* Karl Waclawek <karl@w...> [2005-01-04 18:32]:
> 
> 
> Alan Gutierrez wrote:
> >* Karl Waclawek <karl@w...> [2005-01-04 15:22]:
> >
> >>Alan Gutierrez wrote:
> >>
> >>
> >>>>>I think the API sticks it's nose in. I think API provides a
> >>>>>conduit for XML message events coming in, and it needs to
> >>>>>provide a conduit for error events going, er, where ever.
> >>>>
> >>>>Yes, a conduit, but it should not care about what goes through
> >>>>that conduit.  This would create unnecessary dependencies.
> >>>
> >>>
> >>>  No. I think that means a lot of Object.
> >>>  
> >>>  Attaching a map to the error event?
> >>
> >>The fact that you are adding certain methods with a specific signature
> >>already defines dependencies, even if you use Object everywhere.
> >>Using an abstract class is more open, IMO. It would serve as what
> >>you would call an opaque pointer in other languages.
> >
> >
> >    I don't follow.
> >
> >    public interface StrategyError /* maybe something here */ {
> >        // Stuff here.
> >    }
> >
> >    Suggestions?

> I would use a class, not an interface.
> Sender and receiver agree on a subclass appropriate for their
> problem. The sender passes it as abstract base class, the receiver
> casts it back, the API knows nothing about the subclass.

> You can add some always needed abstract members to the base class,
> like getMessage(), getCause(), etc. But don't add too many,
> as they may create dependencies, and may not always be needed,
> making the design non-minimal.

    If the sender and receiver agree on a class, then they can put
    it in variables map attached to the StrategyError.

    By making StrategyError a /final/ class, I'm enforcing a
    convention, that might be a bad convention, but the idea is that
    Strategy itself does not create an error.
    
    This class is the conduit, or at least the packet in the coduit.
    
    An interface, to me, is only interesting if it specifies a
    behavior, and not simply data members.

--
Alan Gutierrez - alan@e...