| 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 | |
| 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 |
| 21 | along with this program; if not, see <https://www.gnu.org/licenses/>. */ |
| 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 | |
| 27 | /* The algorithm implemented by these routines bears a startling resemblance |
| 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 | |
| 35 | #define DISABLE_SIGN_COMPARE_WARNINGS |
| 36 | |
| 37 | #include "git-compat-util.h" |
| 38 | |
| 39 | #include "kwset.h" |
| 40 | #include "compat/obstack.h" |
| 41 | |
| 42 | #define NCHAR (UCHAR_MAX + 1) |
| 43 | /* adapter for `xmalloc()`, which takes `size_t`, not `long` */ |
| 44 | static void *obstack_chunk_alloc(long size) |
| 45 | { |
| 46 | if (size < 0) |
| 47 | BUG("Cannot allocate a negative amount: %ld", size); |
| 48 | return xmalloc(size); |
| 49 | } |
| 50 | #define obstack_chunk_free free |
| 51 | |
| 52 | #define U(c) ((unsigned char) (c)) |
| 53 | |
| 54 | /* For case-insensitive kwset */ |
| 55 | const unsigned char tolower_trans_tbl[256] = { |
| 56 | 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, |
| 57 | 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, |
| 58 | 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, |
| 59 | 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, |
| 60 | ' ', '!', '"', '#', '$', '%', '&', 0x27, |
| 61 | '(', ')', '*', '+', ',', '-', '.', '/', |
| 62 | '0', '1', '2', '3', '4', '5', '6', '7', |
| 63 | '8', '9', ':', ';', '<', '=', '>', '?', |
| 64 | '@', 'a', 'b', 'c', 'd', 'e', 'f', 'g', |
| 65 | 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', |
| 66 | 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', |
| 67 | 'x', 'y', 'z', '[', 0x5c, ']', '^', '_', |
| 68 | '`', 'a', 'b', 'c', 'd', 'e', 'f', 'g', |
| 69 | 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', |
| 70 | 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', |
| 71 | 'x', 'y', 'z', '{', '|', '}', '~', 0x7f, |
| 72 | 0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, |
| 73 | 0x88, 0x89, 0x8a, 0x8b, 0x8c, 0x8d, 0x8e, 0x8f, |
| 74 | 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, |
| 75 | 0x98, 0x99, 0x9a, 0x9b, 0x9c, 0x9d, 0x9e, 0x9f, |
| 76 | 0xa0, 0xa1, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, |
| 77 | 0xa8, 0xa9, 0xaa, 0xab, 0xac, 0xad, 0xae, 0xaf, |
| 78 | 0xb0, 0xb1, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, 0xb7, |
| 79 | 0xb8, 0xb9, 0xba, 0xbb, 0xbc, 0xbd, 0xbe, 0xbf, |
| 80 | 0xc0, 0xc1, 0xc2, 0xc3, 0xc4, 0xc5, 0xc6, 0xc7, |
| 81 | 0xc8, 0xc9, 0xca, 0xcb, 0xcc, 0xcd, 0xce, 0xcf, |
| 82 | 0xd0, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, |
| 83 | 0xd8, 0xd9, 0xda, 0xdb, 0xdc, 0xdd, 0xde, 0xdf, |
| 84 | 0xe0, 0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, |
| 85 | 0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee, 0xef, |
| 86 | 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, |
| 87 | 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, |
| 88 | }; |
| 89 | |
| 90 | /* Balanced tree of edges and labels leaving a given trie node. */ |
| 91 | struct tree |
| 92 | { |
| 93 | struct tree *llink; /* Left link; MUST be first field. */ |
| 94 | struct tree *rlink; /* Right link (to larger labels). */ |
| 95 | struct trie *trie; /* Trie node pointed to by this edge. */ |
| 96 | unsigned char label; /* Label on this edge. */ |
| 97 | char balance; /* Difference in depths of subtrees. */ |
| 98 | }; |
| 99 | |
| 100 | /* Node of a trie representing a set of reversed keywords. */ |
| 101 | struct trie |
| 102 | { |
| 103 | unsigned int accepting; /* Word index of accepted word, or zero. */ |
| 104 | struct tree *links; /* Tree of edges leaving this node. */ |
| 105 | struct trie *parent; /* Parent of this node. */ |
| 106 | struct trie *next; /* List of all trie nodes in level order. */ |
| 107 | struct trie *fail; /* Aho-Corasick failure function. */ |
| 108 | int depth; /* Depth of this node from the root. */ |
| 109 | int shift; /* Shift function for search failures. */ |
| 110 | int maxshift; /* Max shift of self and descendants. */ |
| 111 | }; |
| 112 | |
| 113 | /* Structure returned opaquely to the caller, containing everything. */ |
| 114 | struct kwset |
| 115 | { |
| 116 | struct obstack obstack; /* Obstack for node allocation. */ |
| 117 | int words; /* Number of words in the trie. */ |
| 118 | struct trie *trie; /* The trie itself. */ |
| 119 | int mind; /* Minimum depth of an accepting node. */ |
| 120 | int maxd; /* Maximum depth of any node. */ |
| 121 | unsigned char delta[NCHAR]; /* Delta table for rapid search. */ |
| 122 | struct trie *next[NCHAR]; /* Table of children of the root. */ |
| 123 | char *target; /* Target string if there's only one. */ |
| 124 | int mind2; /* Used in Boyer-Moore search for one string. */ |
| 125 | unsigned char const *trans; /* Character translation table. */ |
| 126 | }; |
| 127 | |
| 128 | /* Allocate and initialize a keyword set object, returning an opaque |
| 129 | pointer to it. Return NULL if memory is not available. */ |
| 130 | kwset_t |
| 131 | kwsalloc (unsigned char const *trans) |
| 132 | { |
| 133 | struct kwset *kwset; |
| 134 | |
| 135 | kwset = (struct kwset *) xmalloc(sizeof (struct kwset)); |
| 136 | |
| 137 | obstack_init(&kwset->obstack); |
| 138 | kwset->words = 0; |
| 139 | kwset->trie |
| 140 | = (struct trie *) obstack_alloc(&kwset->obstack, sizeof (struct trie)); |
| 141 | if (!kwset->trie) |
| 142 | { |
| 143 | kwsfree((kwset_t) kwset); |
| 144 | return NULL; |
| 145 | } |
| 146 | kwset->trie->accepting = 0; |
| 147 | kwset->trie->links = NULL; |
| 148 | kwset->trie->parent = NULL; |
| 149 | kwset->trie->next = NULL; |
| 150 | kwset->trie->fail = NULL; |
| 151 | kwset->trie->depth = 0; |
| 152 | kwset->trie->shift = 0; |
| 153 | kwset->mind = INT_MAX; |
| 154 | kwset->maxd = -1; |
| 155 | kwset->target = NULL; |
| 156 | kwset->trans = trans; |
| 157 | |
| 158 | return (kwset_t) kwset; |
| 159 | } |
| 160 | |
| 161 | /* This upper bound is valid for CHAR_BIT >= 4 and |
| 162 | exact for CHAR_BIT in { 4..11, 13, 15, 17, 19 }. */ |
| 163 | #define DEPTH_SIZE (CHAR_BIT + CHAR_BIT/2) |
| 164 | |
| 165 | /* Add the given string to the contents of the keyword set. Return NULL |
| 166 | for success, an error message otherwise. */ |
| 167 | const char * |
| 168 | kwsincr (kwset_t kws, char const *text, size_t len) |
| 169 | { |
| 170 | struct kwset *kwset; |
| 171 | register struct trie *trie; |
| 172 | register unsigned char label; |
| 173 | register struct tree *link; |
| 174 | register int depth; |
| 175 | struct tree *links[DEPTH_SIZE]; |
| 176 | enum { L, R } dirs[DEPTH_SIZE]; |
| 177 | struct tree *t, *r, *l, *rl, *lr; |
| 178 | |
| 179 | kwset = (struct kwset *) kws; |
| 180 | trie = kwset->trie; |
| 181 | text += len; |
| 182 | |
| 183 | /* Descend the trie (built of reversed keywords) character-by-character, |
| 184 | installing new nodes when necessary. */ |
| 185 | while (len--) |
| 186 | { |
| 187 | label = kwset->trans ? kwset->trans[U(*--text)] : *--text; |
| 188 | |
| 189 | /* Descend the tree of outgoing links for this trie node, |
| 190 | looking for the current character and keeping track |
| 191 | of the path followed. */ |
| 192 | link = trie->links; |
| 193 | links[0] = (struct tree *) &trie->links; |
| 194 | dirs[0] = L; |
| 195 | depth = 1; |
| 196 | |
| 197 | while (link && label != link->label) |
| 198 | { |
| 199 | links[depth] = link; |
| 200 | if (label < link->label) { |
| 201 | dirs[depth++] = L; |
| 202 | link = link->llink; |
| 203 | } else { |
| 204 | dirs[depth++] = R; |
| 205 | link = link->rlink; |
| 206 | } |
| 207 | } |
| 208 | |
| 209 | /* The current character doesn't have an outgoing link at |
| 210 | this trie node, so build a new trie node and install |
| 211 | a link in the current trie node's tree. */ |
| 212 | if (!link) |
| 213 | { |
| 214 | link = (struct tree *) obstack_alloc(&kwset->obstack, |
| 215 | sizeof (struct tree)); |
| 216 | if (!link) |
| 217 | return "memory exhausted"; |
| 218 | link->llink = NULL; |
| 219 | link->rlink = NULL; |
| 220 | link->trie = (struct trie *) obstack_alloc(&kwset->obstack, |
| 221 | sizeof (struct trie)); |
| 222 | if (!link->trie) |
| 223 | { |
| 224 | obstack_free(&kwset->obstack, link); |
| 225 | return "memory exhausted"; |
| 226 | } |
| 227 | link->trie->accepting = 0; |
| 228 | link->trie->links = NULL; |
| 229 | link->trie->parent = trie; |
| 230 | link->trie->next = NULL; |
| 231 | link->trie->fail = NULL; |
| 232 | link->trie->depth = trie->depth + 1; |
| 233 | link->trie->shift = 0; |
| 234 | link->label = label; |
| 235 | link->balance = 0; |
| 236 | |
| 237 | /* Install the new tree node in its parent. */ |
| 238 | if (dirs[--depth] == L) |
| 239 | links[depth]->llink = link; |
| 240 | else |
| 241 | links[depth]->rlink = link; |
| 242 | |
| 243 | /* Back up the tree fixing the balance flags. */ |
| 244 | while (depth && !links[depth]->balance) |
| 245 | { |
| 246 | if (dirs[depth] == L) |
| 247 | --links[depth]->balance; |
| 248 | else |
| 249 | ++links[depth]->balance; |
| 250 | --depth; |
| 251 | } |
| 252 | |
| 253 | /* Rebalance the tree by pointer rotations if necessary. */ |
| 254 | if (depth && ((dirs[depth] == L && --links[depth]->balance) |
| 255 | || (dirs[depth] == R && ++links[depth]->balance))) |
| 256 | { |
| 257 | switch (links[depth]->balance) |
| 258 | { |
| 259 | case (char) -2: |
| 260 | switch (dirs[depth + 1]) |
| 261 | { |
| 262 | case L: |
| 263 | r = links[depth]; t = r->llink; rl = t->rlink; |
| 264 | t->rlink = r; r->llink = rl; |
| 265 | t->balance = r->balance = 0; |
| 266 | break; |
| 267 | case R: |
| 268 | r = links[depth]; l = r->llink; t = l->rlink; |
| 269 | rl = t->rlink; lr = t->llink; |
| 270 | t->llink = l; l->rlink = lr; t->rlink = r; r->llink = rl; |
| 271 | l->balance = t->balance != 1 ? 0 : -1; |
| 272 | r->balance = t->balance != (char) -1 ? 0 : 1; |
| 273 | t->balance = 0; |
| 274 | break; |
| 275 | default: |
| 276 | abort (); |
| 277 | } |
| 278 | break; |
| 279 | case 2: |
| 280 | switch (dirs[depth + 1]) |
| 281 | { |
| 282 | case R: |
| 283 | l = links[depth]; t = l->rlink; lr = t->llink; |
| 284 | t->llink = l; l->rlink = lr; |
| 285 | t->balance = l->balance = 0; |
| 286 | break; |
| 287 | case L: |
| 288 | l = links[depth]; r = l->rlink; t = r->llink; |
| 289 | lr = t->llink; rl = t->rlink; |
| 290 | t->llink = l; l->rlink = lr; t->rlink = r; r->llink = rl; |
| 291 | l->balance = t->balance != 1 ? 0 : -1; |
| 292 | r->balance = t->balance != (char) -1 ? 0 : 1; |
| 293 | t->balance = 0; |
| 294 | break; |
| 295 | default: |
| 296 | abort (); |
| 297 | } |
| 298 | break; |
| 299 | default: |
| 300 | abort (); |
| 301 | } |
| 302 | |
| 303 | if (dirs[depth - 1] == L) |
| 304 | links[depth - 1]->llink = t; |
| 305 | else |
| 306 | links[depth - 1]->rlink = t; |
| 307 | } |
| 308 | } |
| 309 | |
| 310 | trie = link->trie; |
| 311 | } |
| 312 | |
| 313 | /* Mark the node we finally reached as accepting, encoding the |
| 314 | index number of this word in the keyword set so far. */ |
| 315 | if (!trie->accepting) |
| 316 | trie->accepting = 1 + 2 * kwset->words; |
| 317 | ++kwset->words; |
| 318 | |
| 319 | /* Keep track of the longest and shortest string of the keyword set. */ |
| 320 | if (trie->depth < kwset->mind) |
| 321 | kwset->mind = trie->depth; |
| 322 | if (trie->depth > kwset->maxd) |
| 323 | kwset->maxd = trie->depth; |
| 324 | |
| 325 | return NULL; |
| 326 | } |
| 327 | |
| 328 | /* Enqueue the trie nodes referenced from the given tree in the |
| 329 | given queue. */ |
| 330 | static void |
| 331 | enqueue (struct tree *tree, struct trie **last) |
| 332 | { |
| 333 | if (!tree) |
| 334 | return; |
| 335 | enqueue(tree->llink, last); |
| 336 | enqueue(tree->rlink, last); |
| 337 | (*last) = (*last)->next = tree->trie; |
| 338 | } |
| 339 | |
| 340 | /* Compute the Aho-Corasick failure function for the trie nodes referenced |
| 341 | from the given tree, given the failure function for their parent as |
| 342 | well as a last resort failure node. */ |
| 343 | static void |
| 344 | treefails (register struct tree const *tree, struct trie const *fail, |
| 345 | struct trie *recourse) |
| 346 | { |
| 347 | register struct tree *link; |
| 348 | |
| 349 | if (!tree) |
| 350 | return; |
| 351 | |
| 352 | treefails(tree->llink, fail, recourse); |
| 353 | treefails(tree->rlink, fail, recourse); |
| 354 | |
| 355 | /* Find, in the chain of fails going back to the root, the first |
| 356 | node that has a descendant on the current label. */ |
| 357 | while (fail) |
| 358 | { |
| 359 | link = fail->links; |
| 360 | while (link && tree->label != link->label) |
| 361 | if (tree->label < link->label) |
| 362 | link = link->llink; |
| 363 | else |
| 364 | link = link->rlink; |
| 365 | if (link) |
| 366 | { |
| 367 | tree->trie->fail = link->trie; |
| 368 | return; |
| 369 | } |
| 370 | fail = fail->fail; |
| 371 | } |
| 372 | |
| 373 | tree->trie->fail = recourse; |
| 374 | } |
| 375 | |
| 376 | /* Set delta entries for the links of the given tree such that |
| 377 | the preexisting delta value is larger than the current depth. */ |
| 378 | static void |
| 379 | treedelta (register struct tree const *tree, |
| 380 | register unsigned int depth, |
| 381 | unsigned char delta[]) |
| 382 | { |
| 383 | if (!tree) |
| 384 | return; |
| 385 | treedelta(tree->llink, depth, delta); |
| 386 | treedelta(tree->rlink, depth, delta); |
| 387 | if (depth < delta[tree->label]) |
| 388 | delta[tree->label] = depth; |
| 389 | } |
| 390 | |
| 391 | /* Return true if A has every label in B. */ |
| 392 | static int |
| 393 | hasevery (register struct tree const *a, register struct tree const *b) |
| 394 | { |
| 395 | if (!b) |
| 396 | return 1; |
| 397 | if (!hasevery(a, b->llink)) |
| 398 | return 0; |
| 399 | if (!hasevery(a, b->rlink)) |
| 400 | return 0; |
| 401 | while (a && b->label != a->label) |
| 402 | if (b->label < a->label) |
| 403 | a = a->llink; |
| 404 | else |
| 405 | a = a->rlink; |
| 406 | return !!a; |
| 407 | } |
| 408 | |
| 409 | /* Compute a vector, indexed by character code, of the trie nodes |
| 410 | referenced from the given tree. */ |
| 411 | static void |
| 412 | treenext (struct tree const *tree, struct trie *next[]) |
| 413 | { |
| 414 | if (!tree) |
| 415 | return; |
| 416 | treenext(tree->llink, next); |
| 417 | treenext(tree->rlink, next); |
| 418 | next[tree->label] = tree->trie; |
| 419 | } |
| 420 | |
| 421 | /* Compute the shift for each trie node, as well as the delta |
| 422 | table and next cache for the given keyword set. */ |
| 423 | const char * |
| 424 | kwsprep (kwset_t kws) |
| 425 | { |
| 426 | register struct kwset *kwset; |
| 427 | register int i; |
| 428 | register struct trie *curr; |
| 429 | register unsigned char const *trans; |
| 430 | unsigned char delta[NCHAR]; |
| 431 | |
| 432 | kwset = (struct kwset *) kws; |
| 433 | |
| 434 | /* Initial values for the delta table; will be changed later. The |
| 435 | delta entry for a given character is the smallest depth of any |
| 436 | node at which an outgoing edge is labeled by that character. */ |
| 437 | memset(delta, kwset->mind < UCHAR_MAX ? kwset->mind : UCHAR_MAX, NCHAR); |
| 438 | |
| 439 | /* Check if we can use the simple boyer-moore algorithm, instead |
| 440 | of the hairy commentz-walter algorithm. */ |
| 441 | if (kwset->words == 1 && kwset->trans == NULL) |
| 442 | { |
| 443 | char c; |
| 444 | |
| 445 | /* Looking for just one string. Extract it from the trie. */ |
| 446 | kwset->target = obstack_alloc(&kwset->obstack, kwset->mind); |
| 447 | if (!kwset->target) |
| 448 | return "memory exhausted"; |
| 449 | for (i = kwset->mind - 1, curr = kwset->trie; i >= 0; --i) |
| 450 | { |
| 451 | kwset->target[i] = curr->links->label; |
| 452 | curr = curr->links->trie; |
| 453 | } |
| 454 | /* Build the Boyer Moore delta. Boy that's easy compared to CW. */ |
| 455 | for (i = 0; i < kwset->mind; ++i) |
| 456 | delta[U(kwset->target[i])] = kwset->mind - (i + 1); |
| 457 | /* Find the minimal delta2 shift that we might make after |
| 458 | a backwards match has failed. */ |
| 459 | c = kwset->target[kwset->mind - 1]; |
| 460 | for (i = kwset->mind - 2; i >= 0; --i) |
| 461 | if (kwset->target[i] == c) |
| 462 | break; |
| 463 | kwset->mind2 = kwset->mind - (i + 1); |
| 464 | } |
| 465 | else |
| 466 | { |
| 467 | register struct trie *fail; |
| 468 | struct trie *last, *next[NCHAR]; |
| 469 | |
| 470 | /* Traverse the nodes of the trie in level order, simultaneously |
| 471 | computing the delta table, failure function, and shift function. */ |
| 472 | for (curr = last = kwset->trie; curr; curr = curr->next) |
| 473 | { |
| 474 | /* Enqueue the immediate descendants in the level order queue. */ |
| 475 | enqueue(curr->links, &last); |
| 476 | |
| 477 | curr->shift = kwset->mind; |
| 478 | curr->maxshift = kwset->mind; |
| 479 | |
| 480 | /* Update the delta table for the descendants of this node. */ |
| 481 | treedelta(curr->links, curr->depth, delta); |
| 482 | |
| 483 | /* Compute the failure function for the descendants of this node. */ |
| 484 | treefails(curr->links, curr->fail, kwset->trie); |
| 485 | |
| 486 | /* Update the shifts at each node in the current node's chain |
| 487 | of fails back to the root. */ |
| 488 | for (fail = curr->fail; fail; fail = fail->fail) |
| 489 | { |
| 490 | /* If the current node has some outgoing edge that the fail |
| 491 | doesn't, then the shift at the fail should be no larger |
| 492 | than the difference of their depths. */ |
| 493 | if (!hasevery(fail->links, curr->links)) |
| 494 | if (curr->depth - fail->depth < fail->shift) |
| 495 | fail->shift = curr->depth - fail->depth; |
| 496 | |
| 497 | /* If the current node is accepting then the shift at the |
| 498 | fail and its descendants should be no larger than the |
| 499 | difference of their depths. */ |
| 500 | if (curr->accepting && fail->maxshift > curr->depth - fail->depth) |
| 501 | fail->maxshift = curr->depth - fail->depth; |
| 502 | } |
| 503 | } |
| 504 | |
| 505 | /* Traverse the trie in level order again, fixing up all nodes whose |
| 506 | shift exceeds their inherited maxshift. */ |
| 507 | for (curr = kwset->trie->next; curr; curr = curr->next) |
| 508 | { |
| 509 | if (curr->maxshift > curr->parent->maxshift) |
| 510 | curr->maxshift = curr->parent->maxshift; |
| 511 | if (curr->shift > curr->maxshift) |
| 512 | curr->shift = curr->maxshift; |
| 513 | } |
| 514 | |
| 515 | /* Create a vector, indexed by character code, of the outgoing links |
| 516 | from the root node. */ |
| 517 | for (i = 0; i < NCHAR; ++i) |
| 518 | next[i] = NULL; |
| 519 | treenext(kwset->trie->links, next); |
| 520 | |
| 521 | if ((trans = kwset->trans)) |
| 522 | for (i = 0; i < NCHAR; ++i) |
| 523 | kwset->next[i] = next[U(trans[i])]; |
| 524 | else |
| 525 | COPY_ARRAY(kwset->next, next, NCHAR); |
| 526 | } |
| 527 | |
| 528 | /* Fix things up for any translation table. */ |
| 529 | if ((trans = kwset->trans)) |
| 530 | for (i = 0; i < NCHAR; ++i) |
| 531 | kwset->delta[i] = delta[U(trans[i])]; |
| 532 | else |
| 533 | memcpy(kwset->delta, delta, NCHAR); |
| 534 | |
| 535 | return NULL; |
| 536 | } |
| 537 | |
| 538 | /* Fast boyer-moore search. */ |
| 539 | static size_t |
| 540 | bmexec (kwset_t kws, char const *text, size_t size) |
| 541 | { |
| 542 | struct kwset const *kwset; |
| 543 | register unsigned char const *d1; |
| 544 | register char const *ep, *sp, *tp; |
| 545 | register int d, gc, i, len, md2; |
| 546 | |
| 547 | kwset = (struct kwset const *) kws; |
| 548 | len = kwset->mind; |
| 549 | |
| 550 | if (len == 0) |
| 551 | return 0; |
| 552 | if (len > size) |
| 553 | return -1; |
| 554 | if (len == 1) |
| 555 | { |
| 556 | tp = memchr (text, kwset->target[0], size); |
| 557 | return tp ? tp - text : -1; |
| 558 | } |
| 559 | |
| 560 | d1 = kwset->delta; |
| 561 | sp = kwset->target + len; |
| 562 | gc = U(sp[-2]); |
| 563 | md2 = kwset->mind2; |
| 564 | tp = text + len; |
| 565 | |
| 566 | /* Significance of 12: 1 (initial offset) + 10 (skip loop) + 1 (md2). */ |
| 567 | if (size > 12 * len) |
| 568 | /* 11 is not a bug, the initial offset happens only once. */ |
| 569 | for (ep = text + size - 11 * len;;) |
| 570 | { |
| 571 | while (tp <= ep) |
| 572 | { |
| 573 | d = d1[U(tp[-1])]; tp += d; |
| 574 | d = d1[U(tp[-1])]; tp += d; |
| 575 | if (d == 0) |
| 576 | goto found; |
| 577 | d = d1[U(tp[-1])]; tp += d; |
| 578 | d = d1[U(tp[-1])]; tp += d; |
| 579 | d = d1[U(tp[-1])]; tp += d; |
| 580 | if (d == 0) |
| 581 | goto found; |
| 582 | d = d1[U(tp[-1])]; tp += d; |
| 583 | d = d1[U(tp[-1])]; tp += d; |
| 584 | d = d1[U(tp[-1])]; tp += d; |
| 585 | if (d == 0) |
| 586 | goto found; |
| 587 | d = d1[U(tp[-1])]; tp += d; |
| 588 | d = d1[U(tp[-1])]; tp += d; |
| 589 | } |
| 590 | break; |
| 591 | found: |
| 592 | if (U(tp[-2]) == gc) |
| 593 | { |
| 594 | for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i) |
| 595 | ; |
| 596 | if (i > len) |
| 597 | return tp - len - text; |
| 598 | } |
| 599 | tp += md2; |
| 600 | } |
| 601 | |
| 602 | /* Now we have only a few characters left to search. We |
| 603 | carefully avoid ever producing an out-of-bounds pointer. */ |
| 604 | ep = text + size; |
| 605 | d = d1[U(tp[-1])]; |
| 606 | while (d <= ep - tp) |
| 607 | { |
| 608 | d = d1[U((tp += d)[-1])]; |
| 609 | if (d != 0) |
| 610 | continue; |
| 611 | if (U(tp[-2]) == gc) |
| 612 | { |
| 613 | for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i) |
| 614 | ; |
| 615 | if (i > len) |
| 616 | return tp - len - text; |
| 617 | } |
| 618 | d = md2; |
| 619 | } |
| 620 | |
| 621 | return -1; |
| 622 | } |
| 623 | |
| 624 | /* Hairy multiple string search. */ |
| 625 | static size_t |
| 626 | cwexec (kwset_t kws, char const *text, size_t len, struct kwsmatch *kwsmatch) |
| 627 | { |
| 628 | struct kwset const *kwset; |
| 629 | struct trie * const *next; |
| 630 | struct trie const *trie; |
| 631 | struct trie const *accept; |
| 632 | char const *beg, *lim, *mch, *lmch; |
| 633 | register unsigned char c; |
| 634 | register unsigned char const *delta; |
| 635 | register int d; |
| 636 | register char const *end, *qlim; |
| 637 | register struct tree const *tree; |
| 638 | register unsigned char const *trans; |
| 639 | |
| 640 | accept = NULL; |
| 641 | |
| 642 | /* Initialize register copies and look for easy ways out. */ |
| 643 | kwset = (struct kwset *) kws; |
| 644 | if (len < kwset->mind) |
| 645 | return -1; |
| 646 | next = kwset->next; |
| 647 | delta = kwset->delta; |
| 648 | trans = kwset->trans; |
| 649 | lim = text + len; |
| 650 | end = text; |
| 651 | if ((d = kwset->mind) != 0) |
| 652 | mch = NULL; |
| 653 | else |
| 654 | { |
| 655 | mch = text; |
| 656 | accept = kwset->trie; |
| 657 | goto match; |
| 658 | } |
| 659 | |
| 660 | if (len >= 4 * kwset->mind) |
| 661 | qlim = lim - 4 * kwset->mind; |
| 662 | else |
| 663 | qlim = NULL; |
| 664 | |
| 665 | while (lim - end >= d) |
| 666 | { |
| 667 | if (qlim && end <= qlim) |
| 668 | { |
| 669 | end += d - 1; |
| 670 | while ((d = delta[c = *end]) && end < qlim) |
| 671 | { |
| 672 | end += d; |
| 673 | end += delta[U(*end)]; |
| 674 | end += delta[U(*end)]; |
| 675 | } |
| 676 | ++end; |
| 677 | } |
| 678 | else |
| 679 | d = delta[c = (end += d)[-1]]; |
| 680 | if (d) |
| 681 | continue; |
| 682 | beg = end - 1; |
| 683 | trie = next[c]; |
| 684 | if (trie->accepting) |
| 685 | { |
| 686 | mch = beg; |
| 687 | accept = trie; |
| 688 | } |
| 689 | d = trie->shift; |
| 690 | while (beg > text) |
| 691 | { |
| 692 | c = trans ? trans[U(*--beg)] : *--beg; |
| 693 | tree = trie->links; |
| 694 | while (tree && c != tree->label) |
| 695 | if (c < tree->label) |
| 696 | tree = tree->llink; |
| 697 | else |
| 698 | tree = tree->rlink; |
| 699 | if (tree) |
| 700 | { |
| 701 | trie = tree->trie; |
| 702 | if (trie->accepting) |
| 703 | { |
| 704 | mch = beg; |
| 705 | accept = trie; |
| 706 | } |
| 707 | } |
| 708 | else |
| 709 | break; |
| 710 | d = trie->shift; |
| 711 | } |
| 712 | if (mch) |
| 713 | goto match; |
| 714 | } |
| 715 | return -1; |
| 716 | |
| 717 | match: |
| 718 | /* Given a known match, find the longest possible match anchored |
| 719 | at or before its starting point. This is nearly a verbatim |
| 720 | copy of the preceding main search loops. */ |
| 721 | if (lim - mch > kwset->maxd) |
| 722 | lim = mch + kwset->maxd; |
| 723 | lmch = NULL; |
| 724 | d = 1; |
| 725 | while (lim - end >= d) |
| 726 | { |
| 727 | if ((d = delta[c = (end += d)[-1]]) != 0) |
| 728 | continue; |
| 729 | beg = end - 1; |
| 730 | if (!(trie = next[c])) |
| 731 | { |
| 732 | d = 1; |
| 733 | continue; |
| 734 | } |
| 735 | if (trie->accepting && beg <= mch) |
| 736 | { |
| 737 | lmch = beg; |
| 738 | accept = trie; |
| 739 | } |
| 740 | d = trie->shift; |
| 741 | while (beg > text) |
| 742 | { |
| 743 | c = trans ? trans[U(*--beg)] : *--beg; |
| 744 | tree = trie->links; |
| 745 | while (tree && c != tree->label) |
| 746 | if (c < tree->label) |
| 747 | tree = tree->llink; |
| 748 | else |
| 749 | tree = tree->rlink; |
| 750 | if (tree) |
| 751 | { |
| 752 | trie = tree->trie; |
| 753 | if (trie->accepting && beg <= mch) |
| 754 | { |
| 755 | lmch = beg; |
| 756 | accept = trie; |
| 757 | } |
| 758 | } |
| 759 | else |
| 760 | break; |
| 761 | d = trie->shift; |
| 762 | } |
| 763 | if (lmch) |
| 764 | { |
| 765 | mch = lmch; |
| 766 | goto match; |
| 767 | } |
| 768 | if (!d) |
| 769 | d = 1; |
| 770 | } |
| 771 | |
| 772 | if (kwsmatch) |
| 773 | { |
| 774 | kwsmatch->index = accept->accepting / 2; |
| 775 | kwsmatch->offset[0] = mch - text; |
| 776 | kwsmatch->size[0] = accept->depth; |
| 777 | } |
| 778 | return mch - text; |
| 779 | } |
| 780 | |
| 781 | /* Search through the given text for a match of any member of the |
| 782 | given keyword set. Return a pointer to the first character of |
| 783 | the matching substring, or NULL if no match is found. If FOUNDLEN |
| 784 | is non-NULL store in the referenced location the length of the |
| 785 | matching substring. Similarly, if FOUNDIDX is non-NULL, store |
| 786 | in the referenced location the index number of the particular |
| 787 | keyword matched. */ |
| 788 | size_t |
| 789 | kwsexec (kwset_t kws, char const *text, size_t size, |
| 790 | struct kwsmatch *kwsmatch) |
| 791 | { |
| 792 | struct kwset const *kwset = (struct kwset *) kws; |
| 793 | if (kwset->words == 1 && kwset->trans == NULL) |
| 794 | { |
| 795 | size_t ret = bmexec (kws, text, size); |
| 796 | if (kwsmatch != NULL && ret != (size_t) -1) |
| 797 | { |
| 798 | kwsmatch->index = 0; |
| 799 | kwsmatch->offset[0] = ret; |
| 800 | kwsmatch->size[0] = kwset->mind; |
| 801 | } |
| 802 | return ret; |
| 803 | } |
| 804 | else |
| 805 | return cwexec(kws, text, size, kwsmatch); |
| 806 | } |
| 807 | |
| 808 | /* Free the components of the given keyword set. */ |
| 809 | void |
| 810 | kwsfree (kwset_t kws) |
| 811 | { |
| 812 | struct kwset *kwset; |
| 813 | |
| 814 | kwset = (struct kwset *) kws; |
| 815 | obstack_free(&kwset->obstack, NULL); |
| 816 | free(kws); |
| 817 | } |