This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.h: Shorten, clarify names of internal flags
[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
28d8d7f4 93
785a26d5
YO
94
95 typedef struct regexp_internal {
1f1031fe 96 int name_list_idx; /* Optional data index of an array of paren names */
7122b237
YO
97 union {
98 U32 *offsets; /* offset annotations 20001228 MJD
99 data about mapping the program to the
100 string -
101 offsets[0] is proglen when this is used
102 */
103 U32 proglen;
104 } u;
105
6d5c990f
RGS
106 regnode *regstclass; /* Optional startclass as identified or constructed
107 by the optimiser */
108 struct reg_data *data; /* Additional miscellaneous data used by the program.
109 Used to make it easier to clone and free arbitrary
110 data that the regops need. Often the ARG field of
111 a regop is an index into this structure */
3d2bd50a
DM
112 struct reg_code_block *code_blocks;/* positions of literal (?{}) */
113 int num_code_blocks; /* size of code_blocks[] */
6d5c990f
RGS
114 regnode program[1]; /* Unwarranted chumminess with compiler. */
115} regexp_internal;
116
117#define RXi_SET(x,y) (x)->pprivate = (void*)(y)
118#define RXi_GET(x) ((regexp_internal *)((x)->pprivate))
119#define RXi_GET_DECL(r,ri) regexp_internal *ri = RXi_GET(r)
785a26d5
YO
120/*
121 * Flags stored in regexp->intflags
122 * These are used only internally to the regexp engine
123 *
124 * See regexp.h for flags used externally to the regexp engine
125 */
e3e400ec
YO
126#define RXp_INTFLAGS(rx) ((rx)->intflags)
127#define RX_INTFLAGS(prog) RXp_INTFLAGS(ReANY(prog))
128
785a26d5
YO
129#define PREGf_SKIP 0x00000001
130#define PREGf_IMPLICIT 0x00000002 /* Converted .* to ^.* */
131#define PREGf_NAUGHTY 0x00000004 /* how exponential is this pattern? */
132#define PREGf_VERBARG_SEEN 0x00000008
133#define PREGf_CUTGROUP_SEEN 0x00000010
732caac7 134#define PREGf_USE_RE_EVAL 0x00000020 /* compiled with "use re 'eval'" */
e3e400ec
YO
135/* these used to be extflags, but are now intflags */
136#define PREGf_NOSCAN 0x00000040
33c28ab2 137 /* spare */
8e1490ee
YO
138#define PREGf_GPOS_SEEN 0x00000100
139#define PREGf_GPOS_FLOAT 0x00000200
785a26d5 140
d3d47aac
YO
141#define PREGf_ANCH_MBOL 0x00000400
142#define PREGf_ANCH_SBOL 0x00000800
143#define PREGf_ANCH_GPOS 0x00001000
8e1490ee 144
d3d47aac
YO
145#define PREGf_ANCH \
146 ( PREGf_ANCH_SBOL | PREGf_ANCH_GPOS | PREGf_ANCH_MBOL )
785a26d5
YO
147
148/* this is where the old regcomp.h started */
a687059c 149
c277df42 150struct regnode_string {
cd439c50 151 U8 str_len;
c277df42
IZ
152 U8 type;
153 U16 next_off;
cd439c50 154 char string[1];
c277df42 155};
a687059c 156
6bda09f9
YO
157/* Argument bearing node - workhorse,
158 arg1 is often for the data field */
c277df42
IZ
159struct regnode_1 {
160 U8 flags;
161 U8 type;
162 U16 next_off;
163 U32 arg1;
164};
165
6bda09f9
YO
166/* Similar to a regnode_1 but with an extra signed argument */
167struct regnode_2L {
168 U8 flags;
169 U8 type;
170 U16 next_off;
171 U32 arg1;
172 I32 arg2;
173};
174
175/* 'Two field' -- Two 16 bit unsigned args */
c277df42
IZ
176struct regnode_2 {
177 U8 flags;
178 U8 type;
179 U16 next_off;
180 U16 arg1;
181 U16 arg2;
182};
183
e0a1ff7a
KW
184/* This give the number of code points that can be in the bitmap of an ANYOF
185 * node. The shift number must currently be one of: 8..12. It can't be less
186 * than 8 (256) because some code relies on it being at least that. Above 12
187 * (4096), and you start running into warnings that some data structure widths
188 * have been exceeded, though the test suite as of this writing still passes
189 * for up through 16, which is as high as anyone would ever want to go,
190 * encompassing all of the Unicode BMP, and thus including all the economically
191 * important world scripts. At 12 most of them are: including Arabic,
192 * Cyrillic, Greek, Hebrew, Indian subcontinent, Latin, and Thai; but not Han,
193 * Japanese, nor Korean. (The regarglen structure in regnodes.h is a U8, and
194 * the trie types TRIEC and AHOCORASICKC are larger than U8 for shift values
195 * below above 12.) Be sure to benchmark before changing, as larger sizes do
196 * significantly slow down the test suite */
197#define NUM_ANYOF_CODE_POINTS (1 << 8)
6bda09f9 198
dcb20b36 199#define ANYOF_BITMAP_SIZE (NUM_ANYOF_CODE_POINTS / 8) /* 8 bits/Byte */
936ed897 200
1ee208c4
KW
201/* Note that these form structs which are supersets of the next smaller one, by
202 * appending fields. Alignment problems can occur if one of those optional
203 * fields requires stricter alignment than the base struct. And formal
204 * parameters that can really be two or more of the structs should be
205 * declared as the smallest one it could be. See commit message for
206 * 7dcac5f6a5195002b55c935ee1d67f67e1df280b. Regnode allocation is done
207 * without regard to alignment, and changing it to would also require changing
208 * the code that inserts and deletes regnodes. The basic single-argument
209 * regnode has a U32, which is what reganode() allocates as a unit. Therefore
210 * no field can require stricter alignment than U32. */
211
786e8c11 212/* also used by trie */
936ed897
IZ
213struct regnode_charclass {
214 U8 flags;
215 U8 type;
216 U16 next_off;
9da99c5e 217 U32 arg1; /* set by set_ANYOF_arg() */
19103b91 218 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */
936ed897
IZ
219};
220
bf3f2d85
KW
221/* has runtime (locale) \d, \w, ..., [:posix:] classes */
222struct regnode_charclass_class {
93e92956 223 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
936ed897
IZ
224 U8 type;
225 U16 next_off;
7e817188 226 U32 arg1;
03fa83ba 227 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
9cf78fa0 228 U32 classflags; /* and run-time */
936ed897
IZ
229};
230
85c8e306
KW
231/* A synthetic start class (SSC); is a regnode_charclass_posixl_fold, plus an
232 * extra SV*, used only during its construction and which is not used by
233 * regexec.c. Note that the 'next_off' field is unused, as the SSC stands
234 * alone, so there is never a next node. Also, there is no alignment issue,
b8fda935 235 * because these are declared or allocated as a complete unit so the compiler
85c8e306
KW
236 * takes care of alignment. This is unlike the other regnodes which are
237 * allocated in terms of multiples of a single-argument regnode. SSC nodes can
238 * have a pointer field because there is no alignment issue, and because it is
239 * set to NULL after construction, before any cloning of the pattern */
b8f7bb16 240struct regnode_ssc {
93e92956 241 U8 flags; /* ANYOF_MATCHES_POSIXL bit must go here */
b8f7bb16
KW
242 U8 type;
243 U16 next_off;
7e817188 244 U32 arg1;
03fa83ba 245 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time ... */
631e9161 246 U32 classflags; /* ... and run-time */
85c8e306
KW
247
248 /* Auxiliary, only used during construction; NULL afterwards: list of code
249 * points matched */
250 SV* invlist;
b8f7bb16
KW
251};
252
71068078
KW
253/* We take advantage of 'next_off' not otherwise being used in the SSC by
254 * actually using it: by setting it to 1. This allows us to test and
255 * distinguish between an SSC and other ANYOF node types, as 'next_off' cannot
256 * otherwise be 1, because it is the offset to the next regnode expressed in
257 * units of regnodes. Since an ANYOF node contains extra fields, it adds up
258 * to 12 regnode units on 32-bit systems, (hence the minimum this can be (if
259 * not 0) is 11 there. Even if things get tightly packed on a 64-bit system,
260 * it still would be more than 1. */
261#define set_ANYOF_SYNTHETIC(n) STMT_START{ OP(n) = ANYOF; \
262 NEXT_OFF(n) = 1; \
263 } STMT_END
c1071334 264#define is_ANYOF_SYNTHETIC(n) (PL_regkind[OP(n)] == ANYOF && NEXT_OFF(n) == 1)
71068078 265
9c7e81e8
AD
266/* XXX fix this description.
267 Impose a limit of REG_INFTY on various pattern matching operations
268 to limit stack growth and to avoid "infinite" recursions.
269*/
270/* The default size for REG_INFTY is I16_MAX, which is the same as
271 SHORT_MAX (see perl.h). Unfortunately I16 isn't necessarily 16 bits
272 (see handy.h). On the Cray C90, sizeof(short)==4 and hence I16_MAX is
273 ((1<<31)-1), while on the Cray T90, sizeof(short)==8 and I16_MAX is
274 ((1<<63)-1). To limit stack growth to reasonable sizes, supply a
275 smaller default.
276 --Andy Dougherty 11 June 1998
277*/
278#if SHORTSIZE > 2
279# ifndef REG_INFTY
280# define REG_INFTY ((1<<15)-1)
281# endif
282#endif
283
284#ifndef REG_INFTY
83cfe48c
IZ
285# define REG_INFTY I16_MAX
286#endif
a687059c 287
e91177ed
IZ
288#define ARG_VALUE(arg) (arg)
289#define ARG__SET(arg,val) ((arg) = (val))
c277df42 290
e8e3b0fb
JH
291#undef ARG
292#undef ARG1
293#undef ARG2
294
c277df42
IZ
295#define ARG(p) ARG_VALUE(ARG_LOC(p))
296#define ARG1(p) ARG_VALUE(ARG1_LOC(p))
297#define ARG2(p) ARG_VALUE(ARG2_LOC(p))
6bda09f9 298#define ARG2L(p) ARG_VALUE(ARG2L_LOC(p))
786e8c11 299
c277df42
IZ
300#define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val))
301#define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val))
302#define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val))
6bda09f9 303#define ARG2L_SET(p, val) ARG__SET(ARG2L_LOC(p), (val))
c277df42 304
e8e3b0fb
JH
305#undef NEXT_OFF
306#undef NODE_ALIGN
307
075abff3
AL
308#define NEXT_OFF(p) ((p)->next_off)
309#define NODE_ALIGN(node)
aa48e906
YO
310/* the following define was set to 0xde in 075abff3
311 * as part of some linting logic. I have set it to 0
312 * as otherwise in every place where we /might/ set flags
313 * we have to set it 0 explicitly, which duplicates
314 * assignments and IMO adds an unacceptable level of
315 * surprise to working in the regex engine. If this
316 * is changed from 0 then at the very least make sure
317 * that SBOL for /^/ sets the flags to 0 explicitly.
318 * -- Yves */
319#define NODE_ALIGN_FILL(node) ((node)->flags = 0)
a687059c 320
c277df42
IZ
321#define SIZE_ALIGN NODE_ALIGN
322
e8e3b0fb
JH
323#undef OP
324#undef OPERAND
325#undef MASK
326#undef STRING
327
e91177ed 328#define OP(p) ((p)->type)
b988e673
KW
329#define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \
330 regnode types. For some, it's the \
331 character set of the regnode */
e91177ed 332#define OPERAND(p) (((struct regnode_string *)p)->string)
cd439c50
IZ
333#define MASK(p) ((char*)OPERAND(p))
334#define STR_LEN(p) (((struct regnode_string *)p)->str_len)
335#define STRING(p) (((struct regnode_string *)p)->string)
336#define STR_SZ(l) ((l + sizeof(regnode) - 1) / sizeof(regnode))
337#define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p))+1)
338
e8e3b0fb
JH
339#undef NODE_ALIGN
340#undef ARG_LOC
341#undef NEXTOPER
342#undef PREVOPER
343
e91177ed
IZ
344#define NODE_ALIGN(node)
345#define ARG_LOC(p) (((struct regnode_1 *)p)->arg1)
346#define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1)
347#define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2)
6bda09f9 348#define ARG2L_LOC(p) (((struct regnode_2L *)p)->arg2)
786e8c11 349
e91177ed
IZ
350#define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */
351#define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2)
352
353#define NODE_STEP_B 4
c277df42
IZ
354
355#define NEXTOPER(p) ((p) + NODE_STEP_REGNODE)
356#define PREVOPER(p) ((p) - NODE_STEP_REGNODE)
357
e91177ed 358#define FILL_ADVANCE_NODE(ptr, op) STMT_START { \
c277df42 359 (ptr)->type = op; (ptr)->next_off = 0; (ptr)++; } STMT_END
e91177ed 360#define FILL_ADVANCE_NODE_ARG(ptr, op, arg) STMT_START { \
c277df42 361 ARG_SET(ptr, arg); FILL_ADVANCE_NODE(ptr, op); (ptr) += 1; } STMT_END
9990d5ab
KW
362#define FILL_ADVANCE_NODE_2L_ARG(ptr, op, arg1, arg2) \
363 STMT_START { \
364 ARG_SET(ptr, arg1); \
365 ARG2L_SET(ptr, arg2); \
366 FILL_ADVANCE_NODE(ptr, op); \
367 (ptr) += 2; \
368 } STMT_END
a687059c 369
22c35a8c 370#define REG_MAGIC 0234
a687059c 371
9a81a976 372#define SIZE_ONLY (RExC_emit == (regnode *) & RExC_emit_dummy)
6d15ebf3
KW
373#define PASS1 SIZE_ONLY
374#define PASS2 (! SIZE_ONLY)
c277df42 375
fe8d1b7c
KW
376/* An ANYOF node is basically a bitmap with the index being a code point. If
377 * the bit for that code point is 1, the code point matches; if 0, it doesn't
378 * match (complemented if inverted). There is an additional mechanism to deal
379 * with cases where the bitmap is insufficient in and of itself. This #define
380 * indicates if the bitmap does fully represent what this ANYOF node can match.
381 * The ARG is set to this special value (since 0, 1, ... are legal, but will
382 * never reach this high). */
93e92956 383#define ANYOF_ONLY_HAS_BITMAP ((U32) -1)
137165a6 384
fe8d1b7c
KW
385/* When the bimap isn't completely sufficient for handling the ANYOF node,
386 * flags (in node->flags of the ANYOF node) get set to indicate this. These
387 * are perennially in short supply. Beyond several cases where warnings need
388 * to be raised under certain circumstances, currently, there are six cases
389 * where the bitmap alone isn't sufficient. We could use six flags to
390 * represent the 6 cases, but to save flags bits, we play some games. The
391 * cases are:
d2f465e9 392 *
fe8d1b7c
KW
393 * 1) The bitmap has a compiled-in very finite size. So something else needs
394 * to be used to specify if a code point that is too large for the bitmap
395 * actually matches. The mechanism currently is a swash or inversion
396 * list. ANYOF_ONLY_HAS_BITMAP, described above, being TRUE indicates
397 * there are no matches of too-large code points. But if it is FALSE,
398 * then almost certainly there are matches too large for the bitmap. (The
399 * other cases, described below, either imply this one or are extremely
400 * rare in practice.) So we can just assume that a too-large code point
401 * will need something beyond the bitmap if ANYOF_ONLY_HAS_BITMAP is
402 * FALSE, instead of having a separate flag for this.
403 * 2) A subset of item 1) is if all possible code points outside the bitmap
404 * match. This is a common occurrence when the class is complemented,
405 * like /[^ij]/. Therefore a bit is reserved to indicate this,
e2f5e63d 406 * ANYOF_MATCHES_ALL_ABOVE_BITMAP. If it became necessary, this flag could
fe8d1b7c
KW
407 * be replaced by using the normal swash mechanism, but with a performance
408 * penalty.
409 * 3) Under /d rules, it can happen that code points that are in the upper
410 * latin1 range (\x80-\xFF or their equivalents on EBCDIC platforms) match
411 * only if the runtime target string being matched against is UTF-8. For
412 * example /[\w[:punct:]]/d. This happens only for posix classes (with a
413 * couple of exceptions, like \d), and all such ones also have
414 * above-bitmap matches. Thus, 3) implies 1) as well. Note that /d rules
415 * are no longer encouraged; 'use 5.14' or higher deselects them. But a
416 * flag is required so that they can be properly handled. But it can be a
417 * shared flag: see 5) below.
418 * 4) Also under /d rules, something like /[\Wfoo] will match everything in
419 * the \x80-\xFF range, unless the string being matched against is UTF-8.
420 * A swash could be created for this case, but this is relatively common,
421 * and it turns out that it's all or nothing: if any one of these code
422 * points matches, they all do. Hence a single bit suffices. We use a
e2f5e63d 423 * shared flag that doesn't take up space by itself:
fe8d1b7c
KW
424 * ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER.
425 * This also implies 1), with one exception: [:^cntrl:].
426 * 5) A user-defined \p{} property may not have been defined by the time the
427 * regex is compiled. In this case, we don't know until runtime what it
428 * will match, so we have to assume it could match anything, including
429 * code points that ordinarily would be in the bitmap. A flag bit is
430 * necessary to indicate this , though it can be shared with the item 3)
431 * flag, as that only occurs under /d, and this only occurs under non-d.
432 * This case is quite uncommon in the field, and the /(?[ ...])/ construct
433 * is a better way to accomplish what this feature does. This case also
434 * implies 1).
435 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP
e2f5e63d 436 * is the shared flag.
fe8d1b7c
KW
437 * 6) /[foo]/il may have folds that are only valid if the runtime locale is a
438 * UTF-8 one. These are quite rare, so it would be good to avoid the
439 * expense of looking for them. But /l matching is slow anyway, and we've
e2f5e63d 440 * traditionally not worried too much about its performance. And this
037715a6 441 * condition requires the ANYOFL_FOLD flag to be set, so testing for
fe8d1b7c
KW
442 * that flag would be sufficient to rule out most cases of this. So it is
443 * unclear if this should have a flag or not. But, one is currently
037715a6
KW
444 * allocated for this purpose, ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE (and
445 * the text below indicates how to share it, should another bit be
446 * needed).
d2f465e9 447 *
fe8d1b7c 448 * At the moment, there are no spare bits, but this could be changed by various
e2f5e63d
KW
449 * tricks.
450 *
037715a6
KW
451 * Note that item ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE is not independent of
452 * the ANYOFL_FOLD flag below. Also, the ANYOFL_UTF8_LOCALE_REQD flag is set
453 * only if both these aren't. We can therefore share
454 * ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE with ANYOFL_UTF8_LOCALE_REQD, so what
455 * the shared flag means depends on the ANYOFL_FOLD flag.
d2f465e9 456 *
fe8d1b7c
KW
457 * Beyond that, note that the information may be conveyed by creating new
458 * regnode types. This is not the best solution, as shown later in this
459 * paragraph, but it is something that is feasible. We could have a regnode
460 * for ANYOF_INVERT, for example. A complication of this is that the regexec.c
461 * REGINCLASS macro assumes that it can just use the bitmap if no flags are
462 * set. This would have to be changed to add extra tests for the node type, or
e2f5e63d 463 * a special flag reserved that means unspecified special handling, and then the
fe8d1b7c 464 * node-type would be used internally to sort that out. So we could gain a bit
e2f5e63d 465 * by having an ANYOF_SPECIAL flag, and a node type for INVERT, and another for
fe8d1b7c
KW
466 * POSIXL, and still another for INVERT_POSIXL. This example illustrates one
467 * problem with this, a combinatorial explosion of node types. The one node
468 * type khw can think of that doesn't have this explosion issue is
037715a6 469 * ANYOFL_UTF8_LOCALE_REQD. This flag is a natural candidate for being a separate
fe8d1b7c 470 * node type because it is a specialization of the current ANYOFL, and because
e2f5e63d 471 * no other ANYOFL-only flags are set when it is; also most of its uses are
fe8d1b7c 472 * actually outside the reginclass() function, so this could be done with no
e2f5e63d
KW
473 * performance penalty. But since it can be shared, as noted above, it doesn't
474 * take up space anyway. Another issue when turning a flag into a node type, is
475 * that a SSC may use that flag -- not just a regular ANYOF[DL]?. In the case
037715a6 476 * of ANYOFL_UTF8_LOCALE_REQD, the only likely problem is accurately settting the
e2f5e63d
KW
477 * SSC node-type to the new one, which would likely involve S_ssc_or and
478 * S_ssc_and, and not how the SSC currently gets set to ANYOFL.
fe8d1b7c 479 *
e2f5e63d 480 * Another possibility is to instead rename the ANYOF_POSIXL flag to be
fe8d1b7c
KW
481 * ANYOFL_LARGE, to mean that the ANYOF node has an extra 32 bits beyond what a
482 * regular one does. That's what it effectively means now, with the extra
e2f5e63d
KW
483 * space all for the POSIX class flags. But those classes actually only occupy
484 * 30 bits, so 2 of the locale flags could be moved to that extra space. The
485 * downside of this is that ANYOFL nodes with whichever of the flags get moved
486 * would have to have the extra space always allocated.
fe8d1b7c
KW
487 *
488 * One could completely remove ANYOFL_LARGE and make all ANYOFL nodes large.
489 * The 30 bits in the extra word would indicate if a posix class should be
490 * looked up or not. There isn't an SSC problem as all SSCs are this large
491 * anyway, and the SSC could be set to this node type. REGINCLASS would have
492 * to be modified so that if the node type were this, it would call
493 * reginclass(), as the flag bit that indicates to do this now would be gone.
e2f5e63d
KW
494 * If 2 locale flags are moved to the larger structure, this would free up a
495 * total of 4 bits. If this were done, we could create an ANYOF_INVERT
496 * node-type without a combinatorial explosion, getting us to 5 bits. And,
497 * keep in mind that ANYOF_MATCHES_ALL_ABOVE_BITMAP is solely for performance,
498 * so could be removed. The other performance-related flags are shareable with
499 * flags that are required.
a4525e78 500 *
34fdef84 501 * Several flags are not used in synthetic start class (SSC) nodes, so could be
93e92956
KW
502 * shared should new flags be needed for SSCs, like SSC_MATCHES_EMPTY_STRING
503 * now. */
b8c5462f 504
d2f465e9
KW
505/* If this is set, the result of the match should be complemented. regexec.c
506 * is expecting this to be in the low bit. Never in an SSC */
93e92956 507#define ANYOF_INVERT 0x01
1984f6a0 508
a0dd4231 509/* For the SSC node only, which cannot be inverted, so is shared with that bit.
93e92956
KW
510 * This is used only during regex compilation. */
511#define SSC_MATCHES_EMPTY_STRING ANYOF_INVERT
a0dd4231 512
509d4de0
KW
513/* Set if this is a regnode_charclass_posixl vs a regnode_charclass. This
514 * is used for runtime \d, \w, [:posix:], ..., which are used only in locale
515 * and the optimizer's synthetic start class. Non-locale \d, etc are resolved
516 * at compile-time. Only set under /l; can be in SSC */
517#define ANYOF_MATCHES_POSIXL 0x02
b63f73da
KW
518
519/* The fold is calculated and stored in the bitmap where possible at compile
520 * time. However under locale, the actual folding varies depending on
521 * what the locale is at the time of execution, so it has to be deferred until
d2f465e9 522 * then. Only set under /l; never in an SSC */
037715a6 523#define ANYOFL_FOLD 0x04
b63f73da 524
037715a6 525/* If set, ANYOFL_FOLD is also set, and there are potential matches that
8c9b4e63 526 * will be valid only if the locale is a UTF-8 one. */
037715a6 527#define ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE 0x08
b3b1cf17 528
509d4de0 529/* If set, means to warn if runtime locale isn't a UTF-8 one. Only under /l.
108316fb
KW
530 * If set, none of INVERT, LOC_FOLD, POSIXL,
531 * ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP can
509d4de0 532 * be set. Can be in an SSC */
037715a6 533#define ANYOFL_UTF8_LOCALE_REQD 0x10
c8d3cd88 534
509d4de0
KW
535/* If set, the node matches every code point NUM_ANYOF_CODE_POINTS and above.
536 * Can be in an SSC */
b3b1cf17 537#define ANYOF_MATCHES_ALL_ABOVE_BITMAP 0x20
ef87b810 538
108316fb
KW
539/* Shared bit:
540 * Under /d it means the ANYOFD node matches more things if the target
541 * string is encoded in UTF-8; any such things will be non-ASCII,
542 * characters that are < 256, and can be accessed via the swash.
543 * When not under /d, it means the ANYOF node contains a user-defined
544 * property that wasn't yet defined at the time the regex was compiled,
545 * and so must be looked up at runtime, by creating a swash
546 * (These uses are mutually exclusive because a user-defined property is
547 * specified by \p{}, and \p{} implies /u which deselects /d). The long macro
548 * name is to make sure that you are cautioned about its shared nature. Only
549 * the non-/d meaning can be in an SSC */
550#define ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP 0x40
f240c685
KW
551
552/* Shared bit:
d2f465e9 553 * Under /d it means the ANYOFD node matches all non-ASCII Latin1
f240c685
KW
554 * characters when the target string is not in utf8.
555 * When not under /d, it means the ANYOF node should raise a warning if
556 * matching against an above-Unicode code point.
557 * (These uses are mutually exclusive because the warning requires a \p{}, and
558 * \p{} implies /u which deselects /d). An SSC node only has this bit set if
559 * what is meant is the warning. The long macro name is to make sure that you
560 * are cautioned about its shared nature */
561#define ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER 0x80
7bc66b18 562
3b04b210 563#define ANYOF_FLAGS_ALL (0xff)
653099ff 564
037715a6 565#define ANYOF_LOCALE_FLAGS (ANYOFL_FOLD | ANYOF_MATCHES_POSIXL)
ace6b0e4 566
eff8b7dc
KW
567/* These are the flags that apply to both regular ANYOF nodes and synthetic
568 * start class nodes during construction of the SSC. During finalization of
d2f465e9 569 * the SSC, other of the flags may get added to it */
037715a6 570#define ANYOF_COMMON_FLAGS ANYOFL_UTF8_LOCALE_REQD
eff8b7dc 571
936ed897 572/* Character classes for node->classflags of ANYOF */
653099ff 573/* Should be synchronized with a table in regprop() */
575a8a29 574/* 2n should be the normal one, paired with its complement at 2n+1 */
b8c5462f 575
08ada29e
KW
576#define ANYOF_ALPHA ((_CC_ALPHA) * 2)
577#define ANYOF_NALPHA ((ANYOF_ALPHA) + 1)
e8d596e0
KW
578#define ANYOF_ALPHANUMERIC ((_CC_ALPHANUMERIC) * 2) /* [[:alnum:]] isalnum(3), utf8::IsAlnum */
579#define ANYOF_NALPHANUMERIC ((ANYOF_ALPHANUMERIC) + 1)
08ada29e
KW
580#define ANYOF_ASCII ((_CC_ASCII) * 2)
581#define ANYOF_NASCII ((ANYOF_ASCII) + 1)
e8d596e0
KW
582#define ANYOF_BLANK ((_CC_BLANK) * 2) /* GNU extension: space and tab: non-vertical space */
583#define ANYOF_NBLANK ((ANYOF_BLANK) + 1)
b0d691b2
KW
584#define ANYOF_CASED ((_CC_CASED) * 2) /* Pseudo class for [:lower:] or
585 [:upper:] under /i */
586#define ANYOF_NCASED ((ANYOF_CASED) + 1)
08ada29e
KW
587#define ANYOF_CNTRL ((_CC_CNTRL) * 2)
588#define ANYOF_NCNTRL ((ANYOF_CNTRL) + 1)
e8d596e0
KW
589#define ANYOF_DIGIT ((_CC_DIGIT) * 2) /* \d */
590#define ANYOF_NDIGIT ((ANYOF_DIGIT) + 1)
08ada29e
KW
591#define ANYOF_GRAPH ((_CC_GRAPH) * 2)
592#define ANYOF_NGRAPH ((ANYOF_GRAPH) + 1)
593#define ANYOF_LOWER ((_CC_LOWER) * 2)
594#define ANYOF_NLOWER ((ANYOF_LOWER) + 1)
595#define ANYOF_PRINT ((_CC_PRINT) * 2)
596#define ANYOF_NPRINT ((ANYOF_PRINT) + 1)
597#define ANYOF_PUNCT ((_CC_PUNCT) * 2)
598#define ANYOF_NPUNCT ((ANYOF_PUNCT) + 1)
e8d596e0
KW
599#define ANYOF_SPACE ((_CC_SPACE) * 2) /* \s */
600#define ANYOF_NSPACE ((ANYOF_SPACE) + 1)
08ada29e
KW
601#define ANYOF_UPPER ((_CC_UPPER) * 2)
602#define ANYOF_NUPPER ((ANYOF_UPPER) + 1)
e8d596e0
KW
603#define ANYOF_WORDCHAR ((_CC_WORDCHAR) * 2) /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */
604#define ANYOF_NWORDCHAR ((ANYOF_WORDCHAR) + 1)
08ada29e
KW
605#define ANYOF_XDIGIT ((_CC_XDIGIT) * 2)
606#define ANYOF_NXDIGIT ((ANYOF_XDIGIT) + 1)
b8c5462f 607
87cad4c0 608/* pseudo classes below this, not stored in the class bitmap, but used as flags
e1d1eefb
YO
609 during compilation of char classes */
610
b2ef2e34
KW
611#define ANYOF_VERTWS ((_CC_VERTSPACE) * 2)
612#define ANYOF_NVERTWS ((ANYOF_VERTWS)+1)
e1d1eefb 613
b2ef2e34
KW
614/* It is best if this is the last one, as all above it are stored as bits in a
615 * bitmap, and it isn't part of that bitmap */
616#if _CC_VERTSPACE != _HIGHEST_REGCOMP_DOT_H_SYNC
617# error Problem with handy.h _HIGHEST_REGCOMP_DOT_H_SYNC #define
618#endif
619
8efd3f97
KW
620#define ANYOF_POSIXL_MAX (ANYOF_VERTWS) /* So upper loop limit is written:
621 * '< ANYOF_MAX'
622 * Hence doesn't include VERTWS, as that
623 * is a pseudo class */
624#define ANYOF_MAX ANYOF_POSIXL_MAX
625
626#if (ANYOF_POSIXL_MAX > 32) /* Must fit in 32-bit word */
b2ef2e34 627# error Problem with handy.h _CC_foo #defines
a0947d7b
KW
628#endif
629
8efd3f97
KW
630#define ANYOF_HORIZWS ((ANYOF_POSIXL_MAX)+2) /* = (ANYOF_NVERTWS + 1) */
631#define ANYOF_NHORIZWS ((ANYOF_POSIXL_MAX)+3)
87cad4c0 632
8efd3f97
KW
633#define ANYOF_UNIPROP ((ANYOF_POSIXL_MAX)+4) /* Used to indicate a Unicode
634 property: \p{} or \P{} */
f39a4dc2 635
b8c5462f
JH
636/* Backward source code compatibility. */
637
638#define ANYOF_ALNUML ANYOF_ALNUM
639#define ANYOF_NALNUML ANYOF_NALNUM
640#define ANYOF_SPACEL ANYOF_SPACE
641#define ANYOF_NSPACEL ANYOF_NSPACE
8ebfc29a
KW
642#define ANYOF_ALNUM ANYOF_WORDCHAR
643#define ANYOF_NALNUM ANYOF_NWORDCHAR
b8c5462f
JH
644
645/* Utility macros for the bitmap and classes of ANYOF */
646
936ed897 647#define ANYOF_FLAGS(p) ((p)->flags)
b8c5462f 648
bc51fd78 649#define ANYOF_BIT(c) (1U << ((c) & 7))
b8c5462f 650
d1d040e5 651#define ANYOF_POSIXL_SET(p, c) (((regnode_charclass_posixl*) (p))->classflags |= (1U << (c)))
8efd3f97
KW
652#define ANYOF_CLASS_SET(p, c) ANYOF_POSIXL_SET((p), (c))
653
d1d040e5 654#define ANYOF_POSIXL_CLEAR(p, c) (((regnode_charclass_posixl*) (p))->classflags &= ~ (1U <<(c)))
8efd3f97
KW
655#define ANYOF_CLASS_CLEAR(p, c) ANYOF_POSIXL_CLEAR((p), (c))
656
d1d040e5 657#define ANYOF_POSIXL_TEST(p, c) (((regnode_charclass_posixl*) (p))->classflags & (1U << (c)))
8efd3f97 658#define ANYOF_CLASS_TEST(p, c) ANYOF_POSIXL_TEST((p), (c))
b8c5462f 659
d1d040e5 660#define ANYOF_POSIXL_ZERO(ret) STMT_START { ((regnode_charclass_posixl*) (ret))->classflags = 0; } STMT_END
8efd3f97 661#define ANYOF_CLASS_ZERO(ret) ANYOF_POSIXL_ZERO(ret)
9cf78fa0
KW
662
663/* Shifts a bit to get, eg. 0x4000_0000, then subtracts 1 to get 0x3FFF_FFFF */
d1d040e5 664#define ANYOF_POSIXL_SETALL(ret) STMT_START { ((regnode_charclass_posixl*) (ret))->classflags = ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1; } STMT_END
8efd3f97 665#define ANYOF_CLASS_SETALL(ret) ANYOF_POSIXL_SETALL(ret)
9cf78fa0 666
8efd3f97 667#define ANYOF_POSIXL_TEST_ANY_SET(p) \
93e92956 668 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
d1d040e5 669 && (((regnode_charclass_posixl*)(p))->classflags))
8efd3f97 670#define ANYOF_CLASS_TEST_ANY_SET(p) ANYOF_POSIXL_TEST_ANY_SET(p)
e867ce45 671
e0e1be5f
KW
672/* Since an SSC always has this field, we don't have to test for that; nor do
673 * we want to because the bit isn't set for SSC during its construction */
674#define ANYOF_POSIXL_SSC_TEST_ANY_SET(p) \
675 cBOOL(((regnode_ssc*)(p))->classflags)
676#define ANYOF_POSIXL_SSC_TEST_ALL_SET(p) /* Are all bits set? */ \
677 (((regnode_ssc*) (p))->classflags \
678 == ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1)
679
680#define ANYOF_POSIXL_TEST_ALL_SET(p) \
93e92956 681 ((ANYOF_FLAGS(p) & ANYOF_MATCHES_POSIXL) \
e0e1be5f
KW
682 && ((regnode_charclass_posixl*) (p))->classflags \
683 == ((1U << ((ANYOF_POSIXL_MAX) - 1))) - 1)
fb38762f 684
b27fd97a 685#define ANYOF_POSIXL_OR(source, dest) STMT_START { (dest)->classflags |= (source)->classflags ; } STMT_END
8efd3f97 686#define ANYOF_CLASS_OR(source, dest) ANYOF_POSIXL_OR((source), (dest))
936ed897 687
cdd87c1d
KW
688#define ANYOF_POSIXL_AND(source, dest) STMT_START { (dest)->classflags &= (source)->classflags ; } STMT_END
689
9cf78fa0 690#define ANYOF_BITMAP_ZERO(ret) Zero(((struct regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char)
936ed897 691#define ANYOF_BITMAP(p) (((struct regnode_charclass*)(p))->bitmap)
786e8c11 692#define ANYOF_BITMAP_BYTE(p, c) (ANYOF_BITMAP(p)[(((U8)(c)) >> 3) & 31])
b8c5462f
JH
693#define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c))
694#define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c))
bc51fd78 695#define ANYOF_BITMAP_TEST(p, c) cBOOL(ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c))
b8c5462f 696
f8bef550
NC
697#define ANYOF_BITMAP_SETALL(p) \
698 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE)
699#define ANYOF_BITMAP_CLEARALL(p) \
700 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE)
f8bef550 701
888ee784
KW
702#define ANYOF_SKIP EXTRA_SIZE(struct regnode_charclass)
703#define ANYOF_POSIXL_SKIP EXTRA_SIZE(regnode_charclass_posixl)
3a15e693 704
a687059c
LW
705/*
706 * Utility definitions.
707 */
2304df62 708#ifndef CHARMASK
075abff3 709# define UCHARAT(p) ((int)*(const U8*)(p))
a687059c 710#else
075abff3 711# define UCHARAT(p) ((int)*(p)&CHARMASK)
a687059c 712#endif
a687059c 713
c277df42
IZ
714#define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode))
715
e384d5c1
YO
716#define REG_ZERO_LEN_SEEN 0x00000001
717#define REG_LOOKBEHIND_SEEN 0x00000002
718#define REG_GPOS_SEEN 0x00000004
471f5387 719/* spare */
e384d5c1
YO
720#define REG_RECURSE_SEEN 0x00000020
721#define REG_TOP_LEVEL_BRANCHES_SEEN 0x00000040
722#define REG_VERBARG_SEEN 0x00000080
723#define REG_CUTGROUP_SEEN 0x00000100
724#define REG_RUN_ON_COMMENT_SEEN 0x00000200
725#define REG_UNFOLDED_MULTI_SEEN 0x00000400
726#define REG_GOSTART_SEEN 0x00000800
ee273784
YO
727#define REG_UNBOUNDED_QUANTIFIER_SEEN 0x00001000
728
d09b2d29 729
73c4f7a1
GS
730START_EXTERN_C
731
be8e71aa
YO
732#ifdef PLUGGABLE_RE_EXTENSION
733#include "re_nodes.h"
734#else
d09b2d29 735#include "regnodes.h"
be8e71aa 736#endif
d09b2d29 737
f9f4320a
YO
738#ifndef PLUGGABLE_RE_EXTENSION
739#ifndef DOINIT
740EXTCONST regexp_engine PL_core_reg_engine;
741#else /* DOINIT */
742EXTCONST regexp_engine PL_core_reg_engine = {
fe578d7f 743 Perl_re_compile,
2fdbfb4d 744 Perl_regexec_flags,
f9f4320a
YO
745 Perl_re_intuit_start,
746 Perl_re_intuit_string,
2fdbfb4d
AB
747 Perl_regfree_internal,
748 Perl_reg_numbered_buff_fetch,
749 Perl_reg_numbered_buff_store,
750 Perl_reg_numbered_buff_length,
192b9cd1
AB
751 Perl_reg_named_buff,
752 Perl_reg_named_buff_iter,
49d7dfbc 753 Perl_reg_qr_package,
f9f4320a 754#if defined(USE_ITHREADS)
3c13cae6 755 Perl_regdupe_internal,
f9f4320a 756#endif
3c13cae6 757 Perl_re_op_compile
f9f4320a
YO
758};
759#endif /* DOINIT */
760#endif /* PLUGGABLE_RE_EXTENSION */
761
762
73c4f7a1 763END_EXTERN_C
cad2e5aa 764
cad2e5aa 765
261faec3
MJD
766/* .what is a character array with one character for each member of .data
767 * The character describes the function of the corresponding .data item:
af534a04 768 * a - AV for paren_name_list under DEBUGGING
0111c4fd 769 * f - start-class data for regstclass optimization
68e2671b 770 * l - start op for literal (?{EVAL}) item
d63c20f2 771 * L - start op for literal (?{EVAL}) item, with separate CV (qr//)
b30fcab9 772 * r - pointer to an embedded code-containing qr, e.g. /ab$qr/
38a44b82 773 * s - swash for Unicode-style character class, and the multicharacter
9e55ce06
JH
774 * strings resulting from casefolding the single-character entries
775 * in the character class
a3621e74 776 * t - trie struct
55eed653 777 * u - trie struct's widecharmap (a HV, so can't share, must dup)
2b8b4781 778 * also used for revcharmap and words under DEBUGGING
07be1b83 779 * T - aho-trie struct
81714fb9 780 * S - sv for named capture lookup
261faec3 781 * 20010712 mjd@plover.com
c3613ed7 782 * (Remember to update re_dup() and pregfree() if you add any items.)
261faec3 783 */
cad2e5aa
JH
784struct reg_data {
785 U32 count;
786 U8 *what;
787 void* data[1];
788};
789
a1cac82e
NC
790/* Code in S_to_utf8_substr() and S_to_byte_substr() in regexec.c accesses
791 anchored* and float* via array indexes 0 and 1. */
cad2e5aa 792#define anchored_substr substrs->data[0].substr
33b8afdf 793#define anchored_utf8 substrs->data[0].utf8_substr
cad2e5aa 794#define anchored_offset substrs->data[0].min_offset
1de06328
YO
795#define anchored_end_shift substrs->data[0].end_shift
796
cad2e5aa 797#define float_substr substrs->data[1].substr
33b8afdf 798#define float_utf8 substrs->data[1].utf8_substr
cad2e5aa
JH
799#define float_min_offset substrs->data[1].min_offset
800#define float_max_offset substrs->data[1].max_offset
1de06328
YO
801#define float_end_shift substrs->data[1].end_shift
802
cad2e5aa 803#define check_substr substrs->data[2].substr
33b8afdf 804#define check_utf8 substrs->data[2].utf8_substr
cad2e5aa
JH
805#define check_offset_min substrs->data[2].min_offset
806#define check_offset_max substrs->data[2].max_offset
1de06328 807#define check_end_shift substrs->data[2].end_shift
a3621e74 808
8d919b0a
FC
809#define RX_ANCHORED_SUBSTR(rx) (ReANY(rx)->anchored_substr)
810#define RX_ANCHORED_UTF8(rx) (ReANY(rx)->anchored_utf8)
811#define RX_FLOAT_SUBSTR(rx) (ReANY(rx)->float_substr)
812#define RX_FLOAT_UTF8(rx) (ReANY(rx)->float_utf8)
a3621e74
YO
813
814/* trie related stuff */
5d9a96ca 815
a3621e74
YO
816/* a transition record for the state machine. the
817 check field determines which state "owns" the
818 transition. the char the transition is for is
819 determined by offset from the owning states base
820 field. the next field determines which state
821 is to be transitioned to if any.
822*/
823struct _reg_trie_trans {
824 U32 next;
825 U32 check;
826};
827
828/* a transition list element for the list based representation */
829struct _reg_trie_trans_list_elem {
830 U16 forid;
831 U32 newstate;
832};
833typedef struct _reg_trie_trans_list_elem reg_trie_trans_le;
834
835/* a state for compressed nodes. base is an offset
836 into an array of reg_trie_trans array. If wordnum is
837 nonzero the state is accepting. if base is zero then
838 the state has no children (and will be accepting)
839*/
840struct _reg_trie_state {
841 U16 wordnum;
842 union {
843 U32 base;
844 reg_trie_trans_le* list;
845 } trans;
846};
847
2e64971a
DM
848/* info per word; indexed by wordnum */
849typedef struct {
850 U16 prev; /* previous word in acceptance chain; eg in
851 * zzz|abc|ab/ after matching the chars abc, the
852 * accepted word is #2, and the previous accepted
853 * word is #3 */
854 U32 len; /* how many chars long is this word? */
855 U32 accept; /* accept state for this word */
856} reg_trie_wordinfo;
a3621e74
YO
857
858
a3621e74
YO
859typedef struct _reg_trie_state reg_trie_state;
860typedef struct _reg_trie_trans reg_trie_trans;
861
862
863/* anything in here that needs to be freed later
23eab42c
NC
864 should be dealt with in pregfree.
865 refcount is first in both this and _reg_ac_data to allow a space
866 optimisation in Perl_regdupe. */
a3621e74 867struct _reg_trie_data {
23eab42c 868 U32 refcount; /* number of times this trie is referenced */
786e8c11
YO
869 U32 lasttrans; /* last valid transition element */
870 U16 *charmap; /* byte to charid lookup array */
786e8c11
YO
871 reg_trie_state *states; /* state data */
872 reg_trie_trans *trans; /* array of transition elements */
873 char *bitmap; /* stclass bitmap */
786e8c11
YO
874 U16 *jump; /* optional 1 indexed array of offsets before tail
875 for the node following a given word. */
2e64971a 876 reg_trie_wordinfo *wordinfo; /* array of info per word */
1b5c067c
NC
877 U16 uniquecharcount; /* unique chars in trie (width of trans table) */
878 U32 startstate; /* initial state - used for common prefix optimisation */
879 STRLEN minlen; /* minimum length of words in trie - build/opt only? */
880 STRLEN maxlen; /* maximum length of words in trie - build/opt only? */
2e64971a 881 U32 prefixlen; /* #chars in common prefix */
1e2e3d02
YO
882 U32 statecount; /* Build only - number of states in the states array
883 (including the unused zero state) */
786e8c11 884 U32 wordcount; /* Build only */
a3621e74 885#ifdef DEBUGGING
786e8c11 886 STRLEN charcount; /* Build only */
a3621e74
YO
887#endif
888};
2b8b4781
NC
889/* There is one (3 under DEBUGGING) pointers that logically belong in this
890 structure, but are held outside as they need duplication on thread cloning,
891 whereas the rest of the structure can be read only:
892 HV *widecharmap; code points > 255 to charid
893#ifdef DEBUGGING
894 AV *words; Array of words contained in trie, for dumping
895 AV *revcharmap; Map of each charid back to its character representation
896#endif
897*/
898
899#define TRIE_WORDS_OFFSET 2
900
a3621e74
YO
901typedef struct _reg_trie_data reg_trie_data;
902
23eab42c
NC
903/* refcount is first in both this and _reg_trie_data to allow a space
904 optimisation in Perl_regdupe. */
07be1b83 905struct _reg_ac_data {
23eab42c 906 U32 refcount;
1b5c067c 907 U32 trie;
07be1b83
YO
908 U32 *fail;
909 reg_trie_state *states;
07be1b83
YO
910};
911typedef struct _reg_ac_data reg_ac_data;
912
486ec47a 913/* ANY_BIT doesn't use the structure, so we can borrow it here.
3dab1dad
YO
914 This is simpler than refactoring all of it as wed end up with
915 three different sets... */
916
917#define TRIE_BITMAP(p) (((reg_trie_data *)(p))->bitmap)
786e8c11 918#define TRIE_BITMAP_BYTE(p, c) (TRIE_BITMAP(p)[(((U8)(c)) >> 3) & 31])
5dfe062f
YO
919#define TRIE_BITMAP_SET(p, c) (TRIE_BITMAP_BYTE(p, c) |= ANYOF_BIT((U8)c))
920#define TRIE_BITMAP_CLEAR(p,c) (TRIE_BITMAP_BYTE(p, c) &= ~ANYOF_BIT((U8)c))
921#define TRIE_BITMAP_TEST(p, c) (TRIE_BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
3dab1dad 922
1de06328
YO
923#define IS_ANYOF_TRIE(op) ((op)==TRIEC || (op)==AHOCORASICKC)
924#define IS_TRIE_AC(op) ((op)>=AHOCORASICK)
925
926
786e8c11
YO
927#define BITMAP_BYTE(p, c) (((U8*)p)[(((U8)(c)) >> 3) & 31])
928#define BITMAP_TEST(p, c) (BITMAP_BYTE(p, c) & ANYOF_BIT((U8)c))
3dab1dad 929
a3621e74
YO
930/* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */
931#define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1)
932#define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state))
933#define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state))
934
3dab1dad 935#ifdef DEBUGGING
3dab1dad 936#define TRIE_CHARCOUNT(trie) ((trie)->charcount)
3dab1dad 937#else
3dab1dad 938#define TRIE_CHARCOUNT(trie) (trie_charcount)
3dab1dad 939#endif
a3621e74 940
0111c4fd
RGS
941#define RE_TRIE_MAXBUF_INIT 65536
942#define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF"
a3621e74
YO
943#define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS"
944
be8e71aa
YO
945/*
946
947RE_DEBUG_FLAGS is used to control what debug output is emitted
948its divided into three groups of options, some of which interact.
949The three groups are: Compile, Execute, Extra. There is room for a
950further group, as currently only the low three bytes are used.
951
952 Compile Options:
07be1b83 953
be8e71aa
YO
954 PARSE
955 PEEP
956 TRIE
957 PROGRAM
958 OFFSETS
a3621e74 959
be8e71aa 960 Execute Options:
a3621e74 961
be8e71aa
YO
962 INTUIT
963 MATCH
964 TRIE
a3621e74 965
be8e71aa 966 Extra Options
a3621e74 967
be8e71aa
YO
968 TRIE
969 OFFSETS
970
971If you modify any of these make sure you make corresponding changes to
972re.pm, especially to the documentation.
a3621e74 973
be8e71aa
YO
974*/
975
976
977/* Compile */
978#define RE_DEBUG_COMPILE_MASK 0x0000FF
979#define RE_DEBUG_COMPILE_PARSE 0x000001
980#define RE_DEBUG_COMPILE_OPTIMISE 0x000002
981#define RE_DEBUG_COMPILE_TRIE 0x000004
982#define RE_DEBUG_COMPILE_DUMP 0x000008
f7819f85 983#define RE_DEBUG_COMPILE_FLAGS 0x000010
d9a72fcc 984#define RE_DEBUG_COMPILE_TEST 0x000020
be8e71aa
YO
985
986/* Execute */
987#define RE_DEBUG_EXECUTE_MASK 0x00FF00
988#define RE_DEBUG_EXECUTE_INTUIT 0x000100
989#define RE_DEBUG_EXECUTE_MATCH 0x000200
990#define RE_DEBUG_EXECUTE_TRIE 0x000400
991
992/* Extra */
993#define RE_DEBUG_EXTRA_MASK 0xFF0000
994#define RE_DEBUG_EXTRA_TRIE 0x010000
995#define RE_DEBUG_EXTRA_OFFSETS 0x020000
a5ca303d
YO
996#define RE_DEBUG_EXTRA_OFFDEBUG 0x040000
997#define RE_DEBUG_EXTRA_STATE 0x080000
998#define RE_DEBUG_EXTRA_OPTIMISE 0x100000
e7707071 999#define RE_DEBUG_EXTRA_BUFFERS 0x400000
2c296965 1000#define RE_DEBUG_EXTRA_GPOS 0x800000
24b23f37
YO
1001/* combined */
1002#define RE_DEBUG_EXTRA_STACK 0x280000
be8e71aa 1003
e68ec53f 1004#define RE_DEBUG_FLAG(x) (re_debug_flags & x)
be8e71aa
YO
1005/* Compile */
1006#define DEBUG_COMPILE_r(x) DEBUG_r( \
e68ec53f 1007 if (re_debug_flags & RE_DEBUG_COMPILE_MASK) x )
be8e71aa 1008#define DEBUG_PARSE_r(x) DEBUG_r( \
e68ec53f 1009 if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x )
be8e71aa 1010#define DEBUG_OPTIMISE_r(x) DEBUG_r( \
e68ec53f 1011 if (re_debug_flags & RE_DEBUG_COMPILE_OPTIMISE) x )
be8e71aa 1012#define DEBUG_PARSE_r(x) DEBUG_r( \
e68ec53f 1013 if (re_debug_flags & RE_DEBUG_COMPILE_PARSE) x )
be8e71aa 1014#define DEBUG_DUMP_r(x) DEBUG_r( \
e68ec53f 1015 if (re_debug_flags & RE_DEBUG_COMPILE_DUMP) x )
be8e71aa 1016#define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \
e68ec53f 1017 if (re_debug_flags & RE_DEBUG_COMPILE_TRIE) x )
f7819f85
A
1018#define DEBUG_FLAGS_r(x) DEBUG_r( \
1019 if (re_debug_flags & RE_DEBUG_COMPILE_FLAGS) x )
d9a72fcc
YO
1020#define DEBUG_TEST_r(x) DEBUG_r( \
1021 if (re_debug_flags & RE_DEBUG_COMPILE_TEST) x )
be8e71aa
YO
1022/* Execute */
1023#define DEBUG_EXECUTE_r(x) DEBUG_r( \
e68ec53f 1024 if (re_debug_flags & RE_DEBUG_EXECUTE_MASK) x )
be8e71aa 1025#define DEBUG_INTUIT_r(x) DEBUG_r( \
e68ec53f 1026 if (re_debug_flags & RE_DEBUG_EXECUTE_INTUIT) x )
be8e71aa 1027#define DEBUG_MATCH_r(x) DEBUG_r( \
e68ec53f 1028 if (re_debug_flags & RE_DEBUG_EXECUTE_MATCH) x )
be8e71aa 1029#define DEBUG_TRIE_EXECUTE_r(x) DEBUG_r( \
e68ec53f 1030 if (re_debug_flags & RE_DEBUG_EXECUTE_TRIE) x )
be8e71aa
YO
1031
1032/* Extra */
1033#define DEBUG_EXTRA_r(x) DEBUG_r( \
e68ec53f 1034 if (re_debug_flags & RE_DEBUG_EXTRA_MASK) x )
a5ca303d
YO
1035#define DEBUG_OFFSETS_r(x) DEBUG_r( \
1036 if (re_debug_flags & RE_DEBUG_EXTRA_OFFSETS) x )
ab3bbdeb
YO
1037#define DEBUG_STATE_r(x) DEBUG_r( \
1038 if (re_debug_flags & RE_DEBUG_EXTRA_STATE) x )
24b23f37
YO
1039#define DEBUG_STACK_r(x) DEBUG_r( \
1040 if (re_debug_flags & RE_DEBUG_EXTRA_STACK) x )
e7707071
YO
1041#define DEBUG_BUFFERS_r(x) DEBUG_r( \
1042 if (re_debug_flags & RE_DEBUG_EXTRA_BUFFERS) x )
1043
a5ca303d
YO
1044#define DEBUG_OPTIMISE_MORE_r(x) DEBUG_r( \
1045 if ((RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE) == \
1046 (re_debug_flags & (RE_DEBUG_EXTRA_OPTIMISE|RE_DEBUG_COMPILE_OPTIMISE)) ) x )
07be1b83 1047#define MJD_OFFSET_DEBUG(x) DEBUG_r( \
a5ca303d 1048 if (re_debug_flags & RE_DEBUG_EXTRA_OFFDEBUG) \
be8e71aa
YO
1049 Perl_warn_nocontext x )
1050#define DEBUG_TRIE_COMPILE_MORE_r(x) DEBUG_TRIE_COMPILE_r( \
e68ec53f 1051 if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x )
be8e71aa 1052#define DEBUG_TRIE_EXECUTE_MORE_r(x) DEBUG_TRIE_EXECUTE_r( \
e68ec53f 1053 if (re_debug_flags & RE_DEBUG_EXTRA_TRIE) x )
be8e71aa
YO
1054
1055#define DEBUG_TRIE_r(x) DEBUG_r( \
e68ec53f 1056 if (re_debug_flags & (RE_DEBUG_COMPILE_TRIE \
be8e71aa 1057 | RE_DEBUG_EXECUTE_TRIE )) x )
2c296965
YO
1058#define DEBUG_GPOS_r(x) DEBUG_r( \
1059 if (re_debug_flags & RE_DEBUG_EXTRA_GPOS) x )
be8e71aa
YO
1060
1061/* initialization */
d2d8c711 1062/* get_sv() can return NULL during global destruction. */
e68ec53f
YO
1063#define GET_RE_DEBUG_FLAGS DEBUG_r({ \
1064 SV * re_debug_flags_sv = NULL; \
1065 re_debug_flags_sv = get_sv(RE_DEBUG_FLAGS, 1); \
1066 if (re_debug_flags_sv) { \
8e9a8a48 1067 if (!SvIOK(re_debug_flags_sv)) \
e68ec53f 1068 sv_setuv(re_debug_flags_sv, RE_DEBUG_COMPILE_DUMP | RE_DEBUG_EXECUTE_MASK ); \
8e9a8a48 1069 re_debug_flags=SvIV(re_debug_flags_sv); \
e68ec53f
YO
1070 }\
1071})
a3621e74
YO
1072
1073#ifdef DEBUGGING
ab3bbdeb 1074
cb65b687
DM
1075#define GET_RE_DEBUG_FLAGS_DECL VOL IV re_debug_flags = 0; \
1076 PERL_UNUSED_VAR(re_debug_flags); GET_RE_DEBUG_FLAGS;
ab3bbdeb
YO
1077
1078#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \
1079 const char * const rpv = \
ddc5bc0f 1080 pv_pretty((dsv), (pv), (l), (m), \
ab3bbdeb 1081 PL_colors[(c1)],PL_colors[(c2)], \
c89df6cf 1082 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \
0df25f3d 1083 const int rlen = SvCUR(dsv)
ab3bbdeb
YO
1084
1085#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \
1086 const char * const rpv = \
ddc5bc0f 1087 pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \
ab3bbdeb 1088 PL_colors[(c1)],PL_colors[(c2)], \
c89df6cf 1089 PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) )
ab3bbdeb
YO
1090
1091#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \
1092 const char * const rpv = \
ddc5bc0f 1093 pv_pretty((dsv), (pv), (l), (m), \
ab3bbdeb 1094 PL_colors[0], PL_colors[1], \
c89df6cf 1095 ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \
ab3bbdeb 1096 ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \
0a0e9afc 1097 )
ab3bbdeb
YO
1098
1099#define RE_SV_DUMPLEN(ItEm) (SvCUR(ItEm) - (SvTAIL(ItEm)!=0))
1100#define RE_SV_TAIL(ItEm) (SvTAIL(ItEm) ? "$" : "")
1101
1102#else /* if not DEBUGGING */
1103
a3621e74 1104#define GET_RE_DEBUG_FLAGS_DECL
ab3bbdeb
YO
1105#define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2)
1106#define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m)
1107#define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m)
1108#define RE_SV_DUMPLEN(ItEm)
1109#define RE_SV_TAIL(ItEm)
1110
1111#endif /* DEBUG RELATED DEFINES */
a3621e74 1112
64935bc6
KW
1113typedef enum {
1114 TRADITIONAL_BOUND = _CC_WORDCHAR,
ae3bb8ea 1115 GCB_BOUND,
06ae2722 1116 SB_BOUND,
ae3bb8ea 1117 WB_BOUND
64935bc6
KW
1118} bound_type;
1119
e9a8c099 1120/*
14d04a33 1121 * ex: set ts=8 sts=4 sw=4 et:
e9a8c099 1122 */