]>
Commit | Line | Data |
---|---|---|
fca65d4b FK |
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 | ||
05f3dbba FK |
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 | |
d05b08cd | 21 | along with this program; if not, see <https://www.gnu.org/licenses/>. */ |
05f3dbba FK |
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 | ||
2fec81cb | 27 | /* The algorithm implemented by these routines bears a startling resemblance |
05f3dbba FK |
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 | ||
15db4e7f | 35 | #include "git-compat-util.h" |
fca65d4b | 36 | |
05f3dbba | 37 | #include "kwset.h" |
d190a087 | 38 | #include "compat/obstack.h" |
05f3dbba | 39 | |
05f3dbba | 40 | #define NCHAR (UCHAR_MAX + 1) |
08e04506 JS |
41 | /* adapter for `xmalloc()`, which takes `size_t`, not `long` */ |
42 | static void *obstack_chunk_alloc(long size) | |
43 | { | |
44 | if (size < 0) | |
45 | BUG("Cannot allocate a negative amount: %ld", size); | |
46 | return xmalloc(size); | |
47 | } | |
05f3dbba FK |
48 | #define obstack_chunk_free free |
49 | ||
50 | #define U(c) ((unsigned char) (c)) | |
51 | ||
28aed75a CW |
52 | /* For case-insensitive kwset */ |
53 | const unsigned char tolower_trans_tbl[256] = { | |
54 | 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, | |
55 | 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, | |
56 | 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, | |
57 | 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, | |
58 | ' ', '!', '"', '#', '$', '%', '&', 0x27, | |
59 | '(', ')', '*', '+', ',', '-', '.', '/', | |
60 | '0', '1', '2', '3', '4', '5', '6', '7', | |
61 | '8', '9', ':', ';', '<', '=', '>', '?', | |
62 | '@', 'a', 'b', 'c', 'd', 'e', 'f', 'g', | |
63 | 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', | |
64 | 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', | |
65 | 'x', 'y', 'z', '[', 0x5c, ']', '^', '_', | |
66 | '`', 'a', 'b', 'c', 'd', 'e', 'f', 'g', | |
67 | 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', | |
68 | 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', | |
69 | 'x', 'y', 'z', '{', '|', '}', '~', 0x7f, | |
70 | 0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, | |
71 | 0x88, 0x89, 0x8a, 0x8b, 0x8c, 0x8d, 0x8e, 0x8f, | |
72 | 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, | |
73 | 0x98, 0x99, 0x9a, 0x9b, 0x9c, 0x9d, 0x9e, 0x9f, | |
74 | 0xa0, 0xa1, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, | |
75 | 0xa8, 0xa9, 0xaa, 0xab, 0xac, 0xad, 0xae, 0xaf, | |
76 | 0xb0, 0xb1, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, 0xb7, | |
77 | 0xb8, 0xb9, 0xba, 0xbb, 0xbc, 0xbd, 0xbe, 0xbf, | |
78 | 0xc0, 0xc1, 0xc2, 0xc3, 0xc4, 0xc5, 0xc6, 0xc7, | |
79 | 0xc8, 0xc9, 0xca, 0xcb, 0xcc, 0xcd, 0xce, 0xcf, | |
80 | 0xd0, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, | |
81 | 0xd8, 0xd9, 0xda, 0xdb, 0xdc, 0xdd, 0xde, 0xdf, | |
82 | 0xe0, 0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, | |
83 | 0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee, 0xef, | |
84 | 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, | |
85 | 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, | |
86 | }; | |
87 | ||
05f3dbba FK |
88 | /* Balanced tree of edges and labels leaving a given trie node. */ |
89 | struct tree | |
90 | { | |
91 | struct tree *llink; /* Left link; MUST be first field. */ | |
92 | struct tree *rlink; /* Right link (to larger labels). */ | |
93 | struct trie *trie; /* Trie node pointed to by this edge. */ | |
94 | unsigned char label; /* Label on this edge. */ | |
95 | char balance; /* Difference in depths of subtrees. */ | |
96 | }; | |
97 | ||
98 | /* Node of a trie representing a set of reversed keywords. */ | |
99 | struct trie | |
100 | { | |
101 | unsigned int accepting; /* Word index of accepted word, or zero. */ | |
102 | struct tree *links; /* Tree of edges leaving this node. */ | |
103 | struct trie *parent; /* Parent of this node. */ | |
104 | struct trie *next; /* List of all trie nodes in level order. */ | |
105 | struct trie *fail; /* Aho-Corasick failure function. */ | |
106 | int depth; /* Depth of this node from the root. */ | |
107 | int shift; /* Shift function for search failures. */ | |
749f763d | 108 | int maxshift; /* Max shift of self and descendants. */ |
05f3dbba FK |
109 | }; |
110 | ||
111 | /* Structure returned opaquely to the caller, containing everything. */ | |
112 | struct kwset | |
113 | { | |
114 | struct obstack obstack; /* Obstack for node allocation. */ | |
115 | int words; /* Number of words in the trie. */ | |
116 | struct trie *trie; /* The trie itself. */ | |
117 | int mind; /* Minimum depth of an accepting node. */ | |
118 | int maxd; /* Maximum depth of any node. */ | |
119 | unsigned char delta[NCHAR]; /* Delta table for rapid search. */ | |
120 | struct trie *next[NCHAR]; /* Table of children of the root. */ | |
121 | char *target; /* Target string if there's only one. */ | |
122 | int mind2; /* Used in Boyer-Moore search for one string. */ | |
189c860c | 123 | unsigned char const *trans; /* Character translation table. */ |
05f3dbba FK |
124 | }; |
125 | ||
126 | /* Allocate and initialize a keyword set object, returning an opaque | |
127 | pointer to it. Return NULL if memory is not available. */ | |
128 | kwset_t | |
189c860c | 129 | kwsalloc (unsigned char const *trans) |
05f3dbba FK |
130 | { |
131 | struct kwset *kwset; | |
132 | ||
fca65d4b | 133 | kwset = (struct kwset *) xmalloc(sizeof (struct kwset)); |
05f3dbba FK |
134 | |
135 | obstack_init(&kwset->obstack); | |
136 | kwset->words = 0; | |
137 | kwset->trie | |
138 | = (struct trie *) obstack_alloc(&kwset->obstack, sizeof (struct trie)); | |
139 | if (!kwset->trie) | |
140 | { | |
141 | kwsfree((kwset_t) kwset); | |
142 | return NULL; | |
143 | } | |
144 | kwset->trie->accepting = 0; | |
145 | kwset->trie->links = NULL; | |
146 | kwset->trie->parent = NULL; | |
147 | kwset->trie->next = NULL; | |
148 | kwset->trie->fail = NULL; | |
149 | kwset->trie->depth = 0; | |
150 | kwset->trie->shift = 0; | |
151 | kwset->mind = INT_MAX; | |
152 | kwset->maxd = -1; | |
153 | kwset->target = NULL; | |
154 | kwset->trans = trans; | |
155 | ||
156 | return (kwset_t) kwset; | |
157 | } | |
158 | ||
159 | /* This upper bound is valid for CHAR_BIT >= 4 and | |
160 | exact for CHAR_BIT in { 4..11, 13, 15, 17, 19 }. */ | |
161 | #define DEPTH_SIZE (CHAR_BIT + CHAR_BIT/2) | |
162 | ||
163 | /* Add the given string to the contents of the keyword set. Return NULL | |
164 | for success, an error message otherwise. */ | |
165 | const char * | |
166 | kwsincr (kwset_t kws, char const *text, size_t len) | |
167 | { | |
168 | struct kwset *kwset; | |
169 | register struct trie *trie; | |
170 | register unsigned char label; | |
171 | register struct tree *link; | |
172 | register int depth; | |
173 | struct tree *links[DEPTH_SIZE]; | |
174 | enum { L, R } dirs[DEPTH_SIZE]; | |
175 | struct tree *t, *r, *l, *rl, *lr; | |
176 | ||
177 | kwset = (struct kwset *) kws; | |
178 | trie = kwset->trie; | |
179 | text += len; | |
180 | ||
181 | /* Descend the trie (built of reversed keywords) character-by-character, | |
182 | installing new nodes when necessary. */ | |
183 | while (len--) | |
184 | { | |
185 | label = kwset->trans ? kwset->trans[U(*--text)] : *--text; | |
186 | ||
187 | /* Descend the tree of outgoing links for this trie node, | |
188 | looking for the current character and keeping track | |
189 | of the path followed. */ | |
190 | link = trie->links; | |
191 | links[0] = (struct tree *) &trie->links; | |
192 | dirs[0] = L; | |
193 | depth = 1; | |
194 | ||
195 | while (link && label != link->label) | |
196 | { | |
197 | links[depth] = link; | |
198 | if (label < link->label) | |
199 | dirs[depth++] = L, link = link->llink; | |
200 | else | |
201 | dirs[depth++] = R, link = link->rlink; | |
202 | } | |
203 | ||
204 | /* The current character doesn't have an outgoing link at | |
205 | this trie node, so build a new trie node and install | |
206 | a link in the current trie node's tree. */ | |
207 | if (!link) | |
208 | { | |
209 | link = (struct tree *) obstack_alloc(&kwset->obstack, | |
210 | sizeof (struct tree)); | |
211 | if (!link) | |
fca65d4b | 212 | return "memory exhausted"; |
05f3dbba FK |
213 | link->llink = NULL; |
214 | link->rlink = NULL; | |
215 | link->trie = (struct trie *) obstack_alloc(&kwset->obstack, | |
216 | sizeof (struct trie)); | |
217 | if (!link->trie) | |
218 | { | |
219 | obstack_free(&kwset->obstack, link); | |
fca65d4b | 220 | return "memory exhausted"; |
05f3dbba FK |
221 | } |
222 | link->trie->accepting = 0; | |
223 | link->trie->links = NULL; | |
224 | link->trie->parent = trie; | |
225 | link->trie->next = NULL; | |
226 | link->trie->fail = NULL; | |
227 | link->trie->depth = trie->depth + 1; | |
228 | link->trie->shift = 0; | |
229 | link->label = label; | |
230 | link->balance = 0; | |
231 | ||
232 | /* Install the new tree node in its parent. */ | |
233 | if (dirs[--depth] == L) | |
234 | links[depth]->llink = link; | |
235 | else | |
236 | links[depth]->rlink = link; | |
237 | ||
238 | /* Back up the tree fixing the balance flags. */ | |
239 | while (depth && !links[depth]->balance) | |
240 | { | |
241 | if (dirs[depth] == L) | |
242 | --links[depth]->balance; | |
243 | else | |
244 | ++links[depth]->balance; | |
245 | --depth; | |
246 | } | |
247 | ||
248 | /* Rebalance the tree by pointer rotations if necessary. */ | |
249 | if (depth && ((dirs[depth] == L && --links[depth]->balance) | |
250 | || (dirs[depth] == R && ++links[depth]->balance))) | |
251 | { | |
252 | switch (links[depth]->balance) | |
253 | { | |
254 | case (char) -2: | |
255 | switch (dirs[depth + 1]) | |
256 | { | |
257 | case L: | |
258 | r = links[depth], t = r->llink, rl = t->rlink; | |
259 | t->rlink = r, r->llink = rl; | |
260 | t->balance = r->balance = 0; | |
261 | break; | |
262 | case R: | |
263 | r = links[depth], l = r->llink, t = l->rlink; | |
264 | rl = t->rlink, lr = t->llink; | |
265 | t->llink = l, l->rlink = lr, t->rlink = r, r->llink = rl; | |
266 | l->balance = t->balance != 1 ? 0 : -1; | |
267 | r->balance = t->balance != (char) -1 ? 0 : 1; | |
268 | t->balance = 0; | |
269 | break; | |
270 | default: | |
271 | abort (); | |
272 | } | |
273 | break; | |
274 | case 2: | |
275 | switch (dirs[depth + 1]) | |
276 | { | |
277 | case R: | |
278 | l = links[depth], t = l->rlink, lr = t->llink; | |
279 | t->llink = l, l->rlink = lr; | |
280 | t->balance = l->balance = 0; | |
281 | break; | |
282 | case L: | |
283 | l = links[depth], r = l->rlink, t = r->llink; | |
284 | lr = t->llink, rl = t->rlink; | |
285 | t->llink = l, l->rlink = lr, t->rlink = r, r->llink = rl; | |
286 | l->balance = t->balance != 1 ? 0 : -1; | |
287 | r->balance = t->balance != (char) -1 ? 0 : 1; | |
288 | t->balance = 0; | |
289 | break; | |
290 | default: | |
291 | abort (); | |
292 | } | |
293 | break; | |
294 | default: | |
295 | abort (); | |
296 | } | |
297 | ||
298 | if (dirs[depth - 1] == L) | |
299 | links[depth - 1]->llink = t; | |
300 | else | |
301 | links[depth - 1]->rlink = t; | |
302 | } | |
303 | } | |
304 | ||
305 | trie = link->trie; | |
306 | } | |
307 | ||
308 | /* Mark the node we finally reached as accepting, encoding the | |
309 | index number of this word in the keyword set so far. */ | |
310 | if (!trie->accepting) | |
311 | trie->accepting = 1 + 2 * kwset->words; | |
312 | ++kwset->words; | |
313 | ||
314 | /* Keep track of the longest and shortest string of the keyword set. */ | |
315 | if (trie->depth < kwset->mind) | |
316 | kwset->mind = trie->depth; | |
317 | if (trie->depth > kwset->maxd) | |
318 | kwset->maxd = trie->depth; | |
319 | ||
320 | return NULL; | |
321 | } | |
322 | ||
323 | /* Enqueue the trie nodes referenced from the given tree in the | |
324 | given queue. */ | |
325 | static void | |
326 | enqueue (struct tree *tree, struct trie **last) | |
327 | { | |
328 | if (!tree) | |
329 | return; | |
330 | enqueue(tree->llink, last); | |
331 | enqueue(tree->rlink, last); | |
332 | (*last) = (*last)->next = tree->trie; | |
333 | } | |
334 | ||
335 | /* Compute the Aho-Corasick failure function for the trie nodes referenced | |
336 | from the given tree, given the failure function for their parent as | |
337 | well as a last resort failure node. */ | |
338 | static void | |
339 | treefails (register struct tree const *tree, struct trie const *fail, | |
340 | struct trie *recourse) | |
341 | { | |
342 | register struct tree *link; | |
343 | ||
344 | if (!tree) | |
345 | return; | |
346 | ||
347 | treefails(tree->llink, fail, recourse); | |
348 | treefails(tree->rlink, fail, recourse); | |
349 | ||
350 | /* Find, in the chain of fails going back to the root, the first | |
749f763d | 351 | node that has a descendant on the current label. */ |
05f3dbba FK |
352 | while (fail) |
353 | { | |
354 | link = fail->links; | |
355 | while (link && tree->label != link->label) | |
356 | if (tree->label < link->label) | |
357 | link = link->llink; | |
358 | else | |
359 | link = link->rlink; | |
360 | if (link) | |
361 | { | |
362 | tree->trie->fail = link->trie; | |
363 | return; | |
364 | } | |
365 | fail = fail->fail; | |
366 | } | |
367 | ||
368 | tree->trie->fail = recourse; | |
369 | } | |
370 | ||
371 | /* Set delta entries for the links of the given tree such that | |
372 | the preexisting delta value is larger than the current depth. */ | |
373 | static void | |
374 | treedelta (register struct tree const *tree, | |
375 | register unsigned int depth, | |
376 | unsigned char delta[]) | |
377 | { | |
378 | if (!tree) | |
379 | return; | |
380 | treedelta(tree->llink, depth, delta); | |
381 | treedelta(tree->rlink, depth, delta); | |
382 | if (depth < delta[tree->label]) | |
383 | delta[tree->label] = depth; | |
384 | } | |
385 | ||
386 | /* Return true if A has every label in B. */ | |
387 | static int | |
388 | hasevery (register struct tree const *a, register struct tree const *b) | |
389 | { | |
390 | if (!b) | |
391 | return 1; | |
392 | if (!hasevery(a, b->llink)) | |
393 | return 0; | |
394 | if (!hasevery(a, b->rlink)) | |
395 | return 0; | |
396 | while (a && b->label != a->label) | |
397 | if (b->label < a->label) | |
398 | a = a->llink; | |
399 | else | |
400 | a = a->rlink; | |
401 | return !!a; | |
402 | } | |
403 | ||
404 | /* Compute a vector, indexed by character code, of the trie nodes | |
405 | referenced from the given tree. */ | |
406 | static void | |
407 | treenext (struct tree const *tree, struct trie *next[]) | |
408 | { | |
409 | if (!tree) | |
410 | return; | |
411 | treenext(tree->llink, next); | |
412 | treenext(tree->rlink, next); | |
413 | next[tree->label] = tree->trie; | |
414 | } | |
415 | ||
416 | /* Compute the shift for each trie node, as well as the delta | |
417 | table and next cache for the given keyword set. */ | |
418 | const char * | |
419 | kwsprep (kwset_t kws) | |
420 | { | |
421 | register struct kwset *kwset; | |
422 | register int i; | |
423 | register struct trie *curr; | |
189c860c | 424 | register unsigned char const *trans; |
05f3dbba FK |
425 | unsigned char delta[NCHAR]; |
426 | ||
427 | kwset = (struct kwset *) kws; | |
428 | ||
429 | /* Initial values for the delta table; will be changed later. The | |
430 | delta entry for a given character is the smallest depth of any | |
431 | node at which an outgoing edge is labeled by that character. */ | |
432 | memset(delta, kwset->mind < UCHAR_MAX ? kwset->mind : UCHAR_MAX, NCHAR); | |
433 | ||
434 | /* Check if we can use the simple boyer-moore algorithm, instead | |
435 | of the hairy commentz-walter algorithm. */ | |
436 | if (kwset->words == 1 && kwset->trans == NULL) | |
437 | { | |
438 | char c; | |
439 | ||
440 | /* Looking for just one string. Extract it from the trie. */ | |
441 | kwset->target = obstack_alloc(&kwset->obstack, kwset->mind); | |
442 | if (!kwset->target) | |
fca65d4b | 443 | return "memory exhausted"; |
05f3dbba FK |
444 | for (i = kwset->mind - 1, curr = kwset->trie; i >= 0; --i) |
445 | { | |
446 | kwset->target[i] = curr->links->label; | |
447 | curr = curr->links->trie; | |
448 | } | |
449 | /* Build the Boyer Moore delta. Boy that's easy compared to CW. */ | |
450 | for (i = 0; i < kwset->mind; ++i) | |
451 | delta[U(kwset->target[i])] = kwset->mind - (i + 1); | |
452 | /* Find the minimal delta2 shift that we might make after | |
453 | a backwards match has failed. */ | |
454 | c = kwset->target[kwset->mind - 1]; | |
455 | for (i = kwset->mind - 2; i >= 0; --i) | |
456 | if (kwset->target[i] == c) | |
457 | break; | |
458 | kwset->mind2 = kwset->mind - (i + 1); | |
459 | } | |
460 | else | |
461 | { | |
462 | register struct trie *fail; | |
463 | struct trie *last, *next[NCHAR]; | |
464 | ||
465 | /* Traverse the nodes of the trie in level order, simultaneously | |
466 | computing the delta table, failure function, and shift function. */ | |
467 | for (curr = last = kwset->trie; curr; curr = curr->next) | |
468 | { | |
749f763d | 469 | /* Enqueue the immediate descendants in the level order queue. */ |
05f3dbba FK |
470 | enqueue(curr->links, &last); |
471 | ||
472 | curr->shift = kwset->mind; | |
473 | curr->maxshift = kwset->mind; | |
474 | ||
749f763d | 475 | /* Update the delta table for the descendants of this node. */ |
05f3dbba FK |
476 | treedelta(curr->links, curr->depth, delta); |
477 | ||
2fec81cb | 478 | /* Compute the failure function for the descendants of this node. */ |
05f3dbba FK |
479 | treefails(curr->links, curr->fail, kwset->trie); |
480 | ||
481 | /* Update the shifts at each node in the current node's chain | |
482 | of fails back to the root. */ | |
483 | for (fail = curr->fail; fail; fail = fail->fail) | |
484 | { | |
485 | /* If the current node has some outgoing edge that the fail | |
486 | doesn't, then the shift at the fail should be no larger | |
487 | than the difference of their depths. */ | |
488 | if (!hasevery(fail->links, curr->links)) | |
489 | if (curr->depth - fail->depth < fail->shift) | |
490 | fail->shift = curr->depth - fail->depth; | |
491 | ||
492 | /* If the current node is accepting then the shift at the | |
749f763d | 493 | fail and its descendants should be no larger than the |
05f3dbba FK |
494 | difference of their depths. */ |
495 | if (curr->accepting && fail->maxshift > curr->depth - fail->depth) | |
496 | fail->maxshift = curr->depth - fail->depth; | |
497 | } | |
498 | } | |
499 | ||
500 | /* Traverse the trie in level order again, fixing up all nodes whose | |
501 | shift exceeds their inherited maxshift. */ | |
502 | for (curr = kwset->trie->next; curr; curr = curr->next) | |
503 | { | |
504 | if (curr->maxshift > curr->parent->maxshift) | |
505 | curr->maxshift = curr->parent->maxshift; | |
506 | if (curr->shift > curr->maxshift) | |
507 | curr->shift = curr->maxshift; | |
508 | } | |
509 | ||
510 | /* Create a vector, indexed by character code, of the outgoing links | |
511 | from the root node. */ | |
512 | for (i = 0; i < NCHAR; ++i) | |
513 | next[i] = NULL; | |
514 | treenext(kwset->trie->links, next); | |
515 | ||
afe8a907 | 516 | if ((trans = kwset->trans)) |
05f3dbba FK |
517 | for (i = 0; i < NCHAR; ++i) |
518 | kwset->next[i] = next[U(trans[i])]; | |
519 | else | |
921d49be | 520 | COPY_ARRAY(kwset->next, next, NCHAR); |
05f3dbba FK |
521 | } |
522 | ||
523 | /* Fix things up for any translation table. */ | |
afe8a907 | 524 | if ((trans = kwset->trans)) |
05f3dbba FK |
525 | for (i = 0; i < NCHAR; ++i) |
526 | kwset->delta[i] = delta[U(trans[i])]; | |
527 | else | |
528 | memcpy(kwset->delta, delta, NCHAR); | |
529 | ||
530 | return NULL; | |
531 | } | |
532 | ||
533 | /* Fast boyer-moore search. */ | |
534 | static size_t | |
535 | bmexec (kwset_t kws, char const *text, size_t size) | |
536 | { | |
537 | struct kwset const *kwset; | |
538 | register unsigned char const *d1; | |
539 | register char const *ep, *sp, *tp; | |
540 | register int d, gc, i, len, md2; | |
541 | ||
542 | kwset = (struct kwset const *) kws; | |
543 | len = kwset->mind; | |
544 | ||
545 | if (len == 0) | |
546 | return 0; | |
547 | if (len > size) | |
548 | return -1; | |
549 | if (len == 1) | |
550 | { | |
551 | tp = memchr (text, kwset->target[0], size); | |
552 | return tp ? tp - text : -1; | |
553 | } | |
554 | ||
555 | d1 = kwset->delta; | |
556 | sp = kwset->target + len; | |
557 | gc = U(sp[-2]); | |
558 | md2 = kwset->mind2; | |
559 | tp = text + len; | |
560 | ||
561 | /* Significance of 12: 1 (initial offset) + 10 (skip loop) + 1 (md2). */ | |
562 | if (size > 12 * len) | |
563 | /* 11 is not a bug, the initial offset happens only once. */ | |
564 | for (ep = text + size - 11 * len;;) | |
565 | { | |
566 | while (tp <= ep) | |
567 | { | |
568 | d = d1[U(tp[-1])], tp += d; | |
569 | d = d1[U(tp[-1])], tp += d; | |
570 | if (d == 0) | |
571 | goto found; | |
572 | d = d1[U(tp[-1])], tp += d; | |
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 | } | |
585 | break; | |
586 | found: | |
587 | if (U(tp[-2]) == gc) | |
588 | { | |
589 | for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i) | |
590 | ; | |
591 | if (i > len) | |
592 | return tp - len - text; | |
593 | } | |
594 | tp += md2; | |
595 | } | |
596 | ||
597 | /* Now we have only a few characters left to search. We | |
598 | carefully avoid ever producing an out-of-bounds pointer. */ | |
599 | ep = text + size; | |
600 | d = d1[U(tp[-1])]; | |
601 | while (d <= ep - tp) | |
602 | { | |
603 | d = d1[U((tp += d)[-1])]; | |
604 | if (d != 0) | |
605 | continue; | |
606 | if (U(tp[-2]) == gc) | |
607 | { | |
608 | for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i) | |
609 | ; | |
610 | if (i > len) | |
611 | return tp - len - text; | |
612 | } | |
613 | d = md2; | |
614 | } | |
615 | ||
616 | return -1; | |
617 | } | |
618 | ||
619 | /* Hairy multiple string search. */ | |
620 | static size_t | |
621 | cwexec (kwset_t kws, char const *text, size_t len, struct kwsmatch *kwsmatch) | |
622 | { | |
623 | struct kwset const *kwset; | |
624 | struct trie * const *next; | |
625 | struct trie const *trie; | |
626 | struct trie const *accept; | |
627 | char const *beg, *lim, *mch, *lmch; | |
628 | register unsigned char c; | |
629 | register unsigned char const *delta; | |
630 | register int d; | |
631 | register char const *end, *qlim; | |
632 | register struct tree const *tree; | |
189c860c | 633 | register unsigned char const *trans; |
05f3dbba | 634 | |
05f3dbba | 635 | accept = NULL; |
05f3dbba FK |
636 | |
637 | /* Initialize register copies and look for easy ways out. */ | |
638 | kwset = (struct kwset *) kws; | |
639 | if (len < kwset->mind) | |
640 | return -1; | |
641 | next = kwset->next; | |
642 | delta = kwset->delta; | |
643 | trans = kwset->trans; | |
644 | lim = text + len; | |
645 | end = text; | |
646 | if ((d = kwset->mind) != 0) | |
647 | mch = NULL; | |
648 | else | |
649 | { | |
650 | mch = text, accept = kwset->trie; | |
651 | goto match; | |
652 | } | |
653 | ||
654 | if (len >= 4 * kwset->mind) | |
655 | qlim = lim - 4 * kwset->mind; | |
656 | else | |
657 | qlim = NULL; | |
658 | ||
659 | while (lim - end >= d) | |
660 | { | |
661 | if (qlim && end <= qlim) | |
662 | { | |
663 | end += d - 1; | |
664 | while ((d = delta[c = *end]) && end < qlim) | |
665 | { | |
666 | end += d; | |
667 | end += delta[U(*end)]; | |
668 | end += delta[U(*end)]; | |
669 | } | |
670 | ++end; | |
671 | } | |
672 | else | |
673 | d = delta[c = (end += d)[-1]]; | |
674 | if (d) | |
675 | continue; | |
676 | beg = end - 1; | |
677 | trie = next[c]; | |
678 | if (trie->accepting) | |
679 | { | |
680 | mch = beg; | |
681 | accept = trie; | |
682 | } | |
683 | d = trie->shift; | |
684 | while (beg > text) | |
685 | { | |
686 | c = trans ? trans[U(*--beg)] : *--beg; | |
687 | tree = trie->links; | |
688 | while (tree && c != tree->label) | |
689 | if (c < tree->label) | |
690 | tree = tree->llink; | |
691 | else | |
692 | tree = tree->rlink; | |
693 | if (tree) | |
694 | { | |
695 | trie = tree->trie; | |
696 | if (trie->accepting) | |
697 | { | |
698 | mch = beg; | |
699 | accept = trie; | |
700 | } | |
701 | } | |
702 | else | |
703 | break; | |
704 | d = trie->shift; | |
705 | } | |
706 | if (mch) | |
707 | goto match; | |
708 | } | |
709 | return -1; | |
710 | ||
711 | match: | |
712 | /* Given a known match, find the longest possible match anchored | |
713 | at or before its starting point. This is nearly a verbatim | |
714 | copy of the preceding main search loops. */ | |
715 | if (lim - mch > kwset->maxd) | |
716 | lim = mch + kwset->maxd; | |
a946ef55 | 717 | lmch = NULL; |
05f3dbba FK |
718 | d = 1; |
719 | while (lim - end >= d) | |
720 | { | |
721 | if ((d = delta[c = (end += d)[-1]]) != 0) | |
722 | continue; | |
723 | beg = end - 1; | |
724 | if (!(trie = next[c])) | |
725 | { | |
726 | d = 1; | |
727 | continue; | |
728 | } | |
729 | if (trie->accepting && beg <= mch) | |
730 | { | |
731 | lmch = beg; | |
732 | accept = trie; | |
733 | } | |
734 | d = trie->shift; | |
735 | while (beg > text) | |
736 | { | |
737 | c = trans ? trans[U(*--beg)] : *--beg; | |
738 | tree = trie->links; | |
739 | while (tree && c != tree->label) | |
740 | if (c < tree->label) | |
741 | tree = tree->llink; | |
742 | else | |
743 | tree = tree->rlink; | |
744 | if (tree) | |
745 | { | |
746 | trie = tree->trie; | |
747 | if (trie->accepting && beg <= mch) | |
748 | { | |
749 | lmch = beg; | |
750 | accept = trie; | |
751 | } | |
752 | } | |
753 | else | |
754 | break; | |
755 | d = trie->shift; | |
756 | } | |
757 | if (lmch) | |
758 | { | |
759 | mch = lmch; | |
760 | goto match; | |
761 | } | |
762 | if (!d) | |
763 | d = 1; | |
764 | } | |
765 | ||
766 | if (kwsmatch) | |
767 | { | |
768 | kwsmatch->index = accept->accepting / 2; | |
769 | kwsmatch->offset[0] = mch - text; | |
770 | kwsmatch->size[0] = accept->depth; | |
771 | } | |
772 | return mch - text; | |
773 | } | |
774 | ||
775 | /* Search through the given text for a match of any member of the | |
776 | given keyword set. Return a pointer to the first character of | |
777 | the matching substring, or NULL if no match is found. If FOUNDLEN | |
778 | is non-NULL store in the referenced location the length of the | |
779 | matching substring. Similarly, if FOUNDIDX is non-NULL, store | |
780 | in the referenced location the index number of the particular | |
781 | keyword matched. */ | |
782 | size_t | |
783 | kwsexec (kwset_t kws, char const *text, size_t size, | |
784 | struct kwsmatch *kwsmatch) | |
785 | { | |
786 | struct kwset const *kwset = (struct kwset *) kws; | |
787 | if (kwset->words == 1 && kwset->trans == NULL) | |
788 | { | |
789 | size_t ret = bmexec (kws, text, size); | |
790 | if (kwsmatch != NULL && ret != (size_t) -1) | |
791 | { | |
792 | kwsmatch->index = 0; | |
793 | kwsmatch->offset[0] = ret; | |
794 | kwsmatch->size[0] = kwset->mind; | |
795 | } | |
796 | return ret; | |
797 | } | |
798 | else | |
799 | return cwexec(kws, text, size, kwsmatch); | |
800 | } | |
801 | ||
802 | /* Free the components of the given keyword set. */ | |
803 | void | |
804 | kwsfree (kwset_t kws) | |
805 | { | |
806 | struct kwset *kwset; | |
807 | ||
808 | kwset = (struct kwset *) kws; | |
809 | obstack_free(&kwset->obstack, NULL); | |
810 | free(kws); | |
811 | } |