Hashtable: use dynamic growth and use primes as size

Dynamically increase the hashmap size to not exceed the load factor and
avoid too long chains.

Switch from Separate Chaining to Robin Hood linear probing to improve
cache locality.

Use primes as size to further avoid collisions.

E.g. on a standard kde system the number of entries in the ProcessTable
might be around 650.

Hashtable.c

@@ -8,32 +8,55 @@ in the source distribution for its full text.
 #include "Hashtable.h"
 #include "XUtils.h"
 
-#include <stdlib.h>
 #include <assert.h>
+#include <stdint.h>
+#include <stdlib.h>
 
 
 #ifndef NDEBUG
 
+static void Hashtable_dump(const Hashtable* this) {
+   fprintf(stderr, "Hashtable %p: size=%u items=%u owner=%s\n",
+           (const void*)this,
+           this->size,
+           this->items,
+           this->owner ? "yes" : "no");
+
+   unsigned int items = 0;
+   for (unsigned int i = 0; i < this->size; i++) {
+      fprintf(stderr, "  item %5u: key = %5u probe = %2u value = %p\n",
+              i,
+              this->buckets[i].key,
+              this->buckets[i].probe,
+              this->buckets[i].value ? (const void*)this->buckets[i].value : "(nil)");
+
+      if (this->buckets[i].value)
+         items++;
+   }
+
+   fprintf(stderr, "Hashtable %p: items=%u counted=%u\n",
+           (const void*)this,
+           this->items,
+           items);
+}
+
 static bool Hashtable_isConsistent(const Hashtable* this) {
    unsigned int items = 0;
    for (unsigned int i = 0; i < this->size; i++) {
-      HashtableItem* bucket = this->buckets[i];
+      if (this->buckets[i].value)
-      while (bucket) {
          items++;
-         bucket = bucket->next;
    }
-   }
+   bool res = items == this->items;
-   return items == this->items;
+   if (!res)
+      Hashtable_dump(this);
+   return res;
 }
 
 unsigned int Hashtable_count(const Hashtable* this) {
    unsigned int items = 0;
    for (unsigned int i = 0; i < this->size; i++) {
-      HashtableItem* bucket = this->buckets[i];
+      if (this->buckets[i].value)
-      while (bucket) {
          items++;
-         bucket = bucket->next;
-      }
    }
    assert(items == this->items);
    return items;
@@ -41,14 +64,25 @@ unsigned int Hashtable_count(const Hashtable* this) {
 
 #endif /* NDEBUG */
 
-static HashtableItem* HashtableItem_new(hkey_t key, void* value) {
+/* https://oeis.org/A014234 */
-   HashtableItem* this;
+static const uint64_t OEISprimes[] = {
+   2, 3, 7, 13, 31, 61, 127, 251, 509, 1021, 2039, 4093, 8191,
+   16381, 32749, 65521, 131071, 262139, 524287, 1048573,
+   2097143, 4194301, 8388593, 16777213, 33554393,
+   67108859, 134217689, 268435399, 536870909, 1073741789,
+   2147483647, 4294967291, 8589934583, 17179869143,
+   34359738337, 68719476731, 137438953447
+};
 
-   this = xMalloc(sizeof(HashtableItem));
+static uint64_t nextPrime(unsigned int n) {
-   this->key = key;
+   assert(n <= OEISprimes[ARRAYSIZE(OEISprimes) - 1]);
-   this->value = value;
+
-   this->next = NULL;
+   for (unsigned int i = 0; i < ARRAYSIZE(OEISprimes); i++) {
-   return this;
+      if (n <= OEISprimes[i])
+         return OEISprimes[i];
+   }
+
+   return OEISprimes[ARRAYSIZE(OEISprimes) - 1];
 }
 
 Hashtable* Hashtable_new(unsigned int size, bool owner) {
@@ -56,102 +90,202 @@ Hashtable* Hashtable_new(unsigned int size, bool owner) {
 
    this = xMalloc(sizeof(Hashtable));
    this->items = 0;
-   this->size = size;
+   this->size = size ? nextPrime(size) : 13;
-   this->buckets = (HashtableItem**) xCalloc(size, sizeof(HashtableItem*));
+   this->buckets = (HashtableItem*) xCalloc(this->size, sizeof(HashtableItem));
    this->owner = owner;
    assert(Hashtable_isConsistent(this));
    return this;
 }
 
 void Hashtable_delete(Hashtable* this) {
-   assert(Hashtable_isConsistent(this));
-   for (unsigned int i = 0; i < this->size; i++) {
-      HashtableItem* walk = this->buckets[i];
-      while (walk != NULL) {
-         if (this->owner)
-            free(walk->value);
 
-         HashtableItem* savedWalk = walk;
+   assert(Hashtable_isConsistent(this));
-         walk = savedWalk->next;
+
-         free(savedWalk);
+   if (this->owner) {
-      }
+      for (unsigned int i = 0; i < this->size; i++)
+         free(this->buckets[i].value);
    }
+
    free(this->buckets);
    free(this);
 }
 
-void Hashtable_put(Hashtable* this, hkey_t key, void* value) {
+static void insert(Hashtable* this, hkey_t key, void* value) {
    unsigned int index = key % this->size;
-   HashtableItem** bucketPtr = &(this->buckets[index]);
+   unsigned int probe = 0;
-   while (true)
+#ifndef NDEBUG
-      if (*bucketPtr == NULL) {
+   unsigned int origIndex = index;
-         *bucketPtr = HashtableItem_new(key, value);
+#endif
-         this->items++;
-         break;
-      } else if ((*bucketPtr)->key == key) {
-         if (this->owner && (*bucketPtr)->value != value)
-            free((*bucketPtr)->value);
 
-         (*bucketPtr)->value = value;
+   for (;;) {
-         break;
+      if (!this->buckets[index].value) {
-      } else {
+         this->items++;
-         bucketPtr = &((*bucketPtr)->next);
+         this->buckets[index].key = key;
+         this->buckets[index].probe = probe;
+         this->buckets[index].value = value;
+         return;
       }
 
+      if (this->buckets[index].key == key) {
+         if (this->owner && this->buckets[index].value != value)
+            free(this->buckets[index].value);
+         this->buckets[index].value = value;
+         return;
+      }
+
+      /* Robin Hood swap */
+      if (probe > this->buckets[index].probe) {
+         HashtableItem tmp = this->buckets[index];
+
+         this->buckets[index].key = key;
+         this->buckets[index].probe = probe;
+         this->buckets[index].value = value;
+
+         key = tmp.key;
+         probe = tmp.probe;
+         value = tmp.value;
+      }
+
+      index = (index + 1) % this->size;
+      probe++;
+
+      assert(index != origIndex);
+   }
+}
+
+void Hashtable_setSize(Hashtable* this, unsigned int size) {
+
    assert(Hashtable_isConsistent(this));
+
+   if (size <= this->items)
+      return;
+
+   HashtableItem* oldBuckets = this->buckets;
+   unsigned int oldSize = this->size;
+
+   this->size = nextPrime(size);
+   this->buckets = (HashtableItem*) xCalloc(this->size, sizeof(HashtableItem));
+   this->items = 0;
+
+   /* rehash */
+   for (unsigned int i = 0; i < oldSize; i++) {
+      if (!oldBuckets[i].value)
+         continue;
+
+      insert(this, oldBuckets[i].key, oldBuckets[i].value);
+   }
+
+   free(oldBuckets);
+
+   assert(Hashtable_isConsistent(this));
+}
+
+void Hashtable_put(Hashtable* this, hkey_t key, void* value) {
+
+   assert(Hashtable_isConsistent(this));
+   assert(this->size > 0);
+   assert(value);
+
+   /* grow on load-factor > 0.7 */
+   if (10 * this->items > 7 * this->size)
+      Hashtable_setSize(this, 2 * this->size);
+
+   insert(this, key, value);
+
+   assert(Hashtable_isConsistent(this));
+   assert(Hashtable_get(this, key) != NULL);
+   assert(this->size > this->items);
 }
 
 void* Hashtable_remove(Hashtable* this, hkey_t key) {
    unsigned int index = key % this->size;
+   unsigned int probe = 0;
+#ifndef NDEBUG
+   unsigned int origIndex = index;
+#endif
 
    assert(Hashtable_isConsistent(this));
 
-   HashtableItem** bucket;
+   void* res = NULL;
-   for (bucket = &(this->buckets[index]); *bucket; bucket = &((*bucket)->next) ) {
+
-      if ((*bucket)->key == key) {
+   while (this->buckets[index].value) {
-         void* value = (*bucket)->value;
+      if (this->buckets[index].key == key) {
-         HashtableItem* next = (*bucket)->next;
-         free(*bucket);
-         (*bucket) = next;
-         this->items--;
          if (this->owner) {
-            free(value);
+            free(this->buckets[index].value);
-            assert(Hashtable_isConsistent(this));
-            return NULL;
          } else {
+            res = this->buckets[index].value;
+         }
+
+         unsigned int next = (index + 1) % this->size;
+
+         while (this->buckets[next].value && this->buckets[next].probe > 0) {
+            this->buckets[index] = this->buckets[next];
+            this->buckets[index].probe -= 1;
+
+            index = next;
+            next = (index + 1) % this->size;
+         }
+
+         /* set empty after backward shifting */
+         this->buckets[index].value = NULL;
+         this->items--;
+
+         break;
+      }
+
+      if (this->buckets[index].probe < probe)
+         break;
+
+      index = (index + 1) % this->size;
+      probe++;
+
+      assert(index != origIndex);
+   }
+
    assert(Hashtable_isConsistent(this));
-            return value;
+   assert(Hashtable_get(this, key) == NULL);
-         }
+
-      }
+   /* shrink on load-factor < 0.125 */
-   }
+   if (8 * this->items < this->size)
-   assert(Hashtable_isConsistent(this));
+      Hashtable_setSize(this, this->size / 2);
-   return NULL;
+
+   return res;
 }
 
 void* Hashtable_get(Hashtable* this, hkey_t key) {
    unsigned int index = key % this->size;
-   HashtableItem* bucketPtr = this->buckets[index];
+   unsigned int probe = 0;
-   while (true) {
+   void* res = NULL;
-      if (bucketPtr == NULL) {
+#ifndef NDEBUG
+   unsigned int origIndex = index;
+#endif
+
    assert(Hashtable_isConsistent(this));
-         return NULL;
+
-      } else if (bucketPtr->key == key) {
+   while (this->buckets[index].value) {
-         assert(Hashtable_isConsistent(this));
+      if (this->buckets[index].key == key) {
-         return bucketPtr->value;
+         res = this->buckets[index].value;
-      } else {
+         break;
-         bucketPtr = bucketPtr->next;
       }
+
+      if (this->buckets[index].probe < probe)
+         break;
+
+      index = (index + 1) != this->size ? (index + 1) : 0;
+      probe++;
+
+      assert(index != origIndex);
    }
+
+   return res;
 }
 
 void Hashtable_foreach(Hashtable* this, Hashtable_PairFunction f, void* userData) {
    assert(Hashtable_isConsistent(this));
    for (unsigned int i = 0; i < this->size; i++) {
-      HashtableItem* walk = this->buckets[i];
+      HashtableItem* walk = &this->buckets[i];
-      while (walk != NULL) {
+      if (walk->value)
          f(walk->key, walk->value, userData);
-         walk = walk->next;
-      }
    }
    assert(Hashtable_isConsistent(this));
 }

Hashtable.h

@@ -16,13 +16,13 @@ typedef void(*Hashtable_PairFunction)(hkey_t key, void* value, void* userdata);
 
 typedef struct HashtableItem_ {
    hkey_t key;
+   unsigned int probe;
    void* value;
-   struct HashtableItem_* next;
 } HashtableItem;
 
 typedef struct Hashtable_ {
    unsigned int size;
-   HashtableItem** buckets;
+   HashtableItem* buckets;
    unsigned int items;
    bool owner;
 } Hashtable;
@@ -37,6 +37,8 @@ Hashtable* Hashtable_new(unsigned int size, bool owner);
 
 void Hashtable_delete(Hashtable* this);
 
+void Hashtable_setSize(Hashtable* this, unsigned int size);
+
 void Hashtable_put(Hashtable* this, hkey_t key, void* value);
 
 void* Hashtable_remove(Hashtable* this, hkey_t key);

UsersTable.c

@@ -20,7 +20,7 @@ in the source distribution for its full text.
 UsersTable* UsersTable_new() {
    UsersTable* this;
    this = xMalloc(sizeof(UsersTable));
-   this->users = Hashtable_new(20, true);
+   this->users = Hashtable_new(10, true);
    return this;
 }