On Mar 20, 2006, at 1:16 PM, Aaron Mulder wrote:
On 3/20/06, David Jencks <[EMAIL PROTECTED]> wrote:
I think we are basically thinking of trying to eliminate using
strings or names as intermediate navigation aids and be able to get
the child objects directly from the parents. I think you suggested
doing this early on in the console and I objected :-)... I'm
reconsidering. I don't really see how this would work with remote
kernels, but maybe either it could work or its not important.
I certainly have no fondness for ObjectNames.
My only concern is that as we add clustering, I think we'll want a
single console to be able to manage more than one server, so I don't
think we should drop support for management from remote clients -- at
least, not without a plan for how to bring it back later. Would we be
able to return proxies that work for remote clients? Maybe have some
plumbing to decode the object (proxy) to an ObjectName plus a set of
interfaces (they all support this) on the server side, send that to
the client side, and then re-wrap it with a new proxy on the client
side before returning the value to the caller?
I think we need to make remoteable proxies like EJB uses.
Also, we can't ever drop the String-returning-methods that JSR-77
requires, but we could provide alternatives for all of those and
convert all our extensions to use only object-to-object references
instead of more name references.
I agree. I doubt this is something we can do in 1.1.
-dain
