Gail Badner, thanks for the update!

after discussing on the hibernate-dev mailing list, we agree that this is a bug.

Good news; thanks for looking into this! Maybe once such changes are live I can convince my various project teams to finally upgrade.

It appears that the original database lock in the first transaction blocks the load from the second transaction. Did you see this behavior as well?

In other words, in your testing, did you actually find that the database lock was released after refreshing, and that a second transaction was able to update the entity?

Yes, that is the behavior I find as well. The locking semantics within the actual database transactions are correct - it is just the lock mode within the session/entity manager that is getting out of sync.