Tom Lane wrote:
> Heikki Linnakangas <heikki.linnakangas@enterprisedb.com> writes:
>> But we don't use tuplesort to return tuples from functions, so it's not
>> broken in a user-visible way. Unless you can think of another scenario
>> like that.
>
> (1) create a cursor whose plan involves a sort that will spill to disk
> (2) enter subtransaction
> (3) fetch from cursor (causing the sort to actually happen)
> (4) leave subtransaction
> (5) fetch some more from cursor
>
> Too busy to develop a test case right now, but ISTM it ought to fail.
That was exactly the case that we originally fixed, that caused this
PL/pgSQL issue. It works because cursors run within the portal
ResourceOwner.
-- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com