[news.eclipse.modeling.m2m] Re: [QVTO] Restrictions on EMF resource type

[Sergey Boyko <sboyko@xxxxxxxxxxx>]

Hi Alex ,

Some replies are in-lined, below

Regards,
  Sergey

Alex wrote:
> In general QVTO runtime likely does not care about what kinds of EMF 
> resources are used for models, as long as it can access objects in the 
> EMF resource. However, currently the QVTO launch validation appears to 
> restrict these resources to several types - those that are 
> file/platform/archive based, or those that can be opened via 
> URI.openConnection().
>
> While this likely covers the most common types, it appears to miss out 
> on the flexibility and extendability of the EMF resource framework. For 
> example, custom protocol types can be defined for custom EMF resource 
> implementations - and QVT unfortunately won't accept these as valid 
> models, even though one could easily open them in EMF model editors or 
> manipulate them via EMF APIs.
>
> We've hit this with our own custom implementations of EMF resources, but 
> I am thinking something like CDO should be a good example. If I remember 
> correctly they use cdo:// in their resource URIs, which likely won't be 
> accepted by QVTO.

I got your point.
Actually we have non-EMF models that are wrapped with custom Resource 
implementation and specific protocol defined like follows:
  <extension point="org.eclipse.emf.ecore.protocol_parser">
     <parser
           class="com.xxx.CustomResourceFactory"
           protocolName="xxx"/>
  </extension>

But our case is that these models are input-only. And qvto launch 
configuration allows to select them for such kind.

Agree that existing excessive checking for [inout] and [out] parameter's 
types should be revised as well.


> There may be ways to work around this issue - e.g. using EFS to surface 
> these as platform: resources, implementing custom Java protocol handlers 
> that would enable URL.openConnection(), using proxy files containing 
> pointers to real EMF resources etc, - but directly supporting anything 
> that EMF supports seems to be the best course of action.
>
> Somewhat related: it has been mentioned in this forum that there are 
> plans to support standalone QVTO execution - BTW, is there a bugzilla 
> for this? - and this would likely require some changes to how model 
> resources are handled, but I am not sure if the general issue of EMF 
> resource types is expected to be handled by that activity ...

Just created: https://bugs.eclipse.org/bugs/show_bug.cgi?id=287685

> Alex