andrew cooke wrote:
> I can't even remember what that was about (transport specific endpoint
> element).  It's in one of the Jira issues isn't it?  Can you remember

The discussion was between having arbitrary (untyped) properties on
endpoints or making the allowed values explicit, which would mean specific
endpoint subclasses.

For the rewrite of the javaspaces transport I had exactly this problem,
and decided to put differing endpoint metainfo into the generic
properties. On connect those properties are interpreted by a factory that
knows the endpoints' specific protocol prefix. That keeps the endpoints
generic and I don't have to wrangle the configuration info into the URL,
which would have been *really* ugly in some instances.

I'm somewhat undecided. Making stuff more explicit is better for a checked
configuration but allowing endpoint subclasses may not be helpful
afterwards when one needs to access those values in a generic way - the
consequence would probably be instanceof cascades, downcasting..
However maybe my feeling/assumption that this generic access is needed is
flawed, so it might as well be a non-issue.

-h

---------------------------------------------------------------------
To unsubscribe from this list please visit:

    http://xircles.codehaus.org/manage_email