1 /* kwset.c - search for any of a set of keywords.
2 Copyright 1989, 1998, 2000, 2005-2006, 2010, 2012, 2015 Free Software
5 This program is free software: you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 3 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program. If not, see <http://www.gnu.org/licenses/>. */
18 /* Written August 1989 by Mike Haertel.
19 The author may be reached (Email) at the address mike@ai.mit.edu,
20 or (US mail) as Mike Haertel c/o Free Software Foundation. */
22 /* The algorithm implemented by these routines bears a startling resemblance
23 to one discovered by Beate Commentz-Walter, although it is not identical.
24 See "A String Matching Algorithm Fast on the Average," Technical Report,
25 IBM-Germany, Scientific Center Heidelberg, Tiergartenstrasse 15, D-6900
26 Heidelberg, Germany. See also Aho, A.V., and M. Corasick, "Efficient
27 String Matching: An Aid to Bibliographic Search," CACM June 1975,
28 Vol. 18, No. 6, which describes the failure function used below. */
33 #include <sys/types.h>
39 #define _(str) gettext (str)
42 extern char *xmalloc();
44 # define malloc xmalloc
47 #define NCHAR (UCHAR_MAX + 1)
48 #define obstack_chunk_alloc malloc
49 #define obstack_chunk_free free
51 /* Balanced tree of edges and labels leaving a given trie node. */
54 struct tree *llink; /* Left link; MUST be first field. */
55 struct tree *rlink; /* Right link (to larger labels). */
56 struct trie *trie; /* Trie node pointed to by this edge. */
57 unsigned char label; /* Label on this edge. */
58 char balance; /* Difference in depths of subtrees. */
61 /* Node of a trie representing a set of reversed keywords. */
64 unsigned int accepting; /* Word index of accepted word, or zero. */
65 struct tree *links; /* Tree of edges leaving this node. */
66 struct trie *parent; /* Parent of this node. */
67 struct trie *next; /* List of all trie nodes in level order. */
68 struct trie *fail; /* Aho-Corasick failure function. */
69 int depth; /* Depth of this node from the root. */
70 int shift; /* Shift function for search failures. */
71 int maxshift; /* Max shift of self and descendents. */
74 /* Structure returned opaquely to the caller, containing everything. */
77 struct obstack obstack; /* Obstack for node allocation. */
78 int words; /* Number of words in the trie. */
79 struct trie *trie; /* The trie itself. */
80 int mind; /* Minimum depth of an accepting node. */
81 int maxd; /* Maximum depth of any node. */
82 unsigned char delta[NCHAR]; /* Delta table for rapid search. */
83 struct trie *next[NCHAR]; /* Table of children of the root. */
84 char *target; /* Target string if there's only one. */
85 int mind2; /* Used in Boyer-Moore search for one string. */
86 char const *trans; /* Character translation table. */
89 /* Allocate and initialize a keyword set object, returning an opaque
90 pointer to it. Return NULL if memory is not available. */
92 kwsalloc (char const *trans)
96 kwset = (struct kwset *) malloc (sizeof (struct kwset));
100 obstack_init (&kwset->obstack);
103 = (struct trie *) obstack_alloc (&kwset->obstack, sizeof (struct trie));
109 kwset->trie->accepting = 0;
110 kwset->trie->links = 0;
111 kwset->trie->parent = 0;
112 kwset->trie->next = 0;
113 kwset->trie->fail = 0;
114 kwset->trie->depth = 0;
115 kwset->trie->shift = 0;
116 kwset->mind = INT_MAX;
119 kwset->trans = trans;
124 /* Add the given string to the contents of the keyword set. Return NULL
125 for success, an error message otherwise. */
127 kwsincr (kwset_t kws, char const *text, size_t len)
130 register struct trie *trie;
132 kwset = (struct kwset *) kws;
136 /* Descend the trie (built of reversed keywords) character-by-character,
137 installing new nodes when necessary. */
140 register unsigned char label;
141 register struct tree *link;
143 struct tree *links[12];
144 enum { L, R } dirs[12];
146 label = kwset->trans ? kwset->trans[(unsigned char) *--text] : *--text;
148 /* Descend the tree of outgoing links for this trie node,
149 looking for the current character and keeping track
150 of the path followed. */
152 links[0] = (struct tree *) &trie->links;
156 while (link && label != link->label)
159 if (label < link->label)
160 dirs[depth++] = L, link = link->llink;
162 dirs[depth++] = R, link = link->rlink;
165 /* The current character doesn't have an outgoing link at
166 this trie node, so build a new trie node and install
167 a link in the current trie node's tree. */
170 link = (struct tree *) obstack_alloc (&kwset->obstack,
171 sizeof (struct tree));
173 return _("memory exhausted");
176 link->trie = (struct trie *) obstack_alloc (&kwset->obstack,
177 sizeof (struct trie));
179 return _("memory exhausted");
180 link->trie->accepting = 0;
181 link->trie->links = 0;
182 link->trie->parent = trie;
183 link->trie->next = 0;
184 link->trie->fail = 0;
185 link->trie->depth = trie->depth + 1;
186 link->trie->shift = 0;
190 /* Install the new tree node in its parent. */
191 if (dirs[--depth] == L)
192 links[depth]->llink = link;
194 links[depth]->rlink = link;
196 /* Back up the tree fixing the balance flags. */
197 while (depth && !links[depth]->balance)
199 if (dirs[depth] == L)
200 --links[depth]->balance;
202 ++links[depth]->balance;
206 /* Rebalance the tree by pointer rotations if necessary. */
207 if (depth && ((dirs[depth] == L && --links[depth]->balance)
208 || (dirs[depth] == R && ++links[depth]->balance)))
212 switch (links[depth]->balance)
214 struct tree *r, *l, *rl, *lr;
216 switch (dirs[depth + 1])
219 r = links[depth], t = r->llink, rl = t->rlink;
220 t->rlink = r, r->llink = rl;
221 t->balance = r->balance = 0;
224 r = links[depth], l = r->llink, t = l->rlink;
225 rl = t->rlink, lr = t->llink;
226 t->llink = l, l->rlink = lr, t->rlink = r, r->llink = rl;
227 l->balance = t->balance != 1 ? 0 : -1;
228 r->balance = t->balance != (char) -1 ? 0 : 1;
236 switch (dirs[depth + 1])
239 l = links[depth], t = l->rlink, lr = t->llink;
240 t->llink = l, l->rlink = lr;
241 t->balance = l->balance = 0;
244 l = links[depth], r = l->rlink, t = r->llink;
245 lr = t->llink, rl = t->rlink;
246 t->llink = l, l->rlink = lr, t->rlink = r, r->llink = rl;
247 l->balance = t->balance != 1 ? 0 : -1;
248 r->balance = t->balance != (char) -1 ? 0 : 1;
259 if (dirs[depth - 1] == L)
260 links[depth - 1]->llink = t;
262 links[depth - 1]->rlink = t;
269 /* Mark the node we finally reached as accepting, encoding the
270 index number of this word in the keyword set so far. */
271 if (!trie->accepting)
272 trie->accepting = 1 + 2 * kwset->words;
275 /* Keep track of the longest and shortest string of the keyword set. */
276 if (trie->depth < kwset->mind)
277 kwset->mind = trie->depth;
278 if (trie->depth > kwset->maxd)
279 kwset->maxd = trie->depth;
284 /* Enqueue the trie nodes referenced from the given tree in the
287 enqueue (struct tree *tree, struct trie **last)
291 enqueue (tree->llink, last);
292 enqueue (tree->rlink, last);
293 (*last) = (*last)->next = tree->trie;
296 /* Compute the Aho-Corasick failure function for the trie nodes referenced
297 from the given tree, given the failure function for their parent as
298 well as a last resort failure node. */
300 treefails (register struct tree const *tree, struct trie const *fail,
301 struct trie *recourse)
306 treefails (tree->llink, fail, recourse);
307 treefails (tree->rlink, fail, recourse);
309 /* Find, in the chain of fails going back to the root, the first
310 node that has a descendent on the current label. */
313 register struct tree *link;
316 while (link && tree->label != link->label)
317 if (tree->label < link->label)
323 tree->trie->fail = link->trie;
329 tree->trie->fail = recourse;
332 /* Set delta entries for the links of the given tree such that
333 the preexisting delta value is larger than the current depth. */
335 treedelta (register struct tree const *tree,
336 register unsigned int depth,
337 unsigned char delta[])
341 treedelta (tree->llink, depth, delta);
342 treedelta (tree->rlink, depth, delta);
343 if (depth < delta[tree->label])
344 delta[tree->label] = depth;
347 /* Return true if A has every label in B. */
349 hasevery (register struct tree const *a, register struct tree const *b)
353 if (!hasevery (a, b->llink))
355 if (!hasevery (a, b->rlink))
357 while (a && b->label != a->label)
358 if (b->label < a->label)
365 /* Compute a vector, indexed by character code, of the trie nodes
366 referenced from the given tree. */
368 treenext (struct tree const *tree, struct trie *next[])
372 treenext (tree->llink, next);
373 treenext (tree->rlink, next);
374 next[tree->label] = tree->trie;
377 /* Compute the shift for each trie node, as well as the delta
378 table and next cache for the given keyword set. */
380 kwsprep (kwset_t kwset)
382 unsigned char delta[NCHAR];
384 /* Initial values for the delta table; will be changed later. The
385 delta entry for a given character is the smallest depth of any
386 node at which an outgoing edge is labeled by that character. */
390 if (kwset->mind < 256)
391 for (i = 0; i < NCHAR; ++i)
392 delta[i] = kwset->mind;
394 for (i = 0; i < NCHAR; ++i)
398 /* Check if we can use the simple boyer-moore algorithm, instead
399 of the hairy commentz-walter algorithm. */
400 if (kwset->words == 1 && kwset->trans == 0)
403 register struct trie *curr;
405 /* Looking for just one string. Extract it from the trie. */
406 kwset->target = (char *) obstack_alloc (&kwset->obstack, kwset->mind);
407 for (i = kwset->mind - 1, curr = kwset->trie; i >= 0; --i)
409 kwset->target[i] = curr->links->label;
410 curr = curr->links->trie;
412 /* Build the Boyer Moore delta. Boy that's easy compared to CW. */
413 for (i = 0; i < kwset->mind; ++i)
414 delta[(unsigned char) kwset->target[i]] = kwset->mind - (i + 1);
415 kwset->mind2 = kwset->mind;
416 /* Find the minimal delta2 shift that we might make after
417 a backwards match has failed. */
418 for (i = 0; i < kwset->mind - 1; ++i)
419 if (kwset->target[i] == kwset->target[kwset->mind - 1])
420 kwset->mind2 = kwset->mind - (i + 1);
424 register struct trie *curr;
427 /* Traverse the nodes of the trie in level order, simultaneously
428 computing the delta table, failure function, and shift function. */
429 for (curr = last = kwset->trie; curr; curr = curr->next)
431 register struct trie *fail;
433 /* Enqueue the immediate descendents in the level order queue. */
434 enqueue (curr->links, &last);
436 curr->shift = kwset->mind;
437 curr->maxshift = kwset->mind;
439 /* Update the delta table for the descendents of this node. */
440 treedelta (curr->links, curr->depth, delta);
442 /* Compute the failure function for the descendants of this node. */
443 treefails (curr->links, curr->fail, kwset->trie);
445 /* Update the shifts at each node in the current node's chain
446 of fails back to the root. */
447 for (fail = curr->fail; fail; fail = fail->fail)
449 /* If the current node has some outgoing edge that the fail
450 doesn't, then the shift at the fail should be no larger
451 than the difference of their depths. */
452 if (!hasevery (fail->links, curr->links))
453 if (curr->depth - fail->depth < fail->shift)
454 fail->shift = curr->depth - fail->depth;
456 /* If the current node is accepting then the shift at the
457 fail and its descendents should be no larger than the
458 difference of their depths. */
459 if (curr->accepting && fail->maxshift > curr->depth - fail->depth)
460 fail->maxshift = curr->depth - fail->depth;
464 /* Traverse the trie in level order again, fixing up all nodes whose
465 shift exceeds their inherited maxshift. */
466 for (curr = kwset->trie->next; curr; curr = curr->next)
468 if (curr->maxshift > curr->parent->maxshift)
469 curr->maxshift = curr->parent->maxshift;
470 if (curr->shift > curr->maxshift)
471 curr->shift = curr->maxshift;
474 /* Create a vector, indexed by character code, of the outgoing links
475 from the root node. */
477 struct trie *next[NCHAR];
480 for (i = 0; i < NCHAR; ++i)
482 treenext (kwset->trie->links, next);
485 register char const *trans;
487 if ((trans = kwset->trans) != 0)
488 for (i = 0; i < NCHAR; ++i)
489 kwset->next[i] = next[(unsigned char) trans[i]];
491 for (i = 0; i < NCHAR; ++i)
492 kwset->next[i] = next[i];
497 /* Fix things up for any translation table. */
499 register char const *trans;
502 if ((trans = kwset->trans) != 0)
503 for (i = 0; i < NCHAR; ++i)
504 kwset->delta[i] = delta[(unsigned char) trans[i]];
506 for (i = 0; i < NCHAR; ++i)
507 kwset->delta[i] = delta[i];
513 #define U(C) ((unsigned char) (C))
515 /* Fast boyer-moore search. */
517 bmexec (kwset_t kws, char const *text, size_t size)
519 struct kwset const *kwset;
522 kwset = (struct kwset const *) kws;
531 register char const *tp;
533 tp = (const char *) memchr (text, kwset->target[0], size);
534 return tp ? tp - text : -1;
538 register unsigned char const *d1;
539 register char const *sp;
542 register char const *tp;
545 sp = kwset->target + len;
550 /* Significance of 12: 1 (initial offset) + 10 (skip loop) + 1 (md2). */
553 register char const *ep;
556 /* 11 is not a bug, the initial offset happens only once. */
557 for (ep = text + size - 11 * len;;)
561 d = d1[U(tp[-1])], tp += d;
562 d = d1[U(tp[-1])], tp += d;
565 d = d1[U(tp[-1])], tp += d;
566 d = d1[U(tp[-1])], tp += d;
567 d = d1[U(tp[-1])], tp += d;
570 d = d1[U(tp[-1])], tp += d;
571 d = d1[U(tp[-1])], tp += d;
572 d = d1[U(tp[-1])], tp += d;
575 d = d1[U(tp[-1])], tp += d;
576 d = d1[U(tp[-1])], tp += d;
583 for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i)
586 return tp - len - text;
593 /* Now we have only a few characters left to search. We
594 carefully avoid ever producing an out-of-bounds pointer. */
595 register char const *ep;
602 d = d1[U((tp += d)[-1])];
608 for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i)
611 return tp - len - text;
621 /* Hairy multiple string search. */
623 cwexec (kwset_t kws, char const *text, size_t len, struct kwsmatch *kwsmatch)
625 struct kwset const *kwset;
627 /* Initialize register copies and look for easy ways out. */
628 kwset = (struct kwset *) kws;
629 if (len < kwset->mind)
632 struct trie const *accept;
633 struct trie * const *next;
634 register unsigned char const *delta;
635 register char const *trans;
637 register char const *end;
640 register char const *qlim;
645 delta = kwset->delta;
646 trans = kwset->trans;
649 if ((d = kwset->mind) != 0)
653 mch = text, accept = kwset->trie;
657 if (len >= 4 * kwset->mind)
658 qlim = lim - 4 * kwset->mind;
662 while (lim - end >= d)
665 struct trie const *trie;
668 register unsigned char c;
670 if (qlim && end <= qlim)
673 while ((d = delta[c = *end]) && end < qlim)
676 end += delta[(unsigned char) *end];
677 end += delta[(unsigned char) *end];
682 d = delta[c = (end += d)[-1]];
696 register unsigned char c;
697 register struct tree const *tree;
699 c = trans ? trans[(unsigned char) *--beg] : *--beg;
701 while (tree && c != tree->label)
725 /* Given a known match, find the longest possible match anchored
726 at or before its starting point. This is nearly a verbatim
727 copy of the preceding main search loops. */
731 if (lim - mch > kwset->maxd)
732 lim = mch + kwset->maxd;
735 while (lim - end >= d)
738 struct trie const *trie;
741 register unsigned char c;
743 if ((d = delta[c = (end += d)[-1]]) != 0)
746 if (!(trie = next[c]))
752 if (trie->accepting && beg <= mch)
760 register unsigned char c;
761 register struct tree const *tree;
763 c = trans ? trans[(unsigned char) *--beg] : *--beg;
765 while (tree && c != tree->label)
773 if (trie->accepting && beg <= mch)
795 kwsmatch->index = accept->accepting / 2;
796 kwsmatch->offset[0] = mch - text;
797 kwsmatch->size[0] = accept->depth;
803 /* Search through the given text for a match of any member of the
804 given keyword set. Return a pointer to the first character of
805 the matching substring, or NULL if no match is found. If FOUNDLEN
806 is non-NULL store in the referenced location the length of the
807 matching substring. Similarly, if FOUNDIDX is non-NULL, store
808 in the referenced location the index number of the particular
811 kwsexec (kwset_t kws, char const *text, size_t size,
812 struct kwsmatch *kwsmatch)
814 struct kwset const *kwset = (struct kwset *) kws;
815 if (kwset->words == 1 && kwset->trans == 0)
817 size_t ret = bmexec (kws, text, size);
818 if (kwsmatch != 0 && ret != (size_t) -1)
821 kwsmatch->offset[0] = ret;
822 kwsmatch->size[0] = kwset->mind;
827 return cwexec (kws, text, size, kwsmatch);
830 /* Free the components of the given keyword set. */
832 kwsfree (kwset_t kwset)
834 obstack_free (&kwset->obstack, 0);