Line | Count | Source |
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 | 0 | #define NCHAR (UCHAR_MAX + 1) |
43 | | /* adapter for `xmalloc()`, which takes `size_t`, not `long` */ |
44 | | static void *obstack_chunk_alloc(long size) |
45 | 0 | { |
46 | 0 | if (size < 0) |
47 | 0 | BUG("Cannot allocate a negative amount: %ld", size); |
48 | 0 | return xmalloc(size); |
49 | 0 | } |
50 | 0 | #define obstack_chunk_free free |
51 | | |
52 | 0 | #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 | 0 | { |
133 | 0 | struct kwset *kwset; |
134 | |
|
135 | 0 | kwset = (struct kwset *) xmalloc(sizeof (struct kwset)); |
136 | |
|
137 | 0 | obstack_init(&kwset->obstack); |
138 | 0 | kwset->words = 0; |
139 | 0 | kwset->trie |
140 | 0 | = (struct trie *) obstack_alloc(&kwset->obstack, sizeof (struct trie)); |
141 | 0 | if (!kwset->trie) |
142 | 0 | { |
143 | 0 | kwsfree((kwset_t) kwset); |
144 | 0 | return NULL; |
145 | 0 | } |
146 | 0 | kwset->trie->accepting = 0; |
147 | 0 | kwset->trie->links = NULL; |
148 | 0 | kwset->trie->parent = NULL; |
149 | 0 | kwset->trie->next = NULL; |
150 | 0 | kwset->trie->fail = NULL; |
151 | 0 | kwset->trie->depth = 0; |
152 | 0 | kwset->trie->shift = 0; |
153 | 0 | kwset->mind = INT_MAX; |
154 | 0 | kwset->maxd = -1; |
155 | 0 | kwset->target = NULL; |
156 | 0 | kwset->trans = trans; |
157 | |
|
158 | 0 | return (kwset_t) kwset; |
159 | 0 | } |
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 | 0 | { |
170 | 0 | struct kwset *kwset; |
171 | 0 | register struct trie *trie; |
172 | 0 | register unsigned char label; |
173 | 0 | register struct tree *link; |
174 | 0 | register int depth; |
175 | 0 | struct tree *links[DEPTH_SIZE]; |
176 | 0 | enum { L, R } dirs[DEPTH_SIZE]; |
177 | 0 | struct tree *t, *r, *l, *rl, *lr; |
178 | |
|
179 | 0 | kwset = (struct kwset *) kws; |
180 | 0 | trie = kwset->trie; |
181 | 0 | text += len; |
182 | | |
183 | | /* Descend the trie (built of reversed keywords) character-by-character, |
184 | | installing new nodes when necessary. */ |
185 | 0 | while (len--) |
186 | 0 | { |
187 | 0 | 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 | 0 | link = trie->links; |
193 | 0 | links[0] = (struct tree *) &trie->links; |
194 | 0 | dirs[0] = L; |
195 | 0 | depth = 1; |
196 | |
|
197 | 0 | while (link && label != link->label) |
198 | 0 | { |
199 | 0 | links[depth] = link; |
200 | 0 | if (label < link->label) { |
201 | 0 | dirs[depth++] = L; |
202 | 0 | link = link->llink; |
203 | 0 | } else { |
204 | 0 | dirs[depth++] = R; |
205 | 0 | link = link->rlink; |
206 | 0 | } |
207 | 0 | } |
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 | 0 | if (!link) |
213 | 0 | { |
214 | 0 | link = (struct tree *) obstack_alloc(&kwset->obstack, |
215 | 0 | sizeof (struct tree)); |
216 | 0 | if (!link) |
217 | 0 | return "memory exhausted"; |
218 | 0 | link->llink = NULL; |
219 | 0 | link->rlink = NULL; |
220 | 0 | link->trie = (struct trie *) obstack_alloc(&kwset->obstack, |
221 | 0 | sizeof (struct trie)); |
222 | 0 | if (!link->trie) |
223 | 0 | { |
224 | 0 | obstack_free(&kwset->obstack, link); |
225 | 0 | return "memory exhausted"; |
226 | 0 | } |
227 | 0 | link->trie->accepting = 0; |
228 | 0 | link->trie->links = NULL; |
229 | 0 | link->trie->parent = trie; |
230 | 0 | link->trie->next = NULL; |
231 | 0 | link->trie->fail = NULL; |
232 | 0 | link->trie->depth = trie->depth + 1; |
233 | 0 | link->trie->shift = 0; |
234 | 0 | link->label = label; |
235 | 0 | link->balance = 0; |
236 | | |
237 | | /* Install the new tree node in its parent. */ |
238 | 0 | if (dirs[--depth] == L) |
239 | 0 | links[depth]->llink = link; |
240 | 0 | else |
241 | 0 | links[depth]->rlink = link; |
242 | | |
243 | | /* Back up the tree fixing the balance flags. */ |
244 | 0 | while (depth && !links[depth]->balance) |
245 | 0 | { |
246 | 0 | if (dirs[depth] == L) |
247 | 0 | --links[depth]->balance; |
248 | 0 | else |
249 | 0 | ++links[depth]->balance; |
250 | 0 | --depth; |
251 | 0 | } |
252 | | |
253 | | /* Rebalance the tree by pointer rotations if necessary. */ |
254 | 0 | if (depth && ((dirs[depth] == L && --links[depth]->balance) |
255 | 0 | || (dirs[depth] == R && ++links[depth]->balance))) |
256 | 0 | { |
257 | 0 | switch (links[depth]->balance) |
258 | 0 | { |
259 | 0 | case (char) -2: |
260 | 0 | switch (dirs[depth + 1]) |
261 | 0 | { |
262 | 0 | case L: |
263 | 0 | r = links[depth]; t = r->llink; rl = t->rlink; |
264 | 0 | t->rlink = r; r->llink = rl; |
265 | 0 | t->balance = r->balance = 0; |
266 | 0 | break; |
267 | 0 | case R: |
268 | 0 | r = links[depth]; l = r->llink; t = l->rlink; |
269 | 0 | rl = t->rlink; lr = t->llink; |
270 | 0 | t->llink = l; l->rlink = lr; t->rlink = r; r->llink = rl; |
271 | 0 | l->balance = t->balance != 1 ? 0 : -1; |
272 | 0 | r->balance = t->balance != (char) -1 ? 0 : 1; |
273 | 0 | t->balance = 0; |
274 | 0 | break; |
275 | 0 | default: |
276 | 0 | abort (); |
277 | 0 | } |
278 | 0 | break; |
279 | 0 | case 2: |
280 | 0 | switch (dirs[depth + 1]) |
281 | 0 | { |
282 | 0 | case R: |
283 | 0 | l = links[depth]; t = l->rlink; lr = t->llink; |
284 | 0 | t->llink = l; l->rlink = lr; |
285 | 0 | t->balance = l->balance = 0; |
286 | 0 | break; |
287 | 0 | case L: |
288 | 0 | l = links[depth]; r = l->rlink; t = r->llink; |
289 | 0 | lr = t->llink; rl = t->rlink; |
290 | 0 | t->llink = l; l->rlink = lr; t->rlink = r; r->llink = rl; |
291 | 0 | l->balance = t->balance != 1 ? 0 : -1; |
292 | 0 | r->balance = t->balance != (char) -1 ? 0 : 1; |
293 | 0 | t->balance = 0; |
294 | 0 | break; |
295 | 0 | default: |
296 | 0 | abort (); |
297 | 0 | } |
298 | 0 | break; |
299 | 0 | default: |
300 | 0 | abort (); |
301 | 0 | } |
302 | | |
303 | 0 | if (dirs[depth - 1] == L) |
304 | 0 | links[depth - 1]->llink = t; |
305 | 0 | else |
306 | 0 | links[depth - 1]->rlink = t; |
307 | 0 | } |
308 | 0 | } |
309 | | |
310 | 0 | trie = link->trie; |
311 | 0 | } |
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 | 0 | if (!trie->accepting) |
316 | 0 | trie->accepting = 1 + 2 * kwset->words; |
317 | 0 | ++kwset->words; |
318 | | |
319 | | /* Keep track of the longest and shortest string of the keyword set. */ |
320 | 0 | if (trie->depth < kwset->mind) |
321 | 0 | kwset->mind = trie->depth; |
322 | 0 | if (trie->depth > kwset->maxd) |
323 | 0 | kwset->maxd = trie->depth; |
324 | |
|
325 | 0 | return NULL; |
326 | 0 | } |
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 | 0 | { |
333 | 0 | if (!tree) |
334 | 0 | return; |
335 | 0 | enqueue(tree->llink, last); |
336 | 0 | enqueue(tree->rlink, last); |
337 | 0 | (*last) = (*last)->next = tree->trie; |
338 | 0 | } |
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 | 0 | { |
347 | 0 | register struct tree *link; |
348 | |
|
349 | 0 | if (!tree) |
350 | 0 | return; |
351 | | |
352 | 0 | treefails(tree->llink, fail, recourse); |
353 | 0 | 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 | 0 | while (fail) |
358 | 0 | { |
359 | 0 | link = fail->links; |
360 | 0 | while (link && tree->label != link->label) |
361 | 0 | if (tree->label < link->label) |
362 | 0 | link = link->llink; |
363 | 0 | else |
364 | 0 | link = link->rlink; |
365 | 0 | if (link) |
366 | 0 | { |
367 | 0 | tree->trie->fail = link->trie; |
368 | 0 | return; |
369 | 0 | } |
370 | 0 | fail = fail->fail; |
371 | 0 | } |
372 | | |
373 | 0 | tree->trie->fail = recourse; |
374 | 0 | } |
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 | 0 | { |
383 | 0 | if (!tree) |
384 | 0 | return; |
385 | 0 | treedelta(tree->llink, depth, delta); |
386 | 0 | treedelta(tree->rlink, depth, delta); |
387 | 0 | if (depth < delta[tree->label]) |
388 | 0 | delta[tree->label] = depth; |
389 | 0 | } |
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 | 0 | { |
395 | 0 | if (!b) |
396 | 0 | return 1; |
397 | 0 | if (!hasevery(a, b->llink)) |
398 | 0 | return 0; |
399 | 0 | if (!hasevery(a, b->rlink)) |
400 | 0 | return 0; |
401 | 0 | while (a && b->label != a->label) |
402 | 0 | if (b->label < a->label) |
403 | 0 | a = a->llink; |
404 | 0 | else |
405 | 0 | a = a->rlink; |
406 | 0 | return !!a; |
407 | 0 | } |
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 | 0 | { |
414 | 0 | if (!tree) |
415 | 0 | return; |
416 | 0 | treenext(tree->llink, next); |
417 | 0 | treenext(tree->rlink, next); |
418 | 0 | next[tree->label] = tree->trie; |
419 | 0 | } |
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 | 0 | { |
426 | 0 | register struct kwset *kwset; |
427 | 0 | register int i; |
428 | 0 | register struct trie *curr; |
429 | 0 | register unsigned char const *trans; |
430 | 0 | unsigned char delta[NCHAR]; |
431 | |
|
432 | 0 | 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 | 0 | 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 | 0 | if (kwset->words == 1 && kwset->trans == NULL) |
442 | 0 | { |
443 | 0 | char c; |
444 | | |
445 | | /* Looking for just one string. Extract it from the trie. */ |
446 | 0 | kwset->target = obstack_alloc(&kwset->obstack, kwset->mind); |
447 | 0 | if (!kwset->target) |
448 | 0 | return "memory exhausted"; |
449 | 0 | for (i = kwset->mind - 1, curr = kwset->trie; i >= 0; --i) |
450 | 0 | { |
451 | 0 | kwset->target[i] = curr->links->label; |
452 | 0 | curr = curr->links->trie; |
453 | 0 | } |
454 | | /* Build the Boyer Moore delta. Boy that's easy compared to CW. */ |
455 | 0 | for (i = 0; i < kwset->mind; ++i) |
456 | 0 | 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 | 0 | c = kwset->target[kwset->mind - 1]; |
460 | 0 | for (i = kwset->mind - 2; i >= 0; --i) |
461 | 0 | if (kwset->target[i] == c) |
462 | 0 | break; |
463 | 0 | kwset->mind2 = kwset->mind - (i + 1); |
464 | 0 | } |
465 | 0 | else |
466 | 0 | { |
467 | 0 | register struct trie *fail; |
468 | 0 | 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 | 0 | for (curr = last = kwset->trie; curr; curr = curr->next) |
473 | 0 | { |
474 | | /* Enqueue the immediate descendants in the level order queue. */ |
475 | 0 | enqueue(curr->links, &last); |
476 | |
|
477 | 0 | curr->shift = kwset->mind; |
478 | 0 | curr->maxshift = kwset->mind; |
479 | | |
480 | | /* Update the delta table for the descendants of this node. */ |
481 | 0 | treedelta(curr->links, curr->depth, delta); |
482 | | |
483 | | /* Compute the failure function for the descendants of this node. */ |
484 | 0 | 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 | 0 | for (fail = curr->fail; fail; fail = fail->fail) |
489 | 0 | { |
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 | 0 | if (!hasevery(fail->links, curr->links)) |
494 | 0 | if (curr->depth - fail->depth < fail->shift) |
495 | 0 | 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 | 0 | if (curr->accepting && fail->maxshift > curr->depth - fail->depth) |
501 | 0 | fail->maxshift = curr->depth - fail->depth; |
502 | 0 | } |
503 | 0 | } |
504 | | |
505 | | /* Traverse the trie in level order again, fixing up all nodes whose |
506 | | shift exceeds their inherited maxshift. */ |
507 | 0 | for (curr = kwset->trie->next; curr; curr = curr->next) |
508 | 0 | { |
509 | 0 | if (curr->maxshift > curr->parent->maxshift) |
510 | 0 | curr->maxshift = curr->parent->maxshift; |
511 | 0 | if (curr->shift > curr->maxshift) |
512 | 0 | curr->shift = curr->maxshift; |
513 | 0 | } |
514 | | |
515 | | /* Create a vector, indexed by character code, of the outgoing links |
516 | | from the root node. */ |
517 | 0 | for (i = 0; i < NCHAR; ++i) |
518 | 0 | next[i] = NULL; |
519 | 0 | treenext(kwset->trie->links, next); |
520 | |
|
521 | 0 | if ((trans = kwset->trans)) |
522 | 0 | for (i = 0; i < NCHAR; ++i) |
523 | 0 | kwset->next[i] = next[U(trans[i])]; |
524 | 0 | else |
525 | 0 | COPY_ARRAY(kwset->next, next, NCHAR); |
526 | 0 | } |
527 | | |
528 | | /* Fix things up for any translation table. */ |
529 | 0 | if ((trans = kwset->trans)) |
530 | 0 | for (i = 0; i < NCHAR; ++i) |
531 | 0 | kwset->delta[i] = delta[U(trans[i])]; |
532 | 0 | else |
533 | 0 | memcpy(kwset->delta, delta, NCHAR); |
534 | |
|
535 | 0 | return NULL; |
536 | 0 | } |
537 | | |
538 | | /* Fast boyer-moore search. */ |
539 | | static size_t |
540 | | bmexec (kwset_t kws, char const *text, size_t size) |
541 | 0 | { |
542 | 0 | struct kwset const *kwset; |
543 | 0 | register unsigned char const *d1; |
544 | 0 | register char const *ep, *sp, *tp; |
545 | 0 | register int d, gc, i, len, md2; |
546 | |
|
547 | 0 | kwset = (struct kwset const *) kws; |
548 | 0 | len = kwset->mind; |
549 | |
|
550 | 0 | if (len == 0) |
551 | 0 | return 0; |
552 | 0 | if (len > size) |
553 | 0 | return -1; |
554 | 0 | if (len == 1) |
555 | 0 | { |
556 | 0 | tp = memchr (text, kwset->target[0], size); |
557 | 0 | return tp ? tp - text : -1; |
558 | 0 | } |
559 | | |
560 | 0 | d1 = kwset->delta; |
561 | 0 | sp = kwset->target + len; |
562 | 0 | gc = U(sp[-2]); |
563 | 0 | md2 = kwset->mind2; |
564 | 0 | tp = text + len; |
565 | | |
566 | | /* Significance of 12: 1 (initial offset) + 10 (skip loop) + 1 (md2). */ |
567 | 0 | if (size > 12 * len) |
568 | | /* 11 is not a bug, the initial offset happens only once. */ |
569 | 0 | for (ep = text + size - 11 * len;;) |
570 | 0 | { |
571 | 0 | while (tp <= ep) |
572 | 0 | { |
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 | d = d1[U(tp[-1])]; tp += d; |
585 | 0 | if (d == 0) |
586 | 0 | goto found; |
587 | 0 | d = d1[U(tp[-1])]; tp += d; |
588 | 0 | d = d1[U(tp[-1])]; tp += d; |
589 | 0 | } |
590 | 0 | break; |
591 | 0 | found: |
592 | 0 | if (U(tp[-2]) == gc) |
593 | 0 | { |
594 | 0 | for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i) |
595 | 0 | ; |
596 | 0 | if (i > len) |
597 | 0 | return tp - len - text; |
598 | 0 | } |
599 | 0 | tp += md2; |
600 | 0 | } |
601 | | |
602 | | /* Now we have only a few characters left to search. We |
603 | | carefully avoid ever producing an out-of-bounds pointer. */ |
604 | 0 | ep = text + size; |
605 | 0 | d = d1[U(tp[-1])]; |
606 | 0 | while (d <= ep - tp) |
607 | 0 | { |
608 | 0 | d = d1[U((tp += d)[-1])]; |
609 | 0 | if (d != 0) |
610 | 0 | continue; |
611 | 0 | if (U(tp[-2]) == gc) |
612 | 0 | { |
613 | 0 | for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i) |
614 | 0 | ; |
615 | 0 | if (i > len) |
616 | 0 | return tp - len - text; |
617 | 0 | } |
618 | 0 | d = md2; |
619 | 0 | } |
620 | | |
621 | 0 | return -1; |
622 | 0 | } |
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 | 0 | { |
628 | 0 | struct kwset const *kwset; |
629 | 0 | struct trie * const *next; |
630 | 0 | struct trie const *trie; |
631 | 0 | struct trie const *accept; |
632 | 0 | char const *beg, *lim, *mch, *lmch; |
633 | 0 | register unsigned char c; |
634 | 0 | register unsigned char const *delta; |
635 | 0 | register int d; |
636 | 0 | register char const *end, *qlim; |
637 | 0 | register struct tree const *tree; |
638 | 0 | register unsigned char const *trans; |
639 | |
|
640 | 0 | accept = NULL; |
641 | | |
642 | | /* Initialize register copies and look for easy ways out. */ |
643 | 0 | kwset = (struct kwset *) kws; |
644 | 0 | if (len < kwset->mind) |
645 | 0 | return -1; |
646 | 0 | next = kwset->next; |
647 | 0 | delta = kwset->delta; |
648 | 0 | trans = kwset->trans; |
649 | 0 | lim = text + len; |
650 | 0 | end = text; |
651 | 0 | if ((d = kwset->mind) != 0) |
652 | 0 | mch = NULL; |
653 | 0 | else |
654 | 0 | { |
655 | 0 | mch = text; |
656 | 0 | accept = kwset->trie; |
657 | 0 | goto match; |
658 | 0 | } |
659 | | |
660 | 0 | if (len >= 4 * kwset->mind) |
661 | 0 | qlim = lim - 4 * kwset->mind; |
662 | 0 | else |
663 | 0 | qlim = NULL; |
664 | |
|
665 | 0 | while (lim - end >= d) |
666 | 0 | { |
667 | 0 | if (qlim && end <= qlim) |
668 | 0 | { |
669 | 0 | end += d - 1; |
670 | 0 | while ((d = delta[c = *end]) && end < qlim) |
671 | 0 | { |
672 | 0 | end += d; |
673 | 0 | end += delta[U(*end)]; |
674 | 0 | end += delta[U(*end)]; |
675 | 0 | } |
676 | 0 | ++end; |
677 | 0 | } |
678 | 0 | else |
679 | 0 | d = delta[c = (end += d)[-1]]; |
680 | 0 | if (d) |
681 | 0 | continue; |
682 | 0 | beg = end - 1; |
683 | 0 | trie = next[c]; |
684 | 0 | if (trie->accepting) |
685 | 0 | { |
686 | 0 | mch = beg; |
687 | 0 | accept = trie; |
688 | 0 | } |
689 | 0 | d = trie->shift; |
690 | 0 | while (beg > text) |
691 | 0 | { |
692 | 0 | c = trans ? trans[U(*--beg)] : *--beg; |
693 | 0 | tree = trie->links; |
694 | 0 | while (tree && c != tree->label) |
695 | 0 | if (c < tree->label) |
696 | 0 | tree = tree->llink; |
697 | 0 | else |
698 | 0 | tree = tree->rlink; |
699 | 0 | if (tree) |
700 | 0 | { |
701 | 0 | trie = tree->trie; |
702 | 0 | if (trie->accepting) |
703 | 0 | { |
704 | 0 | mch = beg; |
705 | 0 | accept = trie; |
706 | 0 | } |
707 | 0 | } |
708 | 0 | else |
709 | 0 | break; |
710 | 0 | d = trie->shift; |
711 | 0 | } |
712 | 0 | if (mch) |
713 | 0 | goto match; |
714 | 0 | } |
715 | 0 | return -1; |
716 | | |
717 | 0 | 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 | 0 | if (lim - mch > kwset->maxd) |
722 | 0 | lim = mch + kwset->maxd; |
723 | 0 | lmch = NULL; |
724 | 0 | d = 1; |
725 | 0 | while (lim - end >= d) |
726 | 0 | { |
727 | 0 | if ((d = delta[c = (end += d)[-1]]) != 0) |
728 | 0 | continue; |
729 | 0 | beg = end - 1; |
730 | 0 | if (!(trie = next[c])) |
731 | 0 | { |
732 | 0 | d = 1; |
733 | 0 | continue; |
734 | 0 | } |
735 | 0 | if (trie->accepting && beg <= mch) |
736 | 0 | { |
737 | 0 | lmch = beg; |
738 | 0 | accept = trie; |
739 | 0 | } |
740 | 0 | d = trie->shift; |
741 | 0 | while (beg > text) |
742 | 0 | { |
743 | 0 | c = trans ? trans[U(*--beg)] : *--beg; |
744 | 0 | tree = trie->links; |
745 | 0 | while (tree && c != tree->label) |
746 | 0 | if (c < tree->label) |
747 | 0 | tree = tree->llink; |
748 | 0 | else |
749 | 0 | tree = tree->rlink; |
750 | 0 | if (tree) |
751 | 0 | { |
752 | 0 | trie = tree->trie; |
753 | 0 | if (trie->accepting && beg <= mch) |
754 | 0 | { |
755 | 0 | lmch = beg; |
756 | 0 | accept = trie; |
757 | 0 | } |
758 | 0 | } |
759 | 0 | else |
760 | 0 | break; |
761 | 0 | d = trie->shift; |
762 | 0 | } |
763 | 0 | if (lmch) |
764 | 0 | { |
765 | 0 | mch = lmch; |
766 | 0 | goto match; |
767 | 0 | } |
768 | 0 | if (!d) |
769 | 0 | d = 1; |
770 | 0 | } |
771 | | |
772 | 0 | if (kwsmatch) |
773 | 0 | { |
774 | 0 | kwsmatch->index = accept->accepting / 2; |
775 | 0 | kwsmatch->offset[0] = mch - text; |
776 | 0 | kwsmatch->size[0] = accept->depth; |
777 | 0 | } |
778 | 0 | return mch - text; |
779 | 0 | } |
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 | 0 | { |
792 | 0 | struct kwset const *kwset = (struct kwset *) kws; |
793 | 0 | if (kwset->words == 1 && kwset->trans == NULL) |
794 | 0 | { |
795 | 0 | size_t ret = bmexec (kws, text, size); |
796 | 0 | if (kwsmatch != NULL && ret != (size_t) -1) |
797 | 0 | { |
798 | 0 | kwsmatch->index = 0; |
799 | 0 | kwsmatch->offset[0] = ret; |
800 | 0 | kwsmatch->size[0] = kwset->mind; |
801 | 0 | } |
802 | 0 | return ret; |
803 | 0 | } |
804 | 0 | else |
805 | 0 | return cwexec(kws, text, size, kwsmatch); |
806 | 0 | } |
807 | | |
808 | | /* Free the components of the given keyword set. */ |
809 | | void |
810 | | kwsfree (kwset_t kws) |
811 | 0 | { |
812 | 0 | struct kwset *kwset; |
813 | |
|
814 | 0 | kwset = (struct kwset *) kws; |
815 | 0 | obstack_free(&kwset->obstack, NULL); |
816 | 0 | free(kws); |
817 | 0 | } |