO
893
894
a3621e74
YO
895typedef struct _reg_trie_state reg_trie_state;
896typedef struct _reg_trie_trans reg_trie_trans;
897
898
899/* anything in here that needs to be freed later
23eab42c
NC
900 should be dealt with in pregfree.
901 refcount is first in both this and _reg_ac_data to allow a space
902 optimisation in Perl_regdupe. */
a3621e74 903struct _reg_trie_data {
23eab42c 904 U32 refcount; /* number of times this trie is referenced */
786e8c11
YO
905 U32 lasttrans; /* last valid transition element */
906 U16 *charmap; /* byte to charid lookup array */
786e8c11
YO
907 reg_trie_state *states; /* state data */
908 reg_trie_trans *trans; /* array of transition elements */
909 char *bitmap; /* stclass bitmap */
786e8c11
YO
910 U16 *jump; /* optional 1 indexed array of offsets before tail
911 for the node following a given word. */
2e64971a 912 reg_trie_wordinfo *wordinfo; /* array of info per word */
1b5c067c
NC
913 U16 uniquecharcount; /* unique chars in trie (width of trans table) */
914 U32 startstate; /* initial state - used for common prefix optimisation */
915 STRLEN minlen; /* minimum length of words in trie - build/opt only? */
916 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */
2e64971a 917 U32 prefixlen; /* #chars in common prefix */
1e2e3d02
YO
918 U32 statecount; /* Build only - number of states in the states array
919 (including the unused zero state) */
786e8c11 920 U32 wordcount; /* Build only */
a3621e74 921#ifdef DEBUGGING
786e8c11 922 STRLEN charcount; /* Build only */
a3621e74
YO
923#endif
924};
2b8b4781
NC
925/* There is one (3 under DEBUGGING) pointers that logically belong in this
926 structure, but are held outside as they need duplication on thread cloning,
927 whereas the rest of the structure can be read only:
928 HV *widecharmap; code points > 255 to charid
929#ifdef DEBUGGING
930 AV *words; Array of words contained in trie, for dumping
931 AV *revcharmap; Map of each charid back to its character representation
932#endif
933*/
934
935#define TRIE_WORDS_OFFSET 2
936
a3621e74
YO
937typedef struct _reg_trie_data reg_trie_data;
938
23eab42c
NC
939/* refcount is first in both this and _reg_trie_data to allow a space
940 optimisation in Perl_regdupe. */
07be1b83 941struct _reg_ac_data {
23eab42c 942 U32 refcount;
1b5c067c 943 U32 trie;
07be1b83
YO
944 U32 *fail;
945 reg_trie_state *states;
07be1b83
YO
946};
947typedef struct _reg_ac_data reg_ac_data;
948
486ec47a 949/* ANY_BIT doesn't use the structure, so we can borrow it here.
3dab1dad
YO
950 This is simpler than refactoring all of it as wed end up with
951 three different sets... */
952
953#define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap)
4cbce0a6 954#define TRIE_BITMAP_BYTE(p, c) BITMAP_BYTE(TRIE_BITMAP(p), c)
5dfe062f
YO
955#define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c))
956#define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c))
957#define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
3dab1dad 958
1de06328
YO
959#define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC)
960#define IS_TRIE_AC(op) ((op)>=AHOCORASICK)
961
962
0effac60 963#define BITMAP_BYTE(p, c) (( (U8*) p) [ ( ( (UV) (c)) >> 3) ] )
786e8c11 964#define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
3dab1dad 965
a3621e74
YO
966/* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */
967#define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1)
968#define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state))
969#define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state))
970
3dab1dad 971#ifdef DEBUGGING
3dab1dad 972#define TRIE_CHARCOUNT(trie) ((trie)->charcount)
3dab1dad 973#else
3dab1dad 974#define TRIE_CHARCOUNT(trie) (trie_charcount)
3dab1dad 975#endif
a3621e74 976
0111c4fd
RGS
977#define RE_TRIE_MAXBUF_INIT 65536
978#define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF"
a3621e74
YO
979#define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS"
980
6ef7fe53
KW
981#define RE_COMPILE_RECURSION_INIT 1000
982#define RE_COMPILE_RECURSION_LIMIT "\022E_COMPILE_RECURSION_LIMIT"
983
be8e71aa
YO
984/*
985
986RE_DEBUG_FLAGS is used to control what debug output is emitted
987its divided into three groups of options, some of which interact.
988The three groups are: Compile, Execute, Extra. There is room for a
989further group, as currently only the low three bytes are used.
990
991 Compile Options:
07be1b83 992
be8e71aa
YO
993 PARSE
994 PEEP
995 TRIE
996 PROGRAM
997 OFFSETS
a3621e74 998
be8e71aa 999 Execute Options:
a3621e74 1000
be8e71aa
YO
1001 INTUIT
1002 MATCH
1003 TRIE
a3621e74 1004
be8e71aa 1005 Extra Options
a3621e74 1006
be8e71aa
YO
1007 TRIE
1008 OFFSETS
1009
1010If you modify any of these make sure you make corresponding changes to
1011re.pm, especially to the documentation.
a3621e74 1012
be8e71aa
YO
1013*/
1014
1015
1016/* Compile */
1017#define RE_DEBUG_COMPILE_MASK 0x0000FF
1018#define RE_DEBUG_COMPILE_PARSE 0x000001
1019#define RE_DEBUG_COMPILE_OPTIMISE 0x000002
1020#define RE_DEBUG_COMPILE_TRIE 0x000004
1021#define RE_DEBUG_COMPILE_DUMP 0x000008
f7819f85 1022#define RE_DEBUG_COMPILE_FLAGS 0x000010
d9a72fcc 1023#define RE_DEBUG_COMPILE_TEST 0x000020
be8e71aa
YO
1024
1025/* Execute */
1026#define RE_DEBUG_EXECUTE_MASK 0x00FF00
1027#define RE_DEBUG_EXECUTE_INTUIT 0x000100
1028#define RE_DEBUG_EXECUTE_MATCH 0x000200
1029#define RE_DEBUG_EXECUTE_TRIE 0x000400
1030
1031/* Extra */
51bd1941 1032#define RE_DEBUG_EXTRA_MASK 0x3FF0000
fd41b2d1
KW
1033#define RE_DEBUG_EXTRA_TRIE 0x0010000
1034#define RE_DEBUG_EXTRA_OFFSETS 0x0020000
1035#define RE_DEBUG_EXTRA_OFFDEBUG 0x0040000
1036#define RE_DEBUG_EXTRA_STATE 0x0080000
1037#define RE_DEBUG_EXTRA_OPTIMISE 0x0100000
1038#define RE_DEBUG_EXTRA_BUFFERS 0x0400000
1039#define RE_DEBUG_EXTRA_GPOS 0x0800000
1040#define RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE 0x1000000
51bd1941 1041#define RE_DEBUG_EXTRA_WILDCARD 0x2000000
24b23f37 1042/* combined */
fd41b2d1 1043#define RE_DEBUG_EXTRA_STACK 0x0280000
be8e71aa 1044
55a0b497 1045#define RE_DEBUG_FLAG(x) (re_debug_flags & (x))
be8e71aa
YO
1046/* Compile */
1047#define DEBUG_COMPILE_r(x) DEBUG_r( \
55a0b497 1048 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_MASK)) x )
be8e71aa 1049#define DEBUG_PARSE_r(x) DEBUG_r( \
55a0b497 1050 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_PARSE)) x )
be8e71aa 1051#define DEBUG_OPTIMISE_r(x) DEBUG_r( \
55a0b497 1052 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE)) x )
be8e71aa 1053#define DEBUG_DUMP_r(x) DEBUG_r( \
55a0b497 1054 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_DUMP)) x )
be8e71aa 1055#define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \
55a0b497 1056 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TRIE)) x )
f7819f85 1057#define DEBUG_FLAGS_r(x) DEBUG_r( \
55a0b497 1058 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_FLAGS)) x )
d9a72fcc 1059#define DEBUG_TEST_r(x) DEBUG_r( \
55a0b497 1060 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TEST)) x )
be8e71aa
YO
1061/* Execute */
1062#define DEBUG_EXECUTE_r(x) DEBUG_r( \
55a0b497 1063 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_MASK)) x )
be8e71aa 1064#define DEBUG_INTUIT_r(x) DEBUG_r( \
55a0b497 1065 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_INTUIT)) x )
be8e71aa 1066#define DEBUG_MATCH_r(x) DEBUG_r( \
55a0b497 1067 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_MATCH)) x )
be8e71aa 1068#define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \
55a0b497 1069 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXECUTE_TRIE)) x )
be8e71aa
YO
1070
1071/* Extra */
1072#define DEBUG_EXTRA_r(x) DEBUG_r( \
55a0b497 1073 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_MASK)) x )
a5ca303d 1074#define DEBUG_OFFSETS_r(x) DEBUG_r( \
55a0b497 1075 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_OFFSETS)) x )
ab3bbdeb 1076#define DEBUG_STATE_r(x) DEBUG_r( \
55a0b497 1077 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_STATE)) x )
24b23f37 1078#define DEBUG_STACK_r(x) DEBUG_r( \
55a0b497 1079 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_STACK)) x )
e7707071 1080#define DEBUG_BUFFERS_r(x) DEBUG_r( \
55a0b497 1081 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_BUFFERS)) x )
e7707071 1082
a5ca303d 1083#define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \
dafb2544 1084 if (DEBUG_v_TEST || ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \
55a0b497 1085 RE_DEBUG_FLAG(RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE))) x )
07be1b83 1086#define MJD_OFFSET_DEBUG(x) DEBUG_r( \
55a0b497 1087 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_OFFDEBUG)) \
be8e71aa
YO
1088 Perl_warn_nocontext x )
1089#define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \
55a0b497 1090 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_TRIE)) x )
be8e71aa 1091#define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \
55a0b497 1092 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_TRIE)) x )
be8e71aa
YO
1093
1094#define DEBUG_TRIE_r(x) DEBUG_r( \
55a0b497
KW
1095 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_COMPILE_TRIE \
1096 | RE_DEBUG_EXECUTE_TRIE )) x )
2c296965 1097#define DEBUG_GPOS_r(x) DEBUG_r( \
55a0b497 1098 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_GPOS)) x )
be8e71aa 1099
fd41b2d1 1100#define DEBUG_DUMP_PRE_OPTIMIZE_r(x) DEBUG_r( \
55a0b497 1101 if (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_DUMP_PRE_OPTIMIZE)) x )
fd41b2d1 1102
be8e71aa 1103/* initialization */
1fe1d354
KW
1104/* Get the debug flags for code not in regcomp.c nor regexec.c. This doesn't
1105 * initialize the variable if it isn't already there, instead it just assumes
1106 * the flags are 0 */
1107#define DECLARE_AND_GET_RE_DEBUG_FLAGS_NON_REGEX \
1108 volatile IV re_debug_flags = 0; PERL_UNUSED_VAR(re_debug_flags); \
1109 STMT_START { \
1110 SV * re_debug_flags_sv = NULL; \
1111 /* get_sv() can return NULL during global destruction. */ \
1112 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \
1113 if (re_debug_flags_sv && SvIOK(re_debug_flags_sv)) \
1114 re_debug_flags=SvIV(re_debug_flags_sv); \
1115 } STMT_END
1116
1117
a3621e74 1118#ifdef DEBUGGING
ab3bbdeb 1119
1fe1d354
KW
1120/* For use in regcomp.c and regexec.c, Get the debug flags, and initialize to
1121 * the defaults if not done already */
619d7873
KW
1122#define DECLARE_AND_GET_RE_DEBUG_FLAGS \
1123 volatile IV re_debug_flags = 0; PERL_UNUSED_VAR(re_debug_flags); \
1124 STMT_START { \
1125 SV * re_debug_flags_sv = NULL; \
1126 /* get_sv() can return NULL during global destruction. */ \
1127 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \
1128 if (re_debug_flags_sv) { \
1129 if (!SvIOK(re_debug_flags_sv)) /* If doesnt exist set to default */\
1130 sv_setuv(re_debug_flags_sv, \
1131 /* These defaults should be kept in sync with re.pm */ \
1132 RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \
1133 re_debug_flags=SvIV(re_debug_flags_sv); \
1134 } \
1135 } STMT_END
ab3bbdeb 1136
51bd1941
KW
1137#define isDEBUG_WILDCARD (DEBUG_v_TEST || RE_DEBUG_FLAG(RE_DEBUG_EXTRA_WILDCARD))
1138
eecd4d11
KW
1139#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \
1140 const char * const rpv = \
1141 pv_pretty((dsv), (pv), (l), (m), \
1142 PL_colors[(c1)],PL_colors[(c2)], \
c89df6cf 1143 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \
0df25f3d 1144 const int rlen = SvCUR(dsv)
ab3bbdeb 1145
8ec77c3c 1146/* This is currently unsed in the core */
eecd4d11
KW
1147#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \
1148 const char * const rpv = \
1149 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \
1150 PL_colors[(c1)],PL_colors[(c2)], \
c89df6cf 1151 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) )
ab3bbdeb 1152
eecd4d11
KW
1153#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \
1154 const char * const rpv = \
1155 pv_pretty((dsv), (pv), (l), (m), \
1156 PL_colors[0], PL_colors[1], \
c89df6cf 1157 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \
ab3bbdeb 1158 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \
0a0e9afc 1159 )
ab3bbdeb
YO
1160
1161#define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0))
1162#define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "")
1163
1164#else /* if not DEBUGGING */
1165
82dc0782
KW
1166#define DECLARE_AND_GET_RE_DEBUG_FLAGS dNOOP
1167#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) dNOOP
ab3bbdeb 1168#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m)
82dc0782 1169#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) dNOOP
ab3bbdeb
YO
1170#define RE_SV_DUMPLEN(ItEm)
1171#define RE_SV_TAIL(ItEm)
51bd1941 1172#define isDEBUG_WILDCARD 0
ab3bbdeb
YO
1173
1174#endif /* DEBUG RELATED DEFINES */
a3621e74 1175
0bc0e44f 1176#define FIRST_NON_ASCII_DECIMAL_DIGIT 0x660 /* ARABIC_INDIC_DIGIT_ZERO */
2ca74f27 1177
64935bc6
KW
1178typedef enum {
1179 TRADITIONAL_BOUND = _CC_WORDCHAR,
ae3bb8ea 1180 GCB_BOUND,
6b659339 1181 LB_BOUND,
06ae2722 1182 SB_BOUND,
ae3bb8ea 1183 WB_BOUND
64935bc6
KW
1184} bound_type;
1185
13fcf652 1186/* This unpacks the FLAGS field of ANYOF[HR]x nodes. The value it contains
3146c00a
KW
1187 * gives the strict lower bound for the UTF-8 start byte of any code point
1188 * matchable by the node, and a loose upper bound as well.
1189 *
1190 * The low bound is stored in the upper 6 bits, plus 0xC0.
1191 * The loose upper bound is determined from the lowest 2 bits and the low bound
1192 * (called x) as follows:
1193 *
1194 * 11 The upper limit of the range can be as much as (EF - x) / 8
1195 * 10 The upper limit of the range can be as much as (EF - x) / 4
1196 * 01 The upper limit of the range can be as much as (EF - x) / 2
1197 * 00 The upper limit of the range can be as much as EF
1198 *
0bf4efd7
KW
1199 * For motivation of this design, see commit message in
1200 * 3146c00a633e9cbed741e10146662fbcedfdb8d3 */
3146c00a
KW
1201#ifdef EBCDIC
1202# define MAX_ANYOF_HRx_BYTE 0xF4
1203#else
1204# define MAX_ANYOF_HRx_BYTE 0xEF
1205#endif
1206#define LOWEST_ANYOF_HRx_BYTE(b) (((b) >> 2) + 0xC0)
1207#define HIGHEST_ANYOF_HRx_BYTE(b) \
1208 (LOWEST_ANYOF_HRx_BYTE(b) \
1209 + ((MAX_ANYOF_HRx_BYTE - LOWEST_ANYOF_HRx_BYTE(b)) >> ((b) & 3)))
1210
3dd7db29
JK
1211#endif /* PERL_REGCOMP_H_ */
1212
e9a8c099 1213/*
14d04a33 1214 * ex: set ts=8 sts=4 sw=4 et:
e9a8c099 1215 */