Generic Type Question

[Firen4000]

Hey everyone,

I have a problem dealing with generic type methods. I have the following:

class Point: Is a point in an n-dimensional space. It has a float array containing the values for every dimension

subclasses Point2/ Point3/ Point4: Are specific points with defined number of dimensions and some dimensionspecific methods (for example cross-product etc.)

Now if I create a method Point.add(float pValue), which adds the float argument to every dimension separately, I should put it in the superclass Point since the implementation is the same for all subclasses. But the return-type of this function needs to be Point2/ Point3/ Point4 depending on what kind of subclass I am dealing with. I am not passing a class<T extends Point> or an object of the subclass to the function and thus it does not know what exactly T is.
This is what I have so far:

public <T extends Point> T add(float pValue)
{
    float[] tNewValues = new float[this.getValues().length];
    for (int i = 0; i < tNewValues.length; i++)
    {
        tNewValues[i] = this.getValues()[i] + pValue;
    }
    return (T) createSubClass(tNewValues);
}

The createSubClass() function takes the float array and creates a Point2/ Point3/ Point4 object depending on the length of the array.

Is there any way to change the return type of the function depending on what subclass I am dealing with? So if i have an instance of Point2 and call the add function the return type is also Point2?

I already have two solutions for this problem but they are both some kind of crappy:

1. Create add() methods in all subclasses of Point that call the method of the superclass and have the correct return type, like this:

public class Point2 extends Point
{
    ...
    @Override
    public Point2 add(float pValue)
    {
        return (Point2) super.add(pValue);
    }
    ...
}

2. Whenever i call the add() method with a Point2 object I need to typecast it to Point2 (if I dont I get a compiler error "cannot be applied to given types"):

...
Point2 p1 = new Point2(0.0f, 0.0f);
Point2 p2 = (Point2) p1.add(1.0f);
...

One thing I have noticed is that with solution1 i actually dont have to typecast the return value to Point2:

public Point2 add(float pValue)
{
    return super.add(pValue);
}

This works without an error. But using the add() method like in solution2 i need the typecast orelse i get a compiler error.

I am a bit confused about all this so if someone can explain and help me I would appreciate it.

[pbrockway2]

Defining subclass behaviour in a superclass seems a bit backwards. Are you really sure you need to subclass?

Cross/dot product, translation and enlargement can all be defined for the n dimensional case, for example. [Edit] which you are doing. Sure you would get a compile time check on the dimension, but folk manage to work with arrays (assign, pass, return) without having nanny compiler worry about the array length. Have a PointDimensionException and throw it, as you would ArrayIndexException, if anything goes wrong.

-----

Could you post a SSCCE of the Point class, its add() method and the Point2 subclass, and say what grumbling you are getting from the compiler?

[Firen4000]

As far as I know cross product can only be calculated in three and seven dimensions and quaternion stuff needs 4 dimensions, thats why I made subclasses.

public abstract class Point
{
    private float[] mValues;

    public Point(float[] pValues)
    {
        mValues = pValues;
    }

    public float[] getValues()
    {
        return mValues;
    }
    
    public <T extends Point> T add(float pAdd)
    {
        float[] tNewValues = new float[this.getValues().length];
        for (int i = 0; i < tNewValues.length; i++)
        {
            tNewValues[i] = this.getValues()[i] + pValue;
        }
        return (T) createSubClass(tNewValues);
    }

    public static Point createSubClass(float[] pValues)
    {
        switch (pValues.length)
        {
            case 2:
            {
                return new Point 2(pValues);
            }
            case 3:
            {
                return new Point3(pValues);
            }
            case 4:
            {
                return new Point4(pValues);
            }
            default:
            {
                return null;
            }
        }
    }
}

public class Point3 extends Point
{
    public Point(float[] pValues)
    {
        super(pValues);
    }

    public Point3 cross(Point3 pPoint)
    {
        Point3 tCross;
        ...
        return tCross;
    }
}

Error:

  error: method setLocation in class Bla cannot be applied to given types;
  setLocation(getLocation().add(pDistance));
  required: Point3
  found: Point
  reason: actual argument Point cannot be converted to Point3 by method invocation conversion

Another weird thing is that the add() method of Point class has a return type "Point" although I put <T extends Point> and a class does not extend itself.

EDIT: Also I would like to keep the Point2/ Point3/ Point4 subclasses since they make the whole thing more readable and just nicer..

[pbrockway2]

As far as I know cross product can only be calculated in three and seven dimensions and quaternion stuff needs 4 dimensions, thats why I made subclasses.

You're right, my bad.