Commented on the Jira. It is a regression, though it is not caused specifically by the graph changes. It is caused more by a change in how subsequent selects are fired. If it was an intentional change, it would be good to understand why it was changed so we can make an informed decision about how to best resolve this - whether the subsequent-select changes ought to reverted or if we should adjust the graph handling to account for that change. My concern is that the subsequent-select changes might lead to other such hard-to-diagnose problems. FWIW, changing graph handling wrt `#find` *should be* trivial. I'll work on a PR using this approach to verify it is indeed trivial (after I'm actually back from vacation). P.S. Note that this only affects `#find` operations. Graphs applied to HQL or criteria queries do not have this issue. So that is a viable workaround. On Thu, Dec 27, 2018 at 6:49 AM Guillaume Smet <guillaume.smet(a)gmail.com&gt; wrote: