Re: [m2m-atl-dev] ATL 3.0 architecture & ASMEMFModel Memory leak comes b

[Dennis Wagelaar <dennis.wagelaar@xxxxxxxxx>]

* William Piers <william.piers@xxxxxxx> wrote on 16.12.2008 15:10:11:
> Hello,
> 
> Thanks a lot for your work!
> I just had a remark about the .ecore files into the emf4atl driver. 
> There are ever several "resources directories" in ATL, so we finally 
> need to choose one. Actually we have:
> 1) org.eclipse.m2m.atl
> 2) org.eclipse.m2m.atl.dsls
> 3) org.eclipse.m2m.atl.drivers.emf4atl
> 
> 2) and 3) are problematic because as these plugins contains TCS parsing 
> tools, they depends on the RegularVM.  So org.eclipse.m2m.atl.core.emf 
> (for example) can't refer the embedded ecore files without depending 
> RegularVM...
> 1) depends on nothing, so it's easy to for any plugin to refer it (e.g. 
> emf4atl, org.eclipse.m2m.atl.core.emf).
> This plugin could be improved like org.eclipse.m2m.atl.dsls to provide 
> utilities for retrieving resources.
> 

Anything is fine for me, as long as dependencies remain sane. That means that option (1) is indeed the best, as it does not introduce any additional plugin dependencies.

--
Dennis

> Best regards,
> William
> 
> Dennis Wagelaar a écrit :
> > The ATL_Dynamic_ResourceSet branch is now merged.
> >
> > I've also attached a snapshot of my ATL ant tasks fork based on the 
> > dynamic ModelLoader architecture. It may come in handy when you plan 
> > to port the ant tasks to the new architecture.
> >
> > -- 
> > Dennis
> >
> > William Piers schreef:
> >> I commited the patch, so you can work on it.
> >>
> >> William
> >>
> >> Dennis Wagelaar a écrit :
> >>> Hello William, ATL devs,
> >>>
> >>> First of all, thanks for the quick answer! I put my reply at the 
> >>> bottom:
> >>>
> >>> On 12 Dec 2008, at 15:28, William Piers wrote:
> >>>> Hello Dennis,
> >>>>
> >>>> Answers below:
> >>>> Dennis Wagelaar a écrit :
> >>>>> *snip*
> >>>> I encountered this problem too... so I add an option to clear the 
> >>>> resourceSet after each transformation run. It's definitely not the 
> >>>> better solution...
> >>>>>
> >>>>> What are your opinions on this? William: what do you think of an 
> >>>>> architecture change that involves something like a ModelLoader?
> >>>> I agree with your demonstration. A solution could be to make the 
> >>>> EMFModelFactory "dynamic":
> >>>> * in CoreService, no more registry for it, but a 
> >>>> "createModelFactory(String modelFactoryName)" method.
> >>>> * in the EMFModelFactory, a non-static ResourceSet.
> >>>> So the EMFModelFactory is transparently linked with its ResourceSet 
> >>>> instance, and is recreated for each launch.
> >>>> Note that a programmatic launch (or another launching tool, like 
> >>>> ant) can provide the same behaviour as before is necessary, by 
> >>>> keeping the same factory into chained transformation.
> >>>>
> >>>> Would you agree with that? I tested an implementation and attached 
> >>>> a patch to this mail if you want to see details.
> >>>>
> >>>> Thank you for your feedback!
> >>>>
> >>>> William
> >>>
> >>> I agree with this idea, thanks! As far as I'm concerned, you may 
> >>> commit this change to CVS, after which I can work in the extra 
> >>> changes w.r.t. the finalize() methods and wrapping the updated 
> >>> ASM-based plugins. It's important to get this one right as soon as 
> >>> possible, regarding the global impact that these changes have ;-).
> >>>
> >>> Cheers,
> >>> Dennis
> > ------------------------------------------------------------------------
> >
> > _______________________________________________
> > m2m-atl-dev mailing list
> > m2m-atl-dev@xxxxxxxxxxx
> > https://dev.eclipse.org/mailman/listinfo/m2m-atl-dev
>