Fixes #447 (closed).
The problem was arising when we were asking for a dump (of any kind) at exactly a time where we would have had a snapshot.
That's rare hence went unnoticed. It was triggering a spurious drift_all.
Could you check whether we now pass all your tests successfully?