Josh Berkus <josh(at)agliodbs(dot)com> writes:
> Just got this simple case off IRC today:
> [ hashed versus non-hashed subplan ]
> I'm at a bit of a loss as to what's happening here.
Possibly work_mem is smaller in the second installation?
(If I'm counting on my fingers right, you'd need a setting of at least a
couple MB to let it choose a hashed subplan for this case.)
regards, tom lane