4 typedef OP OP_4tree; /* Will be redefined later. */
7 * The "internal use only" fields in regexp.h are present to pass info from
8 * compile to execute that permits the execute phase to run lots faster on
9 * simple cases. They are:
11 * regstart sv that must begin a match; Nullch if none obvious
12 * reganch is the match anchored (at beginning-of-line only)?
13 * regmust string (pointer into program) that match must include, or NULL
14 * [regmust changed to SV* for bminstr()--law]
15 * regmlen length of regmust string
16 * [regmlen not used currently]
18 * Regstart and reganch permit very fast decisions on suitable starting points
19 * for a match, cutting down the work a lot. Regmust permits fast rejection
20 * of lines that cannot possibly match. The regmust tests are costly enough
21 * that pregcomp() supplies a regmust only if the r.e. contains something
22 * potentially expensive (at present, the only such thing detected is * or +
23 * at the start of the r.e., which can involve a lot of backup). Regmlen is
24 * supplied because the test in pregexec() needs it and pregcomp() is computing
26 * [regmust is now supplied always. The tests that use regmust have a
27 * heuristic that disables the test if it usually matches.]
29 * [In fact, we now use regmust in many cases to locate where the search
30 * starts in the string, so if regback is >= 0, the regmust search is never
31 * wasted effort. The regback variable says how many characters back from
32 * where regmust matched is the earliest possible start of the match.
33 * For instance, /[a-z].foo/ has a regmust of 'foo' and a regback of 2.]
37 * Structure for regexp "program". This is essentially a linear encoding
38 * of a nondeterministic finite-state machine (aka syntax charts or
39 * "railroad normal form" in parsing technology). Each node is an opcode
40 * plus a "next" pointer, possibly plus an operand. "Next" pointers of
41 * all nodes except BRANCH implement concatenation; a "next" pointer with
42 * a BRANCH on both ends of it is connecting two alternatives. (Here we
43 * have one of the subtle syntax dependencies: an individual BRANCH (as
44 * opposed to a collection of them) is never concatenated with anything
45 * because of operator precedence.) The operand of some types of node is
46 * a literal string; for others, it is a node leading into a sub-FSM. In
47 * particular, the operand of a BRANCH node is the first node of the branch.
48 * (NB this is *not* a tree structure: the tail of the branch connects
49 * to the thing following the set of BRANCHes.) The opcodes are:
53 * A node is one char of opcode followed by two chars of "next" pointer.
54 * "Next" pointers are stored as two 8-bit pieces, high order first. The
55 * value is a positive offset from the opcode of the node containing it.
56 * An operand, if any, simply follows the node. (Note that much of the
57 * code generation knows about this implicit relationship.)
59 * Using two bytes for the "next" pointer is vast overkill for most things,
60 * but allows patterns to get big without disasters.
62 * [The "next" pointer is always aligned on an even
63 * boundary, and reads the offset directly as a short. Also, there is no
64 * special test to reverse the sign of BACK pointers since the offset is
68 struct regnode_string {
90 /* XXX fix this description.
91 Impose a limit of REG_INFTY on various pattern matching operations
92 to limit stack growth and to avoid "infinite" recursions.
94 /* The default size for REG_INFTY is I16_MAX, which is the same as
95 SHORT_MAX (see perl.h). Unfortunately I16 isn't necessarily 16 bits
96 (see handy.h). On the Cray C90, sizeof(short)==4 and hence I16_MAX is
97 ((1<<31)-1), while on the Cray T90, sizeof(short)==8 and I16_MAX is
98 ((1<<63)-1). To limit stack growth to reasonable sizes, supply a
100 --Andy Dougherty 11 June 1998
104 # define REG_INFTY ((1<<15)-1)
109 # define REG_INFTY I16_MAX
112 #define ARG_VALUE(arg) (arg)
113 #define ARG__SET(arg,val) ((arg) = (val))
115 #define ARG(p) ARG_VALUE(ARG_LOC(p))
116 #define ARG1(p) ARG_VALUE(ARG1_LOC(p))
117 #define ARG2(p) ARG_VALUE(ARG2_LOC(p))
118 #define ARG_SET(p, val) ARG__SET(ARG_LOC(p), (val))
119 #define ARG1_SET(p, val) ARG__SET(ARG1_LOC(p), (val))
120 #define ARG2_SET(p, val) ARG__SET(ARG2_LOC(p), (val))
123 # define NEXT_OFF(p) ((p)->next_off)
124 # define NODE_ALIGN(node)
125 # define NODE_ALIGN_FILL(node) ((node)->flags = 0xde) /* deadbeef */
127 # define NEXT_OFF(p) 0
128 # define NODE_ALIGN(node)
129 # define NODE_ALIGN_FILL(node)
132 #define SIZE_ALIGN NODE_ALIGN
134 #define OP(p) ((p)->type)
135 #define OPERAND(p) (((struct regnode_string *)p)->string)
136 #define MASK(p) ((char*)OPERAND(p))
137 #define STR_LEN(p) (((struct regnode_string *)p)->str_len)
138 #define STRING(p) (((struct regnode_string *)p)->string)
139 #define STR_SZ(l) ((l + sizeof(regnode) - 1) / sizeof(regnode))
140 #define NODE_SZ_STR(p) (STR_SZ(STR_LEN(p))+1)
142 #define NODE_ALIGN(node)
143 #define ARG_LOC(p) (((struct regnode_1 *)p)->arg1)
144 #define ARG1_LOC(p) (((struct regnode_2 *)p)->arg1)
145 #define ARG2_LOC(p) (((struct regnode_2 *)p)->arg2)
146 #define NODE_STEP_REGNODE 1 /* sizeof(regnode)/sizeof(regnode) */
147 #define EXTRA_STEP_2ARGS EXTRA_SIZE(struct regnode_2)
149 #define NODE_STEP_B 4
151 #define NEXTOPER(p) ((p) + NODE_STEP_REGNODE)
152 #define PREVOPER(p) ((p) - NODE_STEP_REGNODE)
154 #define FILL_ADVANCE_NODE(ptr, op) STMT_START { \
155 (ptr)->type = op; (ptr)->next_off = 0; (ptr)++; } STMT_END
156 #define FILL_ADVANCE_NODE_ARG(ptr, op, arg) STMT_START { \
157 ARG_SET(ptr, arg); FILL_ADVANCE_NODE(ptr, op); (ptr) += 1; } STMT_END
159 #define REG_MAGIC 0234
161 #define SIZE_ONLY (PL_regcode == &PL_regdummy)
163 /* Flags for first parameter byte [0] of ANYOF */
165 #define ANYOF_CLASS 0x08
166 #define ANYOF_INVERT 0x04
167 #define ANYOF_FOLD 0x02
168 #define ANYOF_LOCALE 0x01
170 /* Character classes for bytes [1..4] of ANYOF */
172 #define ANYOF_ALNUM 0 /* \w, utf8::IsWord, isALNUM() */
173 #define ANYOF_NALNUM 1
174 #define ANYOF_SPACE 2
175 #define ANYOF_NSPACE 3
176 #define ANYOF_DIGIT 4
177 #define ANYOF_NDIGIT 5
178 #define ANYOF_ALNUMC 6 /* isalnum(3), utf8::IsAlnum, isALNUMC() */
179 #define ANYOF_NALNUMC 7
180 #define ANYOF_ALPHA 8
181 #define ANYOF_NALPHA 9
182 #define ANYOF_ASCII 10
183 #define ANYOF_NASCII 11
184 #define ANYOF_CNTRL 12
185 #define ANYOF_NCNTRL 13
186 #define ANYOF_GRAPH 14
187 #define ANYOF_NGRAPH 15
188 #define ANYOF_LOWER 16
189 #define ANYOF_NLOWER 17
190 #define ANYOF_PRINT 18
191 #define ANYOF_NPRINT 19
192 #define ANYOF_PUNCT 20
193 #define ANYOF_NPUNCT 21
194 #define ANYOF_UPPER 22
195 #define ANYOF_NUPPER 23
196 #define ANYOF_XDIGIT 24
197 #define ANYOF_NXDIGIT 25
201 /* Backward source code compatibility. */
203 #define ANYOF_ALNUML ANYOF_ALNUM
204 #define ANYOF_NALNUML ANYOF_NALNUM
205 #define ANYOF_SPACEL ANYOF_SPACE
206 #define ANYOF_NSPACEL ANYOF_NSPACE
208 /* Utility macros for the bitmap and classes of ANYOF */
210 #define ANYOF_OPND_SIZE 1
211 #define ANYOF_CLASS_SIZE 4
212 #define ANYOF_BITMAP_SIZE 32 /* 256 b/(8 b/B) */
213 #define ANYOF_SIZE (ANYOF_OPND_SIZE+ANYOF_CLASS_SIZE+ANYOF_BITMAP_SIZE)
215 #define ANYOF_FLAGS(p) ((p)[0])
216 #define ANYOF_FLAGS_ALL 0xff
218 #define ANYOF_BIT(c) (1 << ((c) & 7))
220 #define ANYOF_CLASS_OFFSET ANYOF_OPND_SIZE
221 #define ANYOF_CLASS_BYTE(p, c) ((p)[ANYOF_CLASS_OFFSET + (((c) >> 3) & 3)])
222 #define ANYOF_CLASS_SET(p, c) (ANYOF_CLASS_BYTE(p, c) |= ANYOF_BIT(c))
223 #define ANYOF_CLASS_CLEAR(p, c) (ANYOF_CLASS_BYTE(p, c) &= ~ANYOF_BIT(c))
224 #define ANYOF_CLASS_TEST(p, c) (ANYOF_CLASS_BYTE(p, c) & ANYOF_BIT(c))
226 #define ANYOF_BITMAP_OFFSET (ANYOF_CLASS_OFFSET+ANYOF_CLASS_SIZE)
227 #define ANYOF_BITMAP_BYTE(p, c) ((p)[ANYOF_BITMAP_OFFSET + (((c) >> 3) & 31)])
228 #define ANYOF_BITMAP_SET(p, c) (ANYOF_BITMAP_BYTE(p, c) |= ANYOF_BIT(c))
229 #define ANYOF_BITMAP_CLEAR(p,c) (ANYOF_BITMAP_BYTE(p, c) &= ~ANYOF_BIT(c))
230 #define ANYOF_BITMAP_TEST(p, c) (ANYOF_BITMAP_BYTE(p, c) & ANYOF_BIT(c))
232 #define ANY_SKIP ((ANYOF_SIZE - 1)/sizeof(regnode) + 1)
235 * Utility definitions.
239 #define UCHARAT(p) ((int)*(U8*)(p))
241 #define UCHARAT(p) ((int)*(p)&CHARMASK)
244 #define UCHARAT(p) PL_regdummy
250 SAVEDESTRUCTOR(clear_re,(void*)PL_regcomp_rx); \
251 Perl_croak(aTHX_ "/%.127s/: %s", PL_regprecomp,m); \
254 #define FAIL2(pat,m) \
257 SAVEDESTRUCTOR(clear_re,(void*)PL_regcomp_rx); \
258 S_re_croak2(aTHX_ "/%.127s/: ",pat,PL_regprecomp,m); \
261 #define EXTRA_SIZE(guy) ((sizeof(guy)-1)/sizeof(struct regnode))
263 #define REG_SEEN_ZERO_LEN 1
264 #define REG_SEEN_LOOKBEHIND 2
265 #define REG_SEEN_GPOS 4
266 #define REG_SEEN_EVAL 8
270 #include "regnodes.h"
272 /* The following have no fixed length. U8 so we can do strchr() on it. */
274 EXTCONST U8 PL_varies[];
276 EXTCONST U8 PL_varies[] = {
277 BRANCH, BACK, STAR, PLUS, CURLY, CURLYX, REF, REFF, REFFL,
278 WHILEM, CURLYM, CURLYN, BRANCHJ, IFTHEN, SUSPEND, CLUMP, 0
282 /* The following always have a length of 1. U8 we can do strchr() on it. */
283 /* (Note that length 1 means "one character" under UTF8, not "one octet".) */
285 EXTCONST U8 PL_simple[];
287 EXTCONST U8 PL_simple[] = {
288 REG_ANY, ANYUTF8, SANY, SANYUTF8, ANYOF, ANYOFUTF8,
289 ALNUM, ALNUMUTF8, ALNUML, ALNUMLUTF8,
290 NALNUM, NALNUMUTF8, NALNUML, NALNUMLUTF8,
291 SPACE, SPACEUTF8, SPACEL, SPACELUTF8,
292 NSPACE, NSPACEUTF8, NSPACEL, NSPACELUTF8,
293 DIGIT, DIGITUTF8, NDIGIT, NDIGITUTF8, 0
299 typedef struct re_scream_pos_data_s
301 char **scream_olds; /* match pos */
302 I32 *scream_pos; /* Internal iterator of scream. */
303 } re_scream_pos_data;
311 struct reg_substr_datum {
317 struct reg_substr_data {
318 struct reg_substr_datum data[3]; /* Actual array */
321 #define anchored_substr substrs->data[0].substr
322 #define anchored_offset substrs->data[0].min_offset
323 #define float_substr substrs->data[1].substr
324 #define float_min_offset substrs->data[1].min_offset
325 #define float_max_offset substrs->data[1].max_offset
326 #define check_substr substrs->data[2].substr
327 #define check_offset_min substrs->data[2].min_offset
328 #define check_offset_max substrs->data[2].max_offset