Le Java HashMap sous le capot

1. Vue d'ensemble

Dans cet article, nous allons explorer plus en détail l' implémentation la plus populaire de l' interface Map de Java Collections Framework, en reprenant là où notre article d'introduction s'est arrêté.

Avant de commencer l'implémentation, il est important de souligner que les interfaces de collection principales List et Set étendent Collection, mais pas Map .

En termes simples, le HashMap stocke les valeurs par clé et fournit des API pour ajouter, récupérer et manipuler les données stockées de différentes manières. La mise en œuvre est basée sur les principes d'une table de hachage , ce qui semble un peu complexe au début mais est en fait très facile à comprendre.

Les paires clé-valeur sont stockées dans ce que l'on appelle des buckets qui forment ensemble ce que l'on appelle une table, qui est en fait un tableau interne.

Une fois que nous connaissons la clé sous laquelle un objet est stocké ou doit être stocké, les opérations de stockage et de récupération se produisent en temps constant , O (1) dans une carte de hachage bien dimensionnée.

Pour comprendre comment les cartes de hachage fonctionnent sous le capot, il faut comprendre le mécanisme de stockage et de récupération utilisé par le HashMap. Nous nous concentrerons beaucoup sur ces derniers.

Enfin, les questions liées à HashMap sont assez courantes dans les entretiens , c'est donc un moyen solide de préparer une interview ou de s'y préparer.

2. L' API put ()

Pour stocker une valeur dans une carte de hachage, nous appelons l' API put qui prend deux paramètres; une clé et la valeur correspondante:

V put(K key, V value);

Lorsqu'une valeur est ajoutée à la carte sous une clé, l' API hashCode () de l'objet clé est appelée pour récupérer ce que l'on appelle la valeur de hachage initiale.

Pour voir cela en action, créons un objet qui agira comme une clé. Nous ne créerons qu'un seul attribut à utiliser comme code de hachage pour simuler la première phase de hachage:

public class MyKey { private int id; @Override public int hashCode() { System.out.println("Calling hashCode()"); return id; } // constructor, setters and getters }

Nous pouvons maintenant utiliser cet objet pour mapper une valeur dans la carte de hachage:

@Test public void whenHashCodeIsCalledOnPut_thenCorrect() { MyKey key = new MyKey(1); Map map = new HashMap(); map.put(key, "val"); }

Il ne se passe pas grand-chose dans le code ci-dessus, mais faites attention à la sortie de la console. En effet, la méthode hashCode est appelée:

Calling hashCode()

Ensuite, l' API hash () de la carte de hachage est appelée en interne pour calculer la valeur de hachage finale à l'aide de la valeur de hachage initiale.

Cette valeur de hachage finale se résume finalement à un index dans le tableau interne ou à ce que nous appelons un emplacement de compartiment.

La fonction de hachage de HashMap ressemble à ceci:

static final int hash(Object key) { int h; return (key == null) ? 0 : (h = key.hashCode()) ^ (h >>> 16); }

Ce que nous devons noter ici, c'est uniquement l'utilisation du code de hachage de l'objet clé pour calculer une valeur de hachage finale.

À l'intérieur de la fonction put , la valeur de hachage finale est utilisée comme ceci:

public V put(K key, V value) { return putVal(hash(key), key, value, false, true); }

Notez qu'une fonction putVal interne est appelée et reçoit la valeur de hachage finale comme premier paramètre.

On peut se demander pourquoi la clé est à nouveau utilisée à l'intérieur de cette fonction puisque nous l'avons déjà utilisée pour calculer la valeur de hachage.

La raison en est que les cartes de hachage stockent à la fois la clé et la valeur dans l'emplacement du compartiment en tant qu'objet Map.Entry .

Comme indiqué précédemment, toutes les interfaces du framework de collections Java étendent l' interface de Collection , mais pas Map . Comparez la déclaration de l'interface Map que nous avons vue précédemment à celle de l' interface Set :

public interface Set extends Collection

La raison en est que les cartes ne stockent pas exactement des éléments uniques comme le font d'autres collections, mais plutôt une collection de paires clé-valeur.

Ainsi, les méthodes génériques de l' interface Collection telles que add , toArray n'ont pas de sens quand il s'agit de Map .

Le concept que nous avons couvert dans les trois derniers paragraphes constitue l'une des questions d'entretien les plus populaires de Java Collections Framework . Donc, cela vaut la peine d'être compris.

Un attribut spécial avec la carte de hachage est qu'il accepte les valeurs nulles et les clés nulles:

@Test public void givenNullKeyAndVal_whenAccepts_thenCorrect(){ Map map = new HashMap(); map.put(null, null); }

When a null key is encountered during a put operation, it is automatically assigned a final hash value of 0, which means it becomes the first element of the underlying array.

This also means that when the key is null, there is no hashing operation and therefore, the hashCode API of the key is not invoked, ultimately avoiding a null pointer exception.

During a put operation, when we use a key that was already used previously to store a value, it returns the previous value associated with the key:

@Test public void givenExistingKey_whenPutReturnsPrevValue_thenCorrect() { Map map = new HashMap(); map.put("key1", "val1"); String rtnVal = map.put("key1", "val2"); assertEquals("val1", rtnVal); }

otherwise, it returns null:

@Test public void givenNewKey_whenPutReturnsNull_thenCorrect() { Map map = new HashMap(); String rtnVal = map.put("key1", "val1"); assertNull(rtnVal); }

When put returns null, it could also mean that the previous value associated with the key is null, not necessarily that it's a new key-value mapping:

@Test public void givenNullVal_whenPutReturnsNull_thenCorrect() { Map map = new HashMap(); String rtnVal = map.put("key1", null); assertNull(rtnVal); }

The containsKey API can be used to distinguish between such scenarios as we will see in the next subsection.

3. The get API

To retrieve an object already stored in the hash map, we must know the key under which it was stored. We call the get API and pass to it the key object:

@Test public void whenGetWorks_thenCorrect() { Map map = new HashMap(); map.put("key", "val"); String val = map.get("key"); assertEquals("val", val); }

Internally, the same hashing principle is used. The hashCode() API of the key object is called to obtain the initial hash value:

@Test public void whenHashCodeIsCalledOnGet_thenCorrect() { MyKey key = new MyKey(1); Map map = new HashMap(); map.put(key, "val"); map.get(key); }

This time, the hashCode API of MyKey is called twice; once for put and once for get:

Calling hashCode() Calling hashCode()

This value is then rehashed by calling the internal hash() API to obtain the final hash value.

As we saw in the previous section, this final hash value ultimately boils down to a bucket location or an index of the internal array.

The value object stored in that location is then retrieved and returned to the calling function.

When the returned value is null, it could mean that the key object is not associated with any value in the hash map:

@Test public void givenUnmappedKey_whenGetReturnsNull_thenCorrect() { Map map = new HashMap(); String rtnVal = map.get("key1"); assertNull(rtnVal); }

Or it could simply mean that the key was explicitly mapped to a null instance:

@Test public void givenNullVal_whenRetrieves_thenCorrect() { Map map = new HashMap(); map.put("key", null); String val=map.get("key"); assertNull(val); }

To distinguish between the two scenarios, we can use the containsKey API, to which we pass the key and it returns true if and only if a mapping was created for the specified key in the hash map:

@Test public void whenContainsDistinguishesNullValues_thenCorrect() { Map map = new HashMap(); String val1 = map.get("key"); boolean valPresent = map.containsKey("key"); assertNull(val1); assertFalse(valPresent); map.put("key", null); String val = map.get("key"); valPresent = map.containsKey("key"); assertNull(val); assertTrue(valPresent); }

For both cases in the above test, the return value of the get API call is null but we are able to distinguish which one is which.

4. Collection Views in HashMap

HashMap offers three views that enable us to treat its keys and values as another collection. We can get a set of all keys of the map:

@Test public void givenHashMap_whenRetrievesKeyset_thenCorrect() { Map map = new HashMap(); map.put("name", "baeldung"); map.put("type", "blog"); Set keys = map.keySet(); assertEquals(2, keys.size()); assertTrue(keys.contains("name")); assertTrue(keys.contains("type")); }

The set is backed by the map itself. So any change made to the set is reflected in the map:

@Test public void givenKeySet_whenChangeReflectsInMap_thenCorrect() { Map map = new HashMap(); map.put("name", "baeldung"); map.put("type", "blog"); assertEquals(2, map.size()); Set keys = map.keySet(); keys.remove("name"); assertEquals(1, map.size()); }

We can also obtain a collection view of the values:

@Test public void givenHashMap_whenRetrievesValues_thenCorrect() { Map map = new HashMap(); map.put("name", "baeldung"); map.put("type", "blog"); Collection values = map.values(); assertEquals(2, values.size()); assertTrue(values.contains("baeldung")); assertTrue(values.contains("blog")); }

Just like the key set, any changes made in this collection will be reflected in the underlying map.

Finally, we can obtain a set view of all entries in the map:

@Test public void givenHashMap_whenRetrievesEntries_thenCorrect() { Map map = new HashMap(); map.put("name", "baeldung"); map.put("type", "blog"); Set
    
      entries = map.entrySet(); assertEquals(2, entries.size()); for (Entry e : entries) }
    

Remember that a hash map specifically contains unordered elements, therefore we assume any order when testing the keys and values of entries in the for each loop.

Many times, you will use the collection views in a loop as in the last example, and more specifically using their iterators.

Just remember that the iterators for all the above views are fail-fast.

If any structural modification is made on the map, after the iterator has been created, a concurrent modification exception will be thrown:

@Test(expected = ConcurrentModificationException.class) public void givenIterator_whenFailsFastOnModification_thenCorrect() { Map map = new HashMap(); map.put("name", "baeldung"); map.put("type", "blog"); Set keys = map.keySet(); Iterator it = keys.iterator(); map.remove("type"); while (it.hasNext()) { String key = it.next(); } }

The only allowed structural modification is a remove operation performed through the iterator itself:

public void givenIterator_whenRemoveWorks_thenCorrect() { Map map = new HashMap(); map.put("name", "baeldung"); map.put("type", "blog"); Set keys = map.keySet(); Iterator it = keys.iterator(); while (it.hasNext()) { it.next(); it.remove(); } assertEquals(0, map.size()); }

The final thing to remember about these collection views is the performance of iterations. This is where a hash map performs quite poorly compared with its counterparts linked hash map and tree map.

Iteration over a hash map happens in worst case O(n) where n is the sum of its capacity and the number of entries.

5. HashMap Performance

The performance of a hash map is affected by two parameters: Initial Capacity and Load Factor. The capacity is the number of buckets or the underlying array length and the initial capacity is simply the capacity during creation.

The load factor or LF, in short, is a measure of how full the hash map should be after adding some values before it is resized.

The default initial capacity is 16 and default load factor is 0.75. We can create a hash map with custom values for initial capacity and LF:

Map hashMapWithCapacity=new HashMap(32); Map hashMapWithCapacityAndLF=new HashMap(32, 0.5f);

The default values set by the Java team are well optimized for most cases. However, if you need to use your own values, which is very okay, you need to understand the performance implications so that you know what you are doing.

When the number of hash map entries exceeds the product of LF and capacity, then rehashing occurs i.e. another internal array is created with twice the size of the initial one and all entries are moved over to new bucket locations in the new array.

A low initial capacity reduces space cost but increases the frequency of rehashing. Rehashing is obviously a very expensive process. So as a rule, if you anticipate many entries, you should set a considerably high initial capacity.

On the flip side, if you set the initial capacity too high, you will pay the cost in iteration time. As we saw in the previous section.

So a high initial capacity is good for a large number of entries coupled with little to no iteration.

A low initial capacity is good for few entries with a lot of iteration.

6. Collisions in the HashMap

A collision, or more specifically, a hash code collision in a HashMap, is a situation where two or more key objects produce the same final hash value and hence point to the same bucket location or array index.

This scenario can occur because according to the equals and hashCode contract, two unequal objects in Java can have the same hash code.

It can also happen because of the finite size of the underlying array, that is, before resizing. The smaller this array, the higher the chances of collision.

That said, it's worth mentioning that Java implements a hash code collision resolution technique which we will see using an example.

Keep in mind that it's the hash value of the key that determines the bucket the object will be stored in. And so, if the hash codes of any two keys collide, their entries will still be stored in the same bucket.

And by default, the implementation uses a linked list as the bucket implementation.

The initially constant time O(1)put and get operations will occur in linear time O(n) in the case of a collision. This is because after finding the bucket location with the final hash value, each of the keys at this location will be compared with the provided key object using the equals API.

To simulate this collision resolution technique, let's modify our earlier key object a little:

public class MyKey { private String name; private int id; public MyKey(int id, String name) { this.id = id; this.name = name; } // standard getters and setters @Override public int hashCode() { System.out.println("Calling hashCode()"); return id; } // toString override for pretty logging @Override public boolean equals(Object obj) { System.out.println("Calling equals() for key: " + obj); // generated implementation } }

Notice how we're simply returning the id attribute as the hash code – and thus force a collision to occur.

Also, note that we've added log statements in our equals and hashCode implementations – so that we know exactly when the logic is called.

Let's now go ahead to store and retrieve some objects that collide at some point:

@Test public void whenCallsEqualsOnCollision_thenCorrect() { HashMap map = new HashMap(); MyKey k1 = new MyKey(1, "firstKey"); MyKey k2 = new MyKey(2, "secondKey"); MyKey k3 = new MyKey(2, "thirdKey"); System.out.println("storing value for k1"); map.put(k1, "firstValue"); System.out.println("storing value for k2"); map.put(k2, "secondValue"); System.out.println("storing value for k3"); map.put(k3, "thirdValue"); System.out.println("retrieving value for k1"); String v1 = map.get(k1); System.out.println("retrieving value for k2"); String v2 = map.get(k2); System.out.println("retrieving value for k3"); String v3 = map.get(k3); assertEquals("firstValue", v1); assertEquals("secondValue", v2); assertEquals("thirdValue", v3); }

In the above test, we create three different keys – one has a unique id and the other two have the same id. Since we use id as the initial hash value, there will definitely be a collision during both storage and retrieval of data with these keys.

In addition to that, thanks to the collision resolution technique we saw earlier, we expect each of our stored values to be retrieved correctly, hence the assertions in the last three lines.

When we run the test, it should pass, indicating that collisions were resolved and we will use the logging produced to confirm that the collisions indeed occurred:

storing value for k1 Calling hashCode() storing value for k2 Calling hashCode() storing value for k3 Calling hashCode() Calling equals() for key: MyKey [name=secondKey, id=2] retrieving value for k1 Calling hashCode() retrieving value for k2 Calling hashCode() retrieving value for k3 Calling hashCode() Calling equals() for key: MyKey [name=secondKey, id=2]

Notice that during storage operations, k1 and k2 were successfully mapped to their values using only the hash code.

However, storage of k3 was not so simple, the system detected that its bucket location already contained a mapping for k2. Therefore, equals comparison was used to distinguish them and a linked list was created to contain both mappings.

Any other subsequent mapping whose key hashes to the same bucket location will follow the same route and end up replacing one of the nodes in the linked list or be added to the head of the list if equals comparison returns false for all existing nodes.

Likewise, during retrieval, k3 and k2 were equals-compared to identify the correct key whose value should be retrieved.

On a final note, from Java 8, the linked lists are dynamically replaced with balanced binary search trees in collision resolution after the number of collisions in a given bucket location exceed a certain threshold.

This change offers a performance boost, since, in the case of a collision, storage and retrieval happen in O(log n).

This section is very common in technical interviews, especially after the basic storage and retrieval questions.

7. Conclusion

In this article, we have explored HashMap implementation of Java Map interface.

Le code source complet de tous les exemples utilisés dans cet article se trouve dans le projet GitHub.