| 1 | /* regcomp.c |
| 2 | */ |
| 3 | |
| 4 | /* |
| 5 | * "A fair jaw-cracker dwarf-language must be." --Samwise Gamgee |
| 6 | */ |
| 7 | |
| 8 | /* NOTE: this is derived from Henry Spencer's regexp code, and should not |
| 9 | * confused with the original package (see point 3 below). Thanks, Henry! |
| 10 | */ |
| 11 | |
| 12 | /* Additional note: this code is very heavily munged from Henry's version |
| 13 | * in places. In some spots I've traded clarity for efficiency, so don't |
| 14 | * blame Henry for some of the lack of readability. |
| 15 | */ |
| 16 | |
| 17 | /* The names of the functions have been changed from regcomp and |
| 18 | * regexec to pregcomp and pregexec in order to avoid conflicts |
| 19 | * with the POSIX routines of the same names. |
| 20 | */ |
| 21 | |
| 22 | /*SUPPRESS 112*/ |
| 23 | /* |
| 24 | * pregcomp and pregexec -- regsub and regerror are not used in perl |
| 25 | * |
| 26 | * Copyright (c) 1986 by University of Toronto. |
| 27 | * Written by Henry Spencer. Not derived from licensed software. |
| 28 | * |
| 29 | * Permission is granted to anyone to use this software for any |
| 30 | * purpose on any computer system, and to redistribute it freely, |
| 31 | * subject to the following restrictions: |
| 32 | * |
| 33 | * 1. The author is not responsible for the consequences of use of |
| 34 | * this software, no matter how awful, even if they arise |
| 35 | * from defects in it. |
| 36 | * |
| 37 | * 2. The origin of this software must not be misrepresented, either |
| 38 | * by explicit claim or by omission. |
| 39 | * |
| 40 | * 3. Altered versions must be plainly marked as such, and must not |
| 41 | * be misrepresented as being the original software. |
| 42 | * |
| 43 | * |
| 44 | **** Alterations to Henry's code are... |
| 45 | **** |
| 46 | **** Copyright (c) 1991-1997, Larry Wall |
| 47 | **** |
| 48 | **** You may distribute under the terms of either the GNU General Public |
| 49 | **** License or the Artistic License, as specified in the README file. |
| 50 | |
| 51 | * |
| 52 | * Beware that some of this code is subtly aware of the way operator |
| 53 | * precedence is structured in regular expressions. Serious changes in |
| 54 | * regular-expression syntax might require a total rethink. |
| 55 | */ |
| 56 | #include "EXTERN.h" |
| 57 | #include "perl.h" |
| 58 | #include "INTERN.h" |
| 59 | |
| 60 | #define REG_COMP_C |
| 61 | #include "regcomp.h" |
| 62 | |
| 63 | #ifdef op |
| 64 | #undef op |
| 65 | #endif /* op */ |
| 66 | |
| 67 | static regnode regdummy; |
| 68 | static char * regparse; /* Input-scan pointer. */ |
| 69 | static char * regxend; /* End of input for compile */ |
| 70 | static regnode * regcode; /* Code-emit pointer; ®dummy = don't. */ |
| 71 | static I32 regnaughty; /* How bad is this pattern? */ |
| 72 | static I32 regsawback; /* Did we see \1, ...? */ |
| 73 | |
| 74 | /* This guys appear both in regcomp.c and regexec.c, but there is no |
| 75 | other reason to have them global. */ |
| 76 | static char * regprecomp; /* uncompiled string. */ |
| 77 | static I32 regnpar; /* () count. */ |
| 78 | static I32 regsize; /* Code size. */ |
| 79 | static U16 regflags; /* are we folding, multilining? */ |
| 80 | |
| 81 | #ifdef MSDOS |
| 82 | # if defined(BUGGY_MSC6) |
| 83 | /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */ |
| 84 | # pragma optimize("a",off) |
| 85 | /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/ |
| 86 | # pragma optimize("w",on ) |
| 87 | # endif /* BUGGY_MSC6 */ |
| 88 | #endif /* MSDOS */ |
| 89 | |
| 90 | #ifndef STATIC |
| 91 | #define STATIC static |
| 92 | #endif |
| 93 | |
| 94 | #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?') |
| 95 | #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \ |
| 96 | ((*s) == '{' && regcurly(s))) |
| 97 | #ifdef atarist |
| 98 | #define PERL_META "^$.[()|?+*\\" |
| 99 | #else |
| 100 | #define META "^$.[()|?+*\\" |
| 101 | #endif |
| 102 | |
| 103 | #ifdef SPSTART |
| 104 | #undef SPSTART /* dratted cpp namespace... */ |
| 105 | #endif |
| 106 | /* |
| 107 | * Flags to be passed up and down. |
| 108 | */ |
| 109 | #define WORST 0 /* Worst case. */ |
| 110 | #define HASWIDTH 0x1 /* Known never to match null string. */ |
| 111 | #define SIMPLE 0x2 /* Simple enough to be STAR/PLUS operand. */ |
| 112 | #define SPSTART 0x4 /* Starts with * or +. */ |
| 113 | #define TRYAGAIN 0x8 /* Weeded out a declaration. */ |
| 114 | |
| 115 | /* |
| 116 | * Forward declarations for pregcomp()'s friends. |
| 117 | */ |
| 118 | |
| 119 | static regnode *reg _((I32, I32 *)); |
| 120 | static regnode *reganode _((U8, U32)); |
| 121 | static regnode *regatom _((I32 *)); |
| 122 | static regnode *regbranch _((I32 *, I32)); |
| 123 | static void regc _((U8, char *)); |
| 124 | static regnode *regclass _((void)); |
| 125 | STATIC I32 regcurly _((char *)); |
| 126 | static regnode *reg_node _((U8)); |
| 127 | static regnode *regpiece _((I32 *)); |
| 128 | static void reginsert _((U8, regnode *)); |
| 129 | static void regoptail _((regnode *, regnode *)); |
| 130 | static void regset _((char *, I32)); |
| 131 | static void regtail _((regnode *, regnode *)); |
| 132 | static char* regwhite _((char *, char *)); |
| 133 | static char* nextchar _((void)); |
| 134 | static void re_croak2 _((const char* pat1,const char* pat2,...)) __attribute__((noreturn)); |
| 135 | |
| 136 | static U32 regseen; |
| 137 | static I32 seen_zerolen; |
| 138 | static regexp *rx; |
| 139 | static I32 extralen; |
| 140 | |
| 141 | #ifdef DEBUGGING |
| 142 | static int colorset; |
| 143 | char *colors[4]; |
| 144 | #endif |
| 145 | |
| 146 | /* Length of a variant. */ |
| 147 | |
| 148 | typedef struct { |
| 149 | I32 len_min; |
| 150 | I32 len_delta; |
| 151 | I32 pos_min; |
| 152 | I32 pos_delta; |
| 153 | SV *last_found; |
| 154 | I32 last_end; /* min value, <0 unless valid. */ |
| 155 | I32 last_start_min; |
| 156 | I32 last_start_max; |
| 157 | SV **longest; /* Either &l_fixed, or &l_float. */ |
| 158 | SV *longest_fixed; |
| 159 | I32 offset_fixed; |
| 160 | SV *longest_float; |
| 161 | I32 offset_float_min; |
| 162 | I32 offset_float_max; |
| 163 | I32 flags; |
| 164 | } scan_data_t; |
| 165 | |
| 166 | static scan_data_t zero_scan_data = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; |
| 167 | |
| 168 | #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL) |
| 169 | #define SF_BEFORE_SEOL 0x1 |
| 170 | #define SF_BEFORE_MEOL 0x2 |
| 171 | #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL) |
| 172 | #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL) |
| 173 | |
| 174 | #define SF_FIX_SHIFT_EOL (+2) |
| 175 | #define SF_FL_SHIFT_EOL (+4) |
| 176 | |
| 177 | #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL) |
| 178 | #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL) |
| 179 | |
| 180 | #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL) |
| 181 | #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */ |
| 182 | #define SF_IS_INF 0x40 |
| 183 | #define SF_HAS_PAR 0x80 |
| 184 | #define SF_IN_PAR 0x100 |
| 185 | #define SF_HAS_EVAL 0x200 |
| 186 | #define SCF_DO_SUBSTR 0x400 |
| 187 | |
| 188 | static void |
| 189 | scan_commit(scan_data_t *data) |
| 190 | { |
| 191 | STRLEN l = SvCUR(data->last_found); |
| 192 | STRLEN old_l = SvCUR(*data->longest); |
| 193 | |
| 194 | if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) { |
| 195 | sv_setsv(*data->longest, data->last_found); |
| 196 | if (*data->longest == data->longest_fixed) { |
| 197 | data->offset_fixed = l ? data->last_start_min : data->pos_min; |
| 198 | if (data->flags & SF_BEFORE_EOL) |
| 199 | data->flags |
| 200 | |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL); |
| 201 | else |
| 202 | data->flags &= ~SF_FIX_BEFORE_EOL; |
| 203 | } else { |
| 204 | data->offset_float_min = l ? data->last_start_min : data->pos_min; |
| 205 | data->offset_float_max = (l |
| 206 | ? data->last_start_max |
| 207 | : data->pos_min + data->pos_delta); |
| 208 | if (data->flags & SF_BEFORE_EOL) |
| 209 | data->flags |
| 210 | |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL); |
| 211 | else |
| 212 | data->flags &= ~SF_FL_BEFORE_EOL; |
| 213 | } |
| 214 | } |
| 215 | SvCUR_set(data->last_found, 0); |
| 216 | data->last_end = -1; |
| 217 | data->flags &= ~SF_BEFORE_EOL; |
| 218 | } |
| 219 | |
| 220 | /* Stops at toplevel WHILEM as well as at `last'. At end *scanp is set |
| 221 | to the position after last scanned or to NULL. */ |
| 222 | |
| 223 | static I32 |
| 224 | study_chunk(regnode **scanp, I32 *deltap, regnode *last, scan_data_t *data, U32 flags) |
| 225 | /* scanp: Start here (read-write). */ |
| 226 | /* deltap: Write maxlen-minlen here. */ |
| 227 | /* last: Stop before this one. */ |
| 228 | { |
| 229 | I32 min = 0, pars = 0, code; |
| 230 | regnode *scan = *scanp, *next; |
| 231 | I32 delta = 0; |
| 232 | int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF); |
| 233 | I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0; |
| 234 | scan_data_t data_fake; |
| 235 | |
| 236 | while (scan && OP(scan) != END && scan < last) { |
| 237 | /* Peephole optimizer: */ |
| 238 | |
| 239 | if (regkind[(U8)OP(scan)] == EXACT) { |
| 240 | regnode *n = regnext(scan); |
| 241 | U32 stringok = 1; |
| 242 | #ifdef DEBUGGING |
| 243 | regnode *stop = scan; |
| 244 | #endif |
| 245 | |
| 246 | next = scan + (*OPERAND(scan) + 2 - 1)/sizeof(regnode) + 2; |
| 247 | /* Skip NOTHING, merge EXACT*. */ |
| 248 | while (n && |
| 249 | ( regkind[(U8)OP(n)] == NOTHING || |
| 250 | (stringok && (OP(n) == OP(scan)))) |
| 251 | && NEXT_OFF(n) |
| 252 | && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) { |
| 253 | if (OP(n) == TAIL || n > next) |
| 254 | stringok = 0; |
| 255 | if (regkind[(U8)OP(n)] == NOTHING) { |
| 256 | NEXT_OFF(scan) += NEXT_OFF(n); |
| 257 | next = n + NODE_STEP_REGNODE; |
| 258 | #ifdef DEBUGGING |
| 259 | if (stringok) |
| 260 | stop = n; |
| 261 | #endif |
| 262 | n = regnext(n); |
| 263 | } else { |
| 264 | int oldl = *OPERAND(scan); |
| 265 | regnode *nnext = regnext(n); |
| 266 | |
| 267 | if (oldl + *OPERAND(n) > U8_MAX) |
| 268 | break; |
| 269 | NEXT_OFF(scan) += NEXT_OFF(n); |
| 270 | *OPERAND(scan) += *OPERAND(n); |
| 271 | next = n + (*OPERAND(n) + 2 - 1)/sizeof(regnode) + 2; |
| 272 | /* Now we can overwrite *n : */ |
| 273 | Move(OPERAND(n) + 1, OPERAND(scan) + oldl + 1, |
| 274 | *OPERAND(n) + 1, char); |
| 275 | #ifdef DEBUGGING |
| 276 | if (stringok) |
| 277 | stop = next - 1; |
| 278 | #endif |
| 279 | n = nnext; |
| 280 | } |
| 281 | } |
| 282 | #ifdef DEBUGGING |
| 283 | /* Allow dumping */ |
| 284 | n = scan + (*OPERAND(scan) + 2 - 1)/sizeof(regnode) + 2; |
| 285 | while (n <= stop) { |
| 286 | if (regkind[(U8)OP(n)] != NOTHING || OP(n) == NOTHING) { |
| 287 | OP(n) = OPTIMIZED; |
| 288 | NEXT_OFF(n) = 0; |
| 289 | } |
| 290 | n++; |
| 291 | } |
| 292 | #endif |
| 293 | |
| 294 | } |
| 295 | if (OP(scan) != CURLYX) { |
| 296 | int max = (reg_off_by_arg[OP(scan)] ? I32_MAX : U16_MAX); |
| 297 | int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan)); |
| 298 | int noff; |
| 299 | regnode *n = scan; |
| 300 | |
| 301 | /* Skip NOTHING and LONGJMP. */ |
| 302 | while ((n = regnext(n)) |
| 303 | && ((regkind[(U8)OP(n)] == NOTHING && (noff = NEXT_OFF(n))) |
| 304 | || ((OP(n) == LONGJMP) && (noff = ARG(n)))) |
| 305 | && off + noff < max) |
| 306 | off += noff; |
| 307 | if (reg_off_by_arg[OP(scan)]) |
| 308 | ARG(scan) = off; |
| 309 | else |
| 310 | NEXT_OFF(scan) = off; |
| 311 | } |
| 312 | if (OP(scan) == BRANCH || OP(scan) == BRANCHJ |
| 313 | || OP(scan) == IFTHEN || OP(scan) == SUSPEND) { |
| 314 | next = regnext(scan); |
| 315 | code = OP(scan); |
| 316 | |
| 317 | if (OP(next) == code || code == IFTHEN || code == SUSPEND) { |
| 318 | I32 max1 = 0, min1 = I32_MAX, num = 0; |
| 319 | |
| 320 | if (flags & SCF_DO_SUBSTR) |
| 321 | scan_commit(data); |
| 322 | while (OP(scan) == code) { |
| 323 | I32 deltanext, minnext; |
| 324 | |
| 325 | num++; |
| 326 | data_fake.flags = 0; |
| 327 | next = regnext(scan); |
| 328 | scan = NEXTOPER(scan); |
| 329 | if (code != BRANCH) |
| 330 | scan = NEXTOPER(scan); |
| 331 | /* We suppose the run is continuous, last=next...*/ |
| 332 | minnext = study_chunk(&scan, &deltanext, next, |
| 333 | &data_fake, 0); |
| 334 | if (min1 > minnext) |
| 335 | min1 = minnext; |
| 336 | if (max1 < minnext + deltanext) |
| 337 | max1 = minnext + deltanext; |
| 338 | if (deltanext == I32_MAX) |
| 339 | is_inf = 1; |
| 340 | scan = next; |
| 341 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 342 | pars++; |
| 343 | if (data && (data_fake.flags & SF_HAS_EVAL)) |
| 344 | data->flags |= SF_HAS_EVAL; |
| 345 | if (code == SUSPEND) |
| 346 | break; |
| 347 | } |
| 348 | if (code == IFTHEN && num < 2) /* Empty ELSE branch */ |
| 349 | min1 = 0; |
| 350 | if (flags & SCF_DO_SUBSTR) { |
| 351 | data->pos_min += min1; |
| 352 | data->pos_delta += max1 - min1; |
| 353 | if (max1 != min1 || is_inf) |
| 354 | data->longest = &(data->longest_float); |
| 355 | } |
| 356 | min += min1; |
| 357 | delta += max1 - min1; |
| 358 | } else if (code == BRANCHJ) /* single branch is optimized. */ |
| 359 | scan = NEXTOPER(NEXTOPER(scan)); |
| 360 | else /* single branch is optimized. */ |
| 361 | scan = NEXTOPER(scan); |
| 362 | continue; |
| 363 | } else if (OP(scan) == EXACT) { |
| 364 | min += *OPERAND(scan); |
| 365 | if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */ |
| 366 | I32 l = *OPERAND(scan); |
| 367 | |
| 368 | /* The code below prefers earlier match for fixed |
| 369 | offset, later match for variable offset. */ |
| 370 | if (data->last_end == -1) { /* Update the start info. */ |
| 371 | data->last_start_min = data->pos_min; |
| 372 | data->last_start_max = is_inf |
| 373 | ? I32_MAX : data->pos_min + data->pos_delta; |
| 374 | } |
| 375 | sv_catpvn(data->last_found, (char *)(OPERAND(scan)+1), l); |
| 376 | data->last_end = data->pos_min + l; |
| 377 | data->pos_min += l; /* As in the first entry. */ |
| 378 | data->flags &= ~SF_BEFORE_EOL; |
| 379 | } |
| 380 | } else if (regkind[(U8)OP(scan)] == EXACT) { |
| 381 | if (flags & SCF_DO_SUBSTR) |
| 382 | scan_commit(data); |
| 383 | min += *OPERAND(scan); |
| 384 | if (data && (flags & SCF_DO_SUBSTR)) |
| 385 | data->pos_min += *OPERAND(scan); |
| 386 | } else if (strchr(varies,OP(scan))) { |
| 387 | I32 mincount, maxcount, minnext, deltanext, pos_before, fl; |
| 388 | regnode *oscan = scan; |
| 389 | |
| 390 | switch (regkind[(U8)OP(scan)]) { |
| 391 | case WHILEM: |
| 392 | scan = NEXTOPER(scan); |
| 393 | goto finish; |
| 394 | case PLUS: |
| 395 | if (flags & SCF_DO_SUBSTR) { |
| 396 | next = NEXTOPER(scan); |
| 397 | if (OP(next) == EXACT) { |
| 398 | mincount = 1; |
| 399 | maxcount = REG_INFTY; |
| 400 | next = regnext(scan); |
| 401 | scan = NEXTOPER(scan); |
| 402 | goto do_curly; |
| 403 | } |
| 404 | } |
| 405 | if (flags & SCF_DO_SUBSTR) |
| 406 | data->pos_min++; |
| 407 | min++; |
| 408 | /* Fall through. */ |
| 409 | case STAR: |
| 410 | is_inf = 1; |
| 411 | scan = regnext(scan); |
| 412 | if (flags & SCF_DO_SUBSTR) { |
| 413 | scan_commit(data); |
| 414 | data->longest = &(data->longest_float); |
| 415 | } |
| 416 | goto optimize_curly_tail; |
| 417 | case CURLY: |
| 418 | mincount = ARG1(scan); |
| 419 | maxcount = ARG2(scan); |
| 420 | next = regnext(scan); |
| 421 | scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS; |
| 422 | do_curly: |
| 423 | if (flags & SCF_DO_SUBSTR) { |
| 424 | if (mincount == 0) scan_commit(data); |
| 425 | pos_before = data->pos_min; |
| 426 | } |
| 427 | if (data) { |
| 428 | fl = data->flags; |
| 429 | data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL); |
| 430 | if (is_inf) |
| 431 | data->flags |= SF_IS_INF; |
| 432 | } |
| 433 | /* This will finish on WHILEM, setting scan, or on NULL: */ |
| 434 | minnext = study_chunk(&scan, &deltanext, last, data, |
| 435 | mincount == 0 |
| 436 | ? (flags & ~SCF_DO_SUBSTR) : flags); |
| 437 | if (!scan) /* It was not CURLYX, but CURLY. */ |
| 438 | scan = next; |
| 439 | if (dowarn && (minnext + deltanext == 0) |
| 440 | && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))) |
| 441 | warn("Strange *+?{} on zero-length expression"); |
| 442 | min += minnext * mincount; |
| 443 | is_inf |= (maxcount == REG_INFTY && (minnext + deltanext) > 0 |
| 444 | || deltanext == I32_MAX); |
| 445 | delta += (minnext + deltanext) * maxcount - minnext * mincount; |
| 446 | |
| 447 | /* Try powerful optimization CURLYX => CURLYN. */ |
| 448 | #ifdef REGALIGN_STRUCT |
| 449 | if ( OP(oscan) == CURLYX && data |
| 450 | && data->flags & SF_IN_PAR |
| 451 | && !(data->flags & SF_HAS_EVAL) |
| 452 | && !deltanext && minnext == 1 ) { |
| 453 | /* Try to optimize to CURLYN. */ |
| 454 | regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; |
| 455 | regnode *nxt1 = nxt, *nxt2; |
| 456 | |
| 457 | /* Skip open. */ |
| 458 | nxt = regnext(nxt); |
| 459 | if (!strchr(simple,OP(nxt)) |
| 460 | && !(regkind[(U8)OP(nxt)] == EXACT |
| 461 | && *OPERAND(nxt) == 1)) |
| 462 | goto nogo; |
| 463 | nxt2 = nxt; |
| 464 | nxt = regnext(nxt); |
| 465 | if (OP(nxt) != CLOSE) |
| 466 | goto nogo; |
| 467 | /* Now we know that nxt2 is the only contents: */ |
| 468 | oscan->flags = ARG(nxt); |
| 469 | OP(oscan) = CURLYN; |
| 470 | OP(nxt1) = NOTHING; /* was OPEN. */ |
| 471 | #ifdef DEBUGGING |
| 472 | OP(nxt1 + 1) = OPTIMIZED; /* was count. */ |
| 473 | NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */ |
| 474 | NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */ |
| 475 | OP(nxt) = OPTIMIZED; /* was CLOSE. */ |
| 476 | OP(nxt + 1) = OPTIMIZED; /* was count. */ |
| 477 | NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */ |
| 478 | #endif |
| 479 | } |
| 480 | #endif |
| 481 | nogo: |
| 482 | |
| 483 | /* Try optimization CURLYX => CURLYM. */ |
| 484 | if ( OP(oscan) == CURLYX && data |
| 485 | #ifdef REGALIGN_STRUCT |
| 486 | && !(data->flags & SF_HAS_PAR) |
| 487 | #else |
| 488 | && !(data->flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 489 | #endif |
| 490 | && !(data->flags & SF_HAS_EVAL) |
| 491 | && !deltanext ) { |
| 492 | /* XXXX How to optimize if data == 0? */ |
| 493 | /* Optimize to a simpler form. */ |
| 494 | regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */ |
| 495 | regnode *nxt2; |
| 496 | |
| 497 | OP(oscan) = CURLYM; |
| 498 | while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/ |
| 499 | && (OP(nxt2) != WHILEM)) |
| 500 | nxt = nxt2; |
| 501 | OP(nxt2) = SUCCEED; /* Whas WHILEM */ |
| 502 | #ifdef REGALIGN_STRUCT |
| 503 | /* Need to optimize away parenths. */ |
| 504 | if (data->flags & SF_IN_PAR) { |
| 505 | /* Set the parenth number. */ |
| 506 | regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/ |
| 507 | |
| 508 | if (OP(nxt) != CLOSE) |
| 509 | FAIL("panic opt close"); |
| 510 | oscan->flags = ARG(nxt); |
| 511 | OP(nxt1) = OPTIMIZED; /* was OPEN. */ |
| 512 | OP(nxt) = OPTIMIZED; /* was CLOSE. */ |
| 513 | #ifdef DEBUGGING |
| 514 | OP(nxt1 + 1) = OPTIMIZED; /* was count. */ |
| 515 | OP(nxt + 1) = OPTIMIZED; /* was count. */ |
| 516 | NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */ |
| 517 | NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */ |
| 518 | #endif |
| 519 | #if 0 |
| 520 | while ( nxt1 && (OP(nxt1) != WHILEM)) { |
| 521 | regnode *nnxt = regnext(nxt1); |
| 522 | |
| 523 | if (nnxt == nxt) { |
| 524 | if (reg_off_by_arg[OP(nxt1)]) |
| 525 | ARG_SET(nxt1, nxt2 - nxt1); |
| 526 | else if (nxt2 - nxt1 < U16_MAX) |
| 527 | NEXT_OFF(nxt1) = nxt2 - nxt1; |
| 528 | else |
| 529 | OP(nxt) = NOTHING; /* Cannot beautify */ |
| 530 | } |
| 531 | nxt1 = nnxt; |
| 532 | } |
| 533 | #endif |
| 534 | /* Optimize again: */ |
| 535 | study_chunk(&nxt1, &deltanext, nxt, NULL, 0); |
| 536 | } else |
| 537 | oscan->flags = 0; |
| 538 | #endif |
| 539 | } |
| 540 | if (data && fl & (SF_HAS_PAR|SF_IN_PAR)) |
| 541 | pars++; |
| 542 | if (flags & SCF_DO_SUBSTR) { |
| 543 | SV *last_str = Nullsv; |
| 544 | int counted = mincount != 0; |
| 545 | |
| 546 | if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */ |
| 547 | I32 b = pos_before >= data->last_start_min |
| 548 | ? pos_before : data->last_start_min; |
| 549 | STRLEN l; |
| 550 | char *s = SvPV(data->last_found, l); |
| 551 | |
| 552 | l -= b - data->last_start_min; |
| 553 | /* Get the added string: */ |
| 554 | last_str = newSVpv(s + b - data->last_start_min, l); |
| 555 | if (deltanext == 0 && pos_before == b) { |
| 556 | /* What was added is a constant string */ |
| 557 | if (mincount > 1) { |
| 558 | SvGROW(last_str, (mincount * l) + 1); |
| 559 | repeatcpy(SvPVX(last_str) + l, |
| 560 | SvPVX(last_str), l, mincount - 1); |
| 561 | SvCUR(last_str) *= mincount; |
| 562 | /* Add additional parts. */ |
| 563 | SvCUR_set(data->last_found, |
| 564 | SvCUR(data->last_found) - l); |
| 565 | sv_catsv(data->last_found, last_str); |
| 566 | data->last_end += l * (mincount - 1); |
| 567 | } |
| 568 | } |
| 569 | } |
| 570 | /* It is counted once already... */ |
| 571 | data->pos_min += minnext * (mincount - counted); |
| 572 | data->pos_delta += - counted * deltanext + |
| 573 | (minnext + deltanext) * maxcount - minnext * mincount; |
| 574 | if (mincount != maxcount) { |
| 575 | scan_commit(data); |
| 576 | if (mincount && last_str) { |
| 577 | sv_setsv(data->last_found, last_str); |
| 578 | data->last_end = data->pos_min; |
| 579 | data->last_start_min = |
| 580 | data->pos_min - SvCUR(last_str); |
| 581 | data->last_start_max = is_inf |
| 582 | ? I32_MAX |
| 583 | : data->pos_min + data->pos_delta |
| 584 | - SvCUR(last_str); |
| 585 | } |
| 586 | data->longest = &(data->longest_float); |
| 587 | } |
| 588 | } |
| 589 | if (data && (fl & SF_HAS_EVAL)) |
| 590 | data->flags |= SF_HAS_EVAL; |
| 591 | optimize_curly_tail: |
| 592 | #ifdef REGALIGN |
| 593 | if (OP(oscan) != CURLYX) { |
| 594 | while (regkind[(U8)OP(next = regnext(oscan))] == NOTHING |
| 595 | && NEXT_OFF(next)) |
| 596 | NEXT_OFF(oscan) += NEXT_OFF(next); |
| 597 | } |
| 598 | #endif |
| 599 | continue; |
| 600 | default: /* REF only? */ |
| 601 | if (flags & SCF_DO_SUBSTR) { |
| 602 | scan_commit(data); |
| 603 | data->longest = &(data->longest_float); |
| 604 | } |
| 605 | is_inf = 1; |
| 606 | break; |
| 607 | } |
| 608 | } else if (strchr(simple,OP(scan))) { |
| 609 | if (flags & SCF_DO_SUBSTR) { |
| 610 | scan_commit(data); |
| 611 | data->pos_min++; |
| 612 | } |
| 613 | min++; |
| 614 | } else if (regkind[(U8)OP(scan)] == EOL && flags & SCF_DO_SUBSTR) { |
| 615 | data->flags |= (OP(scan) == MEOL |
| 616 | ? SF_BEFORE_MEOL |
| 617 | : SF_BEFORE_SEOL); |
| 618 | } else if (regkind[(U8)OP(scan)] == BRANCHJ |
| 619 | && (scan->flags || data) |
| 620 | && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) { |
| 621 | I32 deltanext, minnext; |
| 622 | regnode *nscan; |
| 623 | |
| 624 | data_fake.flags = 0; |
| 625 | next = regnext(scan); |
| 626 | nscan = NEXTOPER(NEXTOPER(scan)); |
| 627 | minnext = study_chunk(&nscan, &deltanext, last, &data_fake, 0); |
| 628 | if (scan->flags) { |
| 629 | if (deltanext) { |
| 630 | FAIL("variable length lookbehind not implemented"); |
| 631 | } else if (minnext > U8_MAX) { |
| 632 | FAIL2("lookbehind longer than %d not implemented", U8_MAX); |
| 633 | } |
| 634 | scan->flags = minnext; |
| 635 | } |
| 636 | if (data && data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) |
| 637 | pars++; |
| 638 | if (data && (data_fake.flags & SF_HAS_EVAL)) |
| 639 | data->flags |= SF_HAS_EVAL; |
| 640 | } else if (OP(scan) == OPEN) { |
| 641 | pars++; |
| 642 | } else if (OP(scan) == CLOSE && ARG(scan) == is_par) { |
| 643 | #ifdef REGALIGN_STRUCT |
| 644 | next = regnext(scan); |
| 645 | |
| 646 | if ( next && (OP(next) != WHILEM) && next < last) |
| 647 | #endif |
| 648 | is_par = 0; /* Disable optimization */ |
| 649 | } else if (OP(scan) == EVAL) { |
| 650 | if (data) |
| 651 | data->flags |= SF_HAS_EVAL; |
| 652 | } |
| 653 | /* Else: zero-length, ignore. */ |
| 654 | scan = regnext(scan); |
| 655 | } |
| 656 | |
| 657 | finish: |
| 658 | *scanp = scan; |
| 659 | *deltap = is_inf ? I32_MAX : delta; |
| 660 | if (flags & SCF_DO_SUBSTR && is_inf) |
| 661 | data->pos_delta = I32_MAX - data->pos_min; |
| 662 | if (is_par > U8_MAX) |
| 663 | is_par = 0; |
| 664 | if (is_par && pars==1 && data) { |
| 665 | data->flags |= SF_IN_PAR; |
| 666 | data->flags &= ~SF_HAS_PAR; |
| 667 | } else if (pars && data) { |
| 668 | data->flags |= SF_HAS_PAR; |
| 669 | data->flags &= ~SF_IN_PAR; |
| 670 | } |
| 671 | return min; |
| 672 | } |
| 673 | |
| 674 | static I32 |
| 675 | add_data(I32 n, char *s) |
| 676 | { |
| 677 | if (rx->data) { |
| 678 | Renewc(rx->data, |
| 679 | sizeof(*rx->data) + sizeof(void*) * (rx->data->count + n - 1), |
| 680 | char, struct reg_data); |
| 681 | Renew(rx->data->what, rx->data->count + n, U8); |
| 682 | rx->data->count += n; |
| 683 | } else { |
| 684 | Newc(1207, rx->data, sizeof(*rx->data) + sizeof(void*) * (n - 1), |
| 685 | char, struct reg_data); |
| 686 | New(1208, rx->data->what, n, U8); |
| 687 | rx->data->count = n; |
| 688 | } |
| 689 | Copy(s, rx->data->what + rx->data->count - n, n, U8); |
| 690 | return rx->data->count - n; |
| 691 | } |
| 692 | |
| 693 | /* |
| 694 | - pregcomp - compile a regular expression into internal code |
| 695 | * |
| 696 | * We can't allocate space until we know how big the compiled form will be, |
| 697 | * but we can't compile it (and thus know how big it is) until we've got a |
| 698 | * place to put the code. So we cheat: we compile it twice, once with code |
| 699 | * generation turned off and size counting turned on, and once "for real". |
| 700 | * This also means that we don't allocate space until we are sure that the |
| 701 | * thing really will compile successfully, and we never have to move the |
| 702 | * code and thus invalidate pointers into it. (Note that it has to be in |
| 703 | * one piece because free() must be able to free it all.) [NB: not true in perl] |
| 704 | * |
| 705 | * Beware that the optimization-preparation code in here knows about some |
| 706 | * of the structure of the compiled regexp. [I'll say.] |
| 707 | */ |
| 708 | regexp * |
| 709 | pregcomp(char *exp, char *xend, PMOP *pm) |
| 710 | { |
| 711 | register regexp *r; |
| 712 | regnode *scan; |
| 713 | SV **longest; |
| 714 | SV *longest_fixed; |
| 715 | SV *longest_float; |
| 716 | regnode *first; |
| 717 | I32 flags; |
| 718 | I32 minlen = 0; |
| 719 | I32 sawplus = 0; |
| 720 | I32 sawopen = 0; |
| 721 | |
| 722 | if (exp == NULL) |
| 723 | FAIL("NULL regexp argument"); |
| 724 | |
| 725 | regprecomp = savepvn(exp, xend - exp); |
| 726 | DEBUG_r(PerlIO_printf(Perl_debug_log, "compiling RE `%*s'\n", |
| 727 | xend - exp, regprecomp)); |
| 728 | regflags = pm->op_pmflags; |
| 729 | regsawback = 0; |
| 730 | |
| 731 | regseen = 0; |
| 732 | seen_zerolen = *exp == '^' ? -1 : 0; |
| 733 | extralen = 0; |
| 734 | |
| 735 | /* First pass: determine size, legality. */ |
| 736 | regparse = exp; |
| 737 | regxend = xend; |
| 738 | regnaughty = 0; |
| 739 | regnpar = 1; |
| 740 | regsize = 0L; |
| 741 | regcode = ®dummy; |
| 742 | regc((U8)MAGIC, (char*)regcode); |
| 743 | if (reg(0, &flags) == NULL) { |
| 744 | Safefree(regprecomp); |
| 745 | regprecomp = Nullch; |
| 746 | return(NULL); |
| 747 | } |
| 748 | DEBUG_r(PerlIO_printf(Perl_debug_log, "size %d ", regsize)); |
| 749 | |
| 750 | DEBUG_r( |
| 751 | if (!colorset) { |
| 752 | int i = 0; |
| 753 | char *s = PerlEnv_getenv("TERMCAP_COLORS"); |
| 754 | |
| 755 | colorset = 1; |
| 756 | if (s) { |
| 757 | colors[0] = s = savepv(s); |
| 758 | while (++i < 4) { |
| 759 | s = strchr(s, '\t'); |
| 760 | if (!s) |
| 761 | FAIL("Not enough TABs in TERMCAP_COLORS"); |
| 762 | *s = '\0'; |
| 763 | colors[i] = ++s; |
| 764 | } |
| 765 | } else { |
| 766 | while (i < 4) |
| 767 | colors[i++] = ""; |
| 768 | } |
| 769 | /* Reset colors: */ |
| 770 | PerlIO_printf(Perl_debug_log, "%s%s%s%s", |
| 771 | colors[0],colors[1],colors[2],colors[3]); |
| 772 | } |
| 773 | ); |
| 774 | |
| 775 | /* Small enough for pointer-storage convention? |
| 776 | If extralen==0, this means that we will not need long jumps. */ |
| 777 | #ifndef REGALIGN_STRUCT |
| 778 | if (regsize >= 0x10000L && extralen) |
| 779 | FAIL("regexp too big"); |
| 780 | #else |
| 781 | if (regsize >= 0x10000L && extralen) |
| 782 | regsize += extralen; |
| 783 | else |
| 784 | extralen = 0; |
| 785 | #endif |
| 786 | |
| 787 | /* Allocate space and initialize. */ |
| 788 | Newc(1001, r, sizeof(regexp) + (unsigned)regsize * sizeof(regnode), |
| 789 | char, regexp); |
| 790 | if (r == NULL) |
| 791 | FAIL("regexp out of space"); |
| 792 | r->refcnt = 1; |
| 793 | r->prelen = xend - exp; |
| 794 | r->precomp = regprecomp; |
| 795 | r->subbeg = r->subbase = NULL; |
| 796 | rx = r; |
| 797 | |
| 798 | /* Second pass: emit code. */ |
| 799 | regparse = exp; |
| 800 | regxend = xend; |
| 801 | regnaughty = 0; |
| 802 | regnpar = 1; |
| 803 | regcode = r->program; |
| 804 | regc((U8)MAGIC, (char*) regcode++); |
| 805 | r->data = 0; |
| 806 | if (reg(0, &flags) == NULL) |
| 807 | return(NULL); |
| 808 | |
| 809 | /* Dig out information for optimizations. */ |
| 810 | pm->op_pmflags = regflags; |
| 811 | r->reganch = 0; |
| 812 | r->regstclass = NULL; |
| 813 | r->naughty = regnaughty >= 10; /* Probably an expensive pattern. */ |
| 814 | scan = r->program + 1; /* First BRANCH. */ |
| 815 | |
| 816 | /* XXXX To minimize changes to RE engine we always allocate |
| 817 | 3-units-long substrs field. */ |
| 818 | Newz(1004, r->substrs, 1, struct reg_substr_data); |
| 819 | |
| 820 | if (OP(scan) != BRANCH) { /* Only one top-level choice. */ |
| 821 | scan_data_t data; |
| 822 | I32 fake; |
| 823 | STRLEN longest_float_length, longest_fixed_length; |
| 824 | |
| 825 | StructCopy(&zero_scan_data, &data, scan_data_t); |
| 826 | first = scan; |
| 827 | /* Skip introductions and multiplicators >= 1. */ |
| 828 | while ((OP(first) == OPEN && (sawopen = 1)) || |
| 829 | (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) || |
| 830 | (OP(first) == PLUS) || |
| 831 | (OP(first) == MINMOD) || |
| 832 | (regkind[(U8)OP(first)] == CURLY && ARG1(first) > 0) ) { |
| 833 | if (OP(first) == PLUS) |
| 834 | sawplus = 1; |
| 835 | else |
| 836 | first += regarglen[(U8)OP(first)]; |
| 837 | first = NEXTOPER(first); |
| 838 | } |
| 839 | |
| 840 | /* Starting-point info. */ |
| 841 | again: |
| 842 | if (OP(first) == EXACT); /* Empty, get anchored substr later. */ |
| 843 | else if (strchr(simple+2,OP(first))) |
| 844 | r->regstclass = first; |
| 845 | else if (regkind[(U8)OP(first)] == BOUND || |
| 846 | regkind[(U8)OP(first)] == NBOUND) |
| 847 | r->regstclass = first; |
| 848 | else if (regkind[(U8)OP(first)] == BOL) { |
| 849 | r->reganch |= (OP(first) == MBOL ? ROPT_ANCH_MBOL: ROPT_ANCH_BOL); |
| 850 | first = NEXTOPER(first); |
| 851 | goto again; |
| 852 | } |
| 853 | else if (OP(first) == GPOS) { |
| 854 | r->reganch |= ROPT_ANCH_GPOS; |
| 855 | first = NEXTOPER(first); |
| 856 | goto again; |
| 857 | } |
| 858 | else if ((OP(first) == STAR && |
| 859 | regkind[(U8)OP(NEXTOPER(first))] == ANY) && |
| 860 | !(r->reganch & ROPT_ANCH) ) |
| 861 | { |
| 862 | /* turn .* into ^.* with an implied $*=1 */ |
| 863 | r->reganch |= ROPT_ANCH_BOL | ROPT_IMPLICIT; |
| 864 | first = NEXTOPER(first); |
| 865 | goto again; |
| 866 | } |
| 867 | if (sawplus && (!sawopen || !regsawback)) |
| 868 | r->reganch |= ROPT_SKIP; /* x+ must match 1st of run */ |
| 869 | |
| 870 | /* Scan is after the zeroth branch, first is atomic matcher. */ |
| 871 | DEBUG_r(PerlIO_printf(Perl_debug_log, "first at %d\n", |
| 872 | first - scan + 1)); |
| 873 | /* |
| 874 | * If there's something expensive in the r.e., find the |
| 875 | * longest literal string that must appear and make it the |
| 876 | * regmust. Resolve ties in favor of later strings, since |
| 877 | * the regstart check works with the beginning of the r.e. |
| 878 | * and avoiding duplication strengthens checking. Not a |
| 879 | * strong reason, but sufficient in the absence of others. |
| 880 | * [Now we resolve ties in favor of the earlier string if |
| 881 | * it happens that c_offset_min has been invalidated, since the |
| 882 | * earlier string may buy us something the later one won't.] |
| 883 | */ |
| 884 | minlen = 0; |
| 885 | |
| 886 | data.longest_fixed = newSVpv("",0); |
| 887 | data.longest_float = newSVpv("",0); |
| 888 | data.last_found = newSVpv("",0); |
| 889 | data.longest = &(data.longest_fixed); |
| 890 | first = scan; |
| 891 | |
| 892 | minlen = study_chunk(&first, &fake, scan + regsize, /* Up to end */ |
| 893 | &data, SCF_DO_SUBSTR); |
| 894 | if ( regnpar == 1 && data.longest == &(data.longest_fixed) |
| 895 | && data.last_start_min == 0 && data.last_end > 0 |
| 896 | && !seen_zerolen |
| 897 | && (!(regseen & REG_SEEN_GPOS) || (r->reganch & ROPT_ANCH_GPOS))) |
| 898 | r->reganch |= ROPT_CHECK_ALL; |
| 899 | scan_commit(&data); |
| 900 | SvREFCNT_dec(data.last_found); |
| 901 | |
| 902 | longest_float_length = SvCUR(data.longest_float); |
| 903 | if (longest_float_length |
| 904 | || (data.flags & SF_FL_BEFORE_EOL |
| 905 | && (!(data.flags & SF_FL_BEFORE_MEOL) |
| 906 | || (regflags & PMf_MULTILINE)))) { |
| 907 | if (SvCUR(data.longest_fixed) |
| 908 | && data.offset_fixed == data.offset_float_min) |
| 909 | goto remove; /* Like in (a)+. */ |
| 910 | |
| 911 | r->float_substr = data.longest_float; |
| 912 | r->float_min_offset = data.offset_float_min; |
| 913 | r->float_max_offset = data.offset_float_max; |
| 914 | fbm_compile(r->float_substr, 0); |
| 915 | BmUSEFUL(r->float_substr) = 100; |
| 916 | if (data.flags & SF_FL_BEFORE_EOL /* Cannot have SEOL and MULTI */ |
| 917 | && (!(data.flags & SF_FL_BEFORE_MEOL) |
| 918 | || (regflags & PMf_MULTILINE))) |
| 919 | SvTAIL_on(r->float_substr); |
| 920 | } else { |
| 921 | remove: |
| 922 | r->float_substr = Nullsv; |
| 923 | SvREFCNT_dec(data.longest_float); |
| 924 | longest_float_length = 0; |
| 925 | } |
| 926 | |
| 927 | longest_fixed_length = SvCUR(data.longest_fixed); |
| 928 | if (longest_fixed_length |
| 929 | || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */ |
| 930 | && (!(data.flags & SF_FIX_BEFORE_MEOL) |
| 931 | || (regflags & PMf_MULTILINE)))) { |
| 932 | r->anchored_substr = data.longest_fixed; |
| 933 | r->anchored_offset = data.offset_fixed; |
| 934 | fbm_compile(r->anchored_substr, 0); |
| 935 | BmUSEFUL(r->anchored_substr) = 100; |
| 936 | if (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */ |
| 937 | && (!(data.flags & SF_FIX_BEFORE_MEOL) |
| 938 | || (regflags & PMf_MULTILINE))) |
| 939 | SvTAIL_on(r->anchored_substr); |
| 940 | } else { |
| 941 | r->anchored_substr = Nullsv; |
| 942 | SvREFCNT_dec(data.longest_fixed); |
| 943 | longest_fixed_length = 0; |
| 944 | } |
| 945 | |
| 946 | /* A temporary algorithm prefers floated substr to fixed one to dig more info. */ |
| 947 | if (longest_fixed_length > longest_float_length) { |
| 948 | r->check_substr = r->anchored_substr; |
| 949 | r->check_offset_min = r->check_offset_max = r->anchored_offset; |
| 950 | if (r->reganch & ROPT_ANCH_SINGLE) |
| 951 | r->reganch |= ROPT_NOSCAN; |
| 952 | } else { |
| 953 | r->check_substr = r->float_substr; |
| 954 | r->check_offset_min = data.offset_float_min; |
| 955 | r->check_offset_max = data.offset_float_max; |
| 956 | } |
| 957 | } else { |
| 958 | /* Several toplevels. Best we can is to set minlen. */ |
| 959 | I32 fake; |
| 960 | |
| 961 | DEBUG_r(PerlIO_printf(Perl_debug_log, "\n")); |
| 962 | scan = r->program + 1; |
| 963 | minlen = study_chunk(&scan, &fake, scan + regsize, NULL, 0); |
| 964 | r->check_substr = r->anchored_substr = r->float_substr = Nullsv; |
| 965 | } |
| 966 | |
| 967 | r->nparens = regnpar - 1; |
| 968 | r->minlen = minlen; |
| 969 | if (regseen & REG_SEEN_GPOS) |
| 970 | r->reganch |= ROPT_GPOS_SEEN; |
| 971 | if (regseen & REG_SEEN_LOOKBEHIND) |
| 972 | r->reganch |= ROPT_LOOKBEHIND_SEEN; |
| 973 | Newz(1002, r->startp, regnpar, char*); |
| 974 | Newz(1002, r->endp, regnpar, char*); |
| 975 | DEBUG_r(regdump(r)); |
| 976 | return(r); |
| 977 | } |
| 978 | |
| 979 | /* |
| 980 | - reg - regular expression, i.e. main body or parenthesized thing |
| 981 | * |
| 982 | * Caller must absorb opening parenthesis. |
| 983 | * |
| 984 | * Combining parenthesis handling with the base level of regular expression |
| 985 | * is a trifle forced, but the need to tie the tails of the branches to what |
| 986 | * follows makes it hard to avoid. |
| 987 | */ |
| 988 | static regnode * |
| 989 | reg(I32 paren, I32 *flagp) |
| 990 | /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */ |
| 991 | { |
| 992 | register regnode *ret; /* Will be the head of the group. */ |
| 993 | register regnode *br; |
| 994 | register regnode *lastbr; |
| 995 | register regnode *ender = 0; |
| 996 | register I32 parno = 0; |
| 997 | I32 flags, oregflags = regflags, have_branch = 0, open = 0; |
| 998 | char c; |
| 999 | |
| 1000 | *flagp = HASWIDTH; /* Tentatively. */ |
| 1001 | |
| 1002 | /* Make an OPEN node, if parenthesized. */ |
| 1003 | if (paren) { |
| 1004 | if (*regparse == '?') { |
| 1005 | regparse++; |
| 1006 | paren = *regparse++; |
| 1007 | ret = NULL; /* For look-ahead/behind. */ |
| 1008 | switch (paren) { |
| 1009 | case '<': |
| 1010 | #ifndef REGALIGN_STRUCT |
| 1011 | FAIL("lookbehind non-implemented without REGALIGN_STRUCT"); |
| 1012 | #endif |
| 1013 | regseen |= REG_SEEN_LOOKBEHIND; |
| 1014 | if (*regparse == '!') |
| 1015 | paren = ','; |
| 1016 | if (*regparse != '=' && *regparse != '!') |
| 1017 | goto unknown; |
| 1018 | regparse++; |
| 1019 | case '=': |
| 1020 | case '!': |
| 1021 | seen_zerolen++; |
| 1022 | case ':': |
| 1023 | case '>': |
| 1024 | break; |
| 1025 | case '$': |
| 1026 | case '@': |
| 1027 | FAIL2("Sequence (?%c...) not implemented", (int)paren); |
| 1028 | break; |
| 1029 | case '#': |
| 1030 | while (*regparse && *regparse != ')') |
| 1031 | regparse++; |
| 1032 | if (*regparse != ')') |
| 1033 | FAIL("Sequence (?#... not terminated"); |
| 1034 | nextchar(); |
| 1035 | *flagp = TRYAGAIN; |
| 1036 | return NULL; |
| 1037 | case '{': |
| 1038 | { |
| 1039 | dTHR; |
| 1040 | I32 count = 1, n = 0; |
| 1041 | char c; |
| 1042 | char *s = regparse; |
| 1043 | SV *sv; |
| 1044 | OP_4tree *sop, *rop; |
| 1045 | |
| 1046 | seen_zerolen++; |
| 1047 | while (count && (c = *regparse)) { |
| 1048 | if (c == '\\' && regparse[1]) |
| 1049 | regparse++; |
| 1050 | else if (c == '{') |
| 1051 | count++; |
| 1052 | else if (c == '}') |
| 1053 | count--; |
| 1054 | regparse++; |
| 1055 | } |
| 1056 | if (*regparse != ')') |
| 1057 | FAIL("Sequence (?{...}) not terminated or not {}-balanced"); |
| 1058 | if (!SIZE_ONLY) { |
| 1059 | AV *av; |
| 1060 | |
| 1061 | if (regparse - 1 - s) |
| 1062 | sv = newSVpv(s, regparse - 1 - s); |
| 1063 | else |
| 1064 | sv = newSVpv("", 0); |
| 1065 | |
| 1066 | rop = sv_compile_2op(sv, &sop, "re", &av); |
| 1067 | |
| 1068 | n = add_data(3, "nso"); |
| 1069 | rx->data->data[n] = (void*)rop; |
| 1070 | rx->data->data[n+1] = (void*)av; |
| 1071 | rx->data->data[n+2] = (void*)sop; |
| 1072 | SvREFCNT_dec(sv); |
| 1073 | } else { /* First pass */ |
| 1074 | if (tainted) |
| 1075 | FAIL("Eval-group in insecure regular expression"); |
| 1076 | } |
| 1077 | |
| 1078 | nextchar(); |
| 1079 | return reganode(EVAL, n); |
| 1080 | } |
| 1081 | case '(': |
| 1082 | { |
| 1083 | if (regparse[0] == '?') { |
| 1084 | if (regparse[1] == '=' || regparse[1] == '!' |
| 1085 | || regparse[1] == '<' |
| 1086 | || regparse[1] == '{') { /* Lookahead or eval. */ |
| 1087 | I32 flag; |
| 1088 | |
| 1089 | ret = reg_node(LOGICAL); |
| 1090 | regtail(ret, reg(1, &flag)); |
| 1091 | goto insert_if; |
| 1092 | } |
| 1093 | } else if (regparse[0] >= '1' && regparse[0] <= '9' ) { |
| 1094 | parno = atoi(regparse++); |
| 1095 | |
| 1096 | while (isDIGIT(*regparse)) |
| 1097 | regparse++; |
| 1098 | ret = reganode(GROUPP, parno); |
| 1099 | if ((c = *nextchar()) != ')') |
| 1100 | FAIL2("Switch (?(number%c not recognized", c); |
| 1101 | insert_if: |
| 1102 | regtail(ret, reganode(IFTHEN, 0)); |
| 1103 | br = regbranch(&flags, 1); |
| 1104 | if (br == NULL) |
| 1105 | br = reganode(LONGJMP, 0); |
| 1106 | else |
| 1107 | regtail(br, reganode(LONGJMP, 0)); |
| 1108 | c = *nextchar(); |
| 1109 | if (c == '|') { |
| 1110 | lastbr = reganode(IFTHEN, 0); /* Fake one for optimizer. */ |
| 1111 | regbranch(&flags, 1); |
| 1112 | regtail(ret, lastbr); |
| 1113 | c = *nextchar(); |
| 1114 | } else |
| 1115 | lastbr = NULL; |
| 1116 | if (c != ')') |
| 1117 | FAIL("Switch (?(condition)... contains too many branches"); |
| 1118 | ender = reg_node(TAIL); |
| 1119 | regtail(br, ender); |
| 1120 | if (lastbr) { |
| 1121 | regtail(lastbr, ender); |
| 1122 | regtail(NEXTOPER(NEXTOPER(lastbr)), ender); |
| 1123 | } else |
| 1124 | regtail(ret, ender); |
| 1125 | return ret; |
| 1126 | } else { |
| 1127 | FAIL2("Unknown condition for (?(%.2s", regparse); |
| 1128 | } |
| 1129 | } |
| 1130 | case 0: |
| 1131 | FAIL("Sequence (? incomplete"); |
| 1132 | break; |
| 1133 | default: |
| 1134 | --regparse; |
| 1135 | while (*regparse && strchr("iogcmsx", *regparse)) |
| 1136 | pmflag(®flags, *regparse++); |
| 1137 | unknown: |
| 1138 | if (*regparse != ')') |
| 1139 | FAIL2("Sequence (?%c...) not recognized", *regparse); |
| 1140 | nextchar(); |
| 1141 | *flagp = TRYAGAIN; |
| 1142 | return NULL; |
| 1143 | } |
| 1144 | } |
| 1145 | else { |
| 1146 | parno = regnpar; |
| 1147 | regnpar++; |
| 1148 | ret = reganode(OPEN, parno); |
| 1149 | open = 1; |
| 1150 | } |
| 1151 | } else |
| 1152 | ret = NULL; |
| 1153 | |
| 1154 | /* Pick up the branches, linking them together. */ |
| 1155 | br = regbranch(&flags, 1); |
| 1156 | if (br == NULL) |
| 1157 | return(NULL); |
| 1158 | if (*regparse == '|') { |
| 1159 | if (!SIZE_ONLY && extralen) { |
| 1160 | reginsert(BRANCHJ, br); |
| 1161 | } else |
| 1162 | reginsert(BRANCH, br); |
| 1163 | have_branch = 1; |
| 1164 | if (SIZE_ONLY) |
| 1165 | extralen += 1; /* For BRANCHJ-BRANCH. */ |
| 1166 | } else if (paren == ':') { |
| 1167 | *flagp |= flags&SIMPLE; |
| 1168 | } |
| 1169 | if (open) { /* Starts with OPEN. */ |
| 1170 | regtail(ret, br); /* OPEN -> first. */ |
| 1171 | } else if (paren != '?') /* Not Conditional */ |
| 1172 | ret = br; |
| 1173 | if (!(flags&HASWIDTH)) |
| 1174 | *flagp &= ~HASWIDTH; |
| 1175 | *flagp |= flags&SPSTART; |
| 1176 | lastbr = br; |
| 1177 | while (*regparse == '|') { |
| 1178 | if (!SIZE_ONLY && extralen) { |
| 1179 | ender = reganode(LONGJMP,0); |
| 1180 | regtail(NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */ |
| 1181 | } |
| 1182 | if (SIZE_ONLY) |
| 1183 | extralen += 2; /* Account for LONGJMP. */ |
| 1184 | nextchar(); |
| 1185 | br = regbranch(&flags, 0); |
| 1186 | if (br == NULL) |
| 1187 | return(NULL); |
| 1188 | regtail(lastbr, br); /* BRANCH -> BRANCH. */ |
| 1189 | lastbr = br; |
| 1190 | if (!(flags&HASWIDTH)) |
| 1191 | *flagp &= ~HASWIDTH; |
| 1192 | *flagp |= flags&SPSTART; |
| 1193 | } |
| 1194 | |
| 1195 | if (have_branch || paren != ':') { |
| 1196 | /* Make a closing node, and hook it on the end. */ |
| 1197 | switch (paren) { |
| 1198 | case ':': |
| 1199 | ender = reg_node(TAIL); |
| 1200 | break; |
| 1201 | case 1: |
| 1202 | ender = reganode(CLOSE, parno); |
| 1203 | break; |
| 1204 | case '<': |
| 1205 | case '>': |
| 1206 | case ',': |
| 1207 | case '=': |
| 1208 | case '!': |
| 1209 | ender = reg_node(SUCCEED); |
| 1210 | *flagp &= ~HASWIDTH; |
| 1211 | break; |
| 1212 | case 0: |
| 1213 | ender = reg_node(END); |
| 1214 | break; |
| 1215 | } |
| 1216 | regtail(lastbr, ender); |
| 1217 | |
| 1218 | if (have_branch) { |
| 1219 | /* Hook the tails of the branches to the closing node. */ |
| 1220 | for (br = ret; br != NULL; br = regnext(br)) { |
| 1221 | regoptail(br, ender); |
| 1222 | } |
| 1223 | } |
| 1224 | } |
| 1225 | |
| 1226 | { |
| 1227 | char *p; |
| 1228 | static char parens[] = "=!<,>"; |
| 1229 | |
| 1230 | if (paren && (p = strchr(parens, paren))) { |
| 1231 | int node = ((p - parens) % 2) ? UNLESSM : IFMATCH; |
| 1232 | int flag = (p - parens) > 1; |
| 1233 | |
| 1234 | if (paren == '>') |
| 1235 | node = SUSPEND, flag = 0; |
| 1236 | reginsert(node,ret); |
| 1237 | #ifdef REGALIGN_STRUCT |
| 1238 | ret->flags = flag; |
| 1239 | #endif |
| 1240 | regtail(ret, reg_node(TAIL)); |
| 1241 | } |
| 1242 | } |
| 1243 | |
| 1244 | /* Check for proper termination. */ |
| 1245 | if (paren && (regparse >= regxend || *nextchar() != ')')) { |
| 1246 | FAIL("unmatched () in regexp"); |
| 1247 | } else if (!paren && regparse < regxend) { |
| 1248 | if (*regparse == ')') { |
| 1249 | FAIL("unmatched () in regexp"); |
| 1250 | } else |
| 1251 | FAIL("junk on end of regexp"); /* "Can't happen". */ |
| 1252 | /* NOTREACHED */ |
| 1253 | } |
| 1254 | if (paren != 0) { |
| 1255 | regflags = oregflags; |
| 1256 | } |
| 1257 | |
| 1258 | return(ret); |
| 1259 | } |
| 1260 | |
| 1261 | /* |
| 1262 | - regbranch - one alternative of an | operator |
| 1263 | * |
| 1264 | * Implements the concatenation operator. |
| 1265 | */ |
| 1266 | static regnode * |
| 1267 | regbranch(I32 *flagp, I32 first) |
| 1268 | { |
| 1269 | register regnode *ret; |
| 1270 | register regnode *chain = NULL; |
| 1271 | register regnode *latest; |
| 1272 | I32 flags = 0, c = 0; |
| 1273 | |
| 1274 | if (first) |
| 1275 | ret = NULL; |
| 1276 | else { |
| 1277 | if (!SIZE_ONLY && extralen) |
| 1278 | ret = reganode(BRANCHJ,0); |
| 1279 | else |
| 1280 | ret = reg_node(BRANCH); |
| 1281 | } |
| 1282 | |
| 1283 | if (!first && SIZE_ONLY) |
| 1284 | extralen += 1; /* BRANCHJ */ |
| 1285 | |
| 1286 | *flagp = WORST; /* Tentatively. */ |
| 1287 | |
| 1288 | regparse--; |
| 1289 | nextchar(); |
| 1290 | while (regparse < regxend && *regparse != '|' && *regparse != ')') { |
| 1291 | flags &= ~TRYAGAIN; |
| 1292 | latest = regpiece(&flags); |
| 1293 | if (latest == NULL) { |
| 1294 | if (flags & TRYAGAIN) |
| 1295 | continue; |
| 1296 | return(NULL); |
| 1297 | } else if (ret == NULL) |
| 1298 | ret = latest; |
| 1299 | *flagp |= flags&HASWIDTH; |
| 1300 | if (chain == NULL) /* First piece. */ |
| 1301 | *flagp |= flags&SPSTART; |
| 1302 | else { |
| 1303 | regnaughty++; |
| 1304 | regtail(chain, latest); |
| 1305 | } |
| 1306 | chain = latest; |
| 1307 | c++; |
| 1308 | } |
| 1309 | if (chain == NULL) { /* Loop ran zero times. */ |
| 1310 | chain = reg_node(NOTHING); |
| 1311 | if (ret == NULL) |
| 1312 | ret = chain; |
| 1313 | } |
| 1314 | if (c == 1) { |
| 1315 | *flagp |= flags&SIMPLE; |
| 1316 | } |
| 1317 | |
| 1318 | return(ret); |
| 1319 | } |
| 1320 | |
| 1321 | /* |
| 1322 | - regpiece - something followed by possible [*+?] |
| 1323 | * |
| 1324 | * Note that the branching code sequences used for ? and the general cases |
| 1325 | * of * and + are somewhat optimized: they use the same NOTHING node as |
| 1326 | * both the endmarker for their branch list and the body of the last branch. |
| 1327 | * It might seem that this node could be dispensed with entirely, but the |
| 1328 | * endmarker role is not redundant. |
| 1329 | */ |
| 1330 | static regnode * |
| 1331 | regpiece(I32 *flagp) |
| 1332 | { |
| 1333 | register regnode *ret; |
| 1334 | register char op; |
| 1335 | register char *next; |
| 1336 | I32 flags; |
| 1337 | char *origparse = regparse; |
| 1338 | char *maxpos; |
| 1339 | I32 min; |
| 1340 | I32 max = REG_INFTY; |
| 1341 | |
| 1342 | ret = regatom(&flags); |
| 1343 | if (ret == NULL) { |
| 1344 | if (flags & TRYAGAIN) |
| 1345 | *flagp |= TRYAGAIN; |
| 1346 | return(NULL); |
| 1347 | } |
| 1348 | |
| 1349 | op = *regparse; |
| 1350 | |
| 1351 | if (op == '{' && regcurly(regparse)) { |
| 1352 | next = regparse + 1; |
| 1353 | maxpos = Nullch; |
| 1354 | while (isDIGIT(*next) || *next == ',') { |
| 1355 | if (*next == ',') { |
| 1356 | if (maxpos) |
| 1357 | break; |
| 1358 | else |
| 1359 | maxpos = next; |
| 1360 | } |
| 1361 | next++; |
| 1362 | } |
| 1363 | if (*next == '}') { /* got one */ |
| 1364 | if (!maxpos) |
| 1365 | maxpos = next; |
| 1366 | regparse++; |
| 1367 | min = atoi(regparse); |
| 1368 | if (*maxpos == ',') |
| 1369 | maxpos++; |
| 1370 | else |
| 1371 | maxpos = regparse; |
| 1372 | max = atoi(maxpos); |
| 1373 | if (!max && *maxpos != '0') |
| 1374 | max = REG_INFTY; /* meaning "infinity" */ |
| 1375 | else if (max >= REG_INFTY) |
| 1376 | FAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1); |
| 1377 | regparse = next; |
| 1378 | nextchar(); |
| 1379 | |
| 1380 | do_curly: |
| 1381 | if ((flags&SIMPLE)) { |
| 1382 | regnaughty += 2 + regnaughty / 2; |
| 1383 | reginsert(CURLY, ret); |
| 1384 | } |
| 1385 | else { |
| 1386 | regnaughty += 4 + regnaughty; /* compound interest */ |
| 1387 | regtail(ret, reg_node(WHILEM)); |
| 1388 | if (!SIZE_ONLY && extralen) { |
| 1389 | reginsert(LONGJMP,ret); |
| 1390 | reginsert(NOTHING,ret); |
| 1391 | NEXT_OFF(ret) = 3; /* Go over LONGJMP. */ |
| 1392 | } |
| 1393 | reginsert(CURLYX,ret); |
| 1394 | if (!SIZE_ONLY && extralen) |
| 1395 | NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */ |
| 1396 | regtail(ret, reg_node(NOTHING)); |
| 1397 | if (SIZE_ONLY) |
| 1398 | extralen += 3; |
| 1399 | } |
| 1400 | #ifdef REGALIGN_STRUCT |
| 1401 | ret->flags = 0; |
| 1402 | #endif |
| 1403 | |
| 1404 | if (min > 0) |
| 1405 | *flagp = (WORST|HASWIDTH); |
| 1406 | if (max && max < min) |
| 1407 | FAIL("Can't do {n,m} with n > m"); |
| 1408 | if (!SIZE_ONLY) { |
| 1409 | ARG1_SET(ret, min); |
| 1410 | ARG2_SET(ret, max); |
| 1411 | } |
| 1412 | |
| 1413 | goto nest_check; |
| 1414 | } |
| 1415 | } |
| 1416 | |
| 1417 | if (!ISMULT1(op)) { |
| 1418 | *flagp = flags; |
| 1419 | return(ret); |
| 1420 | } |
| 1421 | |
| 1422 | #if 0 /* Now runtime fix should be reliable. */ |
| 1423 | if (!(flags&HASWIDTH) && op != '?') |
| 1424 | FAIL("regexp *+ operand could be empty"); |
| 1425 | #endif |
| 1426 | |
| 1427 | nextchar(); |
| 1428 | |
| 1429 | *flagp = (op != '+') ? (WORST|SPSTART) : (WORST|HASWIDTH); |
| 1430 | |
| 1431 | if (op == '*' && (flags&SIMPLE)) { |
| 1432 | reginsert(STAR, ret); |
| 1433 | #ifdef REGALIGN_STRUCT |
| 1434 | ret->flags = 0; |
| 1435 | #endif |
| 1436 | regnaughty += 4; |
| 1437 | } |
| 1438 | else if (op == '*') { |
| 1439 | min = 0; |
| 1440 | goto do_curly; |
| 1441 | } else if (op == '+' && (flags&SIMPLE)) { |
| 1442 | reginsert(PLUS, ret); |
| 1443 | #ifdef REGALIGN_STRUCT |
| 1444 | ret->flags = 0; |
| 1445 | #endif |
| 1446 | regnaughty += 3; |
| 1447 | } |
| 1448 | else if (op == '+') { |
| 1449 | min = 1; |
| 1450 | goto do_curly; |
| 1451 | } else if (op == '?') { |
| 1452 | min = 0; max = 1; |
| 1453 | goto do_curly; |
| 1454 | } |
| 1455 | nest_check: |
| 1456 | if (dowarn && !SIZE_ONLY && !(flags&HASWIDTH) && max > 10000) { |
| 1457 | warn("%.*s matches null string many times", |
| 1458 | regparse - origparse, origparse); |
| 1459 | } |
| 1460 | |
| 1461 | if (*regparse == '?') { |
| 1462 | nextchar(); |
| 1463 | reginsert(MINMOD, ret); |
| 1464 | #ifdef REGALIGN |
| 1465 | regtail(ret, ret + NODE_STEP_REGNODE); |
| 1466 | #else |
| 1467 | regtail(ret, ret + 3); |
| 1468 | #endif |
| 1469 | } |
| 1470 | if (ISMULT2(regparse)) |
| 1471 | FAIL("nested *?+ in regexp"); |
| 1472 | |
| 1473 | return(ret); |
| 1474 | } |
| 1475 | |
| 1476 | /* |
| 1477 | - regatom - the lowest level |
| 1478 | * |
| 1479 | * Optimization: gobbles an entire sequence of ordinary characters so that |
| 1480 | * it can turn them into a single node, which is smaller to store and |
| 1481 | * faster to run. Backslashed characters are exceptions, each becoming a |
| 1482 | * separate node; the code is simpler that way and it's not worth fixing. |
| 1483 | * |
| 1484 | * [Yes, it is worth fixing, some scripts can run twice the speed.] |
| 1485 | */ |
| 1486 | static regnode * |
| 1487 | regatom(I32 *flagp) |
| 1488 | { |
| 1489 | register regnode *ret = 0; |
| 1490 | I32 flags; |
| 1491 | |
| 1492 | *flagp = WORST; /* Tentatively. */ |
| 1493 | |
| 1494 | tryagain: |
| 1495 | switch (*regparse) { |
| 1496 | case '^': |
| 1497 | seen_zerolen++; |
| 1498 | nextchar(); |
| 1499 | if (regflags & PMf_MULTILINE) |
| 1500 | ret = reg_node(MBOL); |
| 1501 | else if (regflags & PMf_SINGLELINE) |
| 1502 | ret = reg_node(SBOL); |
| 1503 | else |
| 1504 | ret = reg_node(BOL); |
| 1505 | break; |
| 1506 | case '$': |
| 1507 | if (regparse[1]) |
| 1508 | seen_zerolen++; |
| 1509 | nextchar(); |
| 1510 | if (regflags & PMf_MULTILINE) |
| 1511 | ret = reg_node(MEOL); |
| 1512 | else if (regflags & PMf_SINGLELINE) |
| 1513 | ret = reg_node(SEOL); |
| 1514 | else |
| 1515 | ret = reg_node(EOL); |
| 1516 | break; |
| 1517 | case '.': |
| 1518 | nextchar(); |
| 1519 | if (regflags & PMf_SINGLELINE) |
| 1520 | ret = reg_node(SANY); |
| 1521 | else |
| 1522 | ret = reg_node(ANY); |
| 1523 | regnaughty++; |
| 1524 | *flagp |= HASWIDTH|SIMPLE; |
| 1525 | break; |
| 1526 | case '[': |
| 1527 | regparse++; |
| 1528 | ret = regclass(); |
| 1529 | *flagp |= HASWIDTH|SIMPLE; |
| 1530 | break; |
| 1531 | case '(': |
| 1532 | nextchar(); |
| 1533 | ret = reg(1, &flags); |
| 1534 | if (ret == NULL) { |
| 1535 | if (flags & TRYAGAIN) |
| 1536 | goto tryagain; |
| 1537 | return(NULL); |
| 1538 | } |
| 1539 | *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE); |
| 1540 | break; |
| 1541 | case '|': |
| 1542 | case ')': |
| 1543 | if (flags & TRYAGAIN) { |
| 1544 | *flagp |= TRYAGAIN; |
| 1545 | return NULL; |
| 1546 | } |
| 1547 | FAIL2("internal urp in regexp at /%s/", regparse); |
| 1548 | /* Supposed to be caught earlier. */ |
| 1549 | break; |
| 1550 | case '{': |
| 1551 | if (!regcurly(regparse)) { |
| 1552 | regparse++; |
| 1553 | goto defchar; |
| 1554 | } |
| 1555 | /* FALL THROUGH */ |
| 1556 | case '?': |
| 1557 | case '+': |
| 1558 | case '*': |
| 1559 | FAIL("?+*{} follows nothing in regexp"); |
| 1560 | break; |
| 1561 | case '\\': |
| 1562 | switch (*++regparse) { |
| 1563 | case 'A': |
| 1564 | seen_zerolen++; |
| 1565 | ret = reg_node(SBOL); |
| 1566 | *flagp |= SIMPLE; |
| 1567 | nextchar(); |
| 1568 | break; |
| 1569 | case 'G': |
| 1570 | ret = reg_node(GPOS); |
| 1571 | regseen |= REG_SEEN_GPOS; |
| 1572 | *flagp |= SIMPLE; |
| 1573 | nextchar(); |
| 1574 | break; |
| 1575 | case 'Z': |
| 1576 | ret = reg_node(SEOL); |
| 1577 | *flagp |= SIMPLE; |
| 1578 | nextchar(); |
| 1579 | break; |
| 1580 | case 'w': |
| 1581 | ret = reg_node((regflags & PMf_LOCALE) ? ALNUML : ALNUM); |
| 1582 | *flagp |= HASWIDTH|SIMPLE; |
| 1583 | nextchar(); |
| 1584 | break; |
| 1585 | case 'W': |
| 1586 | ret = reg_node((regflags & PMf_LOCALE) ? NALNUML : NALNUM); |
| 1587 | *flagp |= HASWIDTH|SIMPLE; |
| 1588 | nextchar(); |
| 1589 | break; |
| 1590 | case 'b': |
| 1591 | seen_zerolen++; |
| 1592 | ret = reg_node((regflags & PMf_LOCALE) ? BOUNDL : BOUND); |
| 1593 | *flagp |= SIMPLE; |
| 1594 | nextchar(); |
| 1595 | break; |
| 1596 | case 'B': |
| 1597 | seen_zerolen++; |
| 1598 | ret = reg_node((regflags & PMf_LOCALE) ? NBOUNDL : NBOUND); |
| 1599 | *flagp |= SIMPLE; |
| 1600 | nextchar(); |
| 1601 | break; |
| 1602 | case 's': |
| 1603 | ret = reg_node((regflags & PMf_LOCALE) ? SPACEL : SPACE); |
| 1604 | *flagp |= HASWIDTH|SIMPLE; |
| 1605 | nextchar(); |
| 1606 | break; |
| 1607 | case 'S': |
| 1608 | ret = reg_node((regflags & PMf_LOCALE) ? NSPACEL : NSPACE); |
| 1609 | *flagp |= HASWIDTH|SIMPLE; |
| 1610 | nextchar(); |
| 1611 | break; |
| 1612 | case 'd': |
| 1613 | ret = reg_node(DIGIT); |
| 1614 | *flagp |= HASWIDTH|SIMPLE; |
| 1615 | nextchar(); |
| 1616 | break; |
| 1617 | case 'D': |
| 1618 | ret = reg_node(NDIGIT); |
| 1619 | *flagp |= HASWIDTH|SIMPLE; |
| 1620 | nextchar(); |
| 1621 | break; |
| 1622 | case 'n': |
| 1623 | case 'r': |
| 1624 | case 't': |
| 1625 | case 'f': |
| 1626 | case 'e': |
| 1627 | case 'a': |
| 1628 | case 'x': |
| 1629 | case 'c': |
| 1630 | case '0': |
| 1631 | goto defchar; |
| 1632 | case '1': case '2': case '3': case '4': |
| 1633 | case '5': case '6': case '7': case '8': case '9': |
| 1634 | { |
| 1635 | I32 num = atoi(regparse); |
| 1636 | |
| 1637 | if (num > 9 && num >= regnpar) |
| 1638 | goto defchar; |
| 1639 | else { |
| 1640 | regsawback = 1; |
| 1641 | ret = reganode((regflags & PMf_FOLD) |
| 1642 | ? ((regflags & PMf_LOCALE) ? REFFL : REFF) |
| 1643 | : REF, num); |
| 1644 | *flagp |= HASWIDTH; |
| 1645 | while (isDIGIT(*regparse)) |
| 1646 | regparse++; |
| 1647 | regparse--; |
| 1648 | nextchar(); |
| 1649 | } |
| 1650 | } |
| 1651 | break; |
| 1652 | case '\0': |
| 1653 | if (regparse >= regxend) |
| 1654 | FAIL("trailing \\ in regexp"); |
| 1655 | /* FALL THROUGH */ |
| 1656 | default: |
| 1657 | goto defchar; |
| 1658 | } |
| 1659 | break; |
| 1660 | |
| 1661 | case '#': |
| 1662 | if (regflags & PMf_EXTENDED) { |
| 1663 | while (regparse < regxend && *regparse != '\n') regparse++; |
| 1664 | if (regparse < regxend) |
| 1665 | goto tryagain; |
| 1666 | } |
| 1667 | /* FALL THROUGH */ |
| 1668 | |
| 1669 | default: { |
| 1670 | register I32 len; |
| 1671 | register U8 ender; |
| 1672 | register char *p; |
| 1673 | char *oldp, *s; |
| 1674 | I32 numlen; |
| 1675 | |
| 1676 | regparse++; |
| 1677 | |
| 1678 | defchar: |
| 1679 | ret = reg_node((regflags & PMf_FOLD) |
| 1680 | ? ((regflags & PMf_LOCALE) ? EXACTFL : EXACTF) |
| 1681 | : EXACT); |
| 1682 | s = (char *) OPERAND(ret); |
| 1683 | regc(0, s++); /* save spot for len */ |
| 1684 | for (len = 0, p = regparse - 1; |
| 1685 | len < 127 && p < regxend; |
| 1686 | len++) |
| 1687 | { |
| 1688 | oldp = p; |
| 1689 | |
| 1690 | if (regflags & PMf_EXTENDED) |
| 1691 | p = regwhite(p, regxend); |
| 1692 | switch (*p) { |
| 1693 | case '^': |
| 1694 | case '$': |
| 1695 | case '.': |
| 1696 | case '[': |
| 1697 | case '(': |
| 1698 | case ')': |
| 1699 | case '|': |
| 1700 | goto loopdone; |
| 1701 | case '\\': |
| 1702 | switch (*++p) { |
| 1703 | case 'A': |
| 1704 | case 'G': |
| 1705 | case 'Z': |
| 1706 | case 'w': |
| 1707 | case 'W': |
| 1708 | case 'b': |
| 1709 | case 'B': |
| 1710 | case 's': |
| 1711 | case 'S': |
| 1712 | case 'd': |
| 1713 | case 'D': |
| 1714 | --p; |
| 1715 | goto loopdone; |
| 1716 | case 'n': |
| 1717 | ender = '\n'; |
| 1718 | p++; |
| 1719 | break; |
| 1720 | case 'r': |
| 1721 | ender = '\r'; |
| 1722 | p++; |
| 1723 | break; |
| 1724 | case 't': |
| 1725 | ender = '\t'; |
| 1726 | p++; |
| 1727 | break; |
| 1728 | case 'f': |
| 1729 | ender = '\f'; |
| 1730 | p++; |
| 1731 | break; |
| 1732 | case 'e': |
| 1733 | ender = '\033'; |
| 1734 | p++; |
| 1735 | break; |
| 1736 | case 'a': |
| 1737 | ender = '\007'; |
| 1738 | p++; |
| 1739 | break; |
| 1740 | case 'x': |
| 1741 | ender = scan_hex(++p, 2, &numlen); |
| 1742 | p += numlen; |
| 1743 | break; |
| 1744 | case 'c': |
| 1745 | p++; |
| 1746 | ender = UCHARAT(p++); |
| 1747 | ender = toCTRL(ender); |
| 1748 | break; |
| 1749 | case '0': case '1': case '2': case '3':case '4': |
| 1750 | case '5': case '6': case '7': case '8':case '9': |
| 1751 | if (*p == '0' || |
| 1752 | (isDIGIT(p[1]) && atoi(p) >= regnpar) ) { |
| 1753 | ender = scan_oct(p, 3, &numlen); |
| 1754 | p += numlen; |
| 1755 | } |
| 1756 | else { |
| 1757 | --p; |
| 1758 | goto loopdone; |
| 1759 | } |
| 1760 | break; |
| 1761 | case '\0': |
| 1762 | if (p >= regxend) |
| 1763 | FAIL("trailing \\ in regexp"); |
| 1764 | /* FALL THROUGH */ |
| 1765 | default: |
| 1766 | ender = *p++; |
| 1767 | break; |
| 1768 | } |
| 1769 | break; |
| 1770 | default: |
| 1771 | ender = *p++; |
| 1772 | break; |
| 1773 | } |
| 1774 | if (regflags & PMf_EXTENDED) |
| 1775 | p = regwhite(p, regxend); |
| 1776 | if (ISMULT2(p)) { /* Back off on ?+*. */ |
| 1777 | if (len) |
| 1778 | p = oldp; |
| 1779 | else { |
| 1780 | len++; |
| 1781 | regc(ender, s++); |
| 1782 | } |
| 1783 | break; |
| 1784 | } |
| 1785 | regc(ender, s++); |
| 1786 | } |
| 1787 | loopdone: |
| 1788 | regparse = p - 1; |
| 1789 | nextchar(); |
| 1790 | if (len < 0) |
| 1791 | FAIL("internal disaster in regexp"); |
| 1792 | if (len > 0) |
| 1793 | *flagp |= HASWIDTH; |
| 1794 | if (len == 1) |
| 1795 | *flagp |= SIMPLE; |
| 1796 | if (!SIZE_ONLY) |
| 1797 | *OPERAND(ret) = len; |
| 1798 | regc('\0', s++); |
| 1799 | if (SIZE_ONLY) { |
| 1800 | #ifdef REGALIGN_STRUCT |
| 1801 | regsize += (len + 2 + sizeof(regnode) - 1) / sizeof(regnode); |
| 1802 | #endif |
| 1803 | } else { |
| 1804 | regcode += (len + 2 + sizeof(regnode) - 1) / sizeof(regnode); |
| 1805 | } |
| 1806 | } |
| 1807 | break; |
| 1808 | } |
| 1809 | |
| 1810 | return(ret); |
| 1811 | } |
| 1812 | |
| 1813 | static char * |
| 1814 | regwhite(char *p, char *e) |
| 1815 | { |
| 1816 | while (p < e) { |
| 1817 | if (isSPACE(*p)) |
| 1818 | ++p; |
| 1819 | else if (*p == '#') { |
| 1820 | do { |
| 1821 | p++; |
| 1822 | } while (p < e && *p != '\n'); |
| 1823 | } |
| 1824 | else |
| 1825 | break; |
| 1826 | } |
| 1827 | return p; |
| 1828 | } |
| 1829 | |
| 1830 | static void |
| 1831 | regset(char *opnd, register I32 c) |
| 1832 | { |
| 1833 | if (SIZE_ONLY) |
| 1834 | return; |
| 1835 | c &= 0xFF; |
| 1836 | opnd[1 + (c >> 3)] |= (1 << (c & 7)); |
| 1837 | } |
| 1838 | |
| 1839 | static regnode * |
| 1840 | regclass(void) |
| 1841 | { |
| 1842 | register char *opnd, *s; |
| 1843 | register I32 Class; |
| 1844 | register I32 lastclass = 1234; |
| 1845 | register I32 range = 0; |
| 1846 | register regnode *ret; |
| 1847 | register I32 def; |
| 1848 | I32 numlen; |
| 1849 | |
| 1850 | s = opnd = (char *) OPERAND(regcode); |
| 1851 | ret = reg_node(ANYOF); |
| 1852 | for (Class = 0; Class < 33; Class++) |
| 1853 | regc(0, s++); |
| 1854 | if (*regparse == '^') { /* Complement of range. */ |
| 1855 | regnaughty++; |
| 1856 | regparse++; |
| 1857 | if (!SIZE_ONLY) |
| 1858 | *opnd |= ANYOF_INVERT; |
| 1859 | } |
| 1860 | if (!SIZE_ONLY) { |
| 1861 | regcode += ANY_SKIP; |
| 1862 | if (regflags & PMf_FOLD) |
| 1863 | *opnd |= ANYOF_FOLD; |
| 1864 | if (regflags & PMf_LOCALE) |
| 1865 | *opnd |= ANYOF_LOCALE; |
| 1866 | } else { |
| 1867 | regsize += ANY_SKIP; |
| 1868 | } |
| 1869 | if (*regparse == ']' || *regparse == '-') |
| 1870 | goto skipcond; /* allow 1st char to be ] or - */ |
| 1871 | while (regparse < regxend && *regparse != ']') { |
| 1872 | skipcond: |
| 1873 | Class = UCHARAT(regparse++); |
| 1874 | if (Class == '[' && regparse + 1 < regxend && |
| 1875 | /* I smell either [: or [= or [. -- POSIX has been here, right? */ |
| 1876 | (*regparse == ':' || *regparse == '=' || *regparse == '.')) { |
| 1877 | char posixccc = *regparse; |
| 1878 | char* posixccs = regparse++; |
| 1879 | |
| 1880 | while (regparse < regxend && *regparse != posixccc) |
| 1881 | regparse++; |
| 1882 | if (regparse == regxend) |
| 1883 | /* Grandfather lone [:, [=, [. */ |
| 1884 | regparse = posixccs; |
| 1885 | else { |
| 1886 | regparse++; /* skip over the posixccc */ |
| 1887 | if (*regparse == ']') { |
| 1888 | /* Not Implemented Yet. |
| 1889 | * (POSIX Extended Character Classes, that is) |
| 1890 | * The text between e.g. [: and :] would start |
| 1891 | * at posixccs + 1 and stop at regparse - 2. */ |
| 1892 | if (dowarn && !SIZE_ONLY) |
| 1893 | warn("Character class syntax [%c %c] is reserved for future extensions", posixccc, posixccc); |
| 1894 | regparse++; /* skip over the ending ] */ |
| 1895 | } |
| 1896 | } |
| 1897 | } |
| 1898 | if (Class == '\\') { |
| 1899 | Class = UCHARAT(regparse++); |
| 1900 | switch (Class) { |
| 1901 | case 'w': |
| 1902 | if (regflags & PMf_LOCALE) { |
| 1903 | if (!SIZE_ONLY) |
| 1904 | *opnd |= ANYOF_ALNUML; |
| 1905 | } |
| 1906 | else { |
| 1907 | for (Class = 0; Class < 256; Class++) |
| 1908 | if (isALNUM(Class)) |
| 1909 | regset(opnd, Class); |
| 1910 | } |
| 1911 | lastclass = 1234; |
| 1912 | continue; |
| 1913 | case 'W': |
| 1914 | if (regflags & PMf_LOCALE) { |
| 1915 | if (!SIZE_ONLY) |
| 1916 | *opnd |= ANYOF_NALNUML; |
| 1917 | } |
| 1918 | else { |
| 1919 | for (Class = 0; Class < 256; Class++) |
| 1920 | if (!isALNUM(Class)) |
| 1921 | regset(opnd, Class); |
| 1922 | } |
| 1923 | lastclass = 1234; |
| 1924 | continue; |
| 1925 | case 's': |
| 1926 | if (regflags & PMf_LOCALE) { |
| 1927 | if (!SIZE_ONLY) |
| 1928 | *opnd |= ANYOF_SPACEL; |
| 1929 | } |
| 1930 | else { |
| 1931 | for (Class = 0; Class < 256; Class++) |
| 1932 | if (isSPACE(Class)) |
| 1933 | regset(opnd, Class); |
| 1934 | } |
| 1935 | lastclass = 1234; |
| 1936 | continue; |
| 1937 | case 'S': |
| 1938 | if (regflags & PMf_LOCALE) { |
| 1939 | if (!SIZE_ONLY) |
| 1940 | *opnd |= ANYOF_NSPACEL; |
| 1941 | } |
| 1942 | else { |
| 1943 | for (Class = 0; Class < 256; Class++) |
| 1944 | if (!isSPACE(Class)) |
| 1945 | regset(opnd, Class); |
| 1946 | } |
| 1947 | lastclass = 1234; |
| 1948 | continue; |
| 1949 | case 'd': |
| 1950 | for (Class = '0'; Class <= '9'; Class++) |
| 1951 | regset(opnd, Class); |
| 1952 | lastclass = 1234; |
| 1953 | continue; |
| 1954 | case 'D': |
| 1955 | for (Class = 0; Class < '0'; Class++) |
| 1956 | regset(opnd, Class); |
| 1957 | for (Class = '9' + 1; Class < 256; Class++) |
| 1958 | regset(opnd, Class); |
| 1959 | lastclass = 1234; |
| 1960 | continue; |
| 1961 | case 'n': |
| 1962 | Class = '\n'; |
| 1963 | break; |
| 1964 | case 'r': |
| 1965 | Class = '\r'; |
| 1966 | break; |
| 1967 | case 't': |
| 1968 | Class = '\t'; |
| 1969 | break; |
| 1970 | case 'f': |
| 1971 | Class = '\f'; |
| 1972 | break; |
| 1973 | case 'b': |
| 1974 | Class = '\b'; |
| 1975 | break; |
| 1976 | case 'e': |
| 1977 | Class = '\033'; |
| 1978 | break; |
| 1979 | case 'a': |
| 1980 | Class = '\007'; |
| 1981 | break; |
| 1982 | case 'x': |
| 1983 | Class = scan_hex(regparse, 2, &numlen); |
| 1984 | regparse += numlen; |
| 1985 | break; |
| 1986 | case 'c': |
| 1987 | Class = UCHARAT(regparse++); |
| 1988 | Class = toCTRL(Class); |
| 1989 | break; |
| 1990 | case '0': case '1': case '2': case '3': case '4': |
| 1991 | case '5': case '6': case '7': case '8': case '9': |
| 1992 | Class = scan_oct(--regparse, 3, &numlen); |
| 1993 | regparse += numlen; |
| 1994 | break; |
| 1995 | } |
| 1996 | } |
| 1997 | if (range) { |
| 1998 | if (lastclass > Class) |
| 1999 | FAIL("invalid [] range in regexp"); |
| 2000 | range = 0; |
| 2001 | } |
| 2002 | else { |
| 2003 | lastclass = Class; |
| 2004 | if (*regparse == '-' && regparse+1 < regxend && |
| 2005 | regparse[1] != ']') { |
| 2006 | regparse++; |
| 2007 | range = 1; |
| 2008 | continue; /* do it next time */ |
| 2009 | } |
| 2010 | } |
| 2011 | for ( ; lastclass <= Class; lastclass++) |
| 2012 | regset(opnd, lastclass); |
| 2013 | lastclass = Class; |
| 2014 | } |
| 2015 | if (*regparse != ']') |
| 2016 | FAIL("unmatched [] in regexp"); |
| 2017 | nextchar(); |
| 2018 | return ret; |
| 2019 | } |
| 2020 | |
| 2021 | static char* |
| 2022 | nextchar(void) |
| 2023 | { |
| 2024 | char* retval = regparse++; |
| 2025 | |
| 2026 | for (;;) { |
| 2027 | if (*regparse == '(' && regparse[1] == '?' && |
| 2028 | regparse[2] == '#') { |
| 2029 | while (*regparse && *regparse != ')') |
| 2030 | regparse++; |
| 2031 | regparse++; |
| 2032 | continue; |
| 2033 | } |
| 2034 | if (regflags & PMf_EXTENDED) { |
| 2035 | if (isSPACE(*regparse)) { |
| 2036 | regparse++; |
| 2037 | continue; |
| 2038 | } |
| 2039 | else if (*regparse == '#') { |
| 2040 | while (*regparse && *regparse != '\n') |
| 2041 | regparse++; |
| 2042 | regparse++; |
| 2043 | continue; |
| 2044 | } |
| 2045 | } |
| 2046 | return retval; |
| 2047 | } |
| 2048 | } |
| 2049 | |
| 2050 | /* |
| 2051 | - reg_node - emit a node |
| 2052 | */ |
| 2053 | static regnode * /* Location. */ |
| 2054 | reg_node(U8 op) |
| 2055 | { |
| 2056 | register regnode *ret; |
| 2057 | register regnode *ptr; |
| 2058 | |
| 2059 | ret = regcode; |
| 2060 | if (SIZE_ONLY) { |
| 2061 | SIZE_ALIGN(regsize); |
| 2062 | #ifdef REGALIGN_STRUCT |
| 2063 | regsize += 1; |
| 2064 | #else |
| 2065 | regsize += 3; |
| 2066 | #endif |
| 2067 | return(ret); |
| 2068 | } |
| 2069 | |
| 2070 | NODE_ALIGN_FILL(ret); |
| 2071 | ptr = ret; |
| 2072 | FILL_ADVANCE_NODE(ptr, op); |
| 2073 | regcode = ptr; |
| 2074 | |
| 2075 | return(ret); |
| 2076 | } |
| 2077 | |
| 2078 | /* |
| 2079 | - reganode - emit a node with an argument |
| 2080 | */ |
| 2081 | static regnode * /* Location. */ |
| 2082 | reganode(U8 op, U32 arg) |
| 2083 | { |
| 2084 | register regnode *ret; |
| 2085 | register regnode *ptr; |
| 2086 | |
| 2087 | ret = regcode; |
| 2088 | if (SIZE_ONLY) { |
| 2089 | SIZE_ALIGN(regsize); |
| 2090 | #ifdef REGALIGN |
| 2091 | regsize += 2; |
| 2092 | #else |
| 2093 | regsize += 5; |
| 2094 | #endif |
| 2095 | return(ret); |
| 2096 | } |
| 2097 | |
| 2098 | NODE_ALIGN_FILL(ret); |
| 2099 | ptr = ret; |
| 2100 | FILL_ADVANCE_NODE_ARG(ptr, op, arg); |
| 2101 | regcode = ptr; |
| 2102 | |
| 2103 | return(ret); |
| 2104 | } |
| 2105 | |
| 2106 | /* |
| 2107 | - regc - emit (if appropriate) a byte of code |
| 2108 | */ |
| 2109 | static void |
| 2110 | regc(U8 b, char* s) |
| 2111 | { |
| 2112 | if (!SIZE_ONLY) |
| 2113 | *s = b; |
| 2114 | } |
| 2115 | |
| 2116 | /* |
| 2117 | - reginsert - insert an operator in front of already-emitted operand |
| 2118 | * |
| 2119 | * Means relocating the operand. |
| 2120 | */ |
| 2121 | static void |
| 2122 | reginsert(U8 op, regnode *opnd) |
| 2123 | { |
| 2124 | register regnode *src; |
| 2125 | register regnode *dst; |
| 2126 | register regnode *place; |
| 2127 | register int offset = regarglen[(U8)op]; |
| 2128 | |
| 2129 | /* (regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */ |
| 2130 | |
| 2131 | if (SIZE_ONLY) { |
| 2132 | regsize += NODE_STEP_REGNODE + offset; |
| 2133 | return; |
| 2134 | } |
| 2135 | |
| 2136 | src = regcode; |
| 2137 | regcode += NODE_STEP_REGNODE + offset; |
| 2138 | dst = regcode; |
| 2139 | while (src > opnd) |
| 2140 | StructCopy(--src, --dst, regnode); |
| 2141 | |
| 2142 | place = opnd; /* Op node, where operand used to be. */ |
| 2143 | src = NEXTOPER(place); |
| 2144 | FILL_ADVANCE_NODE(place, op); |
| 2145 | Zero(src, offset, regnode); |
| 2146 | #if defined(REGALIGN) && !defined(REGALIGN_STRUCT) |
| 2147 | src[offset + 1] = '\177'; |
| 2148 | #endif |
| 2149 | } |
| 2150 | |
| 2151 | /* |
| 2152 | - regtail - set the next-pointer at the end of a node chain of p to val. |
| 2153 | */ |
| 2154 | static void |
| 2155 | regtail(regnode *p, regnode *val) |
| 2156 | { |
| 2157 | register regnode *scan; |
| 2158 | register regnode *temp; |
| 2159 | register I32 offset; |
| 2160 | |
| 2161 | if (SIZE_ONLY) |
| 2162 | return; |
| 2163 | |
| 2164 | /* Find last node. */ |
| 2165 | scan = p; |
| 2166 | for (;;) { |
| 2167 | temp = regnext(scan); |
| 2168 | if (temp == NULL) |
| 2169 | break; |
| 2170 | scan = temp; |
| 2171 | } |
| 2172 | |
| 2173 | #ifdef REGALIGN |
| 2174 | # ifdef REGALIGN_STRUCT |
| 2175 | if (reg_off_by_arg[OP(scan)]) { |
| 2176 | ARG_SET(scan, val - scan); |
| 2177 | } else { |
| 2178 | NEXT_OFF(scan) = val - scan; |
| 2179 | } |
| 2180 | # else |
| 2181 | offset = val - scan; |
| 2182 | # ifndef lint |
| 2183 | *(short*)(scan+1) = offset; |
| 2184 | # endif |
| 2185 | #endif |
| 2186 | #else |
| 2187 | if (OP(scan) == BACK) |
| 2188 | offset = scan - val; |
| 2189 | else |
| 2190 | offset = val - scan; |
| 2191 | *(scan+1) = (offset>>8)&0377; |
| 2192 | *(scan+2) = offset&0377; |
| 2193 | #endif |
| 2194 | } |
| 2195 | |
| 2196 | /* |
| 2197 | - regoptail - regtail on operand of first argument; nop if operandless |
| 2198 | */ |
| 2199 | static void |
| 2200 | regoptail(regnode *p, regnode *val) |
| 2201 | { |
| 2202 | /* "Operandless" and "op != BRANCH" are synonymous in practice. */ |
| 2203 | if (p == NULL || SIZE_ONLY) |
| 2204 | return; |
| 2205 | if (regkind[(U8)OP(p)] == BRANCH) { |
| 2206 | regtail(NEXTOPER(p), val); |
| 2207 | } else if ( regkind[(U8)OP(p)] == BRANCHJ) { |
| 2208 | regtail(NEXTOPER(NEXTOPER(p)), val); |
| 2209 | } else |
| 2210 | return; |
| 2211 | } |
| 2212 | |
| 2213 | /* |
| 2214 | - regcurly - a little FSA that accepts {\d+,?\d*} |
| 2215 | */ |
| 2216 | STATIC I32 |
| 2217 | regcurly(register char *s) |
| 2218 | { |
| 2219 | if (*s++ != '{') |
| 2220 | return FALSE; |
| 2221 | if (!isDIGIT(*s)) |
| 2222 | return FALSE; |
| 2223 | while (isDIGIT(*s)) |
| 2224 | s++; |
| 2225 | if (*s == ',') |
| 2226 | s++; |
| 2227 | while (isDIGIT(*s)) |
| 2228 | s++; |
| 2229 | if (*s != '}') |
| 2230 | return FALSE; |
| 2231 | return TRUE; |
| 2232 | } |
| 2233 | |
| 2234 | #ifdef DEBUGGING |
| 2235 | |
| 2236 | static regnode * |
| 2237 | dumpuntil(regnode *start, regnode *node, regnode *last, SV* sv, I32 l) |
| 2238 | { |
| 2239 | register char op = EXACT; /* Arbitrary non-END op. */ |
| 2240 | register regnode *next, *onode; |
| 2241 | |
| 2242 | while (op != END && (!last || node < last)) { |
| 2243 | /* While that wasn't END last time... */ |
| 2244 | |
| 2245 | NODE_ALIGN(node); |
| 2246 | op = OP(node); |
| 2247 | if (op == CLOSE) |
| 2248 | l--; |
| 2249 | next = regnext(node); |
| 2250 | /* Where, what. */ |
| 2251 | if (OP(node) == OPTIMIZED) |
| 2252 | goto after_print; |
| 2253 | regprop(sv, node); |
| 2254 | PerlIO_printf(Perl_debug_log, "%4d%*s%s", node - start, |
| 2255 | 2*l + 1, "", SvPVX(sv)); |
| 2256 | if (next == NULL) /* Next ptr. */ |
| 2257 | PerlIO_printf(Perl_debug_log, "(0)"); |
| 2258 | else |
| 2259 | PerlIO_printf(Perl_debug_log, "(%d)", next - start); |
| 2260 | (void)PerlIO_putc(Perl_debug_log, '\n'); |
| 2261 | after_print: |
| 2262 | if (regkind[(U8)op] == BRANCHJ) { |
| 2263 | register regnode *nnode = (OP(next) == LONGJMP |
| 2264 | ? regnext(next) |
| 2265 | : next); |
| 2266 | if (last && nnode > last) |
| 2267 | nnode = last; |
| 2268 | node = dumpuntil(start, NEXTOPER(NEXTOPER(node)), nnode, sv, l + 1); |
| 2269 | } else if (regkind[(U8)op] == BRANCH) { |
| 2270 | node = dumpuntil(start, NEXTOPER(node), next, sv, l + 1); |
| 2271 | } else if ( op == CURLY) { /* `next' might be very big: optimizer */ |
| 2272 | node = dumpuntil(start, NEXTOPER(node) + EXTRA_STEP_2ARGS, |
| 2273 | NEXTOPER(node) + EXTRA_STEP_2ARGS + 1, sv, l + 1); |
| 2274 | } else if (regkind[(U8)op] == CURLY && op != CURLYX) { |
| 2275 | node = dumpuntil(start, NEXTOPER(node) + EXTRA_STEP_2ARGS, |
| 2276 | next, sv, l + 1); |
| 2277 | } else if ( op == PLUS || op == STAR) { |
| 2278 | node = dumpuntil(start, NEXTOPER(node), NEXTOPER(node) + 1, sv, l + 1); |
| 2279 | } else if (op == ANYOF) { |
| 2280 | node = NEXTOPER(node); |
| 2281 | node += ANY_SKIP; |
| 2282 | } else if (regkind[(U8)op] == EXACT) { |
| 2283 | /* Literal string, where present. */ |
| 2284 | node += ((*OPERAND(node)) + 2 + sizeof(regnode) - 1) / sizeof(regnode); |
| 2285 | node = NEXTOPER(node); |
| 2286 | } else { |
| 2287 | node = NEXTOPER(node); |
| 2288 | node += regarglen[(U8)op]; |
| 2289 | } |
| 2290 | if (op == CURLYX || op == OPEN) |
| 2291 | l++; |
| 2292 | else if (op == WHILEM) |
| 2293 | l--; |
| 2294 | } |
| 2295 | return node; |
| 2296 | } |
| 2297 | |
| 2298 | /* |
| 2299 | - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form |
| 2300 | */ |
| 2301 | void |
| 2302 | regdump(regexp *r) |
| 2303 | { |
| 2304 | SV *sv = sv_newmortal(); |
| 2305 | |
| 2306 | (void)dumpuntil(r->program, r->program + 1, NULL, sv, 0); |
| 2307 | |
| 2308 | /* Header fields of interest. */ |
| 2309 | if (r->anchored_substr) |
| 2310 | PerlIO_printf(Perl_debug_log, "anchored `%s%s%s'%s at %d ", |
| 2311 | colors[0], |
| 2312 | SvPVX(r->anchored_substr), |
| 2313 | colors[1], |
| 2314 | SvTAIL(r->anchored_substr) ? "$" : "", |
| 2315 | r->anchored_offset); |
| 2316 | if (r->float_substr) |
| 2317 | PerlIO_printf(Perl_debug_log, "floating `%s%s%s'%s at %d..%u ", |
| 2318 | colors[0], |
| 2319 | SvPVX(r->float_substr), |
| 2320 | colors[1], |
| 2321 | SvTAIL(r->float_substr) ? "$" : "", |
| 2322 | r->float_min_offset, r->float_max_offset); |
| 2323 | if (r->check_substr) |
| 2324 | PerlIO_printf(Perl_debug_log, |
| 2325 | r->check_substr == r->float_substr |
| 2326 | ? "(checking floating" : "(checking anchored"); |
| 2327 | if (r->reganch & ROPT_NOSCAN) |
| 2328 | PerlIO_printf(Perl_debug_log, " noscan"); |
| 2329 | if (r->reganch & ROPT_CHECK_ALL) |
| 2330 | PerlIO_printf(Perl_debug_log, " isall"); |
| 2331 | if (r->check_substr) |
| 2332 | PerlIO_printf(Perl_debug_log, ") "); |
| 2333 | |
| 2334 | if (r->regstclass) { |
| 2335 | regprop(sv, r->regstclass); |
| 2336 | PerlIO_printf(Perl_debug_log, "stclass `%s' ", SvPVX(sv)); |
| 2337 | } |
| 2338 | if (r->reganch & ROPT_ANCH) { |
| 2339 | PerlIO_printf(Perl_debug_log, "anchored"); |
| 2340 | if (r->reganch & ROPT_ANCH_BOL) |
| 2341 | PerlIO_printf(Perl_debug_log, "(BOL)"); |
| 2342 | if (r->reganch & ROPT_ANCH_MBOL) |
| 2343 | PerlIO_printf(Perl_debug_log, "(MBOL)"); |
| 2344 | if (r->reganch & ROPT_ANCH_GPOS) |
| 2345 | PerlIO_printf(Perl_debug_log, "(GPOS)"); |
| 2346 | PerlIO_putc(Perl_debug_log, ' '); |
| 2347 | } |
| 2348 | if (r->reganch & ROPT_GPOS_SEEN) |
| 2349 | PerlIO_printf(Perl_debug_log, "GPOS "); |
| 2350 | if (r->reganch & ROPT_SKIP) |
| 2351 | PerlIO_printf(Perl_debug_log, "plus "); |
| 2352 | if (r->reganch & ROPT_IMPLICIT) |
| 2353 | PerlIO_printf(Perl_debug_log, "implicit "); |
| 2354 | PerlIO_printf(Perl_debug_log, "minlen %ld ", (long) r->minlen); |
| 2355 | PerlIO_printf(Perl_debug_log, "\n"); |
| 2356 | } |
| 2357 | |
| 2358 | /* |
| 2359 | - regprop - printable representation of opcode |
| 2360 | */ |
| 2361 | void |
| 2362 | regprop(SV *sv, regnode *o) |
| 2363 | { |
| 2364 | register char *p = 0; |
| 2365 | |
| 2366 | sv_setpv(sv, ":"); |
| 2367 | switch (OP(o)) { |
| 2368 | case BOL: |
| 2369 | p = "BOL"; |
| 2370 | break; |
| 2371 | case MBOL: |
| 2372 | p = "MBOL"; |
| 2373 | break; |
| 2374 | case SBOL: |
| 2375 | p = "SBOL"; |
| 2376 | break; |
| 2377 | case EOL: |
| 2378 | p = "EOL"; |
| 2379 | break; |
| 2380 | case MEOL: |
| 2381 | p = "MEOL"; |
| 2382 | break; |
| 2383 | case SEOL: |
| 2384 | p = "SEOL"; |
| 2385 | break; |
| 2386 | case ANY: |
| 2387 | p = "ANY"; |
| 2388 | break; |
| 2389 | case SANY: |
| 2390 | p = "SANY"; |
| 2391 | break; |
| 2392 | case ANYOF: |
| 2393 | p = "ANYOF"; |
| 2394 | break; |
| 2395 | case BRANCH: |
| 2396 | p = "BRANCH"; |
| 2397 | break; |
| 2398 | case EXACT: |
| 2399 | sv_catpvf(sv, "EXACT <%s%s%s>", colors[0], OPERAND(o) + 1, colors[1]); |
| 2400 | break; |
| 2401 | case EXACTF: |
| 2402 | sv_catpvf(sv, "EXACTF <%s%s%s>", colors[0], OPERAND(o) + 1, colors[1]); |
| 2403 | break; |
| 2404 | case EXACTFL: |
| 2405 | sv_catpvf(sv, "EXACTFL <%s%s%s>", colors[0], OPERAND(o) + 1, colors[1]); |
| 2406 | break; |
| 2407 | case NOTHING: |
| 2408 | p = "NOTHING"; |
| 2409 | break; |
| 2410 | case TAIL: |
| 2411 | p = "TAIL"; |
| 2412 | break; |
| 2413 | case BACK: |
| 2414 | p = "BACK"; |
| 2415 | break; |
| 2416 | case END: |
| 2417 | p = "END"; |
| 2418 | break; |
| 2419 | case BOUND: |
| 2420 | p = "BOUND"; |
| 2421 | break; |
| 2422 | case BOUNDL: |
| 2423 | p = "BOUNDL"; |
| 2424 | break; |
| 2425 | case NBOUND: |
| 2426 | p = "NBOUND"; |
| 2427 | break; |
| 2428 | case NBOUNDL: |
| 2429 | p = "NBOUNDL"; |
| 2430 | break; |
| 2431 | case CURLY: |
| 2432 | sv_catpvf(sv, "CURLY {%d,%d}", ARG1(o), ARG2(o)); |
| 2433 | break; |
| 2434 | case CURLYM: |
| 2435 | #ifdef REGALIGN_STRUCT |
| 2436 | sv_catpvf(sv, "CURLYM[%d] {%d,%d}", o->flags, ARG1(o), ARG2(o)); |
| 2437 | #else |
| 2438 | sv_catpvf(sv, "CURLYM {%d,%d}", ARG1(o), ARG2(o)); |
| 2439 | #endif |
| 2440 | break; |
| 2441 | case CURLYN: |
| 2442 | #ifdef REGALIGN_STRUCT |
| 2443 | sv_catpvf(sv, "CURLYN[%d] {%d,%d}", o->flags, ARG1(o), ARG2(o)); |
| 2444 | #else |
| 2445 | sv_catpvf(sv, "CURLYN {%d,%d}", ARG1(o), ARG2(o)); |
| 2446 | #endif |
| 2447 | break; |
| 2448 | case CURLYX: |
| 2449 | sv_catpvf(sv, "CURLYX {%d,%d}", ARG1(o), ARG2(o)); |
| 2450 | break; |
| 2451 | case REF: |
| 2452 | sv_catpvf(sv, "REF%d", ARG(o)); |
| 2453 | break; |
| 2454 | case REFF: |
| 2455 | sv_catpvf(sv, "REFF%d", ARG(o)); |
| 2456 | break; |
| 2457 | case REFFL: |
| 2458 | sv_catpvf(sv, "REFFL%d", ARG(o)); |
| 2459 | break; |
| 2460 | case OPEN: |
| 2461 | sv_catpvf(sv, "OPEN%d", ARG(o)); |
| 2462 | break; |
| 2463 | case CLOSE: |
| 2464 | sv_catpvf(sv, "CLOSE%d", ARG(o)); |
| 2465 | p = NULL; |
| 2466 | break; |
| 2467 | case STAR: |
| 2468 | p = "STAR"; |
| 2469 | break; |
| 2470 | case PLUS: |
| 2471 | p = "PLUS"; |
| 2472 | break; |
| 2473 | case MINMOD: |
| 2474 | p = "MINMOD"; |
| 2475 | break; |
| 2476 | case GPOS: |
| 2477 | p = "GPOS"; |
| 2478 | break; |
| 2479 | case UNLESSM: |
| 2480 | #ifdef REGALIGN_STRUCT |
| 2481 | sv_catpvf(sv, "UNLESSM[-%d]", o->flags); |
| 2482 | #else |
| 2483 | p = "UNLESSM"; |
| 2484 | #endif |
| 2485 | break; |
| 2486 | case IFMATCH: |
| 2487 | #ifdef REGALIGN_STRUCT |
| 2488 | sv_catpvf(sv, "IFMATCH[-%d]", o->flags); |
| 2489 | #else |
| 2490 | p = "IFMATCH"; |
| 2491 | #endif |
| 2492 | break; |
| 2493 | case SUCCEED: |
| 2494 | p = "SUCCEED"; |
| 2495 | break; |
| 2496 | case WHILEM: |
| 2497 | p = "WHILEM"; |
| 2498 | break; |
| 2499 | case DIGIT: |
| 2500 | p = "DIGIT"; |
| 2501 | break; |
| 2502 | case NDIGIT: |
| 2503 | p = "NDIGIT"; |
| 2504 | break; |
| 2505 | case ALNUM: |
| 2506 | p = "ALNUM"; |
| 2507 | break; |
| 2508 | case NALNUM: |
| 2509 | p = "NALNUM"; |
| 2510 | break; |
| 2511 | case SPACE: |
| 2512 | p = "SPACE"; |
| 2513 | break; |
| 2514 | case NSPACE: |
| 2515 | p = "NSPACE"; |
| 2516 | break; |
| 2517 | case ALNUML: |
| 2518 | p = "ALNUML"; |
| 2519 | break; |
| 2520 | case NALNUML: |
| 2521 | p = "NALNUML"; |
| 2522 | break; |
| 2523 | case SPACEL: |
| 2524 | p = "SPACEL"; |
| 2525 | break; |
| 2526 | case NSPACEL: |
| 2527 | p = "NSPACEL"; |
| 2528 | break; |
| 2529 | case EVAL: |
| 2530 | p = "EVAL"; |
| 2531 | break; |
| 2532 | case LONGJMP: |
| 2533 | p = "LONGJMP"; |
| 2534 | break; |
| 2535 | case BRANCHJ: |
| 2536 | p = "BRANCHJ"; |
| 2537 | break; |
| 2538 | case IFTHEN: |
| 2539 | p = "IFTHEN"; |
| 2540 | break; |
| 2541 | case GROUPP: |
| 2542 | sv_catpvf(sv, "GROUPP%d", ARG(o)); |
| 2543 | break; |
| 2544 | case LOGICAL: |
| 2545 | p = "LOGICAL"; |
| 2546 | break; |
| 2547 | case SUSPEND: |
| 2548 | p = "SUSPEND"; |
| 2549 | break; |
| 2550 | case RENUM: |
| 2551 | p = "RENUM"; |
| 2552 | break; |
| 2553 | case OPTIMIZED: |
| 2554 | p = "OPTIMIZED"; |
| 2555 | break; |
| 2556 | default: |
| 2557 | FAIL("corrupted regexp opcode"); |
| 2558 | } |
| 2559 | if (p) |
| 2560 | sv_catpv(sv, p); |
| 2561 | } |
| 2562 | #endif /* DEBUGGING */ |
| 2563 | |
| 2564 | void |
| 2565 | pregfree(struct regexp *r) |
| 2566 | { |
| 2567 | if (!r || (--r->refcnt > 0)) |
| 2568 | return; |
| 2569 | if (r->precomp) |
| 2570 | Safefree(r->precomp); |
| 2571 | if (r->subbase) |
| 2572 | Safefree(r->subbase); |
| 2573 | if (r->substrs) |
| 2574 | Safefree(r->substrs); |
| 2575 | if (r->anchored_substr) |
| 2576 | SvREFCNT_dec(r->anchored_substr); |
| 2577 | if (r->float_substr) |
| 2578 | SvREFCNT_dec(r->float_substr); |
| 2579 | if (r->data) { |
| 2580 | int n = r->data->count; |
| 2581 | while (--n >= 0) { |
| 2582 | switch (r->data->what[n]) { |
| 2583 | case 's': |
| 2584 | SvREFCNT_dec((SV*)r->data->data[n]); |
| 2585 | break; |
| 2586 | case 'o': |
| 2587 | op_free((OP_4tree*)r->data->data[n]); |
| 2588 | break; |
| 2589 | case 'n': |
| 2590 | break; |
| 2591 | default: |
| 2592 | FAIL2("panic: regfree data code '%c'", r->data->what[n]); |
| 2593 | } |
| 2594 | } |
| 2595 | Safefree(r->data->what); |
| 2596 | Safefree(r->data); |
| 2597 | } |
| 2598 | Safefree(r->startp); |
| 2599 | Safefree(r->endp); |
| 2600 | Safefree(r); |
| 2601 | } |
| 2602 | |
| 2603 | /* |
| 2604 | - regnext - dig the "next" pointer out of a node |
| 2605 | * |
| 2606 | * [Note, when REGALIGN is defined there are two places in regmatch() |
| 2607 | * that bypass this code for speed.] |
| 2608 | */ |
| 2609 | regnode * |
| 2610 | regnext(register regnode *p) |
| 2611 | { |
| 2612 | register I32 offset; |
| 2613 | |
| 2614 | if (p == ®dummy) |
| 2615 | return(NULL); |
| 2616 | |
| 2617 | offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p)); |
| 2618 | if (offset == 0) |
| 2619 | return(NULL); |
| 2620 | |
| 2621 | #ifdef REGALIGN |
| 2622 | return(p+offset); |
| 2623 | #else |
| 2624 | if (OP(p) == BACK) |
| 2625 | return(p-offset); |
| 2626 | else |
| 2627 | return(p+offset); |
| 2628 | #endif |
| 2629 | } |
| 2630 | |
| 2631 | #ifdef I_STDARG |
| 2632 | static void |
| 2633 | re_croak2(const char* pat1,const char* pat2,...) |
| 2634 | #else |
| 2635 | /*VARARGS0*/ |
| 2636 | static void |
| 2637 | re_croak2(const char* pat1,const char* pat2, va_alist) |
| 2638 | const char* pat1; |
| 2639 | const char* pat2; |
| 2640 | va_dcl |
| 2641 | #endif |
| 2642 | { |
| 2643 | va_list args; |
| 2644 | STRLEN l1 = strlen(pat1); |
| 2645 | STRLEN l2 = strlen(pat2); |
| 2646 | char buf[512]; |
| 2647 | char *message; |
| 2648 | |
| 2649 | if (l1 > 510) |
| 2650 | l1 = 510; |
| 2651 | if (l1 + l2 > 510) |
| 2652 | l2 = 510 - l1; |
| 2653 | Copy(pat1, buf, l1 , char); |
| 2654 | Copy(pat2, buf + l1, l2 , char); |
| 2655 | buf[l1 + l2 + 1] = '\n'; |
| 2656 | buf[l1 + l2 + 2] = '\0'; |
| 2657 | #ifdef I_STDARG |
| 2658 | va_start(args, pat2); |
| 2659 | #else |
| 2660 | va_start(args); |
| 2661 | #endif |
| 2662 | message = mess(buf, &args); |
| 2663 | va_end(args); |
| 2664 | l1 = strlen(message); |
| 2665 | if (l1 > 512) |
| 2666 | l1 = 512; |
| 2667 | Copy(message, buf, l1 , char); |
| 2668 | buf[l1] = '\0'; /* Overwrite \n */ |
| 2669 | croak("%s", buf); |
| 2670 | } |