What was the issue? Maybe I'm reading the diff wrong, but the membership check should be implicit in the delete statement, so this change shouldn't function differently? Or if it actually does, then there's probably still a bug in the core code.
I think the core should be able to do that check generally: send out
updates only if the element is indeed part of the set. As the
assumption is that all nodes have the same version of the set, that
shouldn't break anything (though it could because in practice that
assumption does not always hold ...)