Comments inline.

Mike Edwards wrote:
Jean-Sebastien,

I've replied separately to your referenced note.

My suggestion is that resolution should be split from loading. Resolution should take place in a cascade from the composites which are at domain level, in response to explicit references discovered there. Only entities in the actually deployed hierarchy will get resolved. Mere presence of some entity in a contribution should be of no significance for resolution. The problem today is that discovery and resolution are tied together in the code.

This needs to look ahead to the point where specific entities in the hierarchy of composition are not destined for the runtime node that is performing the resolution.

Exactly! Resolution is already split from loading by design in the ArtifactProcessor interface, but as you noted, the contribution service ties the two together by calling the resolve methods as part of the contribute(...) processing.

The current implementation is a step stone and one of the first next steps is to move resolve() out of contribute(). It's easy to do, I have that change prototyped and should be able to commit it in the next few days with the other changes covering what I described in [1].

This will address the issues you've outlined. This change is required as well for an even more obvious reason, to support multiple contributions with SCA imports/exports, as when a contribution is read you don't necessarily have the other contributions that it depends on.

Yours,  Mike.

Jean-Sebastien Delfino wrote:
[snip]
Mike Edwards wrote:

Whether resolution occurs for all the resources simply depends on whether there are resolvers for each type of resource.

+1 for improving this, did you see what I proposed in [1], and does it make sense to you? Do you have any suggestions on how to implement this lazy loading + resolution?

[1] http://www.mail-archive.com/tuscany-dev@ws.apache.org/msg18684.html

Thanks

--
Jean-Sebastien


---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to