Ok,
I have had Scott and Bill tell me privately that I must commit and they
would pick up the pieces.
I have had 3 hours of sleep in the last 48 and I am taking a plane in 6
hours, the mode is "panic"
It doesn't work (that is just the way I work).
The stuff compiles but blows NPE on the server side, it is the debugging and
wiring stuff together that I am missing but the classes are there and
compile ok.
The testsuite just doesn't run (at all :) and the cluster stuff needs
rewrite with interceptors so the compile is really without cluster
But the stuff is there.
The design is simple, i was feeling all smug about it on Saturday until I
realized I came up with the servlet design basically for clients. In my
mind that servlet parallel is good and bad, good: it validates it against
previous interceptor work, but this time adapted to the client environment
and dynamic proxies, a really valid and powerful extension. Bad: I don't
get to feel all smug about my design, beaten path really with new eyes.
So the design is:
Interfaces -> Dynamic Proxy -> handler
handler is "ClientContainer"
Client Container changes
invoke(proxy, method,class) to
invoker(invocation) and send it down the next().next() pipe
in invocation, the object we have a InvocationContext that contains ALL THE
INFORMATION for the invocation and is really the "post modern" equivalent of
the server side, where instead of a "container" that is seen from everyone
we have fully stateless interceptors and a contextual information in the
invocation itself, closer to the Turin idea.
****************USE THAT INVOCATION CONTEXT TO STORE YOUR OWN STUFF ACROSS
CALLS**********
OK?
THIS IS ABSOLUTELY EQUIVALENT to HttpSession object you can retrieve in
Request in a servlet and the servlet "filter" is really the same
*implementation*, minus the typing sillyness in the invocation, fuck
wrappers of wrapper of wrappers, Invocation is the word.
So I will commit this in the coming hours, it will be very broken, it will
be very pretty.
I am very sorry for not finishing this work in time but both bill and scott
want it for JavaOne demos and I know a lot of you out there are waiting for
it.
OH, I left the difficult part for Bill/David to do. The difficult part is:
externalizing the client side interceptors in jboss.xml and making that work
in the ProxyFactory, right now I hard code the stack. I am thinking that
client can say
<jboss>
<ejb>...
<client-interceptors>
<remote>
ClientContainer
EntityInterceptor
MyOWNLittleFancyCache
TransactionI...
...</>
One interesting idea also is that THIS SHOULD REALLY LIVE AT THE MBEAN
LEVEL, NOT JUST EJB, here THE EJB NATURE is just SEEN IN ONE LITTLE
INTERCEPTOR, the whole rest is independent and should be use as a
generalized MBeanProxy of sort, bind it in JNDI at start() time or
something, so you can configure JRMP client on the fly for a give MBean and
you can include your interceptors, the second idea would be to have
externalized CONFIGURATION for them and a free flowing format for their
conf, just like we do server side, right now it is the bland () constructor
that is called.
So (bill scott david)
1- dig it
2- fix the server (if I don't in the coming 2 hours)
3- code your own interceptors (security/cluster) and document the examples
for others
4- david, look into the externalization if you can
at's it, I am feeling a bit "smashed" at the moment,
but I still love you guys
marcf
_______________________________________________
Jboss-development mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/jboss-development