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
7 /* kwset.c - search for any of a set of keywords.
8 Copyright 1989, 1998, 2000, 2005 Free Software Foundation, Inc.
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)
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.
20 You should have received a copy of the GNU General Public License
21 along with this program; if not, see <http://www.gnu.org/licenses/>. */
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. */
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. */
38 #include "compat/obstack.h"
40 #define NCHAR (UCHAR_MAX + 1)
41 /* adapter for `xmalloc()`, which takes `size_t`, not `long` */
42 static void *obstack_chunk_alloc(long size)
45 BUG("Cannot allocate a negative amount: %ld", size);
48 #define obstack_chunk_free free
50 #define U(c) ((unsigned char) (c))
52 /* Balanced tree of edges and labels leaving a given trie node. */
55 struct tree *llink; /* Left link; MUST be first field. */
56 struct tree *rlink; /* Right link (to larger labels). */
57 struct trie *trie; /* Trie node pointed to by this edge. */
58 unsigned char label; /* Label on this edge. */
59 char balance; /* Difference in depths of subtrees. */
62 /* Node of a trie representing a set of reversed keywords. */
65 unsigned int accepting; /* Word index of accepted word, or zero. */
66 struct tree *links; /* Tree of edges leaving this node. */
67 struct trie *parent; /* Parent of this node. */
68 struct trie *next; /* List of all trie nodes in level order. */
69 struct trie *fail; /* Aho-Corasick failure function. */
70 int depth; /* Depth of this node from the root. */
71 int shift; /* Shift function for search failures. */
72 int maxshift; /* Max shift of self and descendants. */
75 /* Structure returned opaquely to the caller, containing everything. */
78 struct obstack obstack; /* Obstack for node allocation. */
79 int words; /* Number of words in the trie. */
80 struct trie *trie; /* The trie itself. */
81 int mind; /* Minimum depth of an accepting node. */
82 int maxd; /* Maximum depth of any node. */
83 unsigned char delta[NCHAR]; /* Delta table for rapid search. */
84 struct trie *next[NCHAR]; /* Table of children of the root. */
85 char *target; /* Target string if there's only one. */
86 int mind2; /* Used in Boyer-Moore search for one string. */
87 unsigned char const *trans; /* Character translation table. */
90 /* Allocate and initialize a keyword set object, returning an opaque
91 pointer to it. Return NULL if memory is not available. */
93 kwsalloc (unsigned char const *trans)
97 kwset = (struct kwset *) xmalloc(sizeof (struct kwset));
99 obstack_init(&kwset->obstack);
102 = (struct trie *) obstack_alloc(&kwset->obstack, sizeof (struct trie));
105 kwsfree((kwset_t) kwset);
108 kwset->trie->accepting = 0;
109 kwset->trie->links = NULL;
110 kwset->trie->parent = NULL;
111 kwset->trie->next = NULL;
112 kwset->trie->fail = NULL;
113 kwset->trie->depth = 0;
114 kwset->trie->shift = 0;
115 kwset->mind = INT_MAX;
117 kwset->target = NULL;
118 kwset->trans = trans;
120 return (kwset_t) kwset;
123 /* This upper bound is valid for CHAR_BIT >= 4 and
124 exact for CHAR_BIT in { 4..11, 13, 15, 17, 19 }. */
125 #define DEPTH_SIZE (CHAR_BIT + CHAR_BIT/2)
127 /* Add the given string to the contents of the keyword set. Return NULL
128 for success, an error message otherwise. */
130 kwsincr (kwset_t kws, char const *text, size_t len)
133 register struct trie *trie;
134 register unsigned char label;
135 register struct tree *link;
137 struct tree *links[DEPTH_SIZE];
138 enum { L, R } dirs[DEPTH_SIZE];
139 struct tree *t, *r, *l, *rl, *lr;
141 kwset = (struct kwset *) kws;
145 /* Descend the trie (built of reversed keywords) character-by-character,
146 installing new nodes when necessary. */
149 label = kwset->trans ? kwset->trans[U(*--text)] : *--text;
151 /* Descend the tree of outgoing links for this trie node,
152 looking for the current character and keeping track
153 of the path followed. */
155 links[0] = (struct tree *) &trie->links;
159 while (link && label != link->label)
162 if (label < link->label)
163 dirs[depth++] = L, link = link->llink;
165 dirs[depth++] = R, link = link->rlink;
168 /* The current character doesn't have an outgoing link at
169 this trie node, so build a new trie node and install
170 a link in the current trie node's tree. */
173 link = (struct tree *) obstack_alloc(&kwset->obstack,
174 sizeof (struct tree));
176 return "memory exhausted";
179 link->trie = (struct trie *) obstack_alloc(&kwset->obstack,
180 sizeof (struct trie));
183 obstack_free(&kwset->obstack, link);
184 return "memory exhausted";
186 link->trie->accepting = 0;
187 link->trie->links = NULL;
188 link->trie->parent = trie;
189 link->trie->next = NULL;
190 link->trie->fail = NULL;
191 link->trie->depth = trie->depth + 1;
192 link->trie->shift = 0;
196 /* Install the new tree node in its parent. */
197 if (dirs[--depth] == L)
198 links[depth]->llink = link;
200 links[depth]->rlink = link;
202 /* Back up the tree fixing the balance flags. */
203 while (depth && !links[depth]->balance)
205 if (dirs[depth] == L)
206 --links[depth]->balance;
208 ++links[depth]->balance;
212 /* Rebalance the tree by pointer rotations if necessary. */
213 if (depth && ((dirs[depth] == L && --links[depth]->balance)
214 || (dirs[depth] == R && ++links[depth]->balance)))
216 switch (links[depth]->balance)
219 switch (dirs[depth + 1])
222 r = links[depth], t = r->llink, rl = t->rlink;
223 t->rlink = r, r->llink = rl;
224 t->balance = r->balance = 0;
227 r = links[depth], l = r->llink, t = l->rlink;
228 rl = t->rlink, lr = t->llink;
229 t->llink = l, l->rlink = lr, t->rlink = r, r->llink = rl;
230 l->balance = t->balance != 1 ? 0 : -1;
231 r->balance = t->balance != (char) -1 ? 0 : 1;
239 switch (dirs[depth + 1])
242 l = links[depth], t = l->rlink, lr = t->llink;
243 t->llink = l, l->rlink = lr;
244 t->balance = l->balance = 0;
247 l = links[depth], r = l->rlink, t = r->llink;
248 lr = t->llink, rl = t->rlink;
249 t->llink = l, l->rlink = lr, t->rlink = r, r->llink = rl;
250 l->balance = t->balance != 1 ? 0 : -1;
251 r->balance = t->balance != (char) -1 ? 0 : 1;
262 if (dirs[depth - 1] == L)
263 links[depth - 1]->llink = t;
265 links[depth - 1]->rlink = t;
272 /* Mark the node we finally reached as accepting, encoding the
273 index number of this word in the keyword set so far. */
274 if (!trie->accepting)
275 trie->accepting = 1 + 2 * kwset->words;
278 /* Keep track of the longest and shortest string of the keyword set. */
279 if (trie->depth < kwset->mind)
280 kwset->mind = trie->depth;
281 if (trie->depth > kwset->maxd)
282 kwset->maxd = trie->depth;
287 /* Enqueue the trie nodes referenced from the given tree in the
290 enqueue (struct tree *tree, struct trie **last)
294 enqueue(tree->llink, last);
295 enqueue(tree->rlink, last);
296 (*last) = (*last)->next = tree->trie;
299 /* Compute the Aho-Corasick failure function for the trie nodes referenced
300 from the given tree, given the failure function for their parent as
301 well as a last resort failure node. */
303 treefails (register struct tree const *tree, struct trie const *fail,
304 struct trie *recourse)
306 register struct tree *link;
311 treefails(tree->llink, fail, recourse);
312 treefails(tree->rlink, fail, recourse);
314 /* Find, in the chain of fails going back to the root, the first
315 node that has a descendant on the current label. */
319 while (link && tree->label != link->label)
320 if (tree->label < link->label)
326 tree->trie->fail = link->trie;
332 tree->trie->fail = recourse;
335 /* Set delta entries for the links of the given tree such that
336 the preexisting delta value is larger than the current depth. */
338 treedelta (register struct tree const *tree,
339 register unsigned int depth,
340 unsigned char delta[])
344 treedelta(tree->llink, depth, delta);
345 treedelta(tree->rlink, depth, delta);
346 if (depth < delta[tree->label])
347 delta[tree->label] = depth;
350 /* Return true if A has every label in B. */
352 hasevery (register struct tree const *a, register struct tree const *b)
356 if (!hasevery(a, b->llink))
358 if (!hasevery(a, b->rlink))
360 while (a && b->label != a->label)
361 if (b->label < a->label)
368 /* Compute a vector, indexed by character code, of the trie nodes
369 referenced from the given tree. */
371 treenext (struct tree const *tree, struct trie *next[])
375 treenext(tree->llink, next);
376 treenext(tree->rlink, next);
377 next[tree->label] = tree->trie;
380 /* Compute the shift for each trie node, as well as the delta
381 table and next cache for the given keyword set. */
383 kwsprep (kwset_t kws)
385 register struct kwset *kwset;
387 register struct trie *curr;
388 register unsigned char const *trans;
389 unsigned char delta[NCHAR];
391 kwset = (struct kwset *) kws;
393 /* Initial values for the delta table; will be changed later. The
394 delta entry for a given character is the smallest depth of any
395 node at which an outgoing edge is labeled by that character. */
396 memset(delta, kwset->mind < UCHAR_MAX ? kwset->mind : UCHAR_MAX, NCHAR);
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 == NULL)
404 /* Looking for just one string. Extract it from the trie. */
405 kwset->target = obstack_alloc(&kwset->obstack, kwset->mind);
407 return "memory exhausted";
408 for (i = kwset->mind - 1, curr = kwset->trie; i >= 0; --i)
410 kwset->target[i] = curr->links->label;
411 curr = curr->links->trie;
413 /* Build the Boyer Moore delta. Boy that's easy compared to CW. */
414 for (i = 0; i < kwset->mind; ++i)
415 delta[U(kwset->target[i])] = kwset->mind - (i + 1);
416 /* Find the minimal delta2 shift that we might make after
417 a backwards match has failed. */
418 c = kwset->target[kwset->mind - 1];
419 for (i = kwset->mind - 2; i >= 0; --i)
420 if (kwset->target[i] == c)
422 kwset->mind2 = kwset->mind - (i + 1);
426 register struct trie *fail;
427 struct trie *last, *next[NCHAR];
429 /* Traverse the nodes of the trie in level order, simultaneously
430 computing the delta table, failure function, and shift function. */
431 for (curr = last = kwset->trie; curr; curr = curr->next)
433 /* Enqueue the immediate descendants 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 descendants 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 descendants 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. */
476 for (i = 0; i < NCHAR; ++i)
478 treenext(kwset->trie->links, next);
480 if ((trans = kwset->trans) != NULL)
481 for (i = 0; i < NCHAR; ++i)
482 kwset->next[i] = next[U(trans[i])];
484 COPY_ARRAY(kwset->next, next, NCHAR);
487 /* Fix things up for any translation table. */
488 if ((trans = kwset->trans) != NULL)
489 for (i = 0; i < NCHAR; ++i)
490 kwset->delta[i] = delta[U(trans[i])];
492 memcpy(kwset->delta, delta, NCHAR);
497 /* Fast boyer-moore search. */
499 bmexec (kwset_t kws, char const *text, size_t size)
501 struct kwset const *kwset;
502 register unsigned char const *d1;
503 register char const *ep, *sp, *tp;
504 register int d, gc, i, len, md2;
506 kwset = (struct kwset const *) kws;
515 tp = memchr (text, kwset->target[0], size);
516 return tp ? tp - text : -1;
520 sp = kwset->target + len;
525 /* Significance of 12: 1 (initial offset) + 10 (skip loop) + 1 (md2). */
527 /* 11 is not a bug, the initial offset happens only once. */
528 for (ep = text + size - 11 * len;;)
532 d = d1[U(tp[-1])], tp += d;
533 d = d1[U(tp[-1])], tp += d;
536 d = d1[U(tp[-1])], tp += d;
537 d = d1[U(tp[-1])], tp += d;
538 d = d1[U(tp[-1])], tp += d;
541 d = d1[U(tp[-1])], tp += d;
542 d = d1[U(tp[-1])], tp += d;
543 d = d1[U(tp[-1])], tp += d;
546 d = d1[U(tp[-1])], tp += d;
547 d = d1[U(tp[-1])], tp += d;
553 for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i)
556 return tp - len - text;
561 /* Now we have only a few characters left to search. We
562 carefully avoid ever producing an out-of-bounds pointer. */
567 d = d1[U((tp += d)[-1])];
572 for (i = 3; i <= len && U(tp[-i]) == U(sp[-i]); ++i)
575 return tp - len - text;
583 /* Hairy multiple string search. */
585 cwexec (kwset_t kws, char const *text, size_t len, struct kwsmatch *kwsmatch)
587 struct kwset const *kwset;
588 struct trie * const *next;
589 struct trie const *trie;
590 struct trie const *accept;
591 char const *beg, *lim, *mch, *lmch;
592 register unsigned char c;
593 register unsigned char const *delta;
595 register char const *end, *qlim;
596 register struct tree const *tree;
597 register unsigned char const *trans;
601 /* Initialize register copies and look for easy ways out. */
602 kwset = (struct kwset *) kws;
603 if (len < kwset->mind)
606 delta = kwset->delta;
607 trans = kwset->trans;
610 if ((d = kwset->mind) != 0)
614 mch = text, accept = kwset->trie;
618 if (len >= 4 * kwset->mind)
619 qlim = lim - 4 * kwset->mind;
623 while (lim - end >= d)
625 if (qlim && end <= qlim)
628 while ((d = delta[c = *end]) && end < qlim)
631 end += delta[U(*end)];
632 end += delta[U(*end)];
637 d = delta[c = (end += d)[-1]];
650 c = trans ? trans[U(*--beg)] : *--beg;
652 while (tree && c != tree->label)
676 /* Given a known match, find the longest possible match anchored
677 at or before its starting point. This is nearly a verbatim
678 copy of the preceding main search loops. */
679 if (lim - mch > kwset->maxd)
680 lim = mch + kwset->maxd;
683 while (lim - end >= d)
685 if ((d = delta[c = (end += d)[-1]]) != 0)
688 if (!(trie = next[c]))
693 if (trie->accepting && beg <= mch)
701 c = trans ? trans[U(*--beg)] : *--beg;
703 while (tree && c != tree->label)
711 if (trie->accepting && beg <= mch)
732 kwsmatch->index = accept->accepting / 2;
733 kwsmatch->offset[0] = mch - text;
734 kwsmatch->size[0] = accept->depth;
739 /* Search through the given text for a match of any member of the
740 given keyword set. Return a pointer to the first character of
741 the matching substring, or NULL if no match is found. If FOUNDLEN
742 is non-NULL store in the referenced location the length of the
743 matching substring. Similarly, if FOUNDIDX is non-NULL, store
744 in the referenced location the index number of the particular
747 kwsexec (kwset_t kws, char const *text, size_t size,
748 struct kwsmatch *kwsmatch)
750 struct kwset const *kwset = (struct kwset *) kws;
751 if (kwset->words == 1 && kwset->trans == NULL)
753 size_t ret = bmexec (kws, text, size);
754 if (kwsmatch != NULL && ret != (size_t) -1)
757 kwsmatch->offset[0] = ret;
758 kwsmatch->size[0] = kwset->mind;
763 return cwexec(kws, text, size, kwsmatch);
766 /* Free the components of the given keyword set. */
768 kwsfree (kwset_t kws)
772 kwset = (struct kwset *) kws;
773 obstack_free(&kwset->obstack, NULL);