我应该对应该由子类指定的数据使用抽象方法还是实例变量？

Question

I am writing an abstract class and I have a variable that should be defined by the sub-class (in this case it is an int which is used in the super-class). I can't decide whether to define a protected variable with a default value and let the sub-class change the value either in the constructor or through a setter method. Or to define an abstract method in the parent class such that all sub-classes must implement it and return the value they want to be used (then access the method in the super-class using the abstract method). Can anyone tell me if there are any great reasons why one way should be preferred over the other?

Abstract Method:

public abstract int getValue();

public int getValue() {
    return 5;
}

• Pros:

Forces sub-class to think about this value and choose what they want it to be.

• Cons:

No default value, so even if most of the sub-classes just want to use the same value they still have to implement the method. (The method can be made non-abstract and return the default value, but then you lose the advantage of forcing sub-classes to think about the value)

Protected Variable:

protected int value;

public SubClassImpl() {
    value = 5;
}

• Pros:

Can define a default value and sub-classes can just ignore it if they don't care.

• Cons:

Authors of the sub-class aren't made aware of the existence of the variable, so they might be surprised by the default behaviour.

Answer 1

Well, it depends...

• Does the value have to be able to change during the lifetime of the object? If not, you might want to make it a protected constructor parameter, and make it a final private variable in your base class.

• Might subclasses have reason to compute the variable based on other mutable state? If so, an abstract getter would be appropriate.

• Otherwise, I'd probably use a private variable and a protected setter. I'm generally not a fan of non-private variables. This could allow you to react to changes in the value in your base class, such as changing other computer fields. To avoid the presence of a default value you could combine this with a constructor parameter to force an initial value which the subclass could change later via the setter.

Answer 2

I usually use an abstract method, because that allows you to take advantage of covariant return types but a (mutable) field or property does not.

For example in Guava , this allows ForwardingSortedMap to extend ForwardingMap .