classpath
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: RMI and JOnAS (and Cajo)


From: Mark Wielaard
Subject: Re: RMI and JOnAS (and Cajo)
Date: Sat, 14 Feb 2004 12:25:20 +0100

Hi,

On Sat, 2004-02-14 at 11:36, Andrew Haley wrote:
> JOnAS (http://jonas.objectweb.org/) is, as you may know, the Open
> Source implementation by ObjectWeb of the J2EE(TM) specification.
> 
> We're having trouble with running it.  The basic problem is (quoth
> Simon Nieuviarts) "because Carol is an improvement or a replacement of
> RMI. As such, it needs to plug low in the ORB in functions that are
> not intended to be exported to the end user. AFAIK, without the use of
> internal classes you can not implement mechanisms like interceptors
> which propagate transaction contexts."  To do this, carol uses classes
> in sun.rmi.transport.*.
> 
> What I would like to do is define an interface in Classpath which
> exposes the functionality that JOnAS needs.
> 
> Trouble is, I know zip about RMI.

Good more interest in RMI. I know also almost zip about RMI, but I am
sure when pulling some knowledge together we can get somewhere. Once
upon a time the Orp developers got our RMI implementation working good
enough to support JBoss. And I know the kaffe developers recently fixed
up some little things to make a couple of standard RMI examples work on
kaffe. Also Norbert Frese has been posting a couple of RMI patches on
the gcj mailinglist <http://gcc.gnu.org/ml/java/2004-01/msg00022.html>,
<http://gcc.gnu.org/ml/java-patches/2004-q1/msg00080.html>,
<http://gcc.gnu.org/ml/java-patches/2004-q1/msg00081.html>
and he made a RMI interoperability testsuite
<http://www.scheinwelt.at/~norbertf/gcj/tests/>.

Yesterday have been looking a bit at the Cajo project
<https://cajo.dev.java.net/> which is a distributed object framework
build on top of RMI. (John Catherina has been CCed since he might have
some ideas and clearly knows more about RMI then most of us). Cajo
stress tests RMI and Serialization a lot. And we do have some way to go
before it works. Things I noticed that are currently broken:

- Our rmic implementation doesn't seem to handle 'super-interfaces'
correctly. This doesn't seem to be that hard to fix and can be easily
worked around by implementing the 'super-interface' directly.

- Our java.awt.Font serialization (which the main cajo example uses) is
broken. Not hard to fix, but to make it compatible with other
implementations might be hard since not everything is clearly
documented.

- When extending UnicastRemoteObject a object gets immediatly exported
through its constructor which might be to early since then not all
fields of the object are initialized. This looks like a harder problem
to solve.

BTW Florent Benoit from JOnAS will be at FOSDEM
    http://www.fosdem.org/index/speakers/speakers_benoit

Cheers,

Mark

Attachment: signature.asc
Description: This is a digitally signed message part


reply via email to

[Prev in Thread] Current Thread [Next in Thread]