axis-java-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sanjiva Weerawarana <>
Subject Re: [AXIS2] Proposal for saving the message context
Date Mon, 18 Dec 2006 14:49:15 GMT
On Mon, 2006-12-18 at 09:23 +0000, David Illsley wrote:
> >
> > I agree the proposal has been around for a long time but I've pointed
> > out that its flawed and you're saying "its not flawed for the way I want
> > to use it". But if anyone else does mc.serialize() and tries to bring it
> > back up it won't work.
> I don't buy this argument at all. Anyone who serializes a
> MessageContext and expects a fully restored system just by
> deserialising it again is IMO living in cloud cuckoo land.

?? If that's not the case then what is the point of message context
serialization and deserialization? 

OK how about this- Ann, since you just want to use this feature for your
own persistence model for Sandesha and nothing else, then you don't need
to put this code in the core at all: just write the same writeObject and
readObject methods in a utility class in Sandesha2 and call it yourself.
The only diff is where the code goes.

If that works then there's no problem at all.

> That's not
> to say someone might not, but that there are any number of bits of
> Axis2 that it isn't at all obvious how it works and some additional
> investigation, doc reading and questions is required. I don't think
> the limitations outlined by Ann are in any way an egregious addition
> to those areas, and in addition they're well documented.

I agree its documented but if there's no need to do it I'm confused why
there's insistence on making a core change for something that's focused
on a specific usecase- do the code just for that case then.

> I see this very much in the same scope as the clustering proposal -
> i.e additional function that is:
> 1. Complicated
> 2. Not critical for every Axis2 user
> 3. Not possible to fully isolate from axis2-kernel
> 4. Potentially confusing for basic users

Java object serialization is a fundamental thing in Java and many many
people are aware of it and use it for various things. That's hardly
comparable to someone messing with a clustering API.

> I certainly don't believe that the clustering proposal had all the
> answers or gave a view of an implementation that I want in Axis2.

I don't recall seeing any issues raised by you ... did I miss it?

> However, I think we went the right way in accepting the clustering
> proposal despite these problems as we can simplify, hide and factor
> out over time.
> I believe the same holds for Message Context serialisation. I don't
> fully understand all of it right now but I can see the benefit. Lets
> work with it on the trunk and over time we can improve our
> understanding and most likely some of the code (no offence intended
> Ann!)

I have no objection to this approach in general and we've done that many
times. Iterative development is of course the right way to do things.

However I'd like to understand whether this issue can be dealt with
locally for Sandesha only or not. 

> P.S. I'm actually hopeful that the 2 pieces of work are complimentary
> and that by adding both to head and working on them there we'll end up
> with a much better kernel.

Sorry I didn't grok which 2 pieces of work you're referring to? Are you
talking about using MC serialization for clustering as well? That's a
very good example of someone wanting to use the mc serialization stuff
and it failing totally, even with the proposed solution Ann's now
working on now.

I'm surprised I'm the only one who cares about this to comment!

Sanjiva Weerawarana, Ph.D.
Founder & Director; Lanka Software Foundation;
Founder, Chairman & CEO; WSO2, Inc.;
Director; Open Source Initiative;
Member; Apache Software Foundation;
Visiting Lecturer; University of Moratuwa;

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message