Changeset 0ca7286 in mainline for uspace/lib/c/generic/adt/hash_table.c

Timestamp:

2012-07-21T11:19:27Z (13 years ago)

Author:

Adam Hraska <adam.hraska+hos@…>

Branches:

lfn, master, serial, ticket/834-toolchain-update, topic/msim-upgrade, topic/simplify-dev-export

Children:

2732c94

Parents:

1c1da4b

Message:

Added resizing to user space (single-threaded) hash_table. Resizes in a way to mitigate effects of bad hash functions. Change of interface affected many files.

File:

• uspace/lib/c/generic/adt/hash_table.c (modified) (7 diffs)

uspace/lib/c/generic/adt/hash_table.c

@@ -34 +34 @@
 
 /*
- * This is an implementation of generic chained hash table.
+ * This is an implementation of a generic resizable chained hash table.
+ * 
+ * The table grows to 2*n+1 buckets each time, starting at n == 89, 
+ * per Thomas Wang's recommendation:
+ * http://www.concentric.net/~Ttwang/tech/hashsize.htm
+ * 
+ * This policy produces prime table sizes for the first five resizes
+ * and generally produces table sizes which are either prime or 
+ * have fairly large (prime/odd) divisors. Having a prime table size
+ * mitigates the use of suboptimal hash functions and distributes
+ * items over the whole table.
  */
 
@@ -44 +54 @@
 #include <str.h>
 
+/* Optimal initial bucket count. See comment above. */
+#define HT_MIN_BUCKETS  89
+/* The table is resized when the average load per bucket exceeds this number. */
+#define HT_MAX_LOAD     2
+
+
+static size_t round_up_size(size_t size);
+static bool alloc_table(size_t bucket_cnt, list_t **pbuckets);
+static void item_inserted(hash_table_t *h);
+static void item_removed(hash_table_t *h);
+static inline void remove_item(hash_table_t *h, link_t *item);
+static size_t remove_duplicates(hash_table_t *h, unsigned long key[]);
+static size_t remove_matching(hash_table_t *h, unsigned long key[], size_t key_cnt);
+
+/* Dummy do nothing callback to invoke in place of remove_callback == NULL. */
+static void nop_remove_callback(link_t *item)
+{
+        /* no-op */
+}
+
+
 /** Create chained hash table.
  *
  * @param h        Hash table structure. Will be initialized by this call.
- * @param m        Number of hash table buckets.
+ * @param init_size Initial desired number of hash table buckets. Pass zero
+ *                 if you want the default initial size. 
  * @param max_keys Maximal number of keys needed to identify an item.
- * @param op       Hash table operations structure.
+ * @param op       Hash table operations structure. remove_callback()
+ *                 is optional and can be NULL if no action is to be taken
+ *                 upon removal. equal() is optional if and only if
+ *                 hash_table_insert_unique() will never be invoked.
+ *                 All other operations are mandatory. 
  *
  * @return True on success
  *
  */
-bool hash_table_create(hash_table_t *h, hash_count_t m, hash_count_t max_keys,
-    hash_table_operations_t *op)
+bool hash_table_create(hash_table_t *h, size_t init_size, size_t max_keys,
+    hash_table_ops_t *op)
 {
         assert(h);
-        assert(op && op->hash && op->compare);
+        assert(op && op->hash && op->key_hash && op->match);
         assert(max_keys > 0);
         
-        h->entry = malloc(m * sizeof(list_t));
-        if (!h->entry)
+        h->bucket_cnt = round_up_size(init_size);
+        
+        if (!alloc_table(h->bucket_cnt, &h->bucket))
                 return false;
         
-        memset((void *) h->entry, 0,  m * sizeof(list_t));
-        
-        hash_count_t i;
-        for (i = 0; i < m; i++)
-                list_initialize(&h->entry[i]);
-        
-        h->entries = m;
         h->max_keys = max_keys;
+        h->items = 0;
         h->op = op;
         
+        if (h->op->remove_callback == 0) 
+                h->op->remove_callback = nop_remove_callback;
+        
         return true;
 }
@@ -84 +118 @@
 void hash_table_clear(hash_table_t *h)
 {
-        for (hash_count_t chain = 0; chain < h->entries; ++chain) {
-                link_t *cur;
-                link_t *next;
-                
-                for (cur = h->entry[chain].head.next;
-                    cur != &h->entry[chain].head;
-                    cur = next) {
-                        next = cur->next;
+        for (size_t idx = 0; idx < h->bucket_cnt; ++idx) {
+                list_foreach_safe(h->bucket[idx], cur, next) {
                         list_remove(cur);
                         h->op->remove_callback(cur);
                 }
         }
+        
+        h->items = 0;
+
+        /* Shrink the table to its minimum size if possible. */
+        if (HT_MIN_BUCKETS < h->bucket_cnt) {
+                list_t *new_buckets;
+                if (alloc_table(HT_MIN_BUCKETS, &new_buckets)) {
+                        free(h->bucket);
+                        h->bucket = new_buckets;
+                        h->bucket_cnt = HT_MIN_BUCKETS;
+                }
+        }
 }
 
@@ -106 +146 @@
 {
         assert(h);
-        assert(h->entry);
-        
-        free(h->entry);
+        assert(h->bucket);
+        
+        free(h->bucket);
+
+        h->bucket = 0;
+        h->bucket_cnt = 0;
 }
 
@@ -117 +160 @@
  * @param item Item to be inserted into the hash table.
  */
-void hash_table_insert(hash_table_t *h, unsigned long key[], link_t *item)
+void hash_table_insert(hash_table_t *h, link_t *item)
 {
         assert(item);
-        assert(h && h->op && h->op->hash && h->op->compare);
-        
-        hash_index_t chain = h->op->hash(key);
-        assert(chain < h->entries);
-        
-        list_append(item, &h->entry[chain]);
+        assert(h && h->bucket);
+        assert(h->op && h->op->hash);
+        
+        size_t idx = h->op->hash(item) % h->bucket_cnt;
+        
+        assert(idx < h->bucket_cnt);
+        
+        list_append(item, &h->bucket[idx]);
+        item_inserted(h);
+}
+
+
+/** Insert item into a hash table if not already present.
+ *
+ * @param h    Hash table.
+ * @param key  Array of all keys necessary to compute hash index.
+ * @param item Item to be inserted into the hash table.
+ * 
+ * @return False if such an item had already been inserted. 
+ * @return True if the inserted item was the only item with such a lookup key.
+ */
+bool hash_table_insert_unique(hash_table_t *h, link_t *item)
+{
+        assert(item);
+        assert(h && h->bucket && h->bucket_cnt);
+        assert(h->op && h->op->hash && h->op->equal);
+        
+        size_t item_hash = h->op->hash(item);
+        size_t idx = item_hash % h->bucket_cnt;
+        
+        assert(idx < h->bucket_cnt);
+        
+        /* Check for duplicates. */
+        list_foreach(h->bucket[idx], cur) {
+                /* 
+                 * We could filter out items using their hashes first, but 
+                 * calling equal() might very well be just as fast.
+                 */
+                if (h->op->equal(cur, item))
+                        return false;
+        }
+        
+        list_append(item, &h->bucket[idx]);
+        item_inserted(h);
+        
+        return true;
 }
 
@@ -138 +221 @@
 link_t *hash_table_find(hash_table_t *h, unsigned long key[])
 {
-        assert(h && h->op && h->op->hash && h->op->compare);
-        
-        hash_index_t chain = h->op->hash(key);
-        assert(chain < h->entries);
-        
-        list_foreach(h->entry[chain], cur) {
-                if (h->op->compare(key, h->max_keys, cur)) {
-                        /*
-                         * The entry is there.
+        assert(h && h->bucket);
+        assert(h->op && h->op->key_hash && h->op->match);
+        
+        size_t key_hash = h->op->key_hash(key);
+        size_t idx = key_hash % h->bucket_cnt;
+
+        assert(idx < h->bucket_cnt);
+        
+        list_foreach(h->bucket[idx], cur) {
+                /* 
+                 * Is this is the item we are looking for? We could have first 
+                 * checked if the hashes match but op->match() may very well be 
+                 * just as fast as op->hash().
+                 */
+                if (h->op->match(key, h->max_keys, cur)) {
+                        return cur;
+                }
+        }
+        
+        return NULL;
+}
+
+
+/** Apply function to all items in hash table.
+ *
+ * @param h   Hash table.
+ * @param f   Function to be applied. Return false if no more items 
+ *            should be visited. The functor must not delete the successor
+ *            of the item passed in the first argument.
+ * @param arg Argument to be passed to the function.
+ *
+ */
+void hash_table_apply(hash_table_t *h, bool (*f)(link_t *, void *), void *arg)
+{       
+        for (size_t idx = 0; idx < h->bucket_cnt; ++idx) {
+                list_foreach_safe(h->bucket[idx], cur, next) {
+                        /* 
+                         * The next pointer had already been saved. f() may safely 
+                         * delete cur (but not next!).
                          */
-                        return cur;
-                }
-        }
-        
-        return NULL;
+                        if (!f(cur, arg))
+                                return;
+                }
+        }
 }
 
@@ -163 +275 @@
  *             the hash table.
  * @param keys Number of keys in the 'key' array.
- *
- */
-void hash_table_remove(hash_table_t *h, unsigned long key[], hash_count_t keys)
-{
-        assert(h && h->op && h->op->hash && h->op->compare &&
+ * 
+ * @return Returns the number of removed items.
+ */
+size_t hash_table_remove(hash_table_t *h, unsigned long key[], size_t key_cnt)
+{
+        assert(h && h->bucket);
+        assert(h && h->op && h->op->hash &&
             h->op->remove_callback);
-        assert(keys <= h->max_keys);
-        
-        if (keys == h->max_keys) {
+        assert(key_cnt <= h->max_keys);
+        
+        /* All keys are known, remove from a specific bucket. */
+        if (key_cnt == h->max_keys) {
+                return remove_duplicates(h, key);
+        } else {
                 /*
-                 * All keys are known, hash_table_find() can be used to find the
-                 * entry.
-                 */
-                
-                link_t *cur = hash_table_find(h, key);
-                if (cur) {
-                        list_remove(cur);
-                        h->op->remove_callback(cur);
-                }
-                
-                return;
-        }
-        
+                * Fewer keys were passed.
+                * Any partially matching entries are to be removed.
+                */
+                return remove_matching(h, key, key_cnt);
+        }
+}
+
+/** Removes an item already present in the table. The item must be in the table.*/
+void hash_table_remove_item(hash_table_t *h, link_t *item)
+{
+        assert(item);
+        assert(h && h->bucket);
+        
+        remove_item(h, item);
+}
+
+/** Unlink the item from a bucket, update statistics and resize if needed. */
+static inline void remove_item(hash_table_t *h, link_t *item)
+{
+        assert(item);
+        
+        list_remove(item);
+        item_removed(h);
+        h->op->remove_callback(item);
+}
+
+/** Removes all items matching key in the bucket key hashes to. */
+static size_t remove_duplicates(hash_table_t *h, unsigned long key[])
+{
+        assert(h && h->bucket);
+        assert(h->op && h->op->key_hash && h->op->match);
+        
+        size_t key_hash = h->op->key_hash(key);
+        size_t idx = key_hash % h->bucket_cnt;
+
+        assert(idx < h->bucket_cnt);
+        
+        size_t removed = 0;
+        
+        list_foreach_safe(h->bucket[idx], cur, next) {
+                if (h->op->match(key, h->max_keys, cur)) {
+                        ++removed;
+                        remove_item(h, cur);
+                }
+        }
+        
+        return removed;
+}
+
+/** Removes all items in any bucket in the table that match the partial key. */
+static size_t remove_matching(hash_table_t *h, unsigned long key[], 
+        size_t key_cnt)
+{
+        assert(h && h->bucket);
+        assert(key_cnt < h->max_keys);
+        
+        size_t removed = 0;
         /*
          * Fewer keys were passed.
          * Any partially matching entries are to be removed.
          */
-        hash_index_t chain;
-        for (chain = 0; chain < h->entries; chain++) {
-                for (link_t *cur = h->entry[chain].head.next;
-                    cur != &h->entry[chain].head;
-                    cur = cur->next) {
-                        if (h->op->compare(key, keys, cur)) {
-                                link_t *hlp;
-                                
-                                hlp = cur;
-                                cur = cur->prev;
-                                
-                                list_remove(hlp);
-                                h->op->remove_callback(hlp);
-                                
-                                continue;
+        for (size_t idx = 0; idx < h->bucket_cnt; ++idx) {
+                list_foreach_safe(h->bucket[idx], cur, next) {
+                        if (h->op->match(key, key_cnt, cur)) {
+                                ++removed;
+                                remove_item(h, cur);
                         }
                 }
         }
-}
-
-/** Apply function to all items in hash table.
- *
- * @param h   Hash table.
- * @param f   Function to be applied.
- * @param arg Argument to be passed to the function.
- *
- */
-void hash_table_apply(hash_table_t *h, void (*f)(link_t *, void *), void *arg)
-{       
-        for (hash_index_t bucket = 0; bucket < h->entries; bucket++) {
-                link_t *cur;
-                link_t *next;
-
-                for (cur = h->entry[bucket].head.next; cur != &h->entry[bucket].head;
-                    cur = next) {
-                        /*
-                         * The next pointer must be stored prior to the functor
-                         * call to allow using destructor as the functor (the
-                         * free function could overwrite the cur->next pointer).
-                         */
-                        next = cur->next;
-                        f(cur, arg);
-                }
-        }
-}
+        
+        return removed;
+        
+}
+
+/** Rounds up size to the nearest suitable table size. */
+static size_t round_up_size(size_t size)
+{
+        size_t rounded_size = HT_MIN_BUCKETS;
+        
+        while (rounded_size < size) {
+                rounded_size = 2 * rounded_size + 1;
+        }
+        
+        return rounded_size;
+}
+
+/** Allocates and initializes the desired number of buckets. True if successful.*/
+static bool alloc_table(size_t bucket_cnt, list_t **pbuckets)
+{
+        assert(pbuckets && HT_MIN_BUCKETS <= bucket_cnt);
+                
+        list_t *buckets = malloc(bucket_cnt * sizeof(list_t));
+        if (!buckets)
+                return false;
+        
+        for (size_t i = 0; i < bucket_cnt; i++)
+                list_initialize(&buckets[i]);
+
+        *pbuckets = buckets;
+        return true;
+}
+
+/** Allocates and rehashes items to a new table. Frees the old table. */
+static void resize(hash_table_t *h, size_t new_bucket_cnt) 
+{
+        assert(h && h->bucket);
+        
+        list_t *new_buckets;
+
+        /* Leave the table as is if we cannot resize. */
+        if (!alloc_table(new_bucket_cnt, &new_buckets))
+                return;
+        
+        /* Rehash all the items to the new table. */
+        for (size_t old_idx = 0; old_idx < h->bucket_cnt; ++old_idx) {
+                list_foreach_safe(h->bucket[old_idx], cur, next) {
+                        size_t new_idx = h->op->hash(cur) % new_bucket_cnt;
+                        list_remove(cur);
+                        list_append(cur, &new_buckets[new_idx]);
+                }
+        }
+        
+        free(h->bucket);
+        h->bucket = new_buckets;
+        h->bucket_cnt = new_bucket_cnt;
+}
+
+/** Shrinks the table if needed. */
+static void item_removed(hash_table_t *h)
+{
+        --h->items;
+        
+        if (HT_MIN_BUCKETS < h->items && h->items <= HT_MAX_LOAD * h->bucket_cnt / 4) {
+                /* 
+                 * Keep the bucket_cnt odd (possibly also prime). 
+                 * Shrink from 2n + 1 to n. Integer division discards the +1.
+                 */
+                size_t new_bucket_cnt = h->bucket_cnt / 2;
+                resize(h, new_bucket_cnt);
+        }
+}
+
+/** Grows the table if needed. */
+static void item_inserted(hash_table_t *h)
+{
+        ++h->items;
+        
+        /* Grow the table if the average bucket load exceeds the maximum. */
+        if (HT_MAX_LOAD * h->bucket_cnt < h->items) {
+                /* Keep the bucket_cnt odd (possibly also prime). */
+                size_t new_bucket_cnt = 2 * h->bucket_cnt + 1;
+                resize(h, new_bucket_cnt);
+        }
+}
+
 
 /** @}