Line | Count | Source (jump to first uncovered line) |
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 | | #include "git-compat-util.h" |
36 | | |
37 | | #include "kwset.h" |
38 | | #include "compat/obstack.h" |
39 | | |
40 | 0 | #define NCHAR (UCHAR_MAX + 1) |
41 | | /* adapter for `xmalloc()`, which takes `size_t`, not `long` */ |
42 | | static void *obstack_chunk_alloc(long size) |
43 | 0 | { |
44 | 0 | if (size < 0) |
45 | 0 | BUG("Cannot allocate a negative amount: %ld", size); |
46 | 0 | return xmalloc(size); |
47 | 0 | } |
48 | 0 | #define obstack_chunk_free free |
49 | | |
50 | 0 | #define U(c) ((unsigned char) (c)) |
51 | | |
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 | | |
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. */ |
108 | | int maxshift; /* Max shift of self and descendants. */ |
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. */ |
123 | | unsigned char const *trans; /* Character translation table. */ |
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 |
129 | | kwsalloc (unsigned char const *trans) |
130 | 0 | { |
131 | 0 | struct kwset *kwset; |
132 | |
|
133 | 0 | kwset = (struct kwset *) xmalloc(sizeof (struct kwset)); |
134 | |
|
135 | 0 | obstack_init(&kwset->obstack); |
136 | 0 | kwset->words = 0; |
137 | 0 | kwset->trie |
138 | 0 | = (struct trie *) obstack_alloc(&kwset->obstack, sizeof (struct trie)); |
139 | 0 | if (!kwset->trie) |
140 | 0 | { |
141 | 0 | kwsfree((kwset_t) kwset); |
142 | 0 | return NULL; |
143 | 0 | } |
144 | 0 | kwset->trie->accepting = 0; |
145 | 0 | kwset->trie->links = NULL; |
146 | 0 | kwset->trie->parent = NULL; |
147 | 0 | kwset->trie->next = NULL; |
148 | 0 | kwset->trie->fail = NULL; |
149 | 0 | kwset->trie->depth = 0; |
150 | 0 | kwset->trie->shift = 0; |
151 | 0 | kwset->mind = INT_MAX; |
152 | 0 | kwset->maxd = -1; |
153 | 0 | kwset->target = NULL; |
154 | 0 | kwset->trans = trans; |
155 | |
|
156 | 0 | return (kwset_t) kwset; |
157 | 0 | } |
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 | 0 | { |
168 | 0 | struct kwset *kwset; |
169 | 0 | register struct trie *trie; |
170 | 0 | register unsigned char label; |
171 | 0 | register struct tree *link; |
172 | 0 | register int depth; |
173 | 0 | struct tree *links[DEPTH_SIZE]; |
174 | 0 | enum { L, R } dirs[DEPTH_SIZE]; |
175 | 0 | struct tree *t, *r, *l, *rl, *lr; |
176 | |
|
177 | 0 | kwset = (struct kwset *) kws; |
178 | 0 | trie = kwset->trie; |
179 | 0 | text += len; |
180 | | |
181 | | /* Descend the trie (built of reversed keywords) character-by-character, |
182 | | installing new nodes when necessary. */ |
183 | 0 | while (len--) |
184 | 0 | { |
185 | 0 | 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 | 0 | link = trie->links; |
191 | 0 | links[0] = (struct tree *) &trie->links; |
192 | 0 | dirs[0] = L; |
193 | 0 | depth = 1; |
194 | |
|
195 | 0 | while (link && label != link->label) |
196 | 0 | { |
197 | 0 | links[depth] = link; |
198 | 0 | if (label < link->label) |
199 | 0 | dirs[depth++] = L, link = link->llink; |
200 | 0 | else |
201 | 0 | dirs[depth++] = R, link = link->rlink; |
202 | 0 | } |
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 | 0 | if (!link) |
208 | 0 | { |
209 | 0 | link = (struct tree *) obstack_alloc(&kwset->obstack, |
210 | 0 | sizeof (struct tree)); |
211 | 0 | if (!link) |
212 | 0 | return "memory exhausted"; |
213 | 0 | link->llink = NULL; |
214 | 0 | link->rlink = NULL; |
215 | 0 | link->trie = (struct trie *) obstack_alloc(&kwset->obstack, |
216 | 0 | sizeof (struct trie)); |
217 | 0 | if (!link->trie) |
218 | 0 | { |
219 | 0 | obstack_free(&kwset->obstack, link); |
220 | 0 | return "memory exhausted"; |
221 | 0 | } |
222 | 0 | link->trie->accepting = 0; |
223 | 0 | link->trie->links = NULL; |
224 | 0 | link->trie->parent = trie; |
225 | 0 | link->trie->next = NULL; |
226 | 0 | link->trie->fail = NULL; |
227 | 0 | link->trie->depth = trie->depth + 1; |
228 | 0 | link->trie->shift = 0; |
229 | 0 | link->label = label; |
230 | 0 | link->balance = 0; |
231 | | |
232 | | /* Install the new tree node in its parent. */ |
233 | 0 | if (dirs[--depth] == L) |
234 | 0 | links[depth]->llink = link; |
235 | 0 | else |
236 | 0 | links[depth]->rlink = link; |
237 | | |
238 | | /* Back up the tree fixing the balance flags. */ |
239 | 0 | while (depth && !links[depth]->balance) |
240 | 0 | { |
241 | 0 | if (dirs[depth] == L) |
242 | 0 | --links[depth]->balance; |
243 | 0 | else |
244 | 0 | ++links[depth]->balance; |
245 | 0 | --depth; |
246 | 0 | } |
247 | | |
248 | | /* Rebalance the tree by pointer rotations if necessary. */ |
249 | 0 | if (depth && ((dirs[depth] == L && --links[depth]->balance) |
250 | 0 | || (dirs[depth] == R && ++links[depth]->balance))) |
251 | 0 | { |
252 | 0 | switch (links[depth]->balance) |
253 | 0 | { |
254 | 0 | case (char) -2: |
255 | 0 | switch (dirs[depth + 1]) |
256 | 0 | { |
257 | 0 | case L: |
258 | 0 | r = links[depth], t = r->llink, rl = t->rlink; |
259 | 0 | t->rlink = r, r->llink = rl; |
260 | 0 | t->balance = r->balance = 0; |
261 | 0 | break; |
262 | 0 | case R: |
263 | 0 | r = links[depth], l = r->llink, t = l->rlink; |
264 | 0 | rl = t->rlink, lr = t->llink; |
265 | 0 | t->llink = l, l->rlink = lr, t->rlink = r, r->llink = rl; |
266 | 0 | l->balance = t->balance != 1 ? 0 : -1; |
267 | 0 | r->balance = t->balance != (char) -1 ? 0 : 1; |
268 | 0 | t->balance = 0; |
269 | 0 | break; |
270 | 0 | default: |
271 | 0 | abort (); |
272 | 0 | } |
273 | 0 | break; |
274 | 0 | case 2: |
275 | 0 | switch (dirs[depth + 1]) |
276 | 0 | { |
277 | 0 | case R: |
278 | 0 | l = links[depth], t = l->rlink, lr = t->llink; |
279 | 0 | t->llink = l, l->rlink = lr; |
280 | 0 | t->balance = l->balance = 0; |
281 | 0 | break; |
282 | 0 | case L: |
283 | 0 | l = links[depth], r = l->rlink, t = r->llink; |
284 | 0 | lr = t->llink, rl = t->rlink; |
285 | 0 | t->llink = l, l->rlink = lr, t->rlink = r, r->llink = rl; |
286 | 0 | l->balance = t->balance != 1 ? 0 : -1; |
287 | 0 | r->balance = t->balance != (char) -1 ? 0 : 1; |
288 | 0 | t->balance = 0; |
289 | 0 | break; |
290 | 0 | default: |
291 | 0 | abort (); |
292 | 0 | } |
293 | 0 | break; |
294 | 0 | default: |
295 | 0 | abort (); |
296 | 0 | } |
297 | | |
298 | 0 | if (dirs[depth - 1] == L) |
299 | 0 | links[depth - 1]->llink = t; |
300 | 0 | else |
301 | 0 | links[depth - 1]->rlink = t; |
302 | 0 | } |
303 | 0 | } |
304 | | |
305 | 0 | trie = link->trie; |
306 | 0 | } |
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 | 0 | if (!trie->accepting) |
311 | 0 | trie->accepting = 1 + 2 * kwset->words; |
312 | 0 | ++kwset->words; |
313 | | |
314 | | /* Keep track of the longest and shortest string of the keyword set. */ |
315 | 0 | if (trie->depth < kwset->mind) |
316 | 0 | kwset->mind = trie->depth; |
317 | 0 | if (trie->depth > kwset->maxd) |
318 | 0 | kwset->maxd = trie->depth; |
319 | |
|
320 | 0 | return NULL; |
321 | 0 | } |
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 | 0 | { |
328 | 0 | if (!tree) |
329 | 0 | return; |
330 | 0 | enqueue(tree->llink, last); |
331 | 0 | enqueue(tree->rlink, last); |
332 | 0 | (*last) = (*last)->next = tree->trie; |
333 | 0 | } |
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 | 0 | { |
342 | 0 | register struct tree *link; |
343 | |
|
344 | 0 | if (!tree) |
345 | 0 | return; |
346 | | |
347 | 0 | treefails(tree->llink, fail, recourse); |
348 | 0 | treefails(tree->rlink, fail, recourse); |
349 | | |
350 | | /* Find, in the chain of fails going back to the root, the first |
351 | | node that has a descendant on the current label. */ |
352 | 0 | while (fail) |
353 | 0 | { |
354 | 0 | link = fail->links; |
355 | 0 | while (link && tree->label != link->label) |
356 | 0 | if (tree->label < link->label) |
357 | 0 | link = link->llink; |
358 | 0 | else |
359 | 0 | link = link->rlink; |
360 | 0 | if (link) |
361 | 0 | { |
362 | 0 | tree->trie->fail = link->trie; |
363 | 0 | return; |
364 | 0 | } |
365 | 0 | fail = fail->fail; |
366 | 0 | } |
367 | | |
368 | 0 | tree->trie->fail = recourse; |
369 | 0 | } |
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 | 0 | { |
378 | 0 | if (!tree) |
379 | 0 | return; |
380 | 0 | treedelta(tree->llink, depth, delta); |
381 | 0 | treedelta(tree->rlink, depth, delta); |
382 | 0 | if (depth < delta[tree->label]) |
383 | 0 | delta[tree->label] = depth; |
384 | 0 | } |
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 | 0 | { |
390 | 0 | if (!b) |
391 | 0 | return 1; |
392 | 0 | if (!hasevery(a, b->llink)) |
393 | 0 | return 0; |
394 | 0 | if (!hasevery(a, b->rlink)) |
395 | 0 | return 0; |
396 | 0 | while (a && b->label != a->label) |
397 | 0 | if (b->label < a->label) |
398 | 0 | a = a->llink; |
399 | 0 | else |
400 | 0 | a = a->rlink; |
401 | 0 | return !!a; |
402 | 0 | } |
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 | 0 | { |
409 | 0 | if (!tree) |
410 | 0 | return; |
411 | 0 | treenext(tree->llink, next); |
412 | 0 | treenext(tree->rlink, next); |
413 | 0 | next[tree->label] = tree->trie; |
414 | 0 | } |
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 | 0 | { |
421 | 0 | register struct kwset *kwset; |
422 | 0 | register int i; |
423 | 0 | register struct trie *curr; |
424 | 0 | register unsigned char const *trans; |
425 | 0 | unsigned char delta[NCHAR]; |
426 | |
|
427 | 0 | 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 | 0 | 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 | 0 | if (kwset->words == 1 && kwset->trans == NULL) |
437 | 0 | { |
438 | 0 | char c; |
439 | | |
440 | | /* Looking for just one string. Extract it from the trie. */ |
441 | 0 | kwset->target = obstack_alloc(&kwset->obstack, kwset->mind); |
442 | 0 | if (!kwset->target) |
443 | 0 | return "memory exhausted"; |
444 | 0 | for (i = kwset->mind - 1, curr = kwset->trie; i >= 0; --i) |
445 | 0 | { |
446 | 0 | kwset->target[i] = curr->links->label; |
447 | 0 | curr = curr->links->trie; |
448 | 0 | } |
449 | | /* Build the Boyer Moore delta. Boy that's easy compared to CW. */ |
450 | 0 | for (i = 0; i < kwset->mind; ++i) |
451 | 0 | 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 | 0 | c = kwset->target[kwset->mind - 1]; |
455 | 0 | for (i = kwset->mind - 2; i >= 0; --i) |
456 | 0 | if (kwset->target[i] == c) |
457 | 0 | break; |
458 | 0 | kwset->mind2 = kwset->mind - (i + 1); |
459 | 0 | } |
460 | 0 | else |
461 | 0 | { |
462 | 0 | register struct trie *fail; |
463 | 0 | 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 | 0 | for (curr = last = kwset->trie; curr; curr = curr->next) |
468 | 0 | { |
469 | | /* Enqueue the immediate descendants in the level order queue. */ |
470 | 0 | enqueue(curr->links, &last); |
471 | |
|
472 | 0 | curr->shift = kwset->mind; |
473 | 0 | curr->maxshift = kwset->mind; |
474 | | |
475 | | /* Update the delta table for the descendants of this node. */ |
476 | 0 | treedelta(curr->links, curr->depth, delta); |
477 | | |
478 | | /* Compute the failure function for the descendants of this node. */ |
479 | 0 | 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 | 0 | for (fail = curr->fail; fail; fail = fail->fail) |
484 | 0 | { |
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 | 0 | if (!hasevery(fail->links, curr->links)) |
489 | 0 | if (curr->depth - fail->depth < fail->shift) |
490 | 0 | fail->shift = curr->depth - fail->depth; |
491 | | |
492 | | /* If the current node is accepting then the shift at the |
493 | | fail and its descendants should be no larger than the |
494 | | difference of their depths. */ |
495 | 0 | if (curr->accepting && fail->maxshift > curr->depth - fail->depth) |
496 | 0 | fail->maxshift = curr->depth - fail->depth; |
497 | 0 | } |
498 | 0 | } |
499 | | |
500 | | /* Traverse the trie in level order again, fixing up all nodes whose |
501 | | shift exceeds their inherited maxshift. */ |
502 | 0 | for (curr = kwset->trie->next; curr; curr = curr->next) |
503 | 0 | { |
504 | 0 | if (curr->maxshift > curr->parent->maxshift) |
505 | 0 | curr->maxshift = curr->parent->maxshift; |
506 | 0 | if (curr->shift > curr->maxshift) |
507 | 0 | curr->shift = curr->maxshift; |
508 | 0 | } |
509 | | |
510 | | /* Create a vector, indexed by character code, of the outgoing links |
511 | | from the root node. */ |
512 | 0 | for (i = 0; i < NCHAR; ++i) |
513 | 0 | next[i] = NULL; |
514 | 0 | treenext(kwset->trie->links, next); |
515 | |
|
516 | 0 | if ((trans = kwset->trans)) |
517 | 0 | for (i = 0; i < NCHAR; ++i) |
518 | 0 | kwset->next[i] = next[U(trans[i])]; |
519 | 0 | else |
520 | 0 | COPY_ARRAY(kwset->next, next, NCHAR); |
521 | 0 | } |
522 | | |
523 | | /* Fix things up for any translation table. */ |
524 | 0 | if ((trans = kwset->trans)) |
525 | 0 | for (i = 0; i < NCHAR; ++i) |
526 | 0 | kwset->delta[i] = delta[U(trans[i])]; |
527 | 0 | else |
528 | 0 | memcpy(kwset->delta, delta, NCHAR); |
529 | |
|
530 | 0 | return NULL; |
531 | 0 | } |
532 | | |
533 | | /* Fast boyer-moore search. */ |
534 | | static size_t |
535 | | bmexec (kwset_t kws, char const *text, size_t size) |
536 | 0 | { |
537 | 0 | struct kwset const *kwset; |
538 | 0 | register unsigned char const *d1; |
539 | 0 | register char const *ep, *sp, *tp; |
540 | 0 | register int d, gc, i, len, md2; |
541 | |
|
542 | 0 | kwset = (struct kwset const *) kws; |
543 | 0 | len = kwset->mind; |
544 | |
|
545 | 0 | if (len == 0) |
546 | 0 | return 0; |
547 | 0 | if (len > size) |
548 | 0 | return -1; |
549 | 0 | if (len == 1) |
550 | 0 | { |
551 | 0 | tp = memchr (text, kwset->target[0], size); |
552 | 0 | return tp ? tp - text : -1; |
553 | 0 | } |
554 | | |
555 | 0 | d1 = kwset->delta; |
556 | 0 | sp = kwset->target + len; |
557 | 0 | gc = U(sp[-2]); |
558 | 0 | md2 = kwset->mind2; |
559 | 0 | tp = text + len; |
560 | | |
561 | | /* Significance of 12: 1 (initial offset) + 10 (skip loop) + 1 (md2). */ |
562 | 0 | if (size > 12 * len) |
563 | | /* 11 is not a bug, the initial offset happens only once. */ |
564 | 0 | for (ep = text + size - 11 * len;;) |
565 | 0 | { |
566 | 0 | while (tp <= ep) |
567 | 0 | { |
568 | 0 | d = d1[U(tp[-1])], tp += d; |
569 | 0 | d = d1[U(tp[-1])], tp += d; |
570 | 0 | if (d == 0) |
571 | 0 | goto found; |
572 | 0 | d = d1[U(tp[-1])], tp += d; |
573 | 0 | d = d1[U(tp[-1])], tp += d; |
574 | 0 | d = d1[U(tp[-1])], tp += d; |
575 | 0 | if (d == 0) |
576 | 0 | goto found; |
577 | 0 | d = d1[U(tp[-1])], tp += d; |
578 | 0 | d = d1[U(tp[-1])], tp += d; |
579 | 0 | d = d1[U(tp[-1])], tp += d; |
580 | 0 | if (d == 0) |
581 | 0 | goto found; |
582 | 0 | d = d1[U(tp[-1])], tp += d; |
583 | 0 | d = d1[U(tp[-1])], tp += d; |
584 | 0 | } |
585 | 0 | break; |
586 | 0 | found: |
587 | 0 | if (U(tp[-2]) == gc) |
588 | 0 | { |
589 | 0 | for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i) |
590 | 0 | ; |
591 | 0 | if (i > len) |
592 | 0 | return tp - len - text; |
593 | 0 | } |
594 | 0 | tp += md2; |
595 | 0 | } |
596 | | |
597 | | /* Now we have only a few characters left to search. We |
598 | | carefully avoid ever producing an out-of-bounds pointer. */ |
599 | 0 | ep = text + size; |
600 | 0 | d = d1[U(tp[-1])]; |
601 | 0 | while (d <= ep - tp) |
602 | 0 | { |
603 | 0 | d = d1[U((tp += d)[-1])]; |
604 | 0 | if (d != 0) |
605 | 0 | continue; |
606 | 0 | if (U(tp[-2]) == gc) |
607 | 0 | { |
608 | 0 | for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i) |
609 | 0 | ; |
610 | 0 | if (i > len) |
611 | 0 | return tp - len - text; |
612 | 0 | } |
613 | 0 | d = md2; |
614 | 0 | } |
615 | | |
616 | 0 | return -1; |
617 | 0 | } |
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 | 0 | { |
623 | 0 | struct kwset const *kwset; |
624 | 0 | struct trie * const *next; |
625 | 0 | struct trie const *trie; |
626 | 0 | struct trie const *accept; |
627 | 0 | char const *beg, *lim, *mch, *lmch; |
628 | 0 | register unsigned char c; |
629 | 0 | register unsigned char const *delta; |
630 | 0 | register int d; |
631 | 0 | register char const *end, *qlim; |
632 | 0 | register struct tree const *tree; |
633 | 0 | register unsigned char const *trans; |
634 | |
|
635 | 0 | accept = NULL; |
636 | | |
637 | | /* Initialize register copies and look for easy ways out. */ |
638 | 0 | kwset = (struct kwset *) kws; |
639 | 0 | if (len < kwset->mind) |
640 | 0 | return -1; |
641 | 0 | next = kwset->next; |
642 | 0 | delta = kwset->delta; |
643 | 0 | trans = kwset->trans; |
644 | 0 | lim = text + len; |
645 | 0 | end = text; |
646 | 0 | if ((d = kwset->mind) != 0) |
647 | 0 | mch = NULL; |
648 | 0 | else |
649 | 0 | { |
650 | 0 | mch = text, accept = kwset->trie; |
651 | 0 | goto match; |
652 | 0 | } |
653 | | |
654 | 0 | if (len >= 4 * kwset->mind) |
655 | 0 | qlim = lim - 4 * kwset->mind; |
656 | 0 | else |
657 | 0 | qlim = NULL; |
658 | |
|
659 | 0 | while (lim - end >= d) |
660 | 0 | { |
661 | 0 | if (qlim && end <= qlim) |
662 | 0 | { |
663 | 0 | end += d - 1; |
664 | 0 | while ((d = delta[c = *end]) && end < qlim) |
665 | 0 | { |
666 | 0 | end += d; |
667 | 0 | end += delta[U(*end)]; |
668 | 0 | end += delta[U(*end)]; |
669 | 0 | } |
670 | 0 | ++end; |
671 | 0 | } |
672 | 0 | else |
673 | 0 | d = delta[c = (end += d)[-1]]; |
674 | 0 | if (d) |
675 | 0 | continue; |
676 | 0 | beg = end - 1; |
677 | 0 | trie = next[c]; |
678 | 0 | if (trie->accepting) |
679 | 0 | { |
680 | 0 | mch = beg; |
681 | 0 | accept = trie; |
682 | 0 | } |
683 | 0 | d = trie->shift; |
684 | 0 | while (beg > text) |
685 | 0 | { |
686 | 0 | c = trans ? trans[U(*--beg)] : *--beg; |
687 | 0 | tree = trie->links; |
688 | 0 | while (tree && c != tree->label) |
689 | 0 | if (c < tree->label) |
690 | 0 | tree = tree->llink; |
691 | 0 | else |
692 | 0 | tree = tree->rlink; |
693 | 0 | if (tree) |
694 | 0 | { |
695 | 0 | trie = tree->trie; |
696 | 0 | if (trie->accepting) |
697 | 0 | { |
698 | 0 | mch = beg; |
699 | 0 | accept = trie; |
700 | 0 | } |
701 | 0 | } |
702 | 0 | else |
703 | 0 | break; |
704 | 0 | d = trie->shift; |
705 | 0 | } |
706 | 0 | if (mch) |
707 | 0 | goto match; |
708 | 0 | } |
709 | 0 | return -1; |
710 | | |
711 | 0 | 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 | 0 | if (lim - mch > kwset->maxd) |
716 | 0 | lim = mch + kwset->maxd; |
717 | 0 | lmch = NULL; |
718 | 0 | d = 1; |
719 | 0 | while (lim - end >= d) |
720 | 0 | { |
721 | 0 | if ((d = delta[c = (end += d)[-1]]) != 0) |
722 | 0 | continue; |
723 | 0 | beg = end - 1; |
724 | 0 | if (!(trie = next[c])) |
725 | 0 | { |
726 | 0 | d = 1; |
727 | 0 | continue; |
728 | 0 | } |
729 | 0 | if (trie->accepting && beg <= mch) |
730 | 0 | { |
731 | 0 | lmch = beg; |
732 | 0 | accept = trie; |
733 | 0 | } |
734 | 0 | d = trie->shift; |
735 | 0 | while (beg > text) |
736 | 0 | { |
737 | 0 | c = trans ? trans[U(*--beg)] : *--beg; |
738 | 0 | tree = trie->links; |
739 | 0 | while (tree && c != tree->label) |
740 | 0 | if (c < tree->label) |
741 | 0 | tree = tree->llink; |
742 | 0 | else |
743 | 0 | tree = tree->rlink; |
744 | 0 | if (tree) |
745 | 0 | { |
746 | 0 | trie = tree->trie; |
747 | 0 | if (trie->accepting && beg <= mch) |
748 | 0 | { |
749 | 0 | lmch = beg; |
750 | 0 | accept = trie; |
751 | 0 | } |
752 | 0 | } |
753 | 0 | else |
754 | 0 | break; |
755 | 0 | d = trie->shift; |
756 | 0 | } |
757 | 0 | if (lmch) |
758 | 0 | { |
759 | 0 | mch = lmch; |
760 | 0 | goto match; |
761 | 0 | } |
762 | 0 | if (!d) |
763 | 0 | d = 1; |
764 | 0 | } |
765 | | |
766 | 0 | if (kwsmatch) |
767 | 0 | { |
768 | 0 | kwsmatch->index = accept->accepting / 2; |
769 | 0 | kwsmatch->offset[0] = mch - text; |
770 | 0 | kwsmatch->size[0] = accept->depth; |
771 | 0 | } |
772 | 0 | return mch - text; |
773 | 0 | } |
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 | 0 | { |
786 | 0 | struct kwset const *kwset = (struct kwset *) kws; |
787 | 0 | if (kwset->words == 1 && kwset->trans == NULL) |
788 | 0 | { |
789 | 0 | size_t ret = bmexec (kws, text, size); |
790 | 0 | if (kwsmatch != NULL && ret != (size_t) -1) |
791 | 0 | { |
792 | 0 | kwsmatch->index = 0; |
793 | 0 | kwsmatch->offset[0] = ret; |
794 | 0 | kwsmatch->size[0] = kwset->mind; |
795 | 0 | } |
796 | 0 | return ret; |
797 | 0 | } |
798 | 0 | else |
799 | 0 | return cwexec(kws, text, size, kwsmatch); |
800 | 0 | } |
801 | | |
802 | | /* Free the components of the given keyword set. */ |
803 | | void |
804 | | kwsfree (kwset_t kws) |
805 | 0 | { |
806 | 0 | struct kwset *kwset; |
807 | |
|
808 | 0 | kwset = (struct kwset *) kws; |
809 | 0 | obstack_free(&kwset->obstack, NULL); |
810 | 0 | free(kws); |
811 | 0 | } |