Is this a prerequisite of solving Oszkar's issue?

Yes, it is.

I also doubt that this is an Xtext-specific corner case, it rather seems to be something that is allowed by EMF in general and thus EMF-IncQuery should be prepared to support it.

@istvanrath unresolved != unresolvable. These proxies are unresolvable, which is something we were not prepared for so far. In fact, there does not seem to be any way to detect when these proxies become resolvable, except when they are actually resolved. So far, Xtext seems to be nice enough to proactively resolve them.

Solution attempt failed: neither this resolve message, nor the corresponding "set" is reliable enough.

Culprit is org.eclipse.xtext.linking.lazy.LazyLinkingResource:

    try {
        source.eSetDeliver(false);
        source.eSet(crossRef, target);
    } finally {
        source.eSetDeliver(true);
    }

This is bad news.

I updated my branch but I still have an exception that the previous commit suppose to clean.
Stacktrace: https://gist.github.com/3287073

@OszkarSemerath if you recall our discussion, we have agreed that we cannot solve this issue without patching Xtext. Obviously, the previous commit is not supposed to solve this either, and it does not claim to (the commit comment even calls the attempt doomed). There is nothing surprising in your findings, move along...

I have filed a bug report with Xtext:
https://bugs.eclipse.org/bugs/show_bug.cgi?id=387119

Let's hope for the best.

Migrated to https://bugs.eclipse.org/bugs/show_bug.cgi?id=398782