2000-09-14 19:02:50 +04:00
|
|
|
/*
|
2001-04-16 21:18:24 +04:00
|
|
|
* tree234.c: reasonably generic counted 2-3-4 tree routines.
|
|
|
|
*
|
|
|
|
* This file is copyright 1999-2001 Simon Tatham.
|
|
|
|
*
|
|
|
|
* Permission is hereby granted, free of charge, to any person
|
|
|
|
* obtaining a copy of this software and associated documentation
|
|
|
|
* files (the "Software"), to deal in the Software without
|
|
|
|
* restriction, including without limitation the rights to use,
|
|
|
|
* copy, modify, merge, publish, distribute, sublicense, and/or
|
|
|
|
* sell copies of the Software, and to permit persons to whom the
|
|
|
|
* Software is furnished to do so, subject to the following
|
|
|
|
* conditions:
|
|
|
|
*
|
|
|
|
* The above copyright notice and this permission notice shall be
|
|
|
|
* included in all copies or substantial portions of the Software.
|
|
|
|
*
|
|
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
|
|
|
|
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
|
|
|
|
* OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
|
|
|
|
* NONINFRINGEMENT. IN NO EVENT SHALL SIMON TATHAM BE LIABLE FOR
|
|
|
|
* ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
|
|
|
|
* CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
|
|
|
|
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
|
|
* SOFTWARE.
|
2000-09-14 19:02:50 +04:00
|
|
|
*/
|
|
|
|
|
|
|
|
#include <stdio.h>
|
|
|
|
#include <stdlib.h>
|
2001-04-16 21:18:24 +04:00
|
|
|
#include <assert.h>
|
2000-12-12 13:33:13 +03:00
|
|
|
|
2000-09-14 19:02:50 +04:00
|
|
|
#include "tree234.h"
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
#define smalloc malloc
|
|
|
|
#define sfree free
|
|
|
|
|
2000-12-12 13:33:13 +03:00
|
|
|
#define mknew(typ) ( (typ *) smalloc (sizeof (typ)) )
|
2000-09-14 19:02:50 +04:00
|
|
|
|
|
|
|
#ifdef TEST
|
|
|
|
#define LOG(x) (printf x)
|
|
|
|
#else
|
|
|
|
#define LOG(x)
|
|
|
|
#endif
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
typedef struct node234_Tag node234;
|
|
|
|
|
2000-09-14 19:02:50 +04:00
|
|
|
struct tree234_Tag {
|
|
|
|
node234 *root;
|
|
|
|
cmpfn234 cmp;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct node234_Tag {
|
|
|
|
node234 *parent;
|
|
|
|
node234 *kids[4];
|
2001-04-16 21:18:24 +04:00
|
|
|
int counts[4];
|
2000-09-14 19:02:50 +04:00
|
|
|
void *elems[3];
|
|
|
|
};
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Create a 2-3-4 tree.
|
|
|
|
*/
|
|
|
|
tree234 *newtree234(cmpfn234 cmp) {
|
|
|
|
tree234 *ret = mknew(tree234);
|
|
|
|
LOG(("created tree %p\n", ret));
|
|
|
|
ret->root = NULL;
|
|
|
|
ret->cmp = cmp;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Free a 2-3-4 tree (not including freeing the elements).
|
|
|
|
*/
|
|
|
|
static void freenode234(node234 *n) {
|
|
|
|
if (!n)
|
|
|
|
return;
|
|
|
|
freenode234(n->kids[0]);
|
|
|
|
freenode234(n->kids[1]);
|
|
|
|
freenode234(n->kids[2]);
|
|
|
|
freenode234(n->kids[3]);
|
|
|
|
sfree(n);
|
|
|
|
}
|
|
|
|
void freetree234(tree234 *t) {
|
|
|
|
freenode234(t->root);
|
|
|
|
sfree(t);
|
|
|
|
}
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
/*
|
|
|
|
* Internal function to count a node.
|
|
|
|
*/
|
|
|
|
static int countnode234(node234 *n) {
|
|
|
|
int count = 0;
|
|
|
|
int i;
|
|
|
|
for (i = 0; i < 4; i++)
|
|
|
|
count += n->counts[i];
|
|
|
|
for (i = 0; i < 3; i++)
|
|
|
|
if (n->elems[i])
|
|
|
|
count++;
|
|
|
|
return count;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Count the elements in a tree.
|
|
|
|
*/
|
|
|
|
int count234(tree234 *t) {
|
|
|
|
if (t->root)
|
|
|
|
return countnode234(t->root);
|
|
|
|
else
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2000-09-14 19:02:50 +04:00
|
|
|
/*
|
|
|
|
* Add an element e to a 2-3-4 tree t. Returns e on success, or if
|
|
|
|
* an existing element compares equal, returns that.
|
|
|
|
*/
|
2001-04-16 21:18:24 +04:00
|
|
|
static void *add234_internal(tree234 *t, void *e, int index) {
|
2000-09-14 19:02:50 +04:00
|
|
|
node234 *n, **np, *left, *right;
|
|
|
|
void *orig_e = e;
|
2001-04-16 21:18:24 +04:00
|
|
|
int c, lcount, rcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
|
|
|
|
LOG(("adding node %p to tree %p\n", e, t));
|
|
|
|
if (t->root == NULL) {
|
|
|
|
t->root = mknew(node234);
|
|
|
|
t->root->elems[1] = t->root->elems[2] = NULL;
|
|
|
|
t->root->kids[0] = t->root->kids[1] = NULL;
|
|
|
|
t->root->kids[2] = t->root->kids[3] = NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
t->root->counts[0] = t->root->counts[1] = 0;
|
|
|
|
t->root->counts[2] = t->root->counts[3] = 0;
|
2000-09-14 19:02:50 +04:00
|
|
|
t->root->parent = NULL;
|
|
|
|
t->root->elems[0] = e;
|
|
|
|
LOG((" created root %p\n", t->root));
|
|
|
|
return orig_e;
|
|
|
|
}
|
|
|
|
|
|
|
|
np = &t->root;
|
|
|
|
while (*np) {
|
2001-04-16 21:18:24 +04:00
|
|
|
int childnum;
|
2000-09-14 19:02:50 +04:00
|
|
|
n = *np;
|
2001-04-16 21:18:24 +04:00
|
|
|
LOG((" node %p: %p/%d [%p] %p/%d [%p] %p/%d [%p] %p/%d\n",
|
|
|
|
n,
|
|
|
|
n->kids[0], n->counts[0], n->elems[0],
|
|
|
|
n->kids[1], n->counts[1], n->elems[1],
|
|
|
|
n->kids[2], n->counts[2], n->elems[2],
|
|
|
|
n->kids[3], n->counts[3]));
|
|
|
|
if (index >= 0) {
|
|
|
|
if (!n->kids[0]) {
|
|
|
|
/*
|
|
|
|
* Leaf node. We want to insert at kid position
|
|
|
|
* equal to the index:
|
|
|
|
*
|
|
|
|
* 0 A 1 B 2 C 3
|
|
|
|
*/
|
|
|
|
childnum = index;
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Internal node. We always descend through it (add
|
|
|
|
* always starts at the bottom, never in the
|
|
|
|
* middle).
|
|
|
|
*/
|
|
|
|
do { /* this is a do ... while (0) to allow `break' */
|
|
|
|
if (index <= n->counts[0]) {
|
|
|
|
childnum = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
index -= n->counts[0] + 1;
|
|
|
|
if (index <= n->counts[1]) {
|
|
|
|
childnum = 1;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
index -= n->counts[1] + 1;
|
|
|
|
if (index <= n->counts[2]) {
|
|
|
|
childnum = 2;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
index -= n->counts[2] + 1;
|
|
|
|
if (index <= n->counts[3]) {
|
|
|
|
childnum = 3;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
return NULL; /* error: index out of range */
|
|
|
|
} while (0);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if ((c = t->cmp(e, n->elems[0])) < 0)
|
|
|
|
childnum = 0;
|
|
|
|
else if (c == 0)
|
|
|
|
return n->elems[0]; /* already exists */
|
|
|
|
else if (n->elems[1] == NULL || (c = t->cmp(e, n->elems[1])) < 0)
|
|
|
|
childnum = 1;
|
|
|
|
else if (c == 0)
|
|
|
|
return n->elems[1]; /* already exists */
|
|
|
|
else if (n->elems[2] == NULL || (c = t->cmp(e, n->elems[2])) < 0)
|
|
|
|
childnum = 2;
|
|
|
|
else if (c == 0)
|
|
|
|
return n->elems[2]; /* already exists */
|
|
|
|
else
|
|
|
|
childnum = 3;
|
|
|
|
}
|
|
|
|
np = &n->kids[childnum];
|
|
|
|
LOG((" moving to child %d (%p)\n", childnum, *np));
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We need to insert the new element in n at position np.
|
|
|
|
*/
|
2001-04-16 21:18:24 +04:00
|
|
|
left = NULL; lcount = 0;
|
|
|
|
right = NULL; rcount = 0;
|
2000-09-14 19:02:50 +04:00
|
|
|
while (n) {
|
2001-04-16 21:18:24 +04:00
|
|
|
LOG((" at %p: %p/%d [%p] %p/%d [%p] %p/%d [%p] %p/%d\n",
|
|
|
|
n,
|
|
|
|
n->kids[0], n->counts[0], n->elems[0],
|
|
|
|
n->kids[1], n->counts[1], n->elems[1],
|
|
|
|
n->kids[2], n->counts[2], n->elems[2],
|
|
|
|
n->kids[3], n->counts[3]));
|
|
|
|
LOG((" need to insert %p/%d [%p] %p/%d at position %d\n",
|
|
|
|
left, lcount, e, right, rcount, np - n->kids));
|
2000-09-14 19:02:50 +04:00
|
|
|
if (n->elems[1] == NULL) {
|
|
|
|
/*
|
|
|
|
* Insert in a 2-node; simple.
|
|
|
|
*/
|
|
|
|
if (np == &n->kids[0]) {
|
|
|
|
LOG((" inserting on left of 2-node\n"));
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[2] = n->kids[1]; n->counts[2] = n->counts[1];
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[1] = n->elems[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[1] = right; n->counts[1] = rcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[0] = e;
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[0] = left; n->counts[0] = lcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
} else { /* np == &n->kids[1] */
|
|
|
|
LOG((" inserting on right of 2-node\n"));
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[2] = right; n->counts[2] = rcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[1] = e;
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[1] = left; n->counts[1] = lcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
|
|
|
if (n->kids[0]) n->kids[0]->parent = n;
|
|
|
|
if (n->kids[1]) n->kids[1]->parent = n;
|
|
|
|
if (n->kids[2]) n->kids[2]->parent = n;
|
|
|
|
LOG((" done\n"));
|
|
|
|
break;
|
|
|
|
} else if (n->elems[2] == NULL) {
|
|
|
|
/*
|
|
|
|
* Insert in a 3-node; simple.
|
|
|
|
*/
|
|
|
|
if (np == &n->kids[0]) {
|
|
|
|
LOG((" inserting on left of 3-node\n"));
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[3] = n->kids[2]; n->counts[3] = n->counts[2];
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[2] = n->elems[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[2] = n->kids[1]; n->counts[2] = n->counts[1];
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[1] = n->elems[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[1] = right; n->counts[1] = rcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[0] = e;
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[0] = left; n->counts[0] = lcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
} else if (np == &n->kids[1]) {
|
|
|
|
LOG((" inserting in middle of 3-node\n"));
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[3] = n->kids[2]; n->counts[3] = n->counts[2];
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[2] = n->elems[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[2] = right; n->counts[2] = rcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[1] = e;
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[1] = left; n->counts[1] = lcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
} else { /* np == &n->kids[2] */
|
|
|
|
LOG((" inserting on right of 3-node\n"));
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[3] = right; n->counts[3] = rcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[2] = e;
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[2] = left; n->counts[2] = lcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
|
|
|
if (n->kids[0]) n->kids[0]->parent = n;
|
|
|
|
if (n->kids[1]) n->kids[1]->parent = n;
|
|
|
|
if (n->kids[2]) n->kids[2]->parent = n;
|
|
|
|
if (n->kids[3]) n->kids[3]->parent = n;
|
|
|
|
LOG((" done\n"));
|
|
|
|
break;
|
|
|
|
} else {
|
|
|
|
node234 *m = mknew(node234);
|
|
|
|
m->parent = n->parent;
|
|
|
|
LOG((" splitting a 4-node; created new node %p\n", m));
|
|
|
|
/*
|
|
|
|
* Insert in a 4-node; split into a 2-node and a
|
|
|
|
* 3-node, and move focus up a level.
|
|
|
|
*
|
|
|
|
* I don't think it matters which way round we put the
|
|
|
|
* 2 and the 3. For simplicity, we'll put the 3 first
|
|
|
|
* always.
|
|
|
|
*/
|
|
|
|
if (np == &n->kids[0]) {
|
2001-04-16 21:18:24 +04:00
|
|
|
m->kids[0] = left; m->counts[0] = lcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
m->elems[0] = e;
|
2001-04-16 21:18:24 +04:00
|
|
|
m->kids[1] = right; m->counts[1] = rcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
m->elems[1] = n->elems[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
m->kids[2] = n->kids[1]; m->counts[2] = n->counts[1];
|
2000-09-14 19:02:50 +04:00
|
|
|
e = n->elems[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[0] = n->kids[2]; n->counts[0] = n->counts[2];
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[0] = n->elems[2];
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[1] = n->kids[3]; n->counts[1] = n->counts[3];
|
2000-09-14 19:02:50 +04:00
|
|
|
} else if (np == &n->kids[1]) {
|
2001-04-16 21:18:24 +04:00
|
|
|
m->kids[0] = n->kids[0]; m->counts[0] = n->counts[0];
|
2000-09-14 19:02:50 +04:00
|
|
|
m->elems[0] = n->elems[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
m->kids[1] = left; m->counts[1] = lcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
m->elems[1] = e;
|
2001-04-16 21:18:24 +04:00
|
|
|
m->kids[2] = right; m->counts[2] = rcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
e = n->elems[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[0] = n->kids[2]; n->counts[0] = n->counts[2];
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[0] = n->elems[2];
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[1] = n->kids[3]; n->counts[1] = n->counts[3];
|
2000-09-14 19:02:50 +04:00
|
|
|
} else if (np == &n->kids[2]) {
|
2001-04-16 21:18:24 +04:00
|
|
|
m->kids[0] = n->kids[0]; m->counts[0] = n->counts[0];
|
2000-09-14 19:02:50 +04:00
|
|
|
m->elems[0] = n->elems[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
m->kids[1] = n->kids[1]; m->counts[1] = n->counts[1];
|
2000-09-14 19:02:50 +04:00
|
|
|
m->elems[1] = n->elems[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
m->kids[2] = left; m->counts[2] = lcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
/* e = e; */
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[0] = right; n->counts[0] = rcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[0] = n->elems[2];
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[1] = n->kids[3]; n->counts[1] = n->counts[3];
|
2000-09-14 19:02:50 +04:00
|
|
|
} else { /* np == &n->kids[3] */
|
2001-04-16 21:18:24 +04:00
|
|
|
m->kids[0] = n->kids[0]; m->counts[0] = n->counts[0];
|
2000-09-14 19:02:50 +04:00
|
|
|
m->elems[0] = n->elems[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
m->kids[1] = n->kids[1]; m->counts[1] = n->counts[1];
|
2000-09-14 19:02:50 +04:00
|
|
|
m->elems[1] = n->elems[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
m->kids[2] = n->kids[2]; m->counts[2] = n->counts[2];
|
|
|
|
n->kids[0] = left; n->counts[0] = lcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[0] = e;
|
2001-04-16 21:18:24 +04:00
|
|
|
n->kids[1] = right; n->counts[1] = rcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
e = n->elems[2];
|
|
|
|
}
|
|
|
|
m->kids[3] = n->kids[3] = n->kids[2] = NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
m->counts[3] = n->counts[3] = n->counts[2] = 0;
|
2000-09-14 19:02:50 +04:00
|
|
|
m->elems[2] = n->elems[2] = n->elems[1] = NULL;
|
|
|
|
if (m->kids[0]) m->kids[0]->parent = m;
|
|
|
|
if (m->kids[1]) m->kids[1]->parent = m;
|
|
|
|
if (m->kids[2]) m->kids[2]->parent = m;
|
|
|
|
if (n->kids[0]) n->kids[0]->parent = n;
|
|
|
|
if (n->kids[1]) n->kids[1]->parent = n;
|
2001-04-16 21:18:24 +04:00
|
|
|
LOG((" left (%p): %p/%d [%p] %p/%d [%p] %p/%d\n", m,
|
|
|
|
m->kids[0], m->counts[0], m->elems[0],
|
|
|
|
m->kids[1], m->counts[1], m->elems[1],
|
|
|
|
m->kids[2], m->counts[2]));
|
|
|
|
LOG((" right (%p): %p/%d [%p] %p/%d\n", n,
|
|
|
|
n->kids[0], n->counts[0], n->elems[0],
|
|
|
|
n->kids[1], n->counts[1]));
|
|
|
|
left = m; lcount = countnode234(left);
|
|
|
|
right = n; rcount = countnode234(right);
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
|
|
|
if (n->parent)
|
|
|
|
np = (n->parent->kids[0] == n ? &n->parent->kids[0] :
|
|
|
|
n->parent->kids[1] == n ? &n->parent->kids[1] :
|
|
|
|
n->parent->kids[2] == n ? &n->parent->kids[2] :
|
|
|
|
&n->parent->kids[3]);
|
|
|
|
n = n->parent;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If we've come out of here by `break', n will still be
|
2001-04-16 21:18:24 +04:00
|
|
|
* non-NULL and all we need to do is go back up the tree
|
|
|
|
* updating counts. If we've come here because n is NULL, we
|
|
|
|
* need to create a new root for the tree because the old one
|
|
|
|
* has just split into two. */
|
|
|
|
if (n) {
|
|
|
|
while (n->parent) {
|
|
|
|
int count = countnode234(n);
|
|
|
|
int childnum;
|
|
|
|
childnum = (n->parent->kids[0] == n ? 0 :
|
|
|
|
n->parent->kids[1] == n ? 1 :
|
|
|
|
n->parent->kids[2] == n ? 2 : 3);
|
|
|
|
n->parent->counts[childnum] = count;
|
|
|
|
n = n->parent;
|
|
|
|
}
|
|
|
|
} else {
|
2000-09-14 19:02:50 +04:00
|
|
|
LOG((" root is overloaded, split into two\n"));
|
|
|
|
t->root = mknew(node234);
|
2001-04-16 21:18:24 +04:00
|
|
|
t->root->kids[0] = left; t->root->counts[0] = lcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
t->root->elems[0] = e;
|
2001-04-16 21:18:24 +04:00
|
|
|
t->root->kids[1] = right; t->root->counts[1] = rcount;
|
2000-09-14 19:02:50 +04:00
|
|
|
t->root->elems[1] = NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
t->root->kids[2] = NULL; t->root->counts[2] = 0;
|
2000-09-14 19:02:50 +04:00
|
|
|
t->root->elems[2] = NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
t->root->kids[3] = NULL; t->root->counts[3] = 0;
|
2000-09-14 19:02:50 +04:00
|
|
|
t->root->parent = NULL;
|
|
|
|
if (t->root->kids[0]) t->root->kids[0]->parent = t->root;
|
|
|
|
if (t->root->kids[1]) t->root->kids[1]->parent = t->root;
|
2001-04-16 21:18:24 +04:00
|
|
|
LOG((" new root is %p/%d [%p] %p/%d\n",
|
|
|
|
t->root->kids[0], t->root->counts[0],
|
|
|
|
t->root->elems[0],
|
|
|
|
t->root->kids[1], t->root->counts[1]));
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
return orig_e;
|
|
|
|
}
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
void *add234(tree234 *t, void *e) {
|
|
|
|
if (!t->cmp) /* tree is unsorted */
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return add234_internal(t, e, -1);
|
|
|
|
}
|
|
|
|
void *addpos234(tree234 *t, void *e, int index) {
|
|
|
|
if (index < 0 || /* index out of range */
|
|
|
|
t->cmp) /* tree is sorted */
|
|
|
|
return NULL; /* return failure */
|
|
|
|
|
|
|
|
return add234_internal(t, e, index); /* this checks the upper bound */
|
|
|
|
}
|
|
|
|
|
2000-09-14 19:02:50 +04:00
|
|
|
/*
|
2001-04-16 21:18:24 +04:00
|
|
|
* Look up the element at a given numeric index in a 2-3-4 tree.
|
|
|
|
* Returns NULL if the index is out of range.
|
2000-09-14 19:02:50 +04:00
|
|
|
*/
|
2001-04-16 21:18:24 +04:00
|
|
|
void *index234(tree234 *t, int index) {
|
2000-09-14 19:02:50 +04:00
|
|
|
node234 *n;
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
if (!t->root)
|
|
|
|
return NULL; /* tree is empty */
|
2000-09-14 19:02:50 +04:00
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
if (index < 0 || index >= countnode234(t->root))
|
|
|
|
return NULL; /* out of range */
|
2000-09-14 19:02:50 +04:00
|
|
|
|
|
|
|
n = t->root;
|
2001-04-16 21:18:24 +04:00
|
|
|
|
2000-09-14 19:02:50 +04:00
|
|
|
while (n) {
|
2001-04-16 21:18:24 +04:00
|
|
|
if (index < n->counts[0])
|
2000-09-14 19:02:50 +04:00
|
|
|
n = n->kids[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
else if (index -= n->counts[0] + 1, index < 0)
|
2000-09-14 19:02:50 +04:00
|
|
|
return n->elems[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
else if (index < n->counts[1])
|
2000-09-14 19:02:50 +04:00
|
|
|
n = n->kids[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
else if (index -= n->counts[1] + 1, index < 0)
|
2000-09-14 19:02:50 +04:00
|
|
|
return n->elems[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
else if (index < n->counts[2])
|
2000-09-14 19:02:50 +04:00
|
|
|
n = n->kids[2];
|
2001-04-16 21:18:24 +04:00
|
|
|
else if (index -= n->counts[2] + 1, index < 0)
|
2000-09-14 19:02:50 +04:00
|
|
|
return n->elems[2];
|
|
|
|
else
|
|
|
|
n = n->kids[3];
|
|
|
|
}
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
/* We shouldn't ever get here. I wonder how we did. */
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Find an element e in a sorted 2-3-4 tree t. Returns NULL if not
|
|
|
|
* found. e is always passed as the first argument to cmp, so cmp
|
|
|
|
* can be an asymmetric function if desired. cmp can also be passed
|
|
|
|
* as NULL, in which case the compare function from the tree proper
|
|
|
|
* will be used.
|
|
|
|
*/
|
|
|
|
void *findrelpos234(tree234 *t, void *e, cmpfn234 cmp,
|
|
|
|
int relation, int *index) {
|
|
|
|
node234 *n;
|
|
|
|
void *ret;
|
|
|
|
int c;
|
|
|
|
int idx, ecount, kcount, cmpret;
|
|
|
|
|
|
|
|
if (t->root == NULL)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (cmp == NULL)
|
|
|
|
cmp = t->cmp;
|
|
|
|
|
|
|
|
n = t->root;
|
2000-09-14 19:02:50 +04:00
|
|
|
/*
|
2001-04-16 21:18:24 +04:00
|
|
|
* Attempt to find the element itself.
|
2000-09-14 19:02:50 +04:00
|
|
|
*/
|
2001-04-16 21:18:24 +04:00
|
|
|
idx = 0;
|
|
|
|
ecount = -1;
|
|
|
|
/*
|
|
|
|
* Prepare a fake `cmp' result if e is NULL.
|
|
|
|
*/
|
|
|
|
cmpret = 0;
|
|
|
|
if (e == NULL) {
|
|
|
|
assert(relation == REL234_LT || relation == REL234_GT);
|
|
|
|
if (relation == REL234_LT)
|
|
|
|
cmpret = +1; /* e is a max: always greater */
|
|
|
|
else if (relation == REL234_GT)
|
|
|
|
cmpret = -1; /* e is a min: always smaller */
|
|
|
|
}
|
|
|
|
while (1) {
|
|
|
|
for (kcount = 0; kcount < 4; kcount++) {
|
|
|
|
if (kcount >= 3 || n->elems[kcount] == NULL ||
|
|
|
|
(c = cmpret ? cmpret : cmp(e, n->elems[kcount])) < 0) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (n->kids[kcount]) idx += n->counts[kcount];
|
|
|
|
if (c == 0) {
|
|
|
|
ecount = kcount;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
idx++;
|
|
|
|
}
|
|
|
|
if (ecount >= 0)
|
|
|
|
break;
|
|
|
|
if (n->kids[kcount])
|
|
|
|
n = n->kids[kcount];
|
|
|
|
else
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (ecount >= 0) {
|
|
|
|
/*
|
|
|
|
* We have found the element we're looking for. It's
|
|
|
|
* n->elems[ecount], at tree index idx. If our search
|
|
|
|
* relation is EQ, LE or GE we can now go home.
|
|
|
|
*/
|
|
|
|
if (relation != REL234_LT && relation != REL234_GT) {
|
|
|
|
if (index) *index = idx;
|
|
|
|
return n->elems[ecount];
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Otherwise, we'll do an indexed lookup for the previous
|
|
|
|
* or next element. (It would be perfectly possible to
|
|
|
|
* implement these search types in a non-counted tree by
|
|
|
|
* going back up from where we are, but far more fiddly.)
|
|
|
|
*/
|
|
|
|
if (relation == REL234_LT)
|
|
|
|
idx--;
|
|
|
|
else
|
|
|
|
idx++;
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* We've found our way to the bottom of the tree and we
|
|
|
|
* know where we would insert this node if we wanted to:
|
|
|
|
* we'd put it in in place of the (empty) subtree
|
|
|
|
* n->kids[kcount], and it would have index idx
|
|
|
|
*
|
|
|
|
* But the actual element isn't there. So if our search
|
|
|
|
* relation is EQ, we're doomed.
|
|
|
|
*/
|
|
|
|
if (relation == REL234_EQ)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Otherwise, we must do an index lookup for index idx-1
|
|
|
|
* (if we're going left - LE or LT) or index idx (if we're
|
|
|
|
* going right - GE or GT).
|
|
|
|
*/
|
|
|
|
if (relation == REL234_LT || relation == REL234_LE) {
|
|
|
|
idx--;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We know the index of the element we want; just call index234
|
|
|
|
* to do the rest. This will return NULL if the index is out of
|
|
|
|
* bounds, which is exactly what we want.
|
|
|
|
*/
|
|
|
|
ret = index234(t, idx);
|
|
|
|
if (ret && index) *index = idx;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
void *find234(tree234 *t, void *e, cmpfn234 cmp) {
|
|
|
|
return findrelpos234(t, e, cmp, REL234_EQ, NULL);
|
|
|
|
}
|
|
|
|
void *findrel234(tree234 *t, void *e, cmpfn234 cmp, int relation) {
|
|
|
|
return findrelpos234(t, e, cmp, relation, NULL);
|
|
|
|
}
|
|
|
|
void *findpos234(tree234 *t, void *e, cmpfn234 cmp, int *index) {
|
|
|
|
return findrelpos234(t, e, cmp, REL234_EQ, index);
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Delete an element e in a 2-3-4 tree. Does not free the element,
|
|
|
|
* merely removes all links to it from the tree nodes.
|
|
|
|
*/
|
2001-04-16 21:18:24 +04:00
|
|
|
static void *delpos234_internal(tree234 *t, int index) {
|
2000-09-14 19:02:50 +04:00
|
|
|
node234 *n;
|
2001-04-16 21:18:24 +04:00
|
|
|
void *retval;
|
2000-09-14 19:02:50 +04:00
|
|
|
int ei = -1;
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
retval = 0;
|
|
|
|
|
2000-09-14 19:02:50 +04:00
|
|
|
n = t->root;
|
2001-04-16 21:18:24 +04:00
|
|
|
LOG(("deleting item %d from tree %p\n", index, t));
|
2000-09-14 19:02:50 +04:00
|
|
|
while (1) {
|
|
|
|
while (n) {
|
|
|
|
int c;
|
|
|
|
int ki;
|
|
|
|
node234 *sub;
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
LOG((" node %p: %p/%d [%p] %p/%d [%p] %p/%d [%p] %p/%d index=%d\n",
|
|
|
|
n,
|
|
|
|
n->kids[0], n->counts[0], n->elems[0],
|
|
|
|
n->kids[1], n->counts[1], n->elems[1],
|
|
|
|
n->kids[2], n->counts[2], n->elems[2],
|
|
|
|
n->kids[3], n->counts[3],
|
|
|
|
index));
|
|
|
|
if (index < n->counts[0]) {
|
2000-09-14 19:02:50 +04:00
|
|
|
ki = 0;
|
2001-04-16 21:18:24 +04:00
|
|
|
} else if (index -= n->counts[0]+1, index < 0) {
|
2000-09-14 19:02:50 +04:00
|
|
|
ei = 0; break;
|
2001-04-16 21:18:24 +04:00
|
|
|
} else if (index < n->counts[1]) {
|
2000-09-14 19:02:50 +04:00
|
|
|
ki = 1;
|
2001-04-16 21:18:24 +04:00
|
|
|
} else if (index -= n->counts[1]+1, index < 0) {
|
2000-09-14 19:02:50 +04:00
|
|
|
ei = 1; break;
|
2001-04-16 21:18:24 +04:00
|
|
|
} else if (index < n->counts[2]) {
|
2000-09-14 19:02:50 +04:00
|
|
|
ki = 2;
|
2001-04-16 21:18:24 +04:00
|
|
|
} else if (index -= n->counts[2]+1, index < 0) {
|
2000-09-14 19:02:50 +04:00
|
|
|
ei = 2; break;
|
|
|
|
} else {
|
|
|
|
ki = 3;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* Recurse down to subtree ki. If it has only one element,
|
|
|
|
* we have to do some transformation to start with.
|
|
|
|
*/
|
|
|
|
LOG((" moving to subtree %d\n", ki));
|
|
|
|
sub = n->kids[ki];
|
|
|
|
if (!sub->elems[1]) {
|
|
|
|
LOG((" subtree has only one element!\n", ki));
|
|
|
|
if (ki > 0 && n->kids[ki-1]->elems[1]) {
|
|
|
|
/*
|
|
|
|
* Case 3a, left-handed variant. Child ki has
|
|
|
|
* only one element, but child ki-1 has two or
|
|
|
|
* more. So we need to move a subtree from ki-1
|
|
|
|
* to ki.
|
|
|
|
*
|
|
|
|
* . C . . B .
|
|
|
|
* / \ -> / \
|
|
|
|
* [more] a A b B c d D e [more] a A b c C d D e
|
|
|
|
*/
|
|
|
|
node234 *sib = n->kids[ki-1];
|
|
|
|
int lastelem = (sib->elems[2] ? 2 :
|
|
|
|
sib->elems[1] ? 1 : 0);
|
|
|
|
sub->kids[2] = sub->kids[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
sub->counts[2] = sub->counts[1];
|
2000-09-14 19:02:50 +04:00
|
|
|
sub->elems[1] = sub->elems[0];
|
|
|
|
sub->kids[1] = sub->kids[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
sub->counts[1] = sub->counts[0];
|
2000-09-14 19:02:50 +04:00
|
|
|
sub->elems[0] = n->elems[ki-1];
|
|
|
|
sub->kids[0] = sib->kids[lastelem+1];
|
2001-04-16 21:18:24 +04:00
|
|
|
sub->counts[0] = sib->counts[lastelem+1];
|
2000-09-28 13:48:49 +04:00
|
|
|
if (sub->kids[0]) sub->kids[0]->parent = sub;
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[ki-1] = sib->elems[lastelem];
|
|
|
|
sib->kids[lastelem+1] = NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
sib->counts[lastelem+1] = 0;
|
2000-09-14 19:02:50 +04:00
|
|
|
sib->elems[lastelem] = NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
n->counts[ki] = countnode234(sub);
|
2000-09-14 19:02:50 +04:00
|
|
|
LOG((" case 3a left\n"));
|
2001-04-16 21:18:24 +04:00
|
|
|
LOG((" index and left subtree count before adjustment: %d, %d\n",
|
|
|
|
index, n->counts[ki-1]));
|
|
|
|
index += n->counts[ki-1];
|
|
|
|
n->counts[ki-1] = countnode234(sib);
|
|
|
|
index -= n->counts[ki-1];
|
|
|
|
LOG((" index and left subtree count after adjustment: %d, %d\n",
|
|
|
|
index, n->counts[ki-1]));
|
2000-09-14 19:02:50 +04:00
|
|
|
} else if (ki < 3 && n->kids[ki+1] &&
|
|
|
|
n->kids[ki+1]->elems[1]) {
|
|
|
|
/*
|
|
|
|
* Case 3a, right-handed variant. ki has only
|
|
|
|
* one element but ki+1 has two or more. Move a
|
|
|
|
* subtree from ki+1 to ki.
|
|
|
|
*
|
|
|
|
* . B . . C .
|
|
|
|
* / \ -> / \
|
|
|
|
* a A b c C d D e [more] a A b B c d D e [more]
|
|
|
|
*/
|
|
|
|
node234 *sib = n->kids[ki+1];
|
|
|
|
int j;
|
|
|
|
sub->elems[1] = n->elems[ki];
|
|
|
|
sub->kids[2] = sib->kids[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
sub->counts[2] = sib->counts[0];
|
2000-09-28 13:48:49 +04:00
|
|
|
if (sub->kids[2]) sub->kids[2]->parent = sub;
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[ki] = sib->elems[0];
|
|
|
|
sib->kids[0] = sib->kids[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
sib->counts[0] = sib->counts[1];
|
2000-09-14 19:02:50 +04:00
|
|
|
for (j = 0; j < 2 && sib->elems[j+1]; j++) {
|
|
|
|
sib->kids[j+1] = sib->kids[j+2];
|
2001-04-16 21:18:24 +04:00
|
|
|
sib->counts[j+1] = sib->counts[j+2];
|
2000-09-14 19:02:50 +04:00
|
|
|
sib->elems[j] = sib->elems[j+1];
|
|
|
|
}
|
|
|
|
sib->kids[j+1] = NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
sib->counts[j+1] = 0;
|
2000-09-14 19:02:50 +04:00
|
|
|
sib->elems[j] = NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
n->counts[ki] = countnode234(sub);
|
|
|
|
n->counts[ki+1] = countnode234(sib);
|
2000-09-14 19:02:50 +04:00
|
|
|
LOG((" case 3a right\n"));
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Case 3b. ki has only one element, and has no
|
|
|
|
* neighbour with more than one. So pick a
|
|
|
|
* neighbour and merge it with ki, taking an
|
|
|
|
* element down from n to go in the middle.
|
|
|
|
*
|
|
|
|
* . B . .
|
|
|
|
* / \ -> |
|
|
|
|
* a A b c C d a A b B c C d
|
|
|
|
*
|
|
|
|
* (Since at all points we have avoided
|
|
|
|
* descending to a node with only one element,
|
|
|
|
* we can be sure that n is not reduced to
|
|
|
|
* nothingness by this move, _unless_ it was
|
|
|
|
* the very first node, ie the root of the
|
|
|
|
* tree. In that case we remove the now-empty
|
|
|
|
* root and replace it with its single large
|
|
|
|
* child as shown.)
|
|
|
|
*/
|
|
|
|
node234 *sib;
|
|
|
|
int j;
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
if (ki > 0) {
|
2000-09-14 19:02:50 +04:00
|
|
|
ki--;
|
2001-04-16 21:18:24 +04:00
|
|
|
index += n->counts[ki] + 1;
|
|
|
|
}
|
2000-09-14 19:02:50 +04:00
|
|
|
sib = n->kids[ki];
|
|
|
|
sub = n->kids[ki+1];
|
|
|
|
|
|
|
|
sub->kids[3] = sub->kids[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
sub->counts[3] = sub->counts[1];
|
2000-09-14 19:02:50 +04:00
|
|
|
sub->elems[2] = sub->elems[0];
|
|
|
|
sub->kids[2] = sub->kids[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
sub->counts[2] = sub->counts[0];
|
2000-09-14 19:02:50 +04:00
|
|
|
sub->elems[1] = n->elems[ki];
|
|
|
|
sub->kids[1] = sib->kids[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
sub->counts[1] = sib->counts[1];
|
2000-09-28 13:48:49 +04:00
|
|
|
if (sub->kids[1]) sub->kids[1]->parent = sub;
|
2000-09-14 19:02:50 +04:00
|
|
|
sub->elems[0] = sib->elems[0];
|
|
|
|
sub->kids[0] = sib->kids[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
sub->counts[0] = sib->counts[0];
|
2000-09-28 13:48:49 +04:00
|
|
|
if (sub->kids[0]) sub->kids[0]->parent = sub;
|
2000-09-14 19:02:50 +04:00
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
n->counts[ki+1] = countnode234(sub);
|
|
|
|
|
2000-09-14 19:02:50 +04:00
|
|
|
sfree(sib);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* That's built the big node in sub. Now we
|
|
|
|
* need to remove the reference to sib in n.
|
|
|
|
*/
|
|
|
|
for (j = ki; j < 3 && n->kids[j+1]; j++) {
|
|
|
|
n->kids[j] = n->kids[j+1];
|
2001-04-16 21:18:24 +04:00
|
|
|
n->counts[j] = n->counts[j+1];
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[j] = j<2 ? n->elems[j+1] : NULL;
|
|
|
|
}
|
|
|
|
n->kids[j] = NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
n->counts[j] = 0;
|
2000-09-14 19:02:50 +04:00
|
|
|
if (j < 3) n->elems[j] = NULL;
|
2000-10-02 15:46:10 +04:00
|
|
|
LOG((" case 3b ki=%d\n", ki));
|
2000-09-14 19:02:50 +04:00
|
|
|
|
|
|
|
if (!n->elems[0]) {
|
|
|
|
/*
|
|
|
|
* The root is empty and needs to be
|
|
|
|
* removed.
|
|
|
|
*/
|
|
|
|
LOG((" shifting root!\n"));
|
|
|
|
t->root = sub;
|
|
|
|
sub->parent = NULL;
|
|
|
|
sfree(n);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
n = sub;
|
|
|
|
}
|
2001-04-16 21:18:24 +04:00
|
|
|
if (!retval)
|
|
|
|
retval = n->elems[ei];
|
|
|
|
|
2000-09-14 19:02:50 +04:00
|
|
|
if (ei==-1)
|
2001-04-16 21:18:24 +04:00
|
|
|
return NULL; /* although this shouldn't happen */
|
2000-09-14 19:02:50 +04:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Treat special case: this is the one remaining item in
|
|
|
|
* the tree. n is the tree root (no parent), has one
|
|
|
|
* element (no elems[1]), and has no kids (no kids[0]).
|
|
|
|
*/
|
|
|
|
if (!n->parent && !n->elems[1] && !n->kids[0]) {
|
|
|
|
LOG((" removed last element in tree\n"));
|
|
|
|
sfree(n);
|
|
|
|
t->root = NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
return retval;
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Now we have the element we want, as n->elems[ei], and we
|
|
|
|
* have also arranged for that element not to be the only
|
|
|
|
* one in its node. So...
|
|
|
|
*/
|
|
|
|
|
|
|
|
if (!n->kids[0] && n->elems[1]) {
|
|
|
|
/*
|
|
|
|
* Case 1. n is a leaf node with more than one element,
|
|
|
|
* so it's _really easy_. Just delete the thing and
|
|
|
|
* we're done.
|
|
|
|
*/
|
|
|
|
int i;
|
|
|
|
LOG((" case 1\n"));
|
2000-09-26 15:16:33 +04:00
|
|
|
for (i = ei; i < 2 && n->elems[i+1]; i++)
|
2000-09-14 19:02:50 +04:00
|
|
|
n->elems[i] = n->elems[i+1];
|
|
|
|
n->elems[i] = NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
/*
|
|
|
|
* Having done that to the leaf node, we now go back up
|
|
|
|
* the tree fixing the counts.
|
|
|
|
*/
|
|
|
|
while (n->parent) {
|
|
|
|
int childnum;
|
|
|
|
childnum = (n->parent->kids[0] == n ? 0 :
|
|
|
|
n->parent->kids[1] == n ? 1 :
|
|
|
|
n->parent->kids[2] == n ? 2 : 3);
|
|
|
|
n->parent->counts[childnum]--;
|
|
|
|
n = n->parent;
|
|
|
|
}
|
|
|
|
return retval; /* finished! */
|
2000-09-14 19:02:50 +04:00
|
|
|
} else if (n->kids[ei]->elems[1]) {
|
|
|
|
/*
|
|
|
|
* Case 2a. n is an internal node, and the root of the
|
|
|
|
* subtree to the left of e has more than one element.
|
|
|
|
* So find the predecessor p to e (ie the largest node
|
|
|
|
* in that subtree), place it where e currently is, and
|
|
|
|
* then start the deletion process over again on the
|
|
|
|
* subtree with p as target.
|
|
|
|
*/
|
|
|
|
node234 *m = n->kids[ei];
|
|
|
|
void *target;
|
|
|
|
LOG((" case 2a\n"));
|
|
|
|
while (m->kids[0]) {
|
|
|
|
m = (m->kids[3] ? m->kids[3] :
|
|
|
|
m->kids[2] ? m->kids[2] :
|
|
|
|
m->kids[1] ? m->kids[1] : m->kids[0]);
|
|
|
|
}
|
|
|
|
target = (m->elems[2] ? m->elems[2] :
|
|
|
|
m->elems[1] ? m->elems[1] : m->elems[0]);
|
|
|
|
n->elems[ei] = target;
|
2001-04-16 21:18:24 +04:00
|
|
|
index = n->counts[ei]-1;
|
2000-09-14 19:02:50 +04:00
|
|
|
n = n->kids[ei];
|
|
|
|
} else if (n->kids[ei+1]->elems[1]) {
|
|
|
|
/*
|
|
|
|
* Case 2b, symmetric to 2a but s/left/right/ and
|
|
|
|
* s/predecessor/successor/. (And s/largest/smallest/).
|
|
|
|
*/
|
|
|
|
node234 *m = n->kids[ei+1];
|
|
|
|
void *target;
|
|
|
|
LOG((" case 2b\n"));
|
|
|
|
while (m->kids[0]) {
|
|
|
|
m = m->kids[0];
|
|
|
|
}
|
|
|
|
target = m->elems[0];
|
|
|
|
n->elems[ei] = target;
|
|
|
|
n = n->kids[ei+1];
|
2001-04-16 21:18:24 +04:00
|
|
|
index = 0;
|
2000-09-14 19:02:50 +04:00
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Case 2c. n is an internal node, and the subtrees to
|
|
|
|
* the left and right of e both have only one element.
|
|
|
|
* So combine the two subnodes into a single big node
|
|
|
|
* with their own elements on the left and right and e
|
|
|
|
* in the middle, then restart the deletion process on
|
|
|
|
* that subtree, with e still as target.
|
|
|
|
*/
|
|
|
|
node234 *a = n->kids[ei], *b = n->kids[ei+1];
|
|
|
|
int j;
|
|
|
|
|
|
|
|
LOG((" case 2c\n"));
|
|
|
|
a->elems[1] = n->elems[ei];
|
|
|
|
a->kids[2] = b->kids[0];
|
2001-04-16 21:18:24 +04:00
|
|
|
a->counts[2] = b->counts[0];
|
2000-09-28 13:48:49 +04:00
|
|
|
if (a->kids[2]) a->kids[2]->parent = a;
|
2000-09-14 19:02:50 +04:00
|
|
|
a->elems[2] = b->elems[0];
|
|
|
|
a->kids[3] = b->kids[1];
|
2001-04-16 21:18:24 +04:00
|
|
|
a->counts[3] = b->counts[1];
|
2000-09-28 13:48:49 +04:00
|
|
|
if (a->kids[3]) a->kids[3]->parent = a;
|
2000-09-14 19:02:50 +04:00
|
|
|
sfree(b);
|
2001-04-16 21:18:24 +04:00
|
|
|
n->counts[ei] = countnode234(a);
|
2000-09-14 19:02:50 +04:00
|
|
|
/*
|
|
|
|
* That's built the big node in a, and destroyed b. Now
|
|
|
|
* remove the reference to b (and e) in n.
|
|
|
|
*/
|
|
|
|
for (j = ei; j < 2 && n->elems[j+1]; j++) {
|
|
|
|
n->elems[j] = n->elems[j+1];
|
|
|
|
n->kids[j+1] = n->kids[j+2];
|
2001-04-16 21:18:24 +04:00
|
|
|
n->counts[j+1] = n->counts[j+2];
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
|
|
|
n->elems[j] = NULL;
|
|
|
|
n->kids[j+1] = NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
n->counts[j+1] = 0;
|
2000-10-02 15:47:30 +04:00
|
|
|
/*
|
|
|
|
* It's possible, in this case, that we've just removed
|
|
|
|
* the only element in the root of the tree. If so,
|
|
|
|
* shift the root.
|
|
|
|
*/
|
|
|
|
if (n->elems[0] == NULL) {
|
|
|
|
LOG((" shifting root!\n"));
|
|
|
|
t->root = a;
|
|
|
|
a->parent = NULL;
|
|
|
|
sfree(n);
|
|
|
|
}
|
2000-09-14 19:02:50 +04:00
|
|
|
/*
|
|
|
|
* Now go round the deletion process again, with n
|
|
|
|
* pointing at the new big node and e still the same.
|
|
|
|
*/
|
|
|
|
n = a;
|
2001-04-16 21:18:24 +04:00
|
|
|
index = a->counts[0] + a->counts[1] + 1;
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2001-04-16 21:18:24 +04:00
|
|
|
void *delpos234(tree234 *t, int index) {
|
|
|
|
if (index < 0 || index >= countnode234(t->root))
|
2000-09-14 19:02:50 +04:00
|
|
|
return NULL;
|
2001-04-16 21:18:24 +04:00
|
|
|
return delpos234_internal(t, index);
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
2001-04-16 21:18:24 +04:00
|
|
|
void *del234(tree234 *t, void *e) {
|
|
|
|
int index;
|
|
|
|
if (!findrelpos234(t, e, NULL, REL234_EQ, &index))
|
|
|
|
return NULL; /* it wasn't in there anyway */
|
|
|
|
return delpos234_internal(t, index); /* it's there; delete it. */
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
#ifdef TEST
|
|
|
|
|
2000-10-02 15:46:10 +04:00
|
|
|
/*
|
|
|
|
* Test code for the 2-3-4 tree. This code maintains an alternative
|
|
|
|
* representation of the data in the tree, in an array (using the
|
|
|
|
* obvious and slow insert and delete functions). After each tree
|
2000-10-02 17:57:41 +04:00
|
|
|
* operation, the verify() function is called, which ensures all
|
2001-04-16 21:18:24 +04:00
|
|
|
* the tree properties are preserved:
|
|
|
|
* - node->child->parent always equals node
|
|
|
|
* - tree->root->parent always equals NULL
|
|
|
|
* - number of kids == 0 or number of elements + 1;
|
|
|
|
* - tree has the same depth everywhere
|
|
|
|
* - every node has at least one element
|
|
|
|
* - subtree element counts are accurate
|
|
|
|
* - any NULL kid pointer is accompanied by a zero count
|
|
|
|
* - in a sorted tree: ordering property between elements of a
|
|
|
|
* node and elements of its children is preserved
|
|
|
|
* and also ensures the list represented by the tree is the same
|
|
|
|
* list it should be. (This last check also doubly verifies the
|
|
|
|
* ordering properties, because the `same list it should be' is by
|
|
|
|
* definition correctly ordered. It also ensures all nodes are
|
|
|
|
* distinct, because the enum functions would get caught in a loop
|
|
|
|
* if not.)
|
2000-10-02 15:46:10 +04:00
|
|
|
*/
|
|
|
|
|
|
|
|
#include <stdarg.h>
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
#define srealloc realloc
|
|
|
|
|
2000-10-02 15:46:10 +04:00
|
|
|
/*
|
|
|
|
* Error reporting function.
|
|
|
|
*/
|
|
|
|
void error(char *fmt, ...) {
|
|
|
|
va_list ap;
|
|
|
|
printf("ERROR: ");
|
|
|
|
va_start(ap, fmt);
|
|
|
|
vfprintf(stdout, fmt, ap);
|
|
|
|
va_end(ap);
|
|
|
|
printf("\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
/* The array representation of the data. */
|
|
|
|
void **array;
|
|
|
|
int arraylen, arraysize;
|
|
|
|
cmpfn234 cmp;
|
|
|
|
|
|
|
|
/* The tree representation of the same data. */
|
|
|
|
tree234 *tree;
|
|
|
|
|
|
|
|
typedef struct {
|
|
|
|
int treedepth;
|
|
|
|
int elemcount;
|
|
|
|
} chkctx;
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
int chknode(chkctx *ctx, int level, node234 *node,
|
2000-10-02 15:46:10 +04:00
|
|
|
void *lowbound, void *highbound) {
|
|
|
|
int nkids, nelems;
|
|
|
|
int i;
|
2001-04-16 21:18:24 +04:00
|
|
|
int count;
|
2000-10-02 15:46:10 +04:00
|
|
|
|
|
|
|
/* Count the non-NULL kids. */
|
|
|
|
for (nkids = 0; nkids < 4 && node->kids[nkids]; nkids++);
|
|
|
|
/* Ensure no kids beyond the first NULL are non-NULL. */
|
|
|
|
for (i = nkids; i < 4; i++)
|
|
|
|
if (node->kids[i]) {
|
|
|
|
error("node %p: nkids=%d but kids[%d] non-NULL",
|
|
|
|
node, nkids, i);
|
2001-04-16 21:18:24 +04:00
|
|
|
} else if (node->counts[i]) {
|
|
|
|
error("node %p: kids[%d] NULL but count[%d]=%d nonzero",
|
|
|
|
node, i, i, node->counts[i]);
|
|
|
|
}
|
2000-10-02 15:46:10 +04:00
|
|
|
|
|
|
|
/* Count the non-NULL elements. */
|
|
|
|
for (nelems = 0; nelems < 3 && node->elems[nelems]; nelems++);
|
|
|
|
/* Ensure no elements beyond the first NULL are non-NULL. */
|
|
|
|
for (i = nelems; i < 3; i++)
|
|
|
|
if (node->elems[i]) {
|
|
|
|
error("node %p: nelems=%d but elems[%d] non-NULL",
|
|
|
|
node, nelems, i);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (nkids == 0) {
|
|
|
|
/*
|
|
|
|
* If nkids==0, this is a leaf node; verify that the tree
|
|
|
|
* depth is the same everywhere.
|
|
|
|
*/
|
|
|
|
if (ctx->treedepth < 0)
|
|
|
|
ctx->treedepth = level; /* we didn't know the depth yet */
|
|
|
|
else if (ctx->treedepth != level)
|
|
|
|
error("node %p: leaf at depth %d, previously seen depth %d",
|
|
|
|
node, level, ctx->treedepth);
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* If nkids != 0, then it should be nelems+1, unless nelems
|
|
|
|
* is 0 in which case nkids should also be 0 (and so we
|
|
|
|
* shouldn't be in this condition at all).
|
|
|
|
*/
|
|
|
|
int shouldkids = (nelems ? nelems+1 : 0);
|
|
|
|
if (nkids != shouldkids) {
|
|
|
|
error("node %p: %d elems should mean %d kids but has %d",
|
|
|
|
node, nelems, shouldkids, nkids);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* nelems should be at least 1.
|
|
|
|
*/
|
|
|
|
if (nelems == 0) {
|
|
|
|
error("node %p: no elems", node, nkids);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2001-04-16 21:18:24 +04:00
|
|
|
* Add nelems to the running element count of the whole tree.
|
2000-10-02 15:46:10 +04:00
|
|
|
*/
|
|
|
|
ctx->elemcount += nelems;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Check ordering property: all elements should be strictly >
|
|
|
|
* lowbound, strictly < highbound, and strictly < each other in
|
|
|
|
* sequence. (lowbound and highbound are NULL at edges of tree
|
|
|
|
* - both NULL at root node - and NULL is considered to be <
|
|
|
|
* everything and > everything. IYSWIM.)
|
|
|
|
*/
|
2001-04-16 21:18:24 +04:00
|
|
|
if (cmp) {
|
|
|
|
for (i = -1; i < nelems; i++) {
|
|
|
|
void *lower = (i == -1 ? lowbound : node->elems[i]);
|
|
|
|
void *higher = (i+1 == nelems ? highbound : node->elems[i+1]);
|
|
|
|
if (lower && higher && cmp(lower, higher) >= 0) {
|
|
|
|
error("node %p: kid comparison [%d=%s,%d=%s] failed",
|
|
|
|
node, i, lower, i+1, higher);
|
|
|
|
}
|
|
|
|
}
|
2000-10-02 15:46:10 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Check parent pointers: all non-NULL kids should have a
|
|
|
|
* parent pointer coming back to this node.
|
|
|
|
*/
|
|
|
|
for (i = 0; i < nkids; i++)
|
|
|
|
if (node->kids[i]->parent != node) {
|
|
|
|
error("node %p kid %d: parent ptr is %p not %p",
|
|
|
|
node, i, node->kids[i]->parent, node);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Now (finally!) recurse into subtrees.
|
|
|
|
*/
|
2001-04-16 21:18:24 +04:00
|
|
|
count = nelems;
|
|
|
|
|
2000-10-02 15:46:10 +04:00
|
|
|
for (i = 0; i < nkids; i++) {
|
|
|
|
void *lower = (i == 0 ? lowbound : node->elems[i-1]);
|
|
|
|
void *higher = (i >= nelems ? highbound : node->elems[i]);
|
2001-04-16 21:18:24 +04:00
|
|
|
int subcount = chknode(ctx, level+1, node->kids[i], lower, higher);
|
|
|
|
if (node->counts[i] != subcount) {
|
|
|
|
error("node %p kid %d: count says %d, subtree really has %d",
|
|
|
|
node, i, node->counts[i], subcount);
|
|
|
|
}
|
|
|
|
count += subcount;
|
2000-10-02 15:46:10 +04:00
|
|
|
}
|
2001-04-16 21:18:24 +04:00
|
|
|
|
|
|
|
return count;
|
2000-10-02 15:46:10 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
void verify(void) {
|
|
|
|
chkctx ctx;
|
|
|
|
int i;
|
|
|
|
void *p;
|
|
|
|
|
|
|
|
ctx.treedepth = -1; /* depth unknown yet */
|
|
|
|
ctx.elemcount = 0; /* no elements seen yet */
|
|
|
|
/*
|
|
|
|
* Verify validity of tree properties.
|
|
|
|
*/
|
2001-04-16 21:18:24 +04:00
|
|
|
if (tree->root) {
|
|
|
|
if (tree->root->parent != NULL)
|
|
|
|
error("root->parent is %p should be null", tree->root->parent);
|
2000-10-02 15:46:10 +04:00
|
|
|
chknode(&ctx, 0, tree->root, NULL, NULL);
|
2001-04-16 21:18:24 +04:00
|
|
|
}
|
2000-10-02 15:46:10 +04:00
|
|
|
printf("tree depth: %d\n", ctx.treedepth);
|
|
|
|
/*
|
|
|
|
* Enumerate the tree and ensure it matches up to the array.
|
|
|
|
*/
|
2001-04-16 21:18:24 +04:00
|
|
|
for (i = 0; NULL != (p = index234(tree, i)); i++) {
|
2000-10-02 15:46:10 +04:00
|
|
|
if (i >= arraylen)
|
|
|
|
error("tree contains more than %d elements", arraylen);
|
|
|
|
if (array[i] != p)
|
|
|
|
error("enum at position %d: array says %s, tree says %s",
|
|
|
|
i, array[i], p);
|
|
|
|
}
|
2001-04-16 21:18:24 +04:00
|
|
|
if (ctx.elemcount != i) {
|
2000-10-02 15:46:10 +04:00
|
|
|
error("tree really contains %d elements, enum gave %d",
|
2001-04-16 21:18:24 +04:00
|
|
|
ctx.elemcount, i);
|
2000-10-02 15:46:10 +04:00
|
|
|
}
|
|
|
|
if (i < arraylen) {
|
|
|
|
error("enum gave only %d elements, array has %d", i, arraylen);
|
|
|
|
}
|
2001-04-16 21:18:24 +04:00
|
|
|
i = count234(tree);
|
|
|
|
if (ctx.elemcount != i) {
|
|
|
|
error("tree really contains %d elements, count234 gave %d",
|
|
|
|
ctx.elemcount, i);
|
|
|
|
}
|
2000-10-02 15:46:10 +04:00
|
|
|
}
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
void internal_addtest(void *elem, int index, void *realret) {
|
2000-10-02 15:46:10 +04:00
|
|
|
int i, j;
|
2001-04-16 21:18:24 +04:00
|
|
|
void *retval;
|
2000-10-02 15:46:10 +04:00
|
|
|
|
|
|
|
if (arraysize < arraylen+1) {
|
|
|
|
arraysize = arraylen+1+256;
|
2000-12-12 13:33:13 +03:00
|
|
|
array = (array == NULL ? smalloc(arraysize*sizeof(*array)) :
|
|
|
|
srealloc(array, arraysize*sizeof(*array)));
|
2000-10-02 15:46:10 +04:00
|
|
|
}
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
i = index;
|
2000-10-02 15:46:10 +04:00
|
|
|
/* now i points to the first element >= elem */
|
2001-04-16 21:18:24 +04:00
|
|
|
retval = elem; /* expect elem returned (success) */
|
|
|
|
for (j = arraylen; j > i; j--)
|
|
|
|
array[j] = array[j-1];
|
|
|
|
array[i] = elem; /* add elem to array */
|
|
|
|
arraylen++;
|
2000-10-02 15:46:10 +04:00
|
|
|
|
|
|
|
if (realret != retval) {
|
|
|
|
error("add: retval was %p expected %p", realret, retval);
|
|
|
|
}
|
|
|
|
|
|
|
|
verify();
|
|
|
|
}
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
void addtest(void *elem) {
|
2000-10-02 15:46:10 +04:00
|
|
|
int i;
|
2001-04-16 21:18:24 +04:00
|
|
|
void *realret;
|
|
|
|
|
|
|
|
realret = add234(tree, elem);
|
2000-10-02 15:46:10 +04:00
|
|
|
|
|
|
|
i = 0;
|
|
|
|
while (i < arraylen && cmp(elem, array[i]) > 0)
|
|
|
|
i++;
|
2001-04-16 21:18:24 +04:00
|
|
|
if (i < arraylen && !cmp(elem, array[i])) {
|
|
|
|
void *retval = array[i]; /* expect that returned not elem */
|
|
|
|
if (realret != retval) {
|
|
|
|
error("add: retval was %p expected %p", realret, retval);
|
|
|
|
}
|
|
|
|
} else
|
|
|
|
internal_addtest(elem, i, realret);
|
|
|
|
}
|
|
|
|
|
|
|
|
void addpostest(void *elem, int i) {
|
|
|
|
void *realret;
|
|
|
|
|
|
|
|
realret = addpos234(tree, elem, i);
|
|
|
|
|
|
|
|
internal_addtest(elem, i, realret);
|
|
|
|
}
|
|
|
|
|
|
|
|
void delpostest(int i) {
|
|
|
|
int index = i;
|
|
|
|
void *elem = array[i], *ret;
|
|
|
|
|
|
|
|
/* i points to the right element */
|
|
|
|
while (i < arraylen-1) {
|
|
|
|
array[i] = array[i+1];
|
|
|
|
i++;
|
2000-10-02 15:46:10 +04:00
|
|
|
}
|
2001-04-16 21:18:24 +04:00
|
|
|
arraylen--; /* delete elem from array */
|
|
|
|
|
|
|
|
if (tree->cmp)
|
|
|
|
ret = del234(tree, elem);
|
|
|
|
else
|
|
|
|
ret = delpos234(tree, index);
|
2000-10-02 15:46:10 +04:00
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
if (ret != elem) {
|
|
|
|
error("del returned %p, expected %p", ret, elem);
|
|
|
|
}
|
2000-10-02 15:46:10 +04:00
|
|
|
|
|
|
|
verify();
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
2000-10-02 15:46:10 +04:00
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
void deltest(void *elem) {
|
|
|
|
int i;
|
|
|
|
|
|
|
|
i = 0;
|
|
|
|
while (i < arraylen && cmp(elem, array[i]) > 0)
|
|
|
|
i++;
|
|
|
|
if (i >= arraylen || cmp(elem, array[i]) != 0)
|
|
|
|
return; /* don't do it! */
|
|
|
|
delpostest(i);
|
|
|
|
}
|
|
|
|
|
2000-10-02 15:46:10 +04:00
|
|
|
/* A sample data set and test utility. Designed for pseudo-randomness,
|
|
|
|
* and yet repeatability. */
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This random number generator uses the `portable implementation'
|
|
|
|
* given in ANSI C99 draft N869. It assumes `unsigned' is 32 bits;
|
|
|
|
* change it if not.
|
|
|
|
*/
|
|
|
|
int randomnumber(unsigned *seed) {
|
|
|
|
*seed *= 1103515245;
|
|
|
|
*seed += 12345;
|
|
|
|
return ((*seed) / 65536) % 32768;
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
|
|
|
|
2000-10-02 15:46:10 +04:00
|
|
|
int mycmp(void *av, void *bv) {
|
|
|
|
char const *a = (char const *)av;
|
|
|
|
char const *b = (char const *)bv;
|
2000-09-14 19:02:50 +04:00
|
|
|
return strcmp(a, b);
|
|
|
|
}
|
|
|
|
|
2000-10-02 15:46:10 +04:00
|
|
|
#define lenof(x) ( sizeof((x)) / sizeof(*(x)) )
|
|
|
|
|
|
|
|
char *strings[] = {
|
|
|
|
"a", "ab", "absque", "coram", "de",
|
|
|
|
"palam", "clam", "cum", "ex", "e",
|
|
|
|
"sine", "tenus", "pro", "prae",
|
|
|
|
"banana", "carrot", "cabbage", "broccoli", "onion", "zebra",
|
|
|
|
"penguin", "blancmange", "pangolin", "whale", "hedgehog",
|
|
|
|
"giraffe", "peanut", "bungee", "foo", "bar", "baz", "quux",
|
|
|
|
"murfl", "spoo", "breen", "flarn", "octothorpe",
|
|
|
|
"snail", "tiger", "elephant", "octopus", "warthog", "armadillo",
|
|
|
|
"aardvark", "wyvern", "dragon", "elf", "dwarf", "orc", "goblin",
|
|
|
|
"pixie", "basilisk", "warg", "ape", "lizard", "newt", "shopkeeper",
|
|
|
|
"wand", "ring", "amulet"
|
|
|
|
};
|
|
|
|
|
|
|
|
#define NSTR lenof(strings)
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
int findtest(void) {
|
|
|
|
const static int rels[] = {
|
|
|
|
REL234_EQ, REL234_GE, REL234_LE, REL234_LT, REL234_GT
|
|
|
|
};
|
|
|
|
const static char *const relnames[] = {
|
|
|
|
"EQ", "GE", "LE", "LT", "GT"
|
|
|
|
};
|
|
|
|
int i, j, rel, index;
|
|
|
|
char *p, *ret, *realret, *realret2;
|
|
|
|
int lo, hi, mid, c;
|
|
|
|
|
|
|
|
for (i = 0; i < NSTR; i++) {
|
|
|
|
p = strings[i];
|
|
|
|
for (j = 0; j < sizeof(rels)/sizeof(*rels); j++) {
|
|
|
|
rel = rels[j];
|
|
|
|
|
|
|
|
lo = 0; hi = arraylen-1;
|
|
|
|
while (lo <= hi) {
|
|
|
|
mid = (lo + hi) / 2;
|
|
|
|
c = strcmp(p, array[mid]);
|
|
|
|
if (c < 0)
|
|
|
|
hi = mid-1;
|
|
|
|
else if (c > 0)
|
|
|
|
lo = mid+1;
|
|
|
|
else
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (c == 0) {
|
|
|
|
if (rel == REL234_LT)
|
|
|
|
ret = (mid > 0 ? array[--mid] : NULL);
|
|
|
|
else if (rel == REL234_GT)
|
|
|
|
ret = (mid < arraylen-1 ? array[++mid] : NULL);
|
|
|
|
else
|
|
|
|
ret = array[mid];
|
|
|
|
} else {
|
|
|
|
assert(lo == hi+1);
|
|
|
|
if (rel == REL234_LT || rel == REL234_LE) {
|
|
|
|
mid = hi;
|
|
|
|
ret = (hi >= 0 ? array[hi] : NULL);
|
|
|
|
} else if (rel == REL234_GT || rel == REL234_GE) {
|
|
|
|
mid = lo;
|
|
|
|
ret = (lo < arraylen ? array[lo] : NULL);
|
|
|
|
} else
|
|
|
|
ret = NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
realret = findrelpos234(tree, p, NULL, rel, &index);
|
|
|
|
if (realret != ret) {
|
|
|
|
error("find(\"%s\",%s) gave %s should be %s",
|
|
|
|
p, relnames[j], realret, ret);
|
|
|
|
}
|
|
|
|
if (realret && index != mid) {
|
|
|
|
error("find(\"%s\",%s) gave %d should be %d",
|
|
|
|
p, relnames[j], index, mid);
|
|
|
|
}
|
|
|
|
if (realret && rel == REL234_EQ) {
|
|
|
|
realret2 = index234(tree, index);
|
|
|
|
if (realret2 != realret) {
|
|
|
|
error("find(\"%s\",%s) gave %s(%d) but %d -> %s",
|
|
|
|
p, relnames[j], realret, index, index, realret2);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
#if 0
|
|
|
|
printf("find(\"%s\",%s) gave %s(%d)\n", p, relnames[j],
|
|
|
|
realret, index);
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
realret = findrelpos234(tree, NULL, NULL, REL234_GT, &index);
|
|
|
|
if (arraylen && (realret != array[0] || index != 0)) {
|
|
|
|
error("find(NULL,GT) gave %s(%d) should be %s(0)",
|
|
|
|
realret, index, array[0]);
|
|
|
|
} else if (!arraylen && (realret != NULL)) {
|
|
|
|
error("find(NULL,GT) gave %s(%d) should be NULL",
|
|
|
|
realret, index);
|
|
|
|
}
|
|
|
|
|
|
|
|
realret = findrelpos234(tree, NULL, NULL, REL234_LT, &index);
|
|
|
|
if (arraylen && (realret != array[arraylen-1] || index != arraylen-1)) {
|
|
|
|
error("find(NULL,LT) gave %s(%d) should be %s(0)",
|
|
|
|
realret, index, array[arraylen-1]);
|
|
|
|
} else if (!arraylen && (realret != NULL)) {
|
|
|
|
error("find(NULL,LT) gave %s(%d) should be NULL",
|
|
|
|
realret, index);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2000-09-14 19:02:50 +04:00
|
|
|
int main(void) {
|
2000-10-02 15:46:10 +04:00
|
|
|
int in[NSTR];
|
2001-04-16 21:18:24 +04:00
|
|
|
int i, j, k;
|
2000-10-02 15:46:10 +04:00
|
|
|
unsigned seed = 0;
|
|
|
|
|
|
|
|
for (i = 0; i < NSTR; i++) in[i] = 0;
|
|
|
|
array = NULL;
|
|
|
|
arraylen = arraysize = 0;
|
|
|
|
tree = newtree234(mycmp);
|
|
|
|
cmp = mycmp;
|
|
|
|
|
|
|
|
verify();
|
|
|
|
for (i = 0; i < 10000; i++) {
|
|
|
|
j = randomnumber(&seed);
|
|
|
|
j %= NSTR;
|
|
|
|
printf("trial: %d\n", i);
|
|
|
|
if (in[j]) {
|
|
|
|
printf("deleting %s (%d)\n", strings[j], j);
|
|
|
|
deltest(strings[j]);
|
|
|
|
in[j] = 0;
|
|
|
|
} else {
|
|
|
|
printf("adding %s (%d)\n", strings[j], j);
|
|
|
|
addtest(strings[j]);
|
|
|
|
in[j] = 1;
|
|
|
|
}
|
2001-04-16 21:18:24 +04:00
|
|
|
findtest();
|
2000-10-02 15:46:10 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
while (arraylen > 0) {
|
|
|
|
j = randomnumber(&seed);
|
|
|
|
j %= arraylen;
|
|
|
|
deltest(array[j]);
|
|
|
|
}
|
|
|
|
|
2001-04-16 21:18:24 +04:00
|
|
|
freetree234(tree);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Now try an unsorted tree. We don't really need to test
|
|
|
|
* delpos234 because we know del234 is based on it, so it's
|
|
|
|
* already been tested in the above sorted-tree code; but for
|
|
|
|
* completeness we'll use it to tear down our unsorted tree
|
|
|
|
* once we've built it.
|
|
|
|
*/
|
|
|
|
tree = newtree234(NULL);
|
|
|
|
cmp = NULL;
|
|
|
|
verify();
|
|
|
|
for (i = 0; i < 1000; i++) {
|
|
|
|
printf("trial: %d\n", i);
|
|
|
|
j = randomnumber(&seed);
|
|
|
|
j %= NSTR;
|
|
|
|
k = randomnumber(&seed);
|
|
|
|
k %= count234(tree)+1;
|
|
|
|
printf("adding string %s at index %d\n", strings[j], k);
|
|
|
|
addpostest(strings[j], k);
|
|
|
|
}
|
|
|
|
while (count234(tree) > 0) {
|
|
|
|
printf("cleanup: tree size %d\n", count234(tree));
|
|
|
|
j = randomnumber(&seed);
|
|
|
|
j %= count234(tree);
|
|
|
|
printf("deleting string %s from index %d\n", array[j], j);
|
|
|
|
delpostest(j);
|
|
|
|
}
|
|
|
|
|
2000-10-02 15:46:10 +04:00
|
|
|
return 0;
|
2000-09-14 19:02:50 +04:00
|
|
|
}
|
2000-10-02 15:46:10 +04:00
|
|
|
|
2000-09-14 19:02:50 +04:00
|
|
|
#endif
|