This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.h: Swap struct vs typedef
[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 */
e1d1eefb 10#include "regcharclass.h"
a687059c 11
58e23c8d 12/* Convert branch sequences to more efficient trie ops? */
07be1b83 13#define PERL_ENABLE_TRIE_OPTIMISATION 1
58e23c8d 14
486ec47a 15/* Be really aggressive about optimising patterns with trie sequences? */
07be1b83 16#define PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION 1
58e23c8d
YO
17
18/* Should the optimiser take positive assertions into account? */
68ba3a3f 19#define PERL_ENABLE_POSITIVE_ASSERTION_STUDY 0
58e23c8d
YO
20
21/* Not for production use: */
07be1b83 22#define PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS 0
58e23c8d 23
7122b237
YO
24/* Activate offsets code - set to if 1 to enable */
25#ifdef DEBUGGING
26#define RE_TRACK_PATTERN_OFFSETS
27#endif
28
a687059c
LW
29/*
30 * The "internal use only" fields in regexp.h are present to pass info from
31 * compile to execute that permits the execute phase to run lots faster on
32 * simple cases. They are:
33 *
bd61b366 34 * regstart sv that must begin a match; NULL if none obvious
a687059c
LW
35 * reganch is the match anchored (at beginning-of-line only)?
36 * regmust string (pointer into program) that match must include, or NULL
79072805 37 * [regmust changed to SV* for bminstr()--law]
a687059c
LW
38 * regmlen length of regmust string
39 * [regmlen not used currently]
40 *
41 * Regstart and reganch permit very fast decisions on suitable starting points
42 * for a match, cutting down the work a lot. Regmust permits fast rejection
43 * of lines that cannot possibly match. The regmust tests are costly enough
e50aee73 44 * that pregcomp() supplies a regmust only if the r.e. contains something
a687059c
LW
45 * potentially expensive (at present, the only such thing detected is * or +
46 * at the start of the r.e., which can involve a lot of backup). Regmlen is
e50aee73 47 * supplied because the test in pregexec() needs it and pregcomp() is computing
a687059c
LW
48 * it anyway.
49 * [regmust is now supplied always. The tests that use regmust have a
50 * heuristic that disables the test if it usually matches.]
51 *
52 * [In fact, we now use regmust in many cases to locate where the search
53 * starts in the string, so if regback is >= 0, the regmust search is never
54 * wasted effort. The regback variable says how many characters back from
55 * where regmust matched is the earliest possible start of the match.
56 * For instance, /[a-z].foo/ has a regmust of 'foo' and a regback of 2.]
57 */
58
59/*
60 * Structure for regexp "program". This is essentially a linear encoding
61 * of a nondeterministic finite-state machine (aka syntax charts or
62 * "railroad normal form" in parsing technology). Each node is an opcode
63 * plus a "next" pointer, possibly plus an operand. "Next" pointers of
64 * all nodes except BRANCH implement concatenation; a "next" pointer with
65 * a BRANCH on both ends of it is connecting two alternatives. (Here we
66 * have one of the subtle syntax dependencies: an individual BRANCH (as
67 * opposed to a collection of them) is never concatenated with anything
68 * because of operator precedence.) The operand of some types of node is
69 * a literal string; for others, it is a node leading into a sub-FSM. In
70 * particular, the operand of a BRANCH node is the first node of the branch.
71 * (NB this is *not* a tree structure: the tail of the branch connects
a3621e74
YO
72 * to the thing following the set of BRANCHes.) The opcodes are defined
73 * in regnodes.h which is generated from regcomp.sym by regcomp.pl.
a687059c
LW
74 */
75
a687059c
LW
76/*
77 * A node is one char of opcode followed by two chars of "next" pointer.
78 * "Next" pointers are stored as two 8-bit pieces, high order first. The
79 * value is a positive offset from the opcode of the node containing it.
80 * An operand, if any, simply follows the node. (Note that much of the
81 * code generation knows about this implicit relationship.)
82 *
83 * Using two bytes for the "next" pointer is vast overkill for most things,
84 * but allows patterns to get big without disasters.
85 *
e91177ed 86 * [The "next" pointer is always aligned on an even
62e6ef33 87 * boundary, and reads the offset directly as a short.]
a687059c 88 */
785a26d5
YO
89
90/* This is the stuff that used to live in regexp.h that was truly
91 private to the engine itself. It now lives here. */
92
785a26d5 93 typedef struct regexp_internal {
1f1031fe 94 int name_list_idx; /* Optional data index of an array of paren names */
7122b237
YO
95 union {
96 U32 *offsets; /* offset annotations 20001228 MJD
97 data about mapping the program to the
98 string -
99 offsets[0] is proglen when this is used
100 */
101 U32 proglen;
102 } u;
103
6d5c990f
RGS
104 regnode *regstclass; /* Optional startclass as identified or constructed
105 by the optimiser */
106 struct reg_data *data; /* Additional miscellaneous data used by the program.
107 Used to make it easier to clone and free arbitrary
108 data that the regops need. Often the ARG field of
109 a regop is an index into this structure */
1acab4c5 110 struct reg_code_blocks *code_blocks;/* positions of literal (?{}) */
6d5c990f
RGS
111 regnode program[1]; /* Unwarranted chumminess with compiler. */
112} regexp_internal;
113
114#define RXi_SET(x,y) (x)->pprivate = (void*)(y)
115#define RXi_GET(x) ((regexp_internal *)((x)->pprivate))
116#define RXi_GET_DECL(r,ri) regexp_internal *ri = RXi_GET(r)
785a26d5
YO
117/*
118 * Flags stored in regexp->intflags
119 * These are used only internally to the regexp engine
120 *
121 * See regexp.h for flags used externally to the regexp engine
122 */
e3e400ec
YO
123#define RXp_INTFLAGS(rx) ((rx)->intflags)
124#define RX_INTFLAGS(prog) RXp_INTFLAGS(ReANY(prog))
125
785a26d5
YO
126#define PREGf_SKIP 0x00000001
127#define PREGf_IMPLICIT 0x00000002 /* Converted .* to ^.* */
128#define PREGf_NAUGHTY 0x00000004 /* how exponential is this pattern? */
129#define PREGf_VERBARG_SEEN 0x00000008
130#define PREGf_CUTGROUP_SEEN 0x00000010
732caac7 131#define PREGf_USE_RE_EVAL 0x00000020 /* compiled with "use re 'eval'" */
e3e400ec
YO
132/* these used to be extflags, but are now intflags */
133#define PREGf_NOSCAN 0x00000040
33c28ab2 134 /* spare */
8e1490ee
YO
135#define PREGf_GPOS_SEEN 0x00000100
136#define PREGf_GPOS_FLOAT 0x00000200
785a26d5 137
d3d47aac
YO
138#define PREGf_ANCH_MBOL 0x00000400
139#define PREGf_ANCH_SBOL 0x00000800
140#define PREGf_ANCH_GPOS 0x00001000
d5a00e4a 141#define PREGf_RECURSE_SEEN 0x00002000
8e1490ee 142
d3d47aac
YO
143#define PREGf_ANCH \
144 ( PREGf_ANCH_SBOL | PREGf_ANCH_GPOS | PREGf_ANCH_MBOL )
785a26d5
YO
145
146/* this is where the old regcomp.h started */
a687059c 147
c277df42 148struct regnode_string {
cd439c50 149 U8 str_len;
c277df42
IZ
150 U8 type;
151 U16 next_off;
cd439c50 152 char string[1];
c277df42 153};
a687059c 154
6bda09f9
YO
155/* Argument bearing node - workhorse,
156 arg1 is often for the data field */
c277df42
IZ
157struct regnode_1 {
158 U8 flags;
159 U8 type;
160 U16 next_off;
161 U32 arg1;
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
e0a1ff7a
KW
182/* This give the number of code points that can be in the bitmap of an ANYOF
183 * node. The shift number must currently be one of: 8..12. It can't be less
184 * than 8 (256) because some code relies on it being at least that. Above 12
185 * (4096), and you start running into warnings that some data structure widths
186 * have been exceeded, though the test suite as of this writing still passes
187 * for up through 16, which is as high as anyone would ever want to go,
188 * encompassing all of the Unicode BMP, and thus including all the economically
189 * important world scripts. At 12 most of them are: including Arabic,
190 * Cyrillic, Greek, Hebrew, Indian subcontinent, Latin, and Thai; but not Han,
191 * Japanese, nor Korean. (The regarglen structure in regnodes.h is a U8, and
192 * the trie types TRIEC and AHOCORASICKC are larger than U8 for shift values
193 * below above 12.) Be sure to benchmark before changing, as larger sizes do
194 * significantly slow down the test suite */
195#define NUM_ANYOF_CODE_POINTS (1 << 8)
6bda09f9 196
dcb20b36 197#define ANYOF_BITMAP_SIZE (NUM_ANYOF_CODE_POINTS / 8) /* 8 bits/Byte */
936ed897 198
1ee208c4
KW
199/* Note that these form structs which are supersets of the next smaller one, by
200 * appending fields. Alignment problems can occur if one of those optional
201 * fields requires stricter alignment than the base struct. And formal
202 * parameters that can really be two or more of the structs should be
203 * declared as the smallest one it could be. See commit message for
204 * 7dcac5f6a5195002b55c935ee1d67f67e1df280b. Regnode allocation is done
205 * without regard to alignment, and changing it to would also require changing
206 * the code that inserts and deletes regnodes. The basic single-argument
207 * regnode has a U32, which is what reganode() allocates as a unit. Therefore
208 * no field can require stricter alignment than U32. */
209
786e8c11 210/* also used by trie */
936ed897
IZ
211struct regnode_charclass {
212 U8 flags;
213 U8 type;
214 U16 next_off;
9da99c5e 215 U32 arg1; /* set by set_ANYOF_arg() */
19103b91 216 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */
936ed897
IZ
217};
218
bf3f2d85 219/* has runtime (locale) \d, \w, ..., [:posix:] classes */
ea461dd6 220struct regnode_charclass_posixl {
93e92956 221 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
936ed897
IZ
222 U8 type;
223 U16 next_off;
7e817188 224 U32 arg1;
03fa83ba 225 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
9cf78fa0 226 U32 classflags; /* and run-time */
936ed897
IZ
227};
228
85c8e306
KW
229/* A synthetic start class (SSC); is a regnode_charclass_posixl_fold, plus an
230 * extra SV*, used only during its construction and which is not used by
231 * regexec.c. Note that the 'next_off' field is unused, as the SSC stands
232 * alone, so there is never a next node. Also, there is no alignment issue,
b8fda935 233 * because these are declared or allocated as a complete unit so the compiler
85c8e306
KW
234 * takes care of alignment. This is unlike the other regnodes which are
235 * allocated in terms of multiples of a single-argument regnode. SSC nodes can
236 * have a pointer field because there is no alignment issue, and because it is
237 * set to NULL after construction, before any cloning of the pattern */
b8f7bb16 238struct regnode_ssc {
93e92956 239 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
b8f7bb16
KW
240 U8 type;
241 U16 next_off;
7e817188 242 U32 arg1;
03fa83ba 243 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
631e9161 244 U32 classflags; /* ... and run-time */
85c8e306
KW
245
246 /* Auxiliary, only used during construction; NULL afterwards: list of code
247 * points matched */
248 SV* invlist;
b8f7bb16
KW
249};
250
71068078
KW
251/* We take advantage of 'next_off' not otherwise being used in the SSC by
252 * actually using it: by setting it to 1. This allows us to test and
253 * distinguish between an SSC and other ANYOF node types, as 'next_off' cannot
254 * otherwise be 1, because it is the offset to the next regnode expressed in
255 * units of regnodes. Since an ANYOF node contains extra fields, it adds up
256 * to 12 regnode units on 32-bit systems, (hence the minimum this can be (if
257 * not 0) is 11 there. Even if things get tightly packed on a 64-bit system,
258 * it still would be more than 1. */
259#define set_ANYOF_SYNTHETIC(n) STMT_START{ OP(n) = ANYOF; \
260 NEXT_OFF(n) = 1; \
261 } STMT_END
c1071334 262#define is_ANYOF_SYNTHETIC(n) (PL_regkind[OP(n)] == ANYOF && NEXT_OFF(n) == 1)
71068078 263
9c7e81e8
AD
264/* XXX fix this description.
265 Impose a limit of REG_INFTY on various pattern matching operations
266 to limit stack growth and to avoid "infinite" recursions.
267*/
4288c5b9
KW
268/* The default size for REG_INFTY is U16_MAX, which is the same as
269 USHORT_MAX (see perl.h). Unfortunately U16 isn't necessarily 16 bits
270 (see handy.h). On the Cray C90, sizeof(short)==4 and hence U16_MAX is
271 ((1<<32)-1), while on the Cray T90, sizeof(short)==8 and U16_MAX is
272 ((1<<64)-1). To limit stack growth to reasonable sizes, supply a
9c7e81e8
AD
273 smaller default.
274 --Andy Dougherty 11 June 1998
275*/
276#if SHORTSIZE > 2
277# ifndef REG_INFTY
4288c5b9 278# define REG_INFTY ((1<<16)-1)
9c7e81e8
AD
279# endif
280#endif
281
282#ifndef REG_INFTY
4288c5b9 283# define REG_INFTY U16_MAX
83cfe48c 284#endif
a687059c 285
e91177ed
IZ
286#define ARG_VALUE(arg) (arg)
287#define ARG__SET(arg,val) ((arg) = (val))
c277df42 288
e8e3b0fb
JH
289#undef ARG
290#undef ARG1
291#undef ARG2
292
c277df42
IZ
293#define ARG(p) ARG_VALUE(ARG_LOC(p))
294#define ARG1(p) ARG_VALUE(ARG1_LOC(p))
295#define ARG2(p) ARG_VALUE(ARG2_LOC(p))
6bda09f9 296#define ARG2L(p) ARG_VALUE(ARG2L_LOC(p))
786e8c11 297
c277df42
IZ
298#define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val))
299#define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val))
300#define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val))
6bda09f9 301#define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val))
c277df42 302
e8e3b0fb
JH
303#undef NEXT_OFF
304#undef NODE_ALIGN
305
075abff3
AL
306#define NEXT_OFF(p) ((p)->next_off)
307#define NODE_ALIGN(node)
aa48e906
YO
308/* the following define was set to 0xde in 075abff3
309 * as part of some linting logic. I have set it to 0
310 * as otherwise in every place where we /might/ set flags
311 * we have to set it 0 explicitly, which duplicates
312 * assignments and IMO adds an unacceptable level of
313 * surprise to working in the regex engine. If this
314 * is changed from 0 then at the very least make sure
315 * that SBOL for /^/ sets the flags to 0 explicitly.
316 * -- Yves */
317#define NODE_ALIGN_FILL(node) ((node)->flags = 0)
a687059c 318
c277df42
IZ
319#define SIZE_ALIGN NODE_ALIGN
320
e8e3b0fb
JH
321#undef OP
322#undef OPERAND
323#undef MASK
324#undef STRING
325
e91177ed 326#define OP(p) ((p)->type)
b988e673
KW
327#define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \
328 regnode types. For some, it's the \
329 character set of the regnode */
e91177ed 330#define OPERAND(p) (((struct regnode_string *)p)->string)
cd439c50
IZ
331#define MASK(p) ((char*)OPERAND(p))
332#define STR_LEN(p) (((struct regnode_string *)p)->str_len)
333#define STRING(p) (((struct regnode_string *)p)->string)
334#define STR_SZ(l) ((l + sizeof(regnode) - 1) / sizeof(regnode))
335#define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p))+1)
336
e8e3b0fb
JH
337#undef NODE_ALIGN
338#undef ARG_LOC
339#undef NEXTOPER
340#undef PREVOPER
341
e91177ed
IZ
342#define NODE_ALIGN(node)
343#define ARG_LOC(p) (((struct regnode_1 *)p)->arg1)
344#define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1)
345#define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2)
6bda09f9 346#define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2)
786e8c11 347
e91177ed
IZ
348#define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */
349#define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2)
350
351#define NODE_STEP_B 4
c277df42
IZ
352
353#define NEXTOPER(p) ((p) + NODE_STEP_REGNODE)
354#define PREVOPER(p) ((p) - NODE_STEP_REGNODE)
355
d2a88850
KW
356#define FILL_NODE(ptr, op) \
357 STMT_START { \
358 OP(ptr) = op; \
359 NEXT_OFF(ptr) = 0; \
360 } STMT_END
361#define FILL_ADVANCE_NODE(ptr, op) \
362 STMT_START { \
363 FILL_NODE(ptr, op); \
364 (ptr)++; \
365 } STMT_END
93539530
KW
366#define FILL_ADVANCE_NODE_ARG(ptr, op, arg) \
367 STMT_START { \
368 ARG_SET(ptr, arg); \
369 FILL_ADVANCE_NODE(ptr, op); \
370 (ptr) += 1; \
371 } STMT_END
9990d5ab 372#define FILL_ADVANCE_NODE_2L_ARG(ptr, op, arg1, arg2) \
93539530 373 STMT_START { \
9990d5ab
KW
374 ARG_SET(ptr, arg1); \
375 ARG2L_SET(ptr, arg2); \
376 FILL_ADVANCE_NODE(ptr, op); \
377 (ptr) += 2; \
93539530 378 } STMT_END
a687059c 379
22c35a8c 380#define REG_MAGIC 0234
a687059c 381
9d990559 382#define SIZE_ONLY cBOOL(RExC_emit == (regnode *) & RExC_emit_dummy)
6d15ebf3
KW
383#define PASS1 SIZE_ONLY
384#define PASS2 (! SIZE_ONLY)
c277df42 385
fe8d1b7c
KW
386/* An ANYOF node is basically a bitmap with the index being a code point. If
387 * the bit for that code point is 1, the code point matches; if 0, it doesn't
388 * match (complemented if inverted). There is an additional mechanism to deal
389 * with cases where the bitmap is insufficient in and of itself. This #define
390 * indicates if the bitmap does fully represent what this ANYOF node can match.
391 * The ARG is set to this special value (since 0, 1, ... are legal, but will
392 * never reach this high). */
93e92956 393#define ANYOF_ONLY_HAS_BITMAP ((U32) -1)
137165a6 394
93539530 395/* When the bitmap isn't completely sufficient for handling the ANYOF node,
fe8d1b7c
KW
396 * flags (in node->flags of the ANYOF node) get set to indicate this. These
397 * are perennially in short supply. Beyond several cases where warnings need
398 * to be raised under certain circumstances, currently, there are six cases
399 * where the bitmap alone isn't sufficient. We could use six flags to
400 * represent the 6 cases, but to save flags bits, we play some games. The
401 * cases are:
d2f465e9 402 *
fe8d1b7c
KW
403 * 1) The bitmap has a compiled-in very finite size. So something else needs
404 * to be used to specify if a code point that is too large for the bitmap
405 * actually matches. The mechanism currently is a swash or inversion
406 * list. ANYOF_ONLY_HAS_BITMAP, described above, being TRUE indicates
407 * there are no matches of too-large code points. But if it is FALSE,
408 * then almost certainly there are matches too large for the bitmap. (The
409 * other cases, described below, either imply this one or are extremely
410 * rare in practice.) So we can just assume that a too-large code point
411 * will need something beyond the bitmap if ANYOF_ONLY_HAS_BITMAP is
412 * FALSE, instead of having a separate flag for this.
413 * 2) A subset of item 1) is if all possible code points outside the bitmap
414 * match. This is a common occurrence when the class is complemented,
415 * like /[^ij]/. Therefore a bit is reserved to indicate this,
d1c40ef5
KW
416 * rather than having an expensive swash created,
417 * ANYOF_MATCHES_ALL_ABOVE_BITMAP.
fe8d1b7c
KW
418 * 3) Under /d rules, it can happen that code points that are in the upper
419 * latin1 range (\x80-\xFF or their equivalents on EBCDIC platforms) match
420 * only if the runtime target string being matched against is UTF-8. For
421 * example /[\w[:punct:]]/d. This happens only for posix classes (with a
d1c40ef5
KW
422 * couple of exceptions, like \d where it doesn't happen), and all such
423 * ones also have above-bitmap matches. Thus, 3) implies 1) as well.
424 * Note that /d rules are no longer encouraged; 'use 5.14' or higher
425 * deselects them. But a flag is required so that they can be properly
426 * handled. But it can be a shared flag: see 5) below.
427 * 4) Also under /d rules, something like /[\Wfoo]/ will match everything in
fe8d1b7c
KW
428 * the \x80-\xFF range, unless the string being matched against is UTF-8.
429 * A swash could be created for this case, but this is relatively common,
430 * and it turns out that it's all or nothing: if any one of these code
431 * points matches, they all do. Hence a single bit suffices. We use a
e2f5e63d 432 * shared flag that doesn't take up space by itself:
fe8d1b7c
KW
433 * ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER.
434 * This also implies 1), with one exception: [:^cntrl:].
435 * 5) A user-defined \p{} property may not have been defined by the time the
436 * regex is compiled. In this case, we don't know until runtime what it
437 * will match, so we have to assume it could match anything, including
438 * code points that ordinarily would be in the bitmap. A flag bit is
439 * necessary to indicate this , though it can be shared with the item 3)
440 * flag, as that only occurs under /d, and this only occurs under non-d.
441 * This case is quite uncommon in the field, and the /(?[ ...])/ construct
442 * is a better way to accomplish what this feature does. This case also
443 * implies 1).
444 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP
e2f5e63d 445 * is the shared flag.
fe8d1b7c
KW
446 * 6) /[foo]/il may have folds that are only valid if the runtime locale is a
447 * UTF-8 one. These are quite rare, so it would be good to avoid the
448 * expense of looking for them. But /l matching is slow anyway, and we've
e2f5e63d 449 * traditionally not worried too much about its performance. And this
037715a6 450 * condition requires the ANYOFL_FOLD flag to be set, so testing for
fe8d1b7c 451 * that flag would be sufficient to rule out most cases of this. So it is
d1c40ef5
KW
452 * unclear if this should have a flag or not. But, this flag can be
453 * shared with another, so it doesn't occupy extra space.
d2f465e9 454 *
d1c40ef5 455 * At the moment, there is one spare bit, but this could be increased by
93539530 456 * various tricks:
e2f5e63d 457 *
93539530 458 * If just one more bit is needed, as of this writing it seems to khw that the
d1c40ef5
KW
459 * best choice would be to make ANYOF_MATCHES_ALL_ABOVE_BITMAP not a flag, but
460 * something like
d2f465e9 461 *
d1c40ef5 462 * #define ANYOF_MATCHES_ALL_ABOVE_BITMAP ((U32) -2)
fe8d1b7c 463 *
d1c40ef5
KW
464 * and access it through the ARG like ANYOF_ONLY_HAS_BITMAP is. This flag is
465 * used by all ANYOF node types, and it could be used to avoid calling the
466 * handler function, as the macro REGINCLASS in regexec.c does now for other
467 * cases.
468 *
469 * Another possibility is to instead (or additionally) rename the ANYOF_POSIXL
470 * flag to be ANYOFL_LARGE, to mean that the ANYOF node has an extra 32 bits
471 * beyond what a regular one does. That's what it effectively means now, with
472 * the extra space all for the POSIX class flags. But those classes actually
473 * only occupy 30 bits, so the ANYOFL_FOLD and
474 * ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD flags could be moved
475 * to that extra space. The 30 bits in the extra word would indicate if a
476 * posix class should be looked up or not. The downside of this is that ANYOFL
477 * nodes with folding would always have to have the extra space allocated, even
478 * if they didn't use the 30 posix bits. There isn't an SSC problem as all
479 * SSCs are this large anyway.
fe8d1b7c
KW
480 *
481 * One could completely remove ANYOFL_LARGE and make all ANYOFL nodes large.
d1c40ef5
KW
482 * REGINCLASS would have to be modified so that if the node type were this, it
483 * would call reginclass(), as the flag bit that indicates to do this now would
484 * be gone.
485 *
486 * All told, 5 bits could be available for other uses if all of the above were
487 * done.
a4525e78 488 *
d1c40ef5 489 * Some flags are not used in synthetic start class (SSC) nodes, so could be
93e92956
KW
490 * shared should new flags be needed for SSCs, like SSC_MATCHES_EMPTY_STRING
491 * now. */
b8c5462f 492
d2f465e9
KW
493/* If this is set, the result of the match should be complemented. regexec.c
494 * is expecting this to be in the low bit. Never in an SSC */
93e92956 495#define ANYOF_INVERT 0x01
1984f6a0 496
a0dd4231 497/* For the SSC node only, which cannot be inverted, so is shared with that bit.
93e92956
KW
498 * This is used only during regex compilation. */
499#define SSC_MATCHES_EMPTY_STRING ANYOF_INVERT
a0dd4231 500
509d4de0
KW
501/* Set if this is a regnode_charclass_posixl vs a regnode_charclass. This
502 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale
503 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved
504 * at compile-time. Only set under /l; can be in SSC */
505#define ANYOF_MATCHES_POSIXL 0x02
b63f73da
KW
506
507/* The fold is calculated and stored in the bitmap where possible at compile
508 * time. However under locale, the actual folding varies depending on
509 * what the locale is at the time of execution, so it has to be deferred until
d2f465e9 510 * then. Only set under /l; never in an SSC */
037715a6 511#define ANYOFL_FOLD 0x04
b63f73da 512
d1c40ef5
KW
513/* Shared bit set only with ANYOFL and SSC nodes:
514 * If ANYOFL_FOLD is set, this means there are potential matches valid
515 * only if the locale is a UTF-8 one.
516 * If ANYOFL_FOLD is NOT set, this means to warn if the runtime locale
517 * isn't a UTF-8 one (and the generated node assumes a UTF-8 locale).
518 * None of INVERT, POSIXL,
519 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP
520 * can be set. */
521#define ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD 0x08
522
523/* Convenience macros for teasing apart the meanings when reading the above bit
524 * */
525#define ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags) \
526 ((flags & ( ANYOFL_FOLD /* Both bits are set */ \
527 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \
528 == ( ANYOFL_FOLD \
529 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD))
530
531#define ANYOFL_UTF8_LOCALE_REQD(flags) \
532 ((flags & ( ANYOFL_FOLD /* Only REQD bit is set */ \
533 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)) \
534 == ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD)
535
1744d18d 536/* Spare: Be sure to change ANYOF_FLAGS_ALL if this gets used 0x10 */
c8d3cd88 537
509d4de0
KW
538/* If set, the node matches every code point NUM_ANYOF_CODE_POINTS and above.
539 * Can be in an SSC */
b3b1cf17 540#define ANYOF_MATCHES_ALL_ABOVE_BITMAP 0x20
ef87b810 541
108316fb
KW
542/* Shared bit:
543 * Under /d it means the ANYOFD node matches more things if the target
544 * string is encoded in UTF-8; any such things will be non-ASCII,
545 * characters that are < 256, and can be accessed via the swash.
546 * When not under /d, it means the ANYOF node contains a user-defined
547 * property that wasn't yet defined at the time the regex was compiled,
548 * and so must be looked up at runtime, by creating a swash
549 * (These uses are mutually exclusive because a user-defined property is
550 * specified by \p{}, and \p{} implies /u which deselects /d). The long macro
551 * name is to make sure that you are cautioned about its shared nature. Only
552 * the non-/d meaning can be in an SSC */
553#define ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 0x40
f240c685
KW
554
555/* Shared bit:
d2f465e9 556 * Under /d it means the ANYOFD node matches all non-ASCII Latin1
f240c685
KW
557 * characters when the target string is not in utf8.
558 * When not under /d, it means the ANYOF node should raise a warning if
559 * matching against an above-Unicode code point.
560 * (These uses are mutually exclusive because the warning requires a \p{}, and
561 * \p{} implies /u which deselects /d). An SSC node only has this bit set if
562 * what is meant is the warning. The long macro name is to make sure that you
563 * are cautioned about its shared nature */
564#define ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER 0x80
7bc66b18 565
b271ede1 566#define ANYOF_FLAGS_ALL (0xff & ~0x10)
653099ff 567
037715a6 568#define ANYOF_LOCALE_FLAGS (ANYOFL_FOLD | ANYOF_MATCHES_POSIXL)
ace6b0e4 569
eff8b7dc
KW
570/* These are the flags that apply to both regular ANYOF nodes and synthetic
571 * start class nodes during construction of the SSC. During finalization of
d2f465e9 572 * the SSC, other of the flags may get added to it */
d1c40ef5 573#define ANYOF_COMMON_FLAGS 0
eff8b7dc 574
936ed897 575/* Character classes for node->classflags of ANYOF */
653099ff 576/* Should be synchronized with a table in regprop() */
575a8a29 577/* 2n should be the normal one, paired with its complement at 2n+1 */
b8c5462f 578
08ada29e
KW
579#define ANYOF_ALPHA ((_CC_ALPHA) * 2)
580#define ANYOF_NALPHA ((ANYOF_ALPHA) + 1)
e8d596e0
KW
581#define ANYOF_ALPHANUMERIC ((_CC_ALPHANUMERIC) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */
582#define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1)
08ada29e
KW
583#define ANYOF_ASCII ((_CC_ASCII) * 2)
584#define ANYOF_NASCII ((ANYOF_ASCII) + 1)
e8d596e0
KW
585#define ANYOF_BLANK ((_CC_BLANK) * 2) /* GNU extension: space and tab: non-vertical space */
586#define ANYOF_NBLANK ((ANYOF_BLANK) + 1)
b0d691b2
KW
587#define ANYOF_CASED ((_CC_CASED) * 2) /* Pseudo class for [:lower:] or
588 [:upper:] under /i */
589#define ANYOF_NCASED ((ANYOF_CASED) + 1)
08ada29e
KW
590#define ANYOF_CNTRL ((_CC_CNTRL) * 2)
591#define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1)
e8d596e0
KW
592#define ANYOF_DIGIT ((_CC_DIGIT) * 2) /* \d */
593#define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1)
08ada29e
KW
594#define ANYOF_GRAPH ((_CC_GRAPH) * 2)
595#define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1)
596#define ANYOF_LOWER ((_CC_LOWER) * 2)
597#define ANYOF_NLOWER ((ANYOF_LOWER) + 1)
598#define ANYOF_PRINT ((_CC_PRINT) * 2)
599#define ANYOF_NPRINT ((ANYOF_PRINT) + 1)
600#define ANYOF_PUNCT ((_CC_PUNCT) * 2)
601#define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1)
e8d596e0
KW
602#define ANYOF_SPACE ((_CC_SPACE) * 2) /* \s */
603#define ANYOF_NSPACE ((ANYOF_SPACE) + 1)
08ada29e
KW
604#define ANYOF_UPPER ((_CC_UPPER) * 2)
605#define ANYOF_NUPPER ((ANYOF_UPPER) + 1)
e8d596e0
KW
606#define ANYOF_WORDCHAR ((_CC_WORDCHAR) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */
607#define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1)
08ada29e
KW
608#define ANYOF_XDIGIT ((_CC_XDIGIT) * 2)
609#define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1)
b8c5462f 610
87cad4c0 611/* pseudo classes below this, not stored in the class bitmap, but used as flags
e1d1eefb
YO
612 during compilation of char classes */
613
b2ef2e34
KW
614#define ANYOF_VERTWS ((_CC_VERTSPACE) * 2)
615#define ANYOF_NVERTWS ((ANYOF_VERTWS)+1)
e1d1eefb 616
b2ef2e34
KW
617/* It is best if this is the last one, as all above it are stored as bits in a
618 * bitmap, and it isn't part of that bitmap */
619#if _CC_VERTSPACE != _HIGHEST_REGCOMP_DOT_H_SYNC
620# error Problem with handy.h _HIGHEST_REGCOMP_DOT_H_SYNC #define
621#endif
622
8efd3f97
KW
623#define ANYOF_POSIXL_MAX (ANYOF_VERTWS) /* So upper loop limit is written:
624 * '< ANYOF_MAX'
625 * Hence doesn't include VERTWS, as that
626 * is a pseudo class */
627#define ANYOF_MAX ANYOF_POSIXL_MAX
628
629#if (ANYOF_POSIXL_MAX > 32) /* Must fit in 32-bit word */
b2ef2e34 630# error Problem with handy.h _CC_foo #defines
a0947d7b
KW
631#endif
632
8efd3f97
KW
633#define ANYOF_HORIZWS ((ANYOF_POSIXL_MAX)+2) /* = (ANYOF_NVERTWS + 1) */
634#define ANYOF_NHORIZWS ((ANYOF_POSIXL_MAX)+3)
87cad4c0 635
8efd3f97
KW
636#define ANYOF_UNIPROP ((ANYOF_POSIXL_MAX)+4) /* Used to indicate a Unicode
637 property: \p{} or \P{} */
f39a4dc2 638
b8c5462f
JH
639/* Backward source code compatibility. */
640
641#define ANYOF_ALNUML ANYOF_ALNUM
642#define ANYOF_NALNUML ANYOF_NALNUM
643#define ANYOF_SPACEL ANYOF_SPACE
644#define ANYOF_NSPACEL ANYOF_NSPACE
8ebfc29a
KW
645#define ANYOF_ALNUM ANYOF_WORDCHAR
646#define ANYOF_NALNUM ANYOF_NWORDCHAR
b8c5462f
JH
647
648/* Utility macros for the bitmap and classes of ANYOF */
649
936ed897 650#define ANYOF_FLAGS(p) ((p)->flags)
b8c5462f 651
bc51fd78 652#define ANYOF_BIT(c) (1U << ((c) & 7))
b8c5462f 653
14ed74f6
KW
654#define POSIXL_SET(field, c) ((field) |= (1U << (c)))
655#define ANYOF_POSIXL_SET(p, c) POSIXL_SET(((regnode_charclass_posixl*) (p))->classflags, (c))
8efd3f97 656
14ed74f6
KW
657#define POSIXL_CLEAR(field, c) ((field) &= ~ (1U <<(c)))
658#define ANYOF_POSIXL_CLEAR(p, c) POSIXL_CLEAR(((regnode_charclass_posixl*) (p))->classflags, (c))
8efd3f97 659
14ed74f6
KW
660#define POSIXL_TEST(field, c) ((field) & (1U << (c)))
661#define ANYOF_POSIXL_TEST(p, c) POSIXL_TEST(((regnode_charclass_posixl*) (p))->classflags, (c))
b8c5462f 662
14ed74f6
KW
663#define POSIXL_ZERO(field) STMT_START { (field) = 0; } STMT_END
664#define ANYOF_POSIXL_ZERO(ret) POSIXL_ZERO(((regnode_charclass_posixl*) (ret))->classflags)
665
666#define ANYOF_POSIXL_SET_TO_BITMAP(p, bits) \
667 STMT_START { \
668 ((regnode_charclass_posixl*) (p))->classflags = (bits); \
669 } STMT_END
9cf78fa0
KW
670
671/* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */
d1d040e5 672#define ANYOF_POSIXL_SETALL(ret) STMT_START { ((regnode_charclass_posixl*) (ret))->classflags = ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1; } STMT_END
8efd3f97 673#define ANYOF_CLASS_SETALL(ret) ANYOF_POSIXL_SETALL(ret)
9cf78fa0 674
8efd3f97 675#define ANYOF_POSIXL_TEST_ANY_SET(p) \
93e92956 676 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
d1d040e5 677 && (((regnode_charclass_posixl*)(p))->classflags))
8efd3f97 678#define ANYOF_CLASS_TEST_ANY_SET(p) ANYOF_POSIXL_TEST_ANY_SET(p)
e867ce45 679
e0e1be5f
KW
680/* Since an SSC always has this field, we don't have to test for that; nor do
681 * we want to because the bit isn't set for SSC during its construction */
682#define ANYOF_POSIXL_SSC_TEST_ANY_SET(p) \
683 cBOOL(((regnode_ssc*)(p))->classflags)
684#define ANYOF_POSIXL_SSC_TEST_ALL_SET(p) /* Are all bits set? */ \
685 (((regnode_ssc*) (p))->classflags \
686 == ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1)
687
688#define ANYOF_POSIXL_TEST_ALL_SET(p) \
93e92956 689 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
e0e1be5f
KW
690 && ((regnode_charclass_posixl*) (p))->classflags \
691 == ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1)
fb38762f 692
b27fd97a 693#define ANYOF_POSIXL_OR(source, dest) STMT_START { (dest)->classflags |= (source)->classflags ; } STMT_END
8efd3f97 694#define ANYOF_CLASS_OR(source, dest) ANYOF_POSIXL_OR((source), (dest))
936ed897 695
cdd87c1d
KW
696#define ANYOF_POSIXL_AND(source, dest) STMT_START { (dest)->classflags &= (source)->classflags ; } STMT_END
697
428c1aff
KW
698#define ANYOF_BITMAP_ZERO(ret) Zero(((regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char)
699#define ANYOF_BITMAP(p) ((regnode_charclass*)(p))->bitmap
4cbce0a6 700#define ANYOF_BITMAP_BYTE(p, c) BITMAP_BYTE(ANYOF_BITMAP(p), c)
b8c5462f
JH
701#define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c))
702#define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c))
bc51fd78 703#define ANYOF_BITMAP_TEST(p, c) cBOOL(ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c))
b8c5462f 704
f8bef550
NC
705#define ANYOF_BITMAP_SETALL(p) \
706 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE)
707#define ANYOF_BITMAP_CLEARALL(p) \
708 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE)
f8bef550 709
94b1fc23
KW
710#define ANYOF_SKIP (EXTRA_SIZE(regnode_charclass) \
711 - EXTRA_SIZE(struct regnode_1))
712#define ANYOF_POSIXL_SKIP (EXTRA_SIZE(regnode_charclass_posixl) \
713 - EXTRA_SIZE(struct regnode_1))
3a15e693 714
a687059c
LW
715/*
716 * Utility definitions.
717 */
2304df62 718#ifndef CHARMASK
075abff3 719# define UCHARAT(p) ((int)*(const U8*)(p))
a687059c 720#else
075abff3 721# define UCHARAT(p) ((int)*(p)&CHARMASK)
a687059c 722#endif
a687059c 723
c277df42
IZ
724#define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode))
725
e384d5c1
YO
726#define REG_ZERO_LEN_SEEN 0x00000001
727#define REG_LOOKBEHIND_SEEN 0x00000002
728#define REG_GPOS_SEEN 0x00000004
471f5387 729/* spare */
e384d5c1
YO
730#define REG_RECURSE_SEEN 0x00000020
731#define REG_TOP_LEVEL_BRANCHES_SEEN 0x00000040
732#define REG_VERBARG_SEEN 0x00000080
733#define REG_CUTGROUP_SEEN 0x00000100
734#define REG_RUN_ON_COMMENT_SEEN 0x00000200
735#define REG_UNFOLDED_MULTI_SEEN 0x00000400
d5a00e4a 736/* spare */
ee273784
YO
737#define REG_UNBOUNDED_QUANTIFIER_SEEN 0x00001000
738
d09b2d29 739
73c4f7a1
GS
740START_EXTERN_C
741
be8e71aa
YO
742#ifdef PLUGGABLE_RE_EXTENSION
743#include "re_nodes.h"
744#else
d09b2d29 745#include "regnodes.h"
be8e71aa 746#endif
d09b2d29 747
f9f4320a
YO
748#ifndef PLUGGABLE_RE_EXTENSION
749#ifndef DOINIT
750EXTCONST regexp_engine PL_core_reg_engine;
751#else /* DOINIT */
752EXTCONST regexp_engine PL_core_reg_engine = {
fe578d7f 753 Perl_re_compile,
2fdbfb4d 754 Perl_regexec_flags,
f9f4320a
YO
755 Perl_re_intuit_start,
756 Perl_re_intuit_string,
2fdbfb4d
AB
757 Perl_regfree_internal,
758 Perl_reg_numbered_buff_fetch,
759 Perl_reg_numbered_buff_store,
760 Perl_reg_numbered_buff_length,
192b9cd1
AB
761 Perl_reg_named_buff,
762 Perl_reg_named_buff_iter,
49d7dfbc 763 Perl_reg_qr_package,
f9f4320a 764#if defined(USE_ITHREADS)
3c13cae6 765 Perl_regdupe_internal,
f9f4320a 766#endif
3c13cae6 767 Perl_re_op_compile
f9f4320a
YO
768};
769#endif /* DOINIT */
770#endif /* PLUGGABLE_RE_EXTENSION */
771
772
73c4f7a1 773END_EXTERN_C
cad2e5aa 774
cad2e5aa 775
261faec3
MJD
776/* .what is a character array with one character for each member of .data
777 * The character describes the function of the corresponding .data item:
af534a04 778 * a - AV for paren_name_list under DEBUGGING
0111c4fd 779 * f - start-class data for regstclass optimization
68e2671b 780 * l - start op for literal (?{EVAL}) item
d63c20f2 781 * L - start op for literal (?{EVAL}) item, with separate CV (qr//)
b30fcab9 782 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/
38a44b82 783 * s - swash for Unicode-style character class, and the multicharacter
9e55ce06
JH
784 * strings resulting from casefolding the single-character entries
785 * in the character class
a3621e74 786 * t - trie struct
55eed653 787 * u - trie struct's widecharmap (a HV, so can't share, must dup)
2b8b4781 788 * also used for revcharmap and words under DEBUGGING
07be1b83 789 * T - aho-trie struct
81714fb9 790 * S - sv for named capture lookup
261faec3 791 * 20010712 mjd@plover.com
c3613ed7 792 * (Remember to update re_dup() and pregfree() if you add any items.)
261faec3 793 */
cad2e5aa
JH
794struct reg_data {
795 U32 count;
796 U8 *what;
797 void* data[1];
798};
799
a1cac82e
NC
800/* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses
801 anchored* and float* via array indexes 0 and 1. */
cad2e5aa 802#define anchored_substr substrs->data[0].substr
33b8afdf 803#define anchored_utf8 substrs->data[0].utf8_substr
cad2e5aa 804#define anchored_offset substrs->data[0].min_offset
1de06328
YO
805#define anchored_end_shift substrs->data[0].end_shift
806
cad2e5aa 807#define float_substr substrs->data[1].substr
33b8afdf 808#define float_utf8 substrs->data[1].utf8_substr
cad2e5aa
JH
809#define float_min_offset substrs->data[1].min_offset
810#define float_max_offset substrs->data[1].max_offset
1de06328
YO
811#define float_end_shift substrs->data[1].end_shift
812
cad2e5aa 813#define check_substr substrs->data[2].substr
33b8afdf 814#define check_utf8 substrs->data[2].utf8_substr
cad2e5aa
JH
815#define check_offset_min substrs->data[2].min_offset
816#define check_offset_max substrs->data[2].max_offset
1de06328 817#define check_end_shift substrs->data[2].end_shift
a3621e74 818
8d919b0a
FC
819#define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr)
820#define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8)
821#define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr)
822#define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8)
a3621e74
YO
823
824/* trie related stuff */
5d9a96ca 825
a3621e74
YO
826/* a transition record for the state machine. the
827 check field determines which state "owns" the
828 transition. the char the transition is for is
829 determined by offset from the owning states base
830 field. the next field determines which state
831 is to be transitioned to if any.
832*/
833struct _reg_trie_trans {
834 U32 next;
835 U32 check;
836};
837
838/* a transition list element for the list based representation */
839struct _reg_trie_trans_list_elem {
840 U16 forid;
841 U32 newstate;
842};
843typedef struct _reg_trie_trans_list_elem reg_trie_trans_le;
844
845/* a state for compressed nodes. base is an offset
846 into an array of reg_trie_trans array. If wordnum is
847 nonzero the state is accepting. if base is zero then
848 the state has no children (and will be accepting)
849*/
850struct _reg_trie_state {
851 U16 wordnum;
852 union {
853 U32 base;
854 reg_trie_trans_le* list;
855 } trans;
856};
857
2e64971a
DM
858/* info per word; indexed by wordnum */
859typedef struct {
860 U16 prev; /* previous word in acceptance chain; eg in
861 * zzz|abc|ab/ after matching the chars abc, the
862 * accepted word is #2, and the previous accepted
863 * word is #3 */
864 U32 len; /* how many chars long is this word? */
865 U32 accept; /* accept state for this word */
866} reg_trie_wordinfo;
a3621e74
YO
867
868
a3621e74
YO
869typedef struct _reg_trie_state reg_trie_state;
870typedef struct _reg_trie_trans reg_trie_trans;
871
872
873/* anything in here that needs to be freed later
23eab42c
NC
874 should be dealt with in pregfree.
875 refcount is first in both this and _reg_ac_data to allow a space
876 optimisation in Perl_regdupe. */
a3621e74 877struct _reg_trie_data {
23eab42c 878 U32 refcount; /* number of times this trie is referenced */
786e8c11
YO
879 U32 lasttrans; /* last valid transition element */
880 U16 *charmap; /* byte to charid lookup array */
786e8c11
YO
881 reg_trie_state *states; /* state data */
882 reg_trie_trans *trans; /* array of transition elements */
883 char *bitmap; /* stclass bitmap */
786e8c11
YO
884 U16 *jump; /* optional 1 indexed array of offsets before tail
885 for the node following a given word. */
2e64971a 886 reg_trie_wordinfo *wordinfo; /* array of info per word */
1b5c067c
NC
887 U16 uniquecharcount; /* unique chars in trie (width of trans table) */
888 U32 startstate; /* initial state - used for common prefix optimisation */
889 STRLEN minlen; /* minimum length of words in trie - build/opt only? */
890 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */
2e64971a 891 U32 prefixlen; /* #chars in common prefix */
1e2e3d02
YO
892 U32 statecount; /* Build only - number of states in the states array
893 (including the unused zero state) */
786e8c11 894 U32 wordcount; /* Build only */
a3621e74 895#ifdef DEBUGGING
786e8c11 896 STRLEN charcount; /* Build only */
a3621e74
YO
897#endif
898};
2b8b4781
NC
899/* There is one (3 under DEBUGGING) pointers that logically belong in this
900 structure, but are held outside as they need duplication on thread cloning,
901 whereas the rest of the structure can be read only:
902 HV *widecharmap; code points > 255 to charid
903#ifdef DEBUGGING
904 AV *words; Array of words contained in trie, for dumping
905 AV *revcharmap; Map of each charid back to its character representation
906#endif
907*/
908
909#define TRIE_WORDS_OFFSET 2
910
a3621e74
YO
911typedef struct _reg_trie_data reg_trie_data;
912
23eab42c
NC
913/* refcount is first in both this and _reg_trie_data to allow a space
914 optimisation in Perl_regdupe. */
07be1b83 915struct _reg_ac_data {
23eab42c 916 U32 refcount;
1b5c067c 917 U32 trie;
07be1b83
YO
918 U32 *fail;
919 reg_trie_state *states;
07be1b83
YO
920};
921typedef struct _reg_ac_data reg_ac_data;
922
486ec47a 923/* ANY_BIT doesn't use the structure, so we can borrow it here.
3dab1dad
YO
924 This is simpler than refactoring all of it as wed end up with
925 three different sets... */
926
927#define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap)
4cbce0a6 928#define TRIE_BITMAP_BYTE(p, c) BITMAP_BYTE(TRIE_BITMAP(p), c)
5dfe062f
YO
929#define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c))
930#define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c))
931#define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
3dab1dad 932
1de06328
YO
933#define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC)
934#define IS_TRIE_AC(op) ((op)>=AHOCORASICK)
935
936
0effac60 937#define BITMAP_BYTE(p, c) (( (U8*) p) [ ( ( (UV) (c)) >> 3) ] )
786e8c11 938#define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
3dab1dad 939
a3621e74
YO
940/* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */
941#define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1)
942#define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state))
943#define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state))
944
3dab1dad 945#ifdef DEBUGGING
3dab1dad 946#define TRIE_CHARCOUNT(trie) ((trie)->charcount)
3dab1dad 947#else
3dab1dad 948#define TRIE_CHARCOUNT(trie) (trie_charcount)
3dab1dad 949#endif
a3621e74 950
0111c4fd
RGS
951#define RE_TRIE_MAXBUF_INIT 65536
952#define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF"
a3621e74
YO
953#define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS"
954
be8e71aa
YO
955/*
956
957RE_DEBUG_FLAGS is used to control what debug output is emitted
958its divided into three groups of options, some of which interact.
959The three groups are: Compile, Execute, Extra. There is room for a
960further group, as currently only the low three bytes are used.
961
962 Compile Options:
07be1b83 963
be8e71aa
YO
964 PARSE
965 PEEP
966 TRIE
967 PROGRAM
968 OFFSETS
a3621e74 969
be8e71aa 970 Execute Options:
a3621e74 971
be8e71aa
YO
972 INTUIT
973 MATCH
974 TRIE
a3621e74 975
be8e71aa 976 Extra Options
a3621e74 977
be8e71aa
YO
978 TRIE
979 OFFSETS
980
981If you modify any of these make sure you make corresponding changes to
982re.pm, especially to the documentation.
a3621e74 983
be8e71aa
YO
984*/
985
986
987/* Compile */
988#define RE_DEBUG_COMPILE_MASK 0x0000FF
989#define RE_DEBUG_COMPILE_PARSE 0x000001
990#define RE_DEBUG_COMPILE_OPTIMISE 0x000002
991#define RE_DEBUG_COMPILE_TRIE 0x000004
992#define RE_DEBUG_COMPILE_DUMP 0x000008
f7819f85 993#define RE_DEBUG_COMPILE_FLAGS 0x000010
d9a72fcc 994#define RE_DEBUG_COMPILE_TEST 0x000020
be8e71aa
YO
995
996/* Execute */
997#define RE_DEBUG_EXECUTE_MASK 0x00FF00
998#define RE_DEBUG_EXECUTE_INTUIT 0x000100
999#define RE_DEBUG_EXECUTE_MATCH 0x000200
1000#define RE_DEBUG_EXECUTE_TRIE 0x000400
1001
1002/* Extra */
1003#define RE_DEBUG_EXTRA_MASK 0xFF0000
1004#define RE_DEBUG_EXTRA_TRIE 0x010000
1005#define RE_DEBUG_EXTRA_OFFSETS 0x020000
a5ca303d
YO
1006#define RE_DEBUG_EXTRA_OFFDEBUG 0x040000
1007#define RE_DEBUG_EXTRA_STATE 0x080000
1008#define RE_DEBUG_EXTRA_OPTIMISE 0x100000
e7707071 1009#define RE_DEBUG_EXTRA_BUFFERS 0x400000
2c296965 1010#define RE_DEBUG_EXTRA_GPOS 0x800000
24b23f37
YO
1011/* combined */
1012#define RE_DEBUG_EXTRA_STACK 0x280000
be8e71aa 1013
e68ec53f 1014#define RE_DEBUG_FLAG(x) (re_debug_flags & x)
be8e71aa
YO
1015/* Compile */
1016#define DEBUG_COMPILE_r(x) DEBUG_r( \
e68ec53f 1017 if (re_debug_flags & RE_DEBUG_COMPILE_MASK) x )
be8e71aa 1018#define DEBUG_PARSE_r(x) DEBUG_r( \
e68ec53f 1019 if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x )
be8e71aa 1020#define DEBUG_OPTIMISE_r(x) DEBUG_r( \
e68ec53f 1021 if (re_debug_flags & RE_DEBUG_COMPILE_OPTIMISE) x )
be8e71aa 1022#define DEBUG_PARSE_r(x) DEBUG_r( \
e68ec53f 1023 if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x )
be8e71aa 1024#define DEBUG_DUMP_r(x) DEBUG_r( \
e68ec53f 1025 if (re_debug_flags & RE_DEBUG_COMPILE_DUMP) x )
be8e71aa 1026#define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \
e68ec53f 1027 if (re_debug_flags & RE_DEBUG_COMPILE_TRIE) x )
f7819f85
A
1028#define DEBUG_FLAGS_r(x) DEBUG_r( \
1029 if (re_debug_flags & RE_DEBUG_COMPILE_FLAGS) x )
d9a72fcc
YO
1030#define DEBUG_TEST_r(x) DEBUG_r( \
1031 if (re_debug_flags & RE_DEBUG_COMPILE_TEST) x )
be8e71aa
YO
1032/* Execute */
1033#define DEBUG_EXECUTE_r(x) DEBUG_r( \
e68ec53f 1034 if (re_debug_flags & RE_DEBUG_EXECUTE_MASK) x )
be8e71aa 1035#define DEBUG_INTUIT_r(x) DEBUG_r( \
e68ec53f 1036 if (re_debug_flags & RE_DEBUG_EXECUTE_INTUIT) x )
be8e71aa 1037#define DEBUG_MATCH_r(x) DEBUG_r( \
e68ec53f 1038 if (re_debug_flags & RE_DEBUG_EXECUTE_MATCH) x )
be8e71aa 1039#define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \
e68ec53f 1040 if (re_debug_flags & RE_DEBUG_EXECUTE_TRIE) x )
be8e71aa
YO
1041
1042/* Extra */
1043#define DEBUG_EXTRA_r(x) DEBUG_r( \
e68ec53f 1044 if (re_debug_flags & RE_DEBUG_EXTRA_MASK) x )
a5ca303d
YO
1045#define DEBUG_OFFSETS_r(x) DEBUG_r( \
1046 if (re_debug_flags & RE_DEBUG_EXTRA_OFFSETS) x )
ab3bbdeb
YO
1047#define DEBUG_STATE_r(x) DEBUG_r( \
1048 if (re_debug_flags & RE_DEBUG_EXTRA_STATE) x )
24b23f37
YO
1049#define DEBUG_STACK_r(x) DEBUG_r( \
1050 if (re_debug_flags & RE_DEBUG_EXTRA_STACK) x )
e7707071
YO
1051#define DEBUG_BUFFERS_r(x) DEBUG_r( \
1052 if (re_debug_flags & RE_DEBUG_EXTRA_BUFFERS) x )
1053
a5ca303d
YO
1054#define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \
1055 if ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \
1056 (re_debug_flags & (RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE)) ) x )
07be1b83 1057#define MJD_OFFSET_DEBUG(x) DEBUG_r( \
a5ca303d 1058 if (re_debug_flags & RE_DEBUG_EXTRA_OFFDEBUG) \
be8e71aa
YO
1059 Perl_warn_nocontext x )
1060#define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \
e68ec53f 1061 if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x )
be8e71aa 1062#define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \
e68ec53f 1063 if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x )
be8e71aa
YO
1064
1065#define DEBUG_TRIE_r(x) DEBUG_r( \
e68ec53f 1066 if (re_debug_flags & (RE_DEBUG_COMPILE_TRIE \
be8e71aa 1067 | RE_DEBUG_EXECUTE_TRIE )) x )
2c296965
YO
1068#define DEBUG_GPOS_r(x) DEBUG_r( \
1069 if (re_debug_flags & RE_DEBUG_EXTRA_GPOS) x )
be8e71aa
YO
1070
1071/* initialization */
d2d8c711 1072/* get_sv() can return NULL during global destruction. */
e68ec53f
YO
1073#define GET_RE_DEBUG_FLAGS DEBUG_r({ \
1074 SV * re_debug_flags_sv = NULL; \
243a3b9c 1075 re_debug_flags_sv = PL_curcop ? get_sv(RE_DEBUG_FLAGS, GV_ADD) : NULL; \
e68ec53f 1076 if (re_debug_flags_sv) { \
8e9a8a48 1077 if (!SvIOK(re_debug_flags_sv)) \
e68ec53f 1078 sv_setuv(re_debug_flags_sv, RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \
8e9a8a48 1079 re_debug_flags=SvIV(re_debug_flags_sv); \
e68ec53f
YO
1080 }\
1081})
a3621e74
YO
1082
1083#ifdef DEBUGGING
ab3bbdeb 1084
8162b70e 1085#define GET_RE_DEBUG_FLAGS_DECL volatile IV re_debug_flags = 0; \
cb65b687 1086 PERL_UNUSED_VAR(re_debug_flags); GET_RE_DEBUG_FLAGS;
ab3bbdeb 1087
eecd4d11
KW
1088#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \
1089 const char * const rpv = \
1090 pv_pretty((dsv), (pv), (l), (m), \
1091 PL_colors[(c1)],PL_colors[(c2)], \
c89df6cf 1092 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \
0df25f3d 1093 const int rlen = SvCUR(dsv)
ab3bbdeb 1094
8ec77c3c 1095/* This is currently unsed in the core */
eecd4d11
KW
1096#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \
1097 const char * const rpv = \
1098 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \
1099 PL_colors[(c1)],PL_colors[(c2)], \
c89df6cf 1100 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) )
ab3bbdeb 1101
eecd4d11
KW
1102#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \
1103 const char * const rpv = \
1104 pv_pretty((dsv), (pv), (l), (m), \
1105 PL_colors[0], PL_colors[1], \
c89df6cf 1106 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \
ab3bbdeb 1107 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \
0a0e9afc 1108 )
ab3bbdeb
YO
1109
1110#define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0))
1111#define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "")
1112
1113#else /* if not DEBUGGING */
1114
a3621e74 1115#define GET_RE_DEBUG_FLAGS_DECL
ab3bbdeb
YO
1116#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2)
1117#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m)
1118#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m)
1119#define RE_SV_DUMPLEN(ItEm)
1120#define RE_SV_TAIL(ItEm)
1121
1122#endif /* DEBUG RELATED DEFINES */
a3621e74 1123
0bc0e44f 1124#define FIRST_NON_ASCII_DECIMAL_DIGIT 0x660 /* ARABIC_INDIC_DIGIT_ZERO */
2ca74f27 1125
64935bc6
KW
1126typedef enum {
1127 TRADITIONAL_BOUND = _CC_WORDCHAR,
ae3bb8ea 1128 GCB_BOUND,
6b659339 1129 LB_BOUND,
06ae2722 1130 SB_BOUND,
ae3bb8ea 1131 WB_BOUND
64935bc6
KW
1132} bound_type;
1133
e9a8c099 1134/*
14d04a33 1135 * ex: set ts=8 sts=4 sw=4 et:
e9a8c099 1136 */