The docs of createTuple() and createAssociation() recommend that implementations should not go the datastore but create a transient representation. Indeed that's we do in our dialects with the exception of InfinispanDialect which interacts with ISPN to obtain the required FineGrainedAtomicMap s. I guess that's fine as long as only in-memory is supported. For remote servers it should be done differently.

Either way we can remove that part from the message, as we don't need to repeat the nature of the contract there.