/* Perform one split operation on the given node with the given parent.
*/
static void hattrie_split(hattrie_t* T, node_ptr parent, node_ptr node)
{
/* only buckets may be split */
assert(*node.flag & NODE_TYPE_PURE_BUCKET ||
*node.flag & NODE_TYPE_HYBRID_BUCKET);
assert(*parent.flag & NODE_TYPE_TRIE);
if (*node.flag & NODE_TYPE_PURE_BUCKET) {
/* turn the pure bucket into a hybrid bucket */
parent.t->xs[node.b->c0].t = alloc_trie_node(T, node);
/* if the bucket had an empty key, move it to the new trie node */
value_t* val = ahtable_tryget(node.b, NULL, 0);
if (val) {
parent.t->xs[node.b->c0].t->val = *val;
parent.t->xs[node.b->c0].t->flag |= NODE_HAS_VAL;
*val = 0;
ahtable_del(node.b, NULL, 0);
}
node.b->c0 = 0x00;
node.b->c1 = NODE_MAXCHAR;
node.b->flag = NODE_TYPE_HYBRID_BUCKET;
return;
}
/* This is a hybrid bucket. Perform a proper split. */
/* count the number of occourances of every leading character */
unsigned int cs[NODE_CHILDS]; // occurance count for leading chars
memset(cs, 0, NODE_CHILDS * sizeof(unsigned int));
size_t len;
const char* key;
ahtable_iter_t* i = ahtable_iter_begin(node.b, false);
while (!ahtable_iter_finished(i)) {
key = ahtable_iter_key(i, &len);
assert(len > 0);
cs[(unsigned char) key[0]] += 1;
ahtable_iter_next(i);
}
ahtable_iter_free(i);
/* choose a split point */
unsigned int left_m, right_m, all_m;
unsigned char j = node.b->c0;
all_m = ahtable_size(node.b);
left_m = cs[j];
right_m = all_m - left_m;
int d;
while (j + 1 < node.b->c1) {
d = abs((int) (left_m + cs[j + 1]) - (int) (right_m - cs[j + 1]));
if (d <= abs(left_m - right_m) && left_m + cs[j + 1] < all_m) {
j += 1;
left_m += cs[j];
right_m -= cs[j];
}
else break;
}
/* now split into two node cooresponding to ranges [0, j] and
* [j + 1, NODE_MAXCHAR], respectively. */
/* create new left and right nodes */
/* TODO: Add a special case if either node is a hybrid bucket containing all
* the keys. In such a case, do not build a new table, just use the old one.
* */
size_t num_slots;
for (num_slots = ahtable_initial_size;
(double) left_m > ahtable_max_load_factor * (double) num_slots;
num_slots *= 2);
node_ptr left, right;
left.b = ahtable_create_n(num_slots);
left.b->c0 = node.b->c0;
left.b->c1 = j;
left.b->flag = left.b->c0 == left.b->c1 ?
NODE_TYPE_PURE_BUCKET : NODE_TYPE_HYBRID_BUCKET;
for (num_slots = ahtable_initial_size;
(double) right_m > ahtable_max_load_factor * (double) num_slots;
num_slots *= 2);
right.b = ahtable_create_n(num_slots);
right.b->c0 = j + 1;
right.b->c1 = node.b->c1;
right.b->flag = right.b->c0 == right.b->c1 ?
NODE_TYPE_PURE_BUCKET : NODE_TYPE_HYBRID_BUCKET;
/* update the parent's pointer */
unsigned int c;
for (c = node.b->c0; c <= j; ++c) parent.t->xs[c] = left;
for (; c <= node.b->c1; ++c) parent.t->xs[c] = right;
/* distribute keys to the new left or right node */
value_t* u;
value_t* v;
i = ahtable_iter_begin(node.b, false);
while (!ahtable_iter_finished(i)) {
key = ahtable_iter_key(i, &len);
u = ahtable_iter_val(i);
assert(len > 0);
/* left */
if ((unsigned char) key[0] <= j) {
if (*left.flag & NODE_TYPE_PURE_BUCKET) {
v = ahtable_get(left.b, key + 1, len - 1);
}
else {
v = ahtable_get(left.b, key, len);
}
*v = *u;
}
/* right */
else {
if (*right.flag & NODE_TYPE_PURE_BUCKET) {
v = ahtable_get(right.b, key + 1, len - 1);
}
else {
v = ahtable_get(right.b, key, len);
}
*v = *u;
}
ahtable_iter_next(i);
}
ahtable_iter_free(i);
ahtable_free(node.b);
}
value_t* hattrie_get(hattrie_t* T, const char* key, size_t len)
{
node_ptr parent = T->root;
assert(*parent.flag & NODE_TYPE_TRIE);
if (len == 0) return &parent.t->val;
/* consume all trie nodes, now parent must be trie and child anything */
node_ptr node = hattrie_consume(&parent, &key, &len, 0);
assert(*parent.flag & NODE_TYPE_TRIE);
/* key wasn't consumed and using pure tries */
if (T->bsize == 0) {
node.t = parent.t;
while (len > 0) {
node.t->xs[(unsigned char) *key].t = alloc_empty_node(T);
node = node.t->xs[(unsigned char) *key];
++key;
--len;
}
return hattrie_useval(T, node);
}
/* if the key has been consumed on a trie node, use its value */
if (len == 0) {
if (*node.flag & NODE_TYPE_TRIE) {
return hattrie_useval(T, node);
}
else if (*node.flag & NODE_TYPE_HYBRID_BUCKET) {
return hattrie_useval(T, parent);
}
}
/* preemptively split the bucket if it is full */
while (ahtable_size(node.b) >= T->bsize) {
hattrie_split(T, parent, node);
/* after the split, the node pointer is invalidated, so we search from
* the parent again. */
node = hattrie_consume(&parent, &key, &len, 0);
/* if the key has been consumed on a trie node, use its value */
if (len == 0) {
if (*node.flag & NODE_TYPE_TRIE) {
return hattrie_useval(T, node);
}
else if (*node.flag & NODE_TYPE_HYBRID_BUCKET) {
return hattrie_useval(T, parent);
}
}
}
assert(*node.flag & NODE_TYPE_PURE_BUCKET || *node.flag & NODE_TYPE_HYBRID_BUCKET);
assert(len > 0);
size_t m_old = node.b->m;
value_t* val;
if (*node.flag & NODE_TYPE_PURE_BUCKET) {
val = ahtable_get(node.b, key + 1, len - 1);
}
else {
val = ahtable_get(node.b, key, len);
}
T->m += (node.b->m - m_old);
return val;
}
