为其他对象实现String.intern（）的等效项

Question

I'm trying to implement an equivalent to String.intern(), but for other objets. My goal is the following: I've an object A which I will serialize and then deserialize. If there is another reference to A somewhere, I want the result of the deserialization to be the same reference.

Here is one example of what I would expect.

MyObject A = new MyObject();
A.data1 = 1;
A.data2 = 2;
byte[] serialized = serialize(A);
A.data1 = 3;
MyObject B = deserialize(serialized); // B!=A and B.data1=1, B.data2=2
MyObject C = B.intern(); // Here we should have C == A. Consequently C.data1=3 AND C.data2=2

Here is my implementation atm. (the MyObject class extends InternableObject )

public abstract class InternableObject {

private static final AtomicLong maxObjectId = new AtomicLong();
private static final Map<Long, InternableObject> dataMap = new ConcurrentHashMap<>();
private final long objectId;

public InternableObject() {
    this.objectId = maxObjectId.incrementAndGet();

    dataMap.put(this.objectId, this);
}

@Override
protected void finalize() throws Throwable {
    super.finalize();
    dataMap.remove(this.objectId);
}

public final InternableObject intern() {
    return intern(this);
}

public static InternableObject intern(InternableObject o) {
    InternableObject r = dataMap.get(o.objectId);

    if (r == null) {
        throw new IllegalStateException();
    } else {
        return r;
    }
}
}

My unit test (which fails):

    private static class MyData extends InternableObject implements Serializable {

    public int data;

    public MyData(int data) {
        this.data = data;
    }
}

@Test
public void testIntern() throws Exception {
    MyData data1 = new MyData(7);

    ByteArrayOutputStream baos = new ByteArrayOutputStream();
    ObjectOutputStream oos = new ObjectOutputStream(baos);
    oos.writeObject(data1);
    oos.flush();
    baos.flush();
    oos.close();
    baos.close();
    ByteArrayInputStream bais = new ByteArrayInputStream(baos.toByteArray());
    ObjectInputStream ois = new ObjectInputStream(bais);
    MyData data2 = (MyData) ois.readObject();

    Assert.assertTrue(data1 == data2.intern()); // Fails here
}

The failure is due to the fact that, when deserializing, the constructor of InternableObject is called, and thus objectId will be 2 (even if the serialized data contains "1")

Any idea about how to solve this particular problem or, another approach to handle the high level problem ?

Thanks guys

Answer 1

Do not use the constructor to create instances. Use a factory method that checks if an instance already exists first, only create an instance if there isn't already a matching one.

To get serialization to cooperate, your class will need to make use of readResolve() / writeReplace(). http://docs.oracle.com/javase/7/docs/platform/serialization/spec/serial-arch.html#4539

The way you implemented your constructor, you're leaking a reference during construction, which can lead to very hard to nail down problems. Also, your instance map isn't protected by any locks, so its not thread save.

Answer 2

Typically intern() forms an aspect , and maybe should not be realized as a base class, maybe too restricting its usage in a more complex constellation.

There are two aspects:

1. Sharing the "same" object.

Internalizing an object only gives a profit, when several objects can be "internalized" to the same object. So I think, that InternalableObjecte. with a new sequential number is not really adequate. More important is that the class defines a fitting equals and hashCode.

Then you can do an identity Map<Object, Object> :

public class InternMap {
    private final Map<Object, Object> identityMap = new HashMap<>();
    public static <I extends Internalizable<?>> Object intern(I x) {
        Object first = identityMap.get(x);
        if (first == null) {
           first = x;
           identityMap.put(x, x);
        }
        return first;
    }
}

InternMap could be used for any class, but above we restrict it to Internalizable things.

2. Replacing a dynamically created non-shared object with it's .intern() .

Which in Java 8 could be realised with a defualt method in an interface:

interface Internalizable<T> {
    public static final InternMap interns = new InternMap();
    public default T intern(Class<T> klazz) {
        return klazz.cast(internMap.intern(this));
    }

class C implements Internalizable<C> { ... }

C x = new C();
x = x.intern(C.class);

The Class<T> parameter needed because of type erasure. Concurrency disregarded here.

Prior to Java 8, just use an empty interface Internalizable as _marker: interface, and use a static InternMap.