3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
4 * 2000, 2001, 2002, 2003, 2005 by Larry Wall and others
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.
11 typedef OP OP_4tree; /* Will be redefined later. */
14 * The "internal use only" fields in regexp.h are present to pass info from
15 * compile to execute that permits the execute phase to run lots faster on
16 * simple cases. They are:
18 * regstart sv that must begin a match; NULL if none obvious
19 * reganch is the match anchored (at beginning-of-line only)?
20 * regmust string (pointer into program) that match must include, or NULL
21 * [regmust changed to SV* for bminstr()--law]
22 * regmlen length of regmust string
23 * [regmlen not used currently]
25 * Regstart and reganch permit very fast decisions on suitable starting points
26 * for a match, cutting down the work a lot. Regmust permits fast rejection
27 * of lines that cannot possibly match. The regmust tests are costly enough
28 * that pregcomp() supplies a regmust only if the r.e. contains something
29 * potentially expensive (at present, the only such thing detected is * or +
30 * at the start of the r.e., which can involve a lot of backup). Regmlen is
31 * supplied because the test in pregexec() needs it and pregcomp() is computing
33 * [regmust is now supplied always. The tests that use regmust have a
34 * heuristic that disables the test if it usually matches.]
36 * [In fact, we now use regmust in many cases to locate where the search
37 * starts in the string, so if regback is >= 0, the regmust search is never
38 * wasted effort. The regback variable says how many characters back from
39 * where regmust matched is the earliest possible start of the match.
40 * For instance, /[a-z].foo/ has a regmust of 'foo' and a regback of 2.]
44 * Structure for regexp "program". This is essentially a linear encoding
45 * of a nondeterministic finite-state machine (aka syntax charts or
46 * "railroad normal form" in parsing technology). Each node is an opcode
47 * plus a "next" pointer, possibly plus an operand. "Next" pointers of
48 * all nodes except BRANCH implement concatenation; a "next" pointer with
49 * a BRANCH on both ends of it is connecting two alternatives. (Here we
50 * have one of the subtle syntax dependencies: an individual BRANCH (as
51 * opposed to a collection of them) is never concatenated with anything
52 * because of operator precedence.) The operand of some types of node is
53 * a literal string; for others, it is a node leading into a sub-FSM. In
54 * particular, the operand of a BRANCH node is the first node of the branch.
55 * (NB this is *not* a tree structure: the tail of the branch connects
56 * to the thing following the set of BRANCHes.) The opcodes are defined
57 * in regnodes.h which is generated from regcomp.sym by regcomp.pl.
61 * A node is one char of opcode followed by two chars of "next" pointer.
62 * "Next" pointers are stored as two 8-bit pieces, high order first. The
63 * value is a positive offset from the opcode of the node containing it.
64 * An operand, if any, simply follows the node. (Note that much of the
65 * code generation knows about this implicit relationship.)
67 * Using two bytes for the "next" pointer is vast overkill for most things,
68 * but allows patterns to get big without disasters.
70 * [The "next" pointer is always aligned on an even
71 * boundary, and reads the offset directly as a short. Also, there is no
72 * special test to reverse the sign of BACK pointers since the offset is
76 struct regnode_string {
98 #define ANYOF_BITMAP_SIZE 32 /* 256 b/(8 b/B) */
99 #define ANYOF_CLASSBITMAP_SIZE 4 /* up to 32 (8*4) named classes */
101 struct regnode_charclass {
106 char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */
109 struct regnode_charclass_class { /* has [[:blah:]] classes */
110 U8 flags; /* should have ANYOF_CLASS here */
114 char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time */
115 char classflags[ANYOF_CLASSBITMAP_SIZE]; /* and run-time */
118 /* XXX fix this description.
119 Impose a limit of REG_INFTY on various pattern matching operations
120 to limit stack growth and to avoid "infinite" recursions.
122 /* The default size for REG_INFTY is I16_MAX, which is the same as
123 SHORT_MAX (see perl.h). Unfortunately I16 isn't necessarily 16 bits
124 (see handy.h). On the Cray C90, sizeof(short)==4 and hence I16_MAX is
125 ((1<<31)-1), while on the Cray T90, sizeof(short)==8 and I16_MAX is
126 ((1<<63)-1). To limit stack growth to reasonable sizes, supply a
128 --Andy Dougherty 11 June 1998
132 # define REG_INFTY ((1<<15)-1)
137 # define REG_INFTY I16_MAX
140 #define ARG_VALUE(arg) (arg)
141 #define ARG__SET(arg,val) ((arg) = (val))
147 #define ARG(p) ARG_VALUE(ARG_LOC(p))
148 #define ARG1(p) ARG_VALUE(ARG1_LOC(p))
149 #define ARG2(p) ARG_VALUE(ARG2_LOC(p))
150 #define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val))
151 #define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val))
152 #define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val))
157 #define NEXT_OFF(p) ((p)->next_off)
158 #define NODE_ALIGN(node)
159 #define NODE_ALIGN_FILL(node) ((node)->flags = 0xde) /* deadbeef */
161 #define SIZE_ALIGN NODE_ALIGN
168 #define OP(p) ((p)->type)
169 #define OPERAND(p) (((struct regnode_string *)p)->string)
170 #define MASK(p) ((char*)OPERAND(p))
171 #define STR_LEN(p) (((struct regnode_string *)p)->str_len)
172 #define STRING(p) (((struct regnode_string *)p)->string)
173 #define STR_SZ(l) ((l + sizeof(regnode) - 1) / sizeof(regnode))
174 #define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p))+1)
181 #define NODE_ALIGN(node)
182 #define ARG_LOC(p) (((struct regnode_1 *)p)->arg1)
183 #define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1)
184 #define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2)
185 #define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */
186 #define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2)
188 #define NODE_STEP_B 4
190 #define NEXTOPER(p) ((p) + NODE_STEP_REGNODE)
191 #define PREVOPER(p) ((p) - NODE_STEP_REGNODE)
193 #define FILL_ADVANCE_NODE(ptr, op) STMT_START { \
194 (ptr)->type = op; (ptr)->next_off = 0; (ptr)++; } STMT_END
195 #define FILL_ADVANCE_NODE_ARG(ptr, op, arg) STMT_START { \
196 ARG_SET(ptr, arg); FILL_ADVANCE_NODE(ptr, op); (ptr) += 1; } STMT_END
198 #define REG_MAGIC 0234
200 #define SIZE_ONLY (RExC_emit == &PL_regdummy)
202 /* Flags for node->flags of ANYOF */
204 #define ANYOF_CLASS 0x08 /* has [[:blah:]] classes */
205 #define ANYOF_INVERT 0x04
206 #define ANYOF_FOLD 0x02
207 #define ANYOF_LOCALE 0x01
209 /* Used for regstclass only */
210 #define ANYOF_EOS 0x10 /* Can match an empty string too */
212 /* There is a character or a range past 0xff */
213 #define ANYOF_UNICODE 0x20
214 #define ANYOF_UNICODE_ALL 0x40 /* Can match any char past 0xff */
216 /* size of node is large (includes class pointer) */
217 #define ANYOF_LARGE 0x80
219 /* Are there any runtime flags on in this node? */
220 #define ANYOF_RUNTIME(s) (ANYOF_FLAGS(s) & 0x0f)
222 #define ANYOF_FLAGS_ALL 0xff
224 /* Character classes for node->classflags of ANYOF */
225 /* Should be synchronized with a table in regprop() */
226 /* 2n should pair with 2n+1 */
228 #define ANYOF_ALNUM 0 /* \w, PL_utf8_alnum, utf8::IsWord, ALNUM */
229 #define ANYOF_NALNUM 1
230 #define ANYOF_SPACE 2 /* \s */
231 #define ANYOF_NSPACE 3
232 #define ANYOF_DIGIT 4
233 #define ANYOF_NDIGIT 5
234 #define ANYOF_ALNUMC 6 /* isalnum(3), utf8::IsAlnum, ALNUMC */
235 #define ANYOF_NALNUMC 7
236 #define ANYOF_ALPHA 8
237 #define ANYOF_NALPHA 9
238 #define ANYOF_ASCII 10
239 #define ANYOF_NASCII 11
240 #define ANYOF_CNTRL 12
241 #define ANYOF_NCNTRL 13
242 #define ANYOF_GRAPH 14
243 #define ANYOF_NGRAPH 15
244 #define ANYOF_LOWER 16
245 #define ANYOF_NLOWER 17
246 #define ANYOF_PRINT 18
247 #define ANYOF_NPRINT 19
248 #define ANYOF_PUNCT 20
249 #define ANYOF_NPUNCT 21
250 #define ANYOF_UPPER 22
251 #define ANYOF_NUPPER 23
252 #define ANYOF_XDIGIT 24
253 #define ANYOF_NXDIGIT 25
254 #define ANYOF_PSXSPC 26 /* POSIX space: \s plus the vertical tab */
255 #define ANYOF_NPSXSPC 27
256 #define ANYOF_BLANK 28 /* GNU extension: space and tab: non-vertical space */
257 #define ANYOF_NBLANK 29
261 /* Backward source code compatibility. */
263 #define ANYOF_ALNUML ANYOF_ALNUM
264 #define ANYOF_NALNUML ANYOF_NALNUM
265 #define ANYOF_SPACEL ANYOF_SPACE
266 #define ANYOF_NSPACEL ANYOF_NSPACE
268 /* Utility macros for the bitmap and classes of ANYOF */
270 #define ANYOF_SIZE (sizeof(struct regnode_charclass))
271 #define ANYOF_CLASS_SIZE (sizeof(struct regnode_charclass_class))
273 #define ANYOF_FLAGS(p) ((p)->flags)
275 #define ANYOF_BIT(c) (1 << ((c) & 7))
277 #define ANYOF_CLASS_BYTE(p, c) (((struct regnode_charclass_class*)(p))->classflags[((c) >> 3) & 3])
278 #define ANYOF_CLASS_SET(p, c) (ANYOF_CLASS_BYTE(p, c) |= ANYOF_BIT(c))
279 #define ANYOF_CLASS_CLEAR(p, c) (ANYOF_CLASS_BYTE(p, c) &= ~ANYOF_BIT(c))
280 #define ANYOF_CLASS_TEST(p, c) (ANYOF_CLASS_BYTE(p, c) & ANYOF_BIT(c))
282 #define ANYOF_CLASS_ZERO(ret) Zero(((struct regnode_charclass_class*)(ret))->classflags, ANYOF_CLASSBITMAP_SIZE, char)
283 #define ANYOF_BITMAP_ZERO(ret) Zero(((struct regnode_charclass*)(ret))->bitmap, ANYOF_BITMAP_SIZE, char)
285 #define ANYOF_BITMAP(p) (((struct regnode_charclass*)(p))->bitmap)
286 #define ANYOF_BITMAP_BYTE(p, c) (ANYOF_BITMAP(p)[((c) >> 3) & 31])
287 #define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c))
288 #define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c))
289 #define ANYOF_BITMAP_TEST(p, c) (ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c))
291 #define ANYOF_BITMAP_SETALL(p) \
292 memset (ANYOF_BITMAP(p), 255, ANYOF_BITMAP_SIZE)
293 #define ANYOF_BITMAP_CLEARALL(p) \
294 Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE)
295 /* Check that all 256 bits are all set. Used in S_cl_is_anything() */
296 #define ANYOF_BITMAP_TESTALLSET(p) \
297 memEQ (ANYOF_BITMAP(p), "\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377", ANYOF_BITMAP_SIZE)
299 #define ANYOF_SKIP ((ANYOF_SIZE - 1)/sizeof(regnode))
300 #define ANYOF_CLASS_SKIP ((ANYOF_CLASS_SIZE - 1)/sizeof(regnode))
301 #define ANYOF_CLASS_ADD_SKIP (ANYOF_CLASS_SKIP - ANYOF_SKIP)
304 * Utility definitions.
307 # define UCHARAT(p) ((int)*(const U8*)(p))
309 # define UCHARAT(p) ((int)*(p)&CHARMASK)
312 #define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode))
314 #define REG_SEEN_ZERO_LEN 1
315 #define REG_SEEN_LOOKBEHIND 2
316 #define REG_SEEN_GPOS 4
317 #define REG_SEEN_EVAL 8
318 #define REG_SEEN_CANY 16
319 #define REG_SEEN_SANY REG_SEEN_CANY /* src bckwrd cmpt */
323 #include "regnodes.h"
325 /* The following have no fixed length. U8 so we can do strchr() on it. */
327 EXTCONST U8 PL_varies[];
329 EXTCONST U8 PL_varies[] = {
330 BRANCH, BACK, STAR, PLUS, CURLY, CURLYX, REF, REFF, REFFL,
331 WHILEM, CURLYM, CURLYN, BRANCHJ, IFTHEN, SUSPEND, CLUMP, 0
335 /* The following always have a length of 1. U8 we can do strchr() on it. */
336 /* (Note that length 1 means "one character" under UTF8, not "one octet".) */
338 EXTCONST U8 PL_simple[];
340 EXTCONST U8 PL_simple[] = {
354 typedef struct re_scream_pos_data_s
356 char **scream_olds; /* match pos */
357 I32 *scream_pos; /* Internal iterator of scream. */
358 } re_scream_pos_data;
360 /* .what is a character array with one character for each member of .data
361 * The character describes the function of the corresponding .data item:
362 * f - start-class data for regstclass optimization
363 * n - Root of op tree for (?{EVAL}) item
364 * o - Start op for (?{EVAL}) item
365 * p - Pad for (?{EVAL} item
366 * s - swash for unicode-style character class, and the multicharacter
367 * strings resulting from casefolding the single-character entries
368 * in the character class
370 * 20010712 mjd@plover.com
371 * (Remember to update re_dup() and pregfree() if you add any items.)
379 struct reg_substr_datum {
382 SV *substr; /* non-utf8 variant */
383 SV *utf8_substr; /* utf8 variant */
386 struct reg_substr_data {
387 struct reg_substr_datum data[3]; /* Actual array */
390 #define anchored_substr substrs->data[0].substr
391 #define anchored_utf8 substrs->data[0].utf8_substr
392 #define anchored_offset substrs->data[0].min_offset
393 #define float_substr substrs->data[1].substr
394 #define float_utf8 substrs->data[1].utf8_substr
395 #define float_min_offset substrs->data[1].min_offset
396 #define float_max_offset substrs->data[1].max_offset
397 #define check_substr substrs->data[2].substr
398 #define check_utf8 substrs->data[2].utf8_substr
399 #define check_offset_min substrs->data[2].min_offset
400 #define check_offset_max substrs->data[2].max_offset
404 /* trie related stuff */
406 /* a transition record for the state machine. the
407 check field determines which state "owns" the
408 transition. the char the transition is for is
409 determined by offset from the owning states base
410 field. the next field determines which state
411 is to be transitioned to if any.
413 struct _reg_trie_trans {
418 /* a transition list element for the list based representation */
419 struct _reg_trie_trans_list_elem {
423 typedef struct _reg_trie_trans_list_elem reg_trie_trans_le;
425 /* a state for compressed nodes. base is an offset
426 into an array of reg_trie_trans array. If wordnum is
427 nonzero the state is accepting. if base is zero then
428 the state has no children (and will be accepting)
430 struct _reg_trie_state {
434 reg_trie_trans_le* list;
440 typedef struct _reg_trie_state reg_trie_state;
441 typedef struct _reg_trie_trans reg_trie_trans;
444 /* anything in here that needs to be freed later
445 should be dealt with in pregfree */
446 struct _reg_trie_data {
454 reg_trie_state *states;
455 reg_trie_trans *trans;
463 typedef struct _reg_trie_data reg_trie_data;
465 /* these defines assume uniquecharcount is the correct variable, and state may be evaluated twice */
466 #define TRIE_NODENUM(state) (((state)-1)/(trie->uniquecharcount)+1)
467 #define SAFE_TRIE_NODENUM(state) ((state) ? (((state)-1)/(trie->uniquecharcount)+1) : (state))
468 #define TRIE_NODEIDX(state) ((state) ? (((state)-1)*(trie->uniquecharcount)+1) : (state))
473 #define RE_TRIE_MAXBUF_INIT 65536
474 #define RE_TRIE_MAXBUF_NAME "\022E_TRIE_MAXBUF"
475 #define RE_DEBUG_FLAGS "\022E_DEBUG_FLAGS"
477 /* If you change these be sure to update ext/re/re.pm as well */
478 #define RE_DEBUG_COMPILE 1
479 #define RE_DEBUG_EXECUTE 2
480 #define RE_DEBUG_TRIE_COMPILE 4
481 #define RE_DEBUG_TRIE_EXECUTE 8
482 #define RE_DEBUG_TRIE_MORE 16
483 #define RE_DEBUG_OPTIMISE 32
484 #define RE_DEBUG_OFFSETS 64
486 #define DEBUG_OPTIMISE_r(x) DEBUG_r( if (SvIV(re_debug_flags) & RE_DEBUG_OPTIMISE) x )
487 #define DEBUG_EXECUTE_r(x) DEBUG_r( if (SvIV(re_debug_flags) & RE_DEBUG_EXECUTE) x )
488 #define DEBUG_COMPILE_r(x) DEBUG_r( if (SvIV(re_debug_flags) & RE_DEBUG_COMPILE) x )
489 #define DEBUG_OFFSETS_r(x) DEBUG_r( if (SvIV(re_debug_flags) & RE_DEBUG_OFFSETS) x )
490 #define DEBUG_TRIE_r(x) DEBUG_r( \
491 if (SvIV(re_debug_flags) & RE_DEBUG_TRIE_COMPILE \
492 || SvIV(re_debug_flags) & RE_DEBUG_TRIE_EXECUTE ) \
495 #define DEBUG_TRIE_EXECUTE_r(x) \
496 DEBUG_r( if (SvIV(re_debug_flags) & RE_DEBUG_TRIE_EXECUTE) x )
498 #define DEBUG_TRIE_COMPILE_r(x) \
499 DEBUG_r( if (SvIV(re_debug_flags) & RE_DEBUG_TRIE_COMPILE) x )
501 #define DEBUG_TRIE_EXECUTE_MORE_r(x) \
502 DEBUG_TRIE_EXECUTE_r( if (SvIV(re_debug_flags) & RE_DEBUG_TRIE_MORE) x )
504 #define DEBUG_TRIE_COMPILE_MORE_r(x) \
505 DEBUG_TRIE_COMPILE_r( if (SvIV(re_debug_flags) & RE_DEBUG_TRIE_MORE) x )
507 #define GET_RE_DEBUG_FLAGS DEBUG_r( \
508 re_debug_flags=get_sv(RE_DEBUG_FLAGS, 1); \
509 if (!SvIOK(re_debug_flags)) { \
510 sv_setiv(re_debug_flags, RE_DEBUG_COMPILE | RE_DEBUG_EXECUTE | RE_DEBUG_OFFSETS); \
516 #define GET_RE_DEBUG_FLAGS_DECL SV *re_debug_flags = NULL; GET_RE_DEBUG_FLAGS;
518 #define GET_RE_DEBUG_FLAGS_DECL