tags:

views:

399

answers:

8

In a Java program, I have multiple subclasses inheriting from a parent (which is abstract). I wanted to express that every child should have a member that is set once only (which I was planning to do from the constructor). My plan was to code s.th. like this:

public abstract class Parent {
    protected final String birthmark;
}

public class Child extends Parent {
    public Child(String s) {
        this.birthmark = s;
    }
}

However, this seems to not please the Java gods. In the parent class, I get the message that birthmark "might not have been initialized", in the child class I get "The final field birthmark cannot be accessed".

So what's the Java way for this? What am I missing?

+7  A: 

Pass it to the parent constructor:

public abstract class Parent {
    private final String birthmark;
    public Parent(String s) {
        birthmark = s;
    }
}

public class Child extends Parent {
    public Child(String s) {
        super(s);
    }
}
McDowell
+7  A: 

You can't do it because while comparing the parent class, the compiler can't be sure that the subclass will initialize it. You'll have to initialize it in the parent's constructor, and have the child call the parent's constructor:

public abstract class Parent {
    protected final String birthmark;
    protected Parent(String s) {
        birthmark = s;
    }
}

public class Child extends Parent {
    public Child(String s) {
        super(s);
        ...
    }
}
Adam Rosenfield
+1 for making the Parent ctor protected!
Paul Tomblin
If I would skip declaring the child's constructor (and, in that case, probably keep the parent's ctor public), I could do 'new Child("set this")' and get what I want, right?
Hanno Fietz
+1  A: 

Another Java-ish way to do this is probably to have the parent class to define an abstract "getter", and have the children implement it. It's not a great way to do it in this case, but it in some cases it can be exactly what you want.

Paul Tomblin
calling the abstract getter method in the constructor can be potentially dangerous however, a good rule is to never call an overrideable method, either directlry or indirectly, from a constructor.
TofuBeer
A: 

Yes, the final members are to be assigned in the class in which they are declared. You need to add a constructor with a String argument to Parent.

Maurice Perry
A: 

Declare a constructor in the superclass that's called by the subclass. You must set the field in the superclass to make sure it's initialized, or the compiler can't be sure the field is initialized.

Jorn
A: 

You probably want to have a Parent(String birthmark) constructor so that you can ensure in your Parent class that final is always initialized. Then you can call super(birthmark) from your Child() constructor.

Cody Casterline
A: 

Why not delegate initialization to a method. Then override the method in the parent class.

public class Parent {
   public final Object x = getValueOfX();
   public Object getValueOfX() {
      return y;
   }
}
public class Child {
  @Override
  public Object getValueOfX() {
     // whatever ...
  }
}

This should allow custom initialization.

ng
This idiom is a really bad idea... if a user makes getValueofX() depend on any field of the Child class, there will be a NullPointerException, since the Child has not yet actually been constructed. It can work if you are very careful with docs, but stupid/malicious users can break it in a heartbeat.
Chamelaeon
@Chamelaeon: This is such a stupid comment, what would a malicious or stupid user be doing writing code for your project? A malicious or stupid user could write System.exit(0)?
ng
+1  A: 

I would do it like this:

public abstract class Parent 
{
    protected final String birthmark;

    protected Parent(final String mark)
    {
        // only if this makes sense.
        if(mark == null)
        {
            throw new IllegalArgumentException("mark cannot be null");
        }

        birthmark = mark;
    }
}

public class Child 
    extends Parent 
{
    public Child(final String s) 
    {
        super(s);
    }
}

final means that the variable can be initialized once per instance. The compiler isn't able to make sure that every subclass will provide the assignment to birthmark so it forces the assignment to happen in the constructor of the parent class.

I added the checking for null just to sho wthat you also get the benefit of being able to check the arguments in one place rather than each cosntructor.

TofuBeer
Is there a reason to mark the parameters in the declaration of the function final, too? The compiler let me get away without.
Hanno Fietz
I do it out of habit... stops you from doing param = instanceVariable. If hotspot is ever able to optimize final vars better (I don't think it does now) then my code will magically be faster too :-)
TofuBeer