Re: [architecture] Geronimo and OSGi

["Richard S. Hall" <heavy@xxxxxxxxxxxxxx>]

Jean-Bernard Stefani wrote:

> Well, my reading of the "Fractal deployment effort" you mention was 
> not in any way a means to prevent having
> a way to package Fractal structures in OSGI bundles, but an effort to 
> obtain a programming-environment independent specification of loading 
> and local deployment, much as the Fractal specification is programming
> language independent. For instance, we have developments taking place 
> in Think, which can be viewed as a C implementation of Fractal. In 
> that context, having an OSGI-based loader is not really useful, but 
> still the loading and deployment issue exists. When working in the 
> Java environment, I agree with you that a path of least energy (in 
> particular in terms of mind share) is the one you suggest: using OSGI 
> as a super class loader for Fractal deployment. But that does not 
> apply to other environments.


I agree in a general sense, that OSGi would not be a solution for a C 
version of Fractal, but since we were dealing with the Julia 
implementation of Fractal, it seems like it would have been a good fit. 
The interoperability of different implementations of Fractal is a 
complicated issue and one that is not really addressed anywhere.

> Concerning the view that Fractal should be 'at the bottom of things', 
> I can only say that that is precisely my belief, and therefore, apart 
> from the tactical agreement with you I mentioned above (it's probably 
> best to employ OSGI when it comes to local deployment issues because 
> of OSGI mind share with Eclipse), we are in disagreement in that 
> respect. I would definitely push to have Fractal at the bottom of 
> things if one wants to build a software infrastructure with maximum 
> flexibilty. Said otherwise, I would say that Fractal is indeed a 
> "super" plug-in architecture, and I would argue that it (Fractal) be 
> used as the basis of componentization in ObjectWeb middleware. As a 
> plug-in architecture it would be in direct competition with OSGI, and 
> there lies the perceived conflict.


I think that is still makes sense to view Fractal as a "super plug-in 
architecture" and OSGi as the packaging/deployment mechanism for those 
super plug-ins. The notion that Fractal should somehow lie beneath class 
loading is the only aspect I disagree with from your comments.

> 1/ the disagreement above is a legitimate one: the implication that 
> there would be a kind of rampant 'Fractal imperialism' which would 
> prevent sensible choices, as certain parts of the discussion seem to 
> imply, is simply something I cannot bear, for nothing is further from 
> the truth. Besides, said 'sensible choices' may not be  ones that I, 
> as a Fractal proponent, would deem that sensible architecturally.


I agree totally and I would hope that my comments did not add to that 
misunderstanding. The issue is just a legitimate difference of views 
over how/where Fractal belongs, which at worst has led to an impasse, 
but that is all.

> 2/ If we had failed opportunities, it was more opportunities to debate 
> wide ranging architectural choices
> within OW, and in particular for the evolution of OW's application 
> server architecture. If the debate had taken place in full, we would 
> have considered eg the implications of having 'Fractal at the bottom' 
> vs other, possibly standards-based, approaches. 


Unfortunately, the debates were always piecemeal and I don't know the 
reason for that.

-> richard