Question

I’d like to extend ArrayList to add a few methods for a specific class whose instances would be held by the extended ArrayList. A simplified illustrative code sample is below.

This seems sensible to me, but I’m very new to Java and I see other questions which discourage extending ArrayList, for example Extending ArrayList and Creating new methods. I don’t know enough Java to understand the objections.

In my prior attempt, I ending up creating a number of methods in ThingContainer that were essentially pass-throughs to ArrayList, so extending seemed easier.

Is there a better way to do what I’m trying to do? If so, how should it be implemented?

import java.util.*;

class Thing {
    public String name;
    public int amt;

    public Thing(String name, int amt) {
        this.name = name;
        this.amt = amt;
    }

    public String toString() {
        return String.format("%s: %d", name, amt);
    }

    public int getAmt() {
        return amt;
    }
}

class ThingContainer extends ArrayList<Thing> {
    public void report() {
        for(int i=0; i < size(); i++) {
            System.out.println(get(i));
        }
    }

    public int total() {
        int tot = 0;
        for(int i=0; i < size(); i++) {
            tot += ((Thing)get(i)).getAmt();
        }
        return tot;
    }

}

public class Tester {
    public static void main(String[] args) {
        ThingContainer blue = new ThingContainer();

        Thing a = new Thing("A", 2);
        Thing b = new Thing("B", 4);

        blue.add(a);
        blue.add(b);

        blue.report();
        System.out.println(blue.total());

        for (Thing tc: blue) {
            System.out.println(tc);
        }
    }
}

Answer 1

Nothing in that answer discourages extending ArrayList; there was a syntax issue. Class extension exists so we may re-use code.

The normal objections to extending a class is the “favor composition over inheritance” discussion. Extension isn’t always the preferred mechanism, but it depends on what you’re actually doing.

Edit for composition example as requested.

public class ThingContainer implements List<Thing> { // Or Collection based on your needs.
    List<Thing> things;
    public boolean add(Thing thing) { things.add(thing); }
    public void clear() { things.clear(); }
    public Iterator<Thing> iterator() { things.iterator(); }
    // Etc., and create the list in the constructor
}

You wouldn’t necessarily need to expose a full list interface, just collection, or none at all. Exposing none of the functionality greatly reduces the general usefulness, though.

In Groovy you can just use the @Delegate annotation to build the methods automagically. Java can use Project Lombok‘s @Delegate annotation to do the same thing. I’m not sure how Lombok would expose the interface, or if it does.

I’m with glowcoder, I don’t see anything fundamentally wrong with extension in this case–it’s really a matter of which solution fits the problem better.

Edit for details regarding how inheritance can violate encapsulation

See Bloch’s Effective Java, Item 16 for more details.

If a subclass relies on superclass behavior, and the superclass’s behavior changes, the subclass may break. If we don’t control the superclass, this can be bad.

Here’s a concrete example, lifted from the book (sorry Josh!), in pseudo-code, and heavily paraphrased (all errors are mine).

class CountingHashSet extends HashSet {
    private int count = 0;
    boolean add(Object o) {
        count++;
        return super.add(o);
    }
    boolean addAll(Collection c) {
        count += c.size();
        return super.addAll(c);
    }
    int getCount() { return count; }
}

Then we use it:

s = new CountingHashSet();
s.addAll(Arrays.asList("bar", "baz", "plugh");

And it returns… three? Nope. Six. Why?

HashSet.addAll() is implemented on HashSet.add(), but that’s an internal implementation detail. Our subclass addAll() adds three, calls super.addAll(), which invokes add(), which also increments count.

We could remove the subclass’s addAll(), but now we’re relying on superclass implementation details, which could change. We could modify our addAll() to iterate and call add() on each element, but now we’re reimplementing superclass behavior, which defeats the purpose, and might not always be possible, if superclass behavior depends on access to private members.

Or a superclass might implement a new method that our subclass doesn’t, meaning a user of our class could unintentionally bypass intended behavior by directly calling the superclass method, so we have to track the superclass API to determine when, and if, the subclass should change.