> So another option might be to instead cache the service contracts we have loaded and use those as a check. This would work

Yeah, that was my alternative proposal but the issue with this is that we keep the services around. The direction we want to take nowadays is being able to get rid of useless services as soon as they are not used anymore.

> But given what real tests might show, another option might be to instead cache the service contracts we have loaded and use those as a check.

Interesting proposal. I like the approach and I think we should throw an exception, otherwise we will totally miss it if we do something wrong in ORM proper. Just a bit worried that we could potentially break applications doing weird things.

I just tried this approach and our core test suite mostly passes, except for the artificial test mentioned above and a bunch of tests (12), using MetadataSources.buildMetadata() which loads the MetadataSourcesContributor services. Not sure they will all be easy to fix.

James Perkins what is your timeline for this fix? You want it for the future versions of WF or you want it backported to 5.3 (which is the current version in WF/EAP)?