On Friday, August 8, 2003, at 01:53 pm, Dhananjay Nene wrote:
Assuming one started development without any existing code, AOP wouldOn Friday, August 8, 2003, at 09:51 am, Stefan Arentz wrote:
On Thursday, August 7, 2003, at 19:30, James Strachan wrote:
Personally I don't care much (yet) about a service for users of the servers but I think it should be given some serious thoughts for container development.
Agreed. Though lets get the J2EE server ready first then lets figure out how to weave J2EE aspects nicely into POJOs later on - or in a separate parallel project.
Incidentally if there was gonna be some J2EE aspects, they should
hopefully be mostly J2EE standard apsects and so work with any J2EE
container right? So maybe its time for a separate AspectJ2ee project?
I'm sure Geronimo could eventually supply some custom Geronimo-specific
aspects later on down the road.
deliver two advantages.
a. For Developers - Offer a better mechanism of expressing cross cutting
design constructs and thus the code
FWIW (whenever you get to see it) the initial Geronimo code already has an interceptor stack - so its quite ready for use in an AOP framework.
b. For Developers and End users - Offer a better mechanism of expressing
geronimo features which can be perceived to be cross cutting for the users.
The most obvious ones here are transaction / security / logging /
persistence
etc.
An aspect based design would support a scenario where geronimo ships with
j2ee aspects, and would allow "unsatisfied" users to tweak the aspects or
write their own.
I agree in principle, Geronimo is already gonna have several modules (core container, web container connector, WS connector etc). So why don't we start an experimental AOP module? Then the folks developing the AOP module can focus on AOP stuff and the Geronimo server developers can focus on getting the J2EE server working. If some of the AOP work ends up getting used inside the core container - cool. Otherwise they can be separate, parallel modules that work together.
I believe it would be important to have a considered
strategy on whether such a feature would be required and if so how and when
should it be brought in.
AOP is certainly not required for J2EE compliance; though it is a nice feature for folks wanting to use POJOs and so forth. So I'd recommend a separate project somwhere or a parallel separate module in Geronimo.
The implementation could be phased, but if the
approach isnt in place, then it may never take off. Hence I dont think one
can "defer" this issue.
I disagree. If you look at AspectWerkz and Nanning, already they've gone a long way to adding many J2EE style aspects to any POJOs quite easily with very little development. I don't see why a group of folks can't get together, do a groovy J2EE aspects project - try and work with most J2EE servers and by all means write some Geronimo-specific extras - but I don't think we need to interfere with the general Geronimo server development to do this.
You do bring out an interesting option of a separate AspectJ2EE project, but
I think it would be useful to consider how such a project could cooperate
with geronimo for example.
We could start off as a module of Geronimo. Or it could start elsewhere too. (I'd like to get the AspectWerkz & Nanning guys involved too if we can).
I suspect making aspects into a separate project
and hoping that they would cooperate nicely with geronimo would be a lot
harder than it seems.
I disagree - many folks on this list work on various separate projects which work together. Whether the AOP project is a module in Geronimo or a separate project elsewhere I'm sure we'd work closely together to make it work. I'm sure plenty of folks on both sides of the fence can understand each others perspectives and try to help each other. (I myself might swap hats occasionally, sitting in both camps since I think both projects are very useful). However I don't want AOP considerations to slow down Geronimo getting a fully working & tested J2EE stack.
I ask myself - how would I build a new container from scratch today, and I
am unable to really have a satisfactory answer which does not involve
aspects.
Aspects and an interceptor stack are quite similar in many ways - from a containers perspective. I don't see any reason why the Geronimo interceptor stack can't be easily integrated into any J2EE AOP project.
James ------- http://radio.weblogs.com/0112098/
