Fredrik Kuivinen | fca65d4 | 2011-08-21 00:41:41 +0200 | [diff] [blame] | 1 | /* |
| 2 | * This file has been copied from commit e7ac713d^ in the GNU grep git |
| 3 | * repository. A few small changes have been made to adapt the code to |
| 4 | * Git. |
| 5 | */ |
| 6 | |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 7 | /* kwset.c - search for any of a set of keywords. |
| 8 | Copyright 1989, 1998, 2000, 2005 Free Software Foundation, Inc. |
| 9 | |
| 10 | This program is free software; you can redistribute it and/or modify |
| 11 | it under the terms of the GNU General Public License as published by |
| 12 | the Free Software Foundation; either version 2, or (at your option) |
| 13 | any later version. |
| 14 | |
| 15 | This program is distributed in the hope that it will be useful, |
| 16 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 18 | GNU General Public License for more details. |
| 19 | |
| 20 | You should have received a copy of the GNU General Public License |
Todd Zullinger | 4842579 | 2017-11-07 00:39:33 -0500 | [diff] [blame] | 21 | along with this program; if not, see <http://www.gnu.org/licenses/>. */ |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 22 | |
| 23 | /* Written August 1989 by Mike Haertel. |
| 24 | The author may be reached (Email) at the address mike@ai.mit.edu, |
| 25 | or (US mail) as Mike Haertel c/o Free Software Foundation. */ |
| 26 | |
Stefano Lattarini | 2fec81c | 2013-04-12 00:36:10 +0200 | [diff] [blame] | 27 | /* The algorithm implemented by these routines bears a startling resemblance |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 28 | to one discovered by Beate Commentz-Walter, although it is not identical. |
| 29 | See "A String Matching Algorithm Fast on the Average," Technical Report, |
| 30 | IBM-Germany, Scientific Center Heidelberg, Tiergartenstrasse 15, D-6900 |
| 31 | Heidelberg, Germany. See also Aho, A.V., and M. Corasick, "Efficient |
| 32 | String Matching: An Aid to Bibliographic Search," CACM June 1975, |
| 33 | Vol. 18, No. 6, which describes the failure function used below. */ |
| 34 | |
Fredrik Kuivinen | fca65d4 | 2011-08-21 00:41:41 +0200 | [diff] [blame] | 35 | #include "cache.h" |
| 36 | |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 37 | #include "kwset.h" |
Fredrik Kuivinen | d190a08 | 2011-08-28 22:08:46 +0200 | [diff] [blame] | 38 | #include "compat/obstack.h" |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 39 | |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 40 | #define NCHAR (UCHAR_MAX + 1) |
Fredrik Kuivinen | fca65d4 | 2011-08-21 00:41:41 +0200 | [diff] [blame] | 41 | #define obstack_chunk_alloc xmalloc |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 42 | #define obstack_chunk_free free |
| 43 | |
| 44 | #define U(c) ((unsigned char) (c)) |
| 45 | |
| 46 | /* Balanced tree of edges and labels leaving a given trie node. */ |
| 47 | struct tree |
| 48 | { |
| 49 | struct tree *llink; /* Left link; MUST be first field. */ |
| 50 | struct tree *rlink; /* Right link (to larger labels). */ |
| 51 | struct trie *trie; /* Trie node pointed to by this edge. */ |
| 52 | unsigned char label; /* Label on this edge. */ |
| 53 | char balance; /* Difference in depths of subtrees. */ |
| 54 | }; |
| 55 | |
| 56 | /* Node of a trie representing a set of reversed keywords. */ |
| 57 | struct trie |
| 58 | { |
| 59 | unsigned int accepting; /* Word index of accepted word, or zero. */ |
| 60 | struct tree *links; /* Tree of edges leaving this node. */ |
| 61 | struct trie *parent; /* Parent of this node. */ |
| 62 | struct trie *next; /* List of all trie nodes in level order. */ |
| 63 | struct trie *fail; /* Aho-Corasick failure function. */ |
| 64 | int depth; /* Depth of this node from the root. */ |
| 65 | int shift; /* Shift function for search failures. */ |
Ondřej Bílka | 749f763 | 2013-07-22 23:02:23 +0200 | [diff] [blame] | 66 | int maxshift; /* Max shift of self and descendants. */ |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 67 | }; |
| 68 | |
| 69 | /* Structure returned opaquely to the caller, containing everything. */ |
| 70 | struct kwset |
| 71 | { |
| 72 | struct obstack obstack; /* Obstack for node allocation. */ |
| 73 | int words; /* Number of words in the trie. */ |
| 74 | struct trie *trie; /* The trie itself. */ |
| 75 | int mind; /* Minimum depth of an accepting node. */ |
| 76 | int maxd; /* Maximum depth of any node. */ |
| 77 | unsigned char delta[NCHAR]; /* Delta table for rapid search. */ |
| 78 | struct trie *next[NCHAR]; /* Table of children of the root. */ |
| 79 | char *target; /* Target string if there's only one. */ |
| 80 | int mind2; /* Used in Boyer-Moore search for one string. */ |
Ben Walton | 189c860 | 2015-03-02 19:22:31 +0000 | [diff] [blame] | 81 | unsigned char const *trans; /* Character translation table. */ |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 82 | }; |
| 83 | |
| 84 | /* Allocate and initialize a keyword set object, returning an opaque |
| 85 | pointer to it. Return NULL if memory is not available. */ |
| 86 | kwset_t |
Ben Walton | 189c860 | 2015-03-02 19:22:31 +0000 | [diff] [blame] | 87 | kwsalloc (unsigned char const *trans) |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 88 | { |
| 89 | struct kwset *kwset; |
| 90 | |
Fredrik Kuivinen | fca65d4 | 2011-08-21 00:41:41 +0200 | [diff] [blame] | 91 | kwset = (struct kwset *) xmalloc(sizeof (struct kwset)); |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 92 | |
| 93 | obstack_init(&kwset->obstack); |
| 94 | kwset->words = 0; |
| 95 | kwset->trie |
| 96 | = (struct trie *) obstack_alloc(&kwset->obstack, sizeof (struct trie)); |
| 97 | if (!kwset->trie) |
| 98 | { |
| 99 | kwsfree((kwset_t) kwset); |
| 100 | return NULL; |
| 101 | } |
| 102 | kwset->trie->accepting = 0; |
| 103 | kwset->trie->links = NULL; |
| 104 | kwset->trie->parent = NULL; |
| 105 | kwset->trie->next = NULL; |
| 106 | kwset->trie->fail = NULL; |
| 107 | kwset->trie->depth = 0; |
| 108 | kwset->trie->shift = 0; |
| 109 | kwset->mind = INT_MAX; |
| 110 | kwset->maxd = -1; |
| 111 | kwset->target = NULL; |
| 112 | kwset->trans = trans; |
| 113 | |
| 114 | return (kwset_t) kwset; |
| 115 | } |
| 116 | |
| 117 | /* This upper bound is valid for CHAR_BIT >= 4 and |
| 118 | exact for CHAR_BIT in { 4..11, 13, 15, 17, 19 }. */ |
| 119 | #define DEPTH_SIZE (CHAR_BIT + CHAR_BIT/2) |
| 120 | |
| 121 | /* Add the given string to the contents of the keyword set. Return NULL |
| 122 | for success, an error message otherwise. */ |
| 123 | const char * |
| 124 | kwsincr (kwset_t kws, char const *text, size_t len) |
| 125 | { |
| 126 | struct kwset *kwset; |
| 127 | register struct trie *trie; |
| 128 | register unsigned char label; |
| 129 | register struct tree *link; |
| 130 | register int depth; |
| 131 | struct tree *links[DEPTH_SIZE]; |
| 132 | enum { L, R } dirs[DEPTH_SIZE]; |
| 133 | struct tree *t, *r, *l, *rl, *lr; |
| 134 | |
| 135 | kwset = (struct kwset *) kws; |
| 136 | trie = kwset->trie; |
| 137 | text += len; |
| 138 | |
| 139 | /* Descend the trie (built of reversed keywords) character-by-character, |
| 140 | installing new nodes when necessary. */ |
| 141 | while (len--) |
| 142 | { |
| 143 | label = kwset->trans ? kwset->trans[U(*--text)] : *--text; |
| 144 | |
| 145 | /* Descend the tree of outgoing links for this trie node, |
| 146 | looking for the current character and keeping track |
| 147 | of the path followed. */ |
| 148 | link = trie->links; |
| 149 | links[0] = (struct tree *) &trie->links; |
| 150 | dirs[0] = L; |
| 151 | depth = 1; |
| 152 | |
| 153 | while (link && label != link->label) |
| 154 | { |
| 155 | links[depth] = link; |
| 156 | if (label < link->label) |
| 157 | dirs[depth++] = L, link = link->llink; |
| 158 | else |
| 159 | dirs[depth++] = R, link = link->rlink; |
| 160 | } |
| 161 | |
| 162 | /* The current character doesn't have an outgoing link at |
| 163 | this trie node, so build a new trie node and install |
| 164 | a link in the current trie node's tree. */ |
| 165 | if (!link) |
| 166 | { |
| 167 | link = (struct tree *) obstack_alloc(&kwset->obstack, |
| 168 | sizeof (struct tree)); |
| 169 | if (!link) |
Fredrik Kuivinen | fca65d4 | 2011-08-21 00:41:41 +0200 | [diff] [blame] | 170 | return "memory exhausted"; |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 171 | link->llink = NULL; |
| 172 | link->rlink = NULL; |
| 173 | link->trie = (struct trie *) obstack_alloc(&kwset->obstack, |
| 174 | sizeof (struct trie)); |
| 175 | if (!link->trie) |
| 176 | { |
| 177 | obstack_free(&kwset->obstack, link); |
Fredrik Kuivinen | fca65d4 | 2011-08-21 00:41:41 +0200 | [diff] [blame] | 178 | return "memory exhausted"; |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 179 | } |
| 180 | link->trie->accepting = 0; |
| 181 | link->trie->links = NULL; |
| 182 | link->trie->parent = trie; |
| 183 | link->trie->next = NULL; |
| 184 | link->trie->fail = NULL; |
| 185 | link->trie->depth = trie->depth + 1; |
| 186 | link->trie->shift = 0; |
| 187 | link->label = label; |
| 188 | link->balance = 0; |
| 189 | |
| 190 | /* Install the new tree node in its parent. */ |
| 191 | if (dirs[--depth] == L) |
| 192 | links[depth]->llink = link; |
| 193 | else |
| 194 | links[depth]->rlink = link; |
| 195 | |
| 196 | /* Back up the tree fixing the balance flags. */ |
| 197 | while (depth && !links[depth]->balance) |
| 198 | { |
| 199 | if (dirs[depth] == L) |
| 200 | --links[depth]->balance; |
| 201 | else |
| 202 | ++links[depth]->balance; |
| 203 | --depth; |
| 204 | } |
| 205 | |
| 206 | /* Rebalance the tree by pointer rotations if necessary. */ |
| 207 | if (depth && ((dirs[depth] == L && --links[depth]->balance) |
| 208 | || (dirs[depth] == R && ++links[depth]->balance))) |
| 209 | { |
| 210 | switch (links[depth]->balance) |
| 211 | { |
| 212 | case (char) -2: |
| 213 | switch (dirs[depth + 1]) |
| 214 | { |
| 215 | case L: |
| 216 | r = links[depth], t = r->llink, rl = t->rlink; |
| 217 | t->rlink = r, r->llink = rl; |
| 218 | t->balance = r->balance = 0; |
| 219 | break; |
| 220 | case R: |
| 221 | r = links[depth], l = r->llink, t = l->rlink; |
| 222 | rl = t->rlink, lr = t->llink; |
| 223 | t->llink = l, l->rlink = lr, t->rlink = r, r->llink = rl; |
| 224 | l->balance = t->balance != 1 ? 0 : -1; |
| 225 | r->balance = t->balance != (char) -1 ? 0 : 1; |
| 226 | t->balance = 0; |
| 227 | break; |
| 228 | default: |
| 229 | abort (); |
| 230 | } |
| 231 | break; |
| 232 | case 2: |
| 233 | switch (dirs[depth + 1]) |
| 234 | { |
| 235 | case R: |
| 236 | l = links[depth], t = l->rlink, lr = t->llink; |
| 237 | t->llink = l, l->rlink = lr; |
| 238 | t->balance = l->balance = 0; |
| 239 | break; |
| 240 | case L: |
| 241 | l = links[depth], r = l->rlink, t = r->llink; |
| 242 | lr = t->llink, rl = t->rlink; |
| 243 | t->llink = l, l->rlink = lr, t->rlink = r, r->llink = rl; |
| 244 | l->balance = t->balance != 1 ? 0 : -1; |
| 245 | r->balance = t->balance != (char) -1 ? 0 : 1; |
| 246 | t->balance = 0; |
| 247 | break; |
| 248 | default: |
| 249 | abort (); |
| 250 | } |
| 251 | break; |
| 252 | default: |
| 253 | abort (); |
| 254 | } |
| 255 | |
| 256 | if (dirs[depth - 1] == L) |
| 257 | links[depth - 1]->llink = t; |
| 258 | else |
| 259 | links[depth - 1]->rlink = t; |
| 260 | } |
| 261 | } |
| 262 | |
| 263 | trie = link->trie; |
| 264 | } |
| 265 | |
| 266 | /* Mark the node we finally reached as accepting, encoding the |
| 267 | index number of this word in the keyword set so far. */ |
| 268 | if (!trie->accepting) |
| 269 | trie->accepting = 1 + 2 * kwset->words; |
| 270 | ++kwset->words; |
| 271 | |
| 272 | /* Keep track of the longest and shortest string of the keyword set. */ |
| 273 | if (trie->depth < kwset->mind) |
| 274 | kwset->mind = trie->depth; |
| 275 | if (trie->depth > kwset->maxd) |
| 276 | kwset->maxd = trie->depth; |
| 277 | |
| 278 | return NULL; |
| 279 | } |
| 280 | |
| 281 | /* Enqueue the trie nodes referenced from the given tree in the |
| 282 | given queue. */ |
| 283 | static void |
| 284 | enqueue (struct tree *tree, struct trie **last) |
| 285 | { |
| 286 | if (!tree) |
| 287 | return; |
| 288 | enqueue(tree->llink, last); |
| 289 | enqueue(tree->rlink, last); |
| 290 | (*last) = (*last)->next = tree->trie; |
| 291 | } |
| 292 | |
| 293 | /* Compute the Aho-Corasick failure function for the trie nodes referenced |
| 294 | from the given tree, given the failure function for their parent as |
| 295 | well as a last resort failure node. */ |
| 296 | static void |
| 297 | treefails (register struct tree const *tree, struct trie const *fail, |
| 298 | struct trie *recourse) |
| 299 | { |
| 300 | register struct tree *link; |
| 301 | |
| 302 | if (!tree) |
| 303 | return; |
| 304 | |
| 305 | treefails(tree->llink, fail, recourse); |
| 306 | treefails(tree->rlink, fail, recourse); |
| 307 | |
| 308 | /* Find, in the chain of fails going back to the root, the first |
Ondřej Bílka | 749f763 | 2013-07-22 23:02:23 +0200 | [diff] [blame] | 309 | node that has a descendant on the current label. */ |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 310 | while (fail) |
| 311 | { |
| 312 | link = fail->links; |
| 313 | while (link && tree->label != link->label) |
| 314 | if (tree->label < link->label) |
| 315 | link = link->llink; |
| 316 | else |
| 317 | link = link->rlink; |
| 318 | if (link) |
| 319 | { |
| 320 | tree->trie->fail = link->trie; |
| 321 | return; |
| 322 | } |
| 323 | fail = fail->fail; |
| 324 | } |
| 325 | |
| 326 | tree->trie->fail = recourse; |
| 327 | } |
| 328 | |
| 329 | /* Set delta entries for the links of the given tree such that |
| 330 | the preexisting delta value is larger than the current depth. */ |
| 331 | static void |
| 332 | treedelta (register struct tree const *tree, |
| 333 | register unsigned int depth, |
| 334 | unsigned char delta[]) |
| 335 | { |
| 336 | if (!tree) |
| 337 | return; |
| 338 | treedelta(tree->llink, depth, delta); |
| 339 | treedelta(tree->rlink, depth, delta); |
| 340 | if (depth < delta[tree->label]) |
| 341 | delta[tree->label] = depth; |
| 342 | } |
| 343 | |
| 344 | /* Return true if A has every label in B. */ |
| 345 | static int |
| 346 | hasevery (register struct tree const *a, register struct tree const *b) |
| 347 | { |
| 348 | if (!b) |
| 349 | return 1; |
| 350 | if (!hasevery(a, b->llink)) |
| 351 | return 0; |
| 352 | if (!hasevery(a, b->rlink)) |
| 353 | return 0; |
| 354 | while (a && b->label != a->label) |
| 355 | if (b->label < a->label) |
| 356 | a = a->llink; |
| 357 | else |
| 358 | a = a->rlink; |
| 359 | return !!a; |
| 360 | } |
| 361 | |
| 362 | /* Compute a vector, indexed by character code, of the trie nodes |
| 363 | referenced from the given tree. */ |
| 364 | static void |
| 365 | treenext (struct tree const *tree, struct trie *next[]) |
| 366 | { |
| 367 | if (!tree) |
| 368 | return; |
| 369 | treenext(tree->llink, next); |
| 370 | treenext(tree->rlink, next); |
| 371 | next[tree->label] = tree->trie; |
| 372 | } |
| 373 | |
| 374 | /* Compute the shift for each trie node, as well as the delta |
| 375 | table and next cache for the given keyword set. */ |
| 376 | const char * |
| 377 | kwsprep (kwset_t kws) |
| 378 | { |
| 379 | register struct kwset *kwset; |
| 380 | register int i; |
| 381 | register struct trie *curr; |
Ben Walton | 189c860 | 2015-03-02 19:22:31 +0000 | [diff] [blame] | 382 | register unsigned char const *trans; |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 383 | unsigned char delta[NCHAR]; |
| 384 | |
| 385 | kwset = (struct kwset *) kws; |
| 386 | |
| 387 | /* Initial values for the delta table; will be changed later. The |
| 388 | delta entry for a given character is the smallest depth of any |
| 389 | node at which an outgoing edge is labeled by that character. */ |
| 390 | memset(delta, kwset->mind < UCHAR_MAX ? kwset->mind : UCHAR_MAX, NCHAR); |
| 391 | |
| 392 | /* Check if we can use the simple boyer-moore algorithm, instead |
| 393 | of the hairy commentz-walter algorithm. */ |
| 394 | if (kwset->words == 1 && kwset->trans == NULL) |
| 395 | { |
| 396 | char c; |
| 397 | |
| 398 | /* Looking for just one string. Extract it from the trie. */ |
| 399 | kwset->target = obstack_alloc(&kwset->obstack, kwset->mind); |
| 400 | if (!kwset->target) |
Fredrik Kuivinen | fca65d4 | 2011-08-21 00:41:41 +0200 | [diff] [blame] | 401 | return "memory exhausted"; |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 402 | for (i = kwset->mind - 1, curr = kwset->trie; i >= 0; --i) |
| 403 | { |
| 404 | kwset->target[i] = curr->links->label; |
| 405 | curr = curr->links->trie; |
| 406 | } |
| 407 | /* Build the Boyer Moore delta. Boy that's easy compared to CW. */ |
| 408 | for (i = 0; i < kwset->mind; ++i) |
| 409 | delta[U(kwset->target[i])] = kwset->mind - (i + 1); |
| 410 | /* Find the minimal delta2 shift that we might make after |
| 411 | a backwards match has failed. */ |
| 412 | c = kwset->target[kwset->mind - 1]; |
| 413 | for (i = kwset->mind - 2; i >= 0; --i) |
| 414 | if (kwset->target[i] == c) |
| 415 | break; |
| 416 | kwset->mind2 = kwset->mind - (i + 1); |
| 417 | } |
| 418 | else |
| 419 | { |
| 420 | register struct trie *fail; |
| 421 | struct trie *last, *next[NCHAR]; |
| 422 | |
| 423 | /* Traverse the nodes of the trie in level order, simultaneously |
| 424 | computing the delta table, failure function, and shift function. */ |
| 425 | for (curr = last = kwset->trie; curr; curr = curr->next) |
| 426 | { |
Ondřej Bílka | 749f763 | 2013-07-22 23:02:23 +0200 | [diff] [blame] | 427 | /* Enqueue the immediate descendants in the level order queue. */ |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 428 | enqueue(curr->links, &last); |
| 429 | |
| 430 | curr->shift = kwset->mind; |
| 431 | curr->maxshift = kwset->mind; |
| 432 | |
Ondřej Bílka | 749f763 | 2013-07-22 23:02:23 +0200 | [diff] [blame] | 433 | /* Update the delta table for the descendants of this node. */ |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 434 | treedelta(curr->links, curr->depth, delta); |
| 435 | |
Stefano Lattarini | 2fec81c | 2013-04-12 00:36:10 +0200 | [diff] [blame] | 436 | /* Compute the failure function for the descendants of this node. */ |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 437 | treefails(curr->links, curr->fail, kwset->trie); |
| 438 | |
| 439 | /* Update the shifts at each node in the current node's chain |
| 440 | of fails back to the root. */ |
| 441 | for (fail = curr->fail; fail; fail = fail->fail) |
| 442 | { |
| 443 | /* If the current node has some outgoing edge that the fail |
| 444 | doesn't, then the shift at the fail should be no larger |
| 445 | than the difference of their depths. */ |
| 446 | if (!hasevery(fail->links, curr->links)) |
| 447 | if (curr->depth - fail->depth < fail->shift) |
| 448 | fail->shift = curr->depth - fail->depth; |
| 449 | |
| 450 | /* If the current node is accepting then the shift at the |
Ondřej Bílka | 749f763 | 2013-07-22 23:02:23 +0200 | [diff] [blame] | 451 | fail and its descendants should be no larger than the |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 452 | difference of their depths. */ |
| 453 | if (curr->accepting && fail->maxshift > curr->depth - fail->depth) |
| 454 | fail->maxshift = curr->depth - fail->depth; |
| 455 | } |
| 456 | } |
| 457 | |
| 458 | /* Traverse the trie in level order again, fixing up all nodes whose |
| 459 | shift exceeds their inherited maxshift. */ |
| 460 | for (curr = kwset->trie->next; curr; curr = curr->next) |
| 461 | { |
| 462 | if (curr->maxshift > curr->parent->maxshift) |
| 463 | curr->maxshift = curr->parent->maxshift; |
| 464 | if (curr->shift > curr->maxshift) |
| 465 | curr->shift = curr->maxshift; |
| 466 | } |
| 467 | |
| 468 | /* Create a vector, indexed by character code, of the outgoing links |
| 469 | from the root node. */ |
| 470 | for (i = 0; i < NCHAR; ++i) |
| 471 | next[i] = NULL; |
| 472 | treenext(kwset->trie->links, next); |
| 473 | |
| 474 | if ((trans = kwset->trans) != NULL) |
| 475 | for (i = 0; i < NCHAR; ++i) |
| 476 | kwset->next[i] = next[U(trans[i])]; |
| 477 | else |
| 478 | memcpy(kwset->next, next, NCHAR * sizeof(struct trie *)); |
| 479 | } |
| 480 | |
| 481 | /* Fix things up for any translation table. */ |
| 482 | if ((trans = kwset->trans) != NULL) |
| 483 | for (i = 0; i < NCHAR; ++i) |
| 484 | kwset->delta[i] = delta[U(trans[i])]; |
| 485 | else |
| 486 | memcpy(kwset->delta, delta, NCHAR); |
| 487 | |
| 488 | return NULL; |
| 489 | } |
| 490 | |
| 491 | /* Fast boyer-moore search. */ |
| 492 | static size_t |
| 493 | bmexec (kwset_t kws, char const *text, size_t size) |
| 494 | { |
| 495 | struct kwset const *kwset; |
| 496 | register unsigned char const *d1; |
| 497 | register char const *ep, *sp, *tp; |
| 498 | register int d, gc, i, len, md2; |
| 499 | |
| 500 | kwset = (struct kwset const *) kws; |
| 501 | len = kwset->mind; |
| 502 | |
| 503 | if (len == 0) |
| 504 | return 0; |
| 505 | if (len > size) |
| 506 | return -1; |
| 507 | if (len == 1) |
| 508 | { |
| 509 | tp = memchr (text, kwset->target[0], size); |
| 510 | return tp ? tp - text : -1; |
| 511 | } |
| 512 | |
| 513 | d1 = kwset->delta; |
| 514 | sp = kwset->target + len; |
| 515 | gc = U(sp[-2]); |
| 516 | md2 = kwset->mind2; |
| 517 | tp = text + len; |
| 518 | |
| 519 | /* Significance of 12: 1 (initial offset) + 10 (skip loop) + 1 (md2). */ |
| 520 | if (size > 12 * len) |
| 521 | /* 11 is not a bug, the initial offset happens only once. */ |
| 522 | for (ep = text + size - 11 * len;;) |
| 523 | { |
| 524 | while (tp <= ep) |
| 525 | { |
| 526 | d = d1[U(tp[-1])], tp += d; |
| 527 | d = d1[U(tp[-1])], tp += d; |
| 528 | if (d == 0) |
| 529 | goto found; |
| 530 | d = d1[U(tp[-1])], tp += d; |
| 531 | d = d1[U(tp[-1])], tp += d; |
| 532 | d = d1[U(tp[-1])], tp += d; |
| 533 | if (d == 0) |
| 534 | goto found; |
| 535 | d = d1[U(tp[-1])], tp += d; |
| 536 | d = d1[U(tp[-1])], tp += d; |
| 537 | d = d1[U(tp[-1])], tp += d; |
| 538 | if (d == 0) |
| 539 | goto found; |
| 540 | d = d1[U(tp[-1])], tp += d; |
| 541 | d = d1[U(tp[-1])], tp += d; |
| 542 | } |
| 543 | break; |
| 544 | found: |
| 545 | if (U(tp[-2]) == gc) |
| 546 | { |
| 547 | for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i) |
| 548 | ; |
| 549 | if (i > len) |
| 550 | return tp - len - text; |
| 551 | } |
| 552 | tp += md2; |
| 553 | } |
| 554 | |
| 555 | /* Now we have only a few characters left to search. We |
| 556 | carefully avoid ever producing an out-of-bounds pointer. */ |
| 557 | ep = text + size; |
| 558 | d = d1[U(tp[-1])]; |
| 559 | while (d <= ep - tp) |
| 560 | { |
| 561 | d = d1[U((tp += d)[-1])]; |
| 562 | if (d != 0) |
| 563 | continue; |
| 564 | if (U(tp[-2]) == gc) |
| 565 | { |
| 566 | for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i) |
| 567 | ; |
| 568 | if (i > len) |
| 569 | return tp - len - text; |
| 570 | } |
| 571 | d = md2; |
| 572 | } |
| 573 | |
| 574 | return -1; |
| 575 | } |
| 576 | |
| 577 | /* Hairy multiple string search. */ |
| 578 | static size_t |
| 579 | cwexec (kwset_t kws, char const *text, size_t len, struct kwsmatch *kwsmatch) |
| 580 | { |
| 581 | struct kwset const *kwset; |
| 582 | struct trie * const *next; |
| 583 | struct trie const *trie; |
| 584 | struct trie const *accept; |
| 585 | char const *beg, *lim, *mch, *lmch; |
| 586 | register unsigned char c; |
| 587 | register unsigned char const *delta; |
| 588 | register int d; |
| 589 | register char const *end, *qlim; |
| 590 | register struct tree const *tree; |
Ben Walton | 189c860 | 2015-03-02 19:22:31 +0000 | [diff] [blame] | 591 | register unsigned char const *trans; |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 592 | |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 593 | accept = NULL; |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 594 | |
| 595 | /* Initialize register copies and look for easy ways out. */ |
| 596 | kwset = (struct kwset *) kws; |
| 597 | if (len < kwset->mind) |
| 598 | return -1; |
| 599 | next = kwset->next; |
| 600 | delta = kwset->delta; |
| 601 | trans = kwset->trans; |
| 602 | lim = text + len; |
| 603 | end = text; |
| 604 | if ((d = kwset->mind) != 0) |
| 605 | mch = NULL; |
| 606 | else |
| 607 | { |
| 608 | mch = text, accept = kwset->trie; |
| 609 | goto match; |
| 610 | } |
| 611 | |
| 612 | if (len >= 4 * kwset->mind) |
| 613 | qlim = lim - 4 * kwset->mind; |
| 614 | else |
| 615 | qlim = NULL; |
| 616 | |
| 617 | while (lim - end >= d) |
| 618 | { |
| 619 | if (qlim && end <= qlim) |
| 620 | { |
| 621 | end += d - 1; |
| 622 | while ((d = delta[c = *end]) && end < qlim) |
| 623 | { |
| 624 | end += d; |
| 625 | end += delta[U(*end)]; |
| 626 | end += delta[U(*end)]; |
| 627 | } |
| 628 | ++end; |
| 629 | } |
| 630 | else |
| 631 | d = delta[c = (end += d)[-1]]; |
| 632 | if (d) |
| 633 | continue; |
| 634 | beg = end - 1; |
| 635 | trie = next[c]; |
| 636 | if (trie->accepting) |
| 637 | { |
| 638 | mch = beg; |
| 639 | accept = trie; |
| 640 | } |
| 641 | d = trie->shift; |
| 642 | while (beg > text) |
| 643 | { |
| 644 | c = trans ? trans[U(*--beg)] : *--beg; |
| 645 | tree = trie->links; |
| 646 | while (tree && c != tree->label) |
| 647 | if (c < tree->label) |
| 648 | tree = tree->llink; |
| 649 | else |
| 650 | tree = tree->rlink; |
| 651 | if (tree) |
| 652 | { |
| 653 | trie = tree->trie; |
| 654 | if (trie->accepting) |
| 655 | { |
| 656 | mch = beg; |
| 657 | accept = trie; |
| 658 | } |
| 659 | } |
| 660 | else |
| 661 | break; |
| 662 | d = trie->shift; |
| 663 | } |
| 664 | if (mch) |
| 665 | goto match; |
| 666 | } |
| 667 | return -1; |
| 668 | |
| 669 | match: |
| 670 | /* Given a known match, find the longest possible match anchored |
| 671 | at or before its starting point. This is nearly a verbatim |
| 672 | copy of the preceding main search loops. */ |
| 673 | if (lim - mch > kwset->maxd) |
| 674 | lim = mch + kwset->maxd; |
Ramsay Jones | a946ef5 | 2011-09-11 20:25:48 +0100 | [diff] [blame] | 675 | lmch = NULL; |
Fredrik Kuivinen | 05f3dbb | 2011-08-21 00:41:11 +0200 | [diff] [blame] | 676 | d = 1; |
| 677 | while (lim - end >= d) |
| 678 | { |
| 679 | if ((d = delta[c = (end += d)[-1]]) != 0) |
| 680 | continue; |
| 681 | beg = end - 1; |
| 682 | if (!(trie = next[c])) |
| 683 | { |
| 684 | d = 1; |
| 685 | continue; |
| 686 | } |
| 687 | if (trie->accepting && beg <= mch) |
| 688 | { |
| 689 | lmch = beg; |
| 690 | accept = trie; |
| 691 | } |
| 692 | d = trie->shift; |
| 693 | while (beg > text) |
| 694 | { |
| 695 | c = trans ? trans[U(*--beg)] : *--beg; |
| 696 | tree = trie->links; |
| 697 | while (tree && c != tree->label) |
| 698 | if (c < tree->label) |
| 699 | tree = tree->llink; |
| 700 | else |
| 701 | tree = tree->rlink; |
| 702 | if (tree) |
| 703 | { |
| 704 | trie = tree->trie; |
| 705 | if (trie->accepting && beg <= mch) |
| 706 | { |
| 707 | lmch = beg; |
| 708 | accept = trie; |
| 709 | } |
| 710 | } |
| 711 | else |
| 712 | break; |
| 713 | d = trie->shift; |
| 714 | } |
| 715 | if (lmch) |
| 716 | { |
| 717 | mch = lmch; |
| 718 | goto match; |
| 719 | } |
| 720 | if (!d) |
| 721 | d = 1; |
| 722 | } |
| 723 | |
| 724 | if (kwsmatch) |
| 725 | { |
| 726 | kwsmatch->index = accept->accepting / 2; |
| 727 | kwsmatch->offset[0] = mch - text; |
| 728 | kwsmatch->size[0] = accept->depth; |
| 729 | } |
| 730 | return mch - text; |
| 731 | } |
| 732 | |
| 733 | /* Search through the given text for a match of any member of the |
| 734 | given keyword set. Return a pointer to the first character of |
| 735 | the matching substring, or NULL if no match is found. If FOUNDLEN |
| 736 | is non-NULL store in the referenced location the length of the |
| 737 | matching substring. Similarly, if FOUNDIDX is non-NULL, store |
| 738 | in the referenced location the index number of the particular |
| 739 | keyword matched. */ |
| 740 | size_t |
| 741 | kwsexec (kwset_t kws, char const *text, size_t size, |
| 742 | struct kwsmatch *kwsmatch) |
| 743 | { |
| 744 | struct kwset const *kwset = (struct kwset *) kws; |
| 745 | if (kwset->words == 1 && kwset->trans == NULL) |
| 746 | { |
| 747 | size_t ret = bmexec (kws, text, size); |
| 748 | if (kwsmatch != NULL && ret != (size_t) -1) |
| 749 | { |
| 750 | kwsmatch->index = 0; |
| 751 | kwsmatch->offset[0] = ret; |
| 752 | kwsmatch->size[0] = kwset->mind; |
| 753 | } |
| 754 | return ret; |
| 755 | } |
| 756 | else |
| 757 | return cwexec(kws, text, size, kwsmatch); |
| 758 | } |
| 759 | |
| 760 | /* Free the components of the given keyword set. */ |
| 761 | void |
| 762 | kwsfree (kwset_t kws) |
| 763 | { |
| 764 | struct kwset *kwset; |
| 765 | |
| 766 | kwset = (struct kwset *) kws; |
| 767 | obstack_free(&kwset->obstack, NULL); |
| 768 | free(kws); |
| 769 | } |