2009/11/9 Andrew Gierth <andrew@tao11.riddles.org.uk>:
>>>>>> "Heikki" == Heikki Linnakangas <heikki.linnakangas@enterprisedb.com> writes:
>
> Heikki> Yeah, we need the reinitialization support to handle the generic case.
>
> Heikki> One idea is to take a copy of the state datum after each row.
>
> Some existing aggregates, notably array_agg, rely on the assumption
> that this never happens.
I'm carefully coding to keep the existing behavior that when frame
start is UNBOUNDED PRECEDING aggregate state is cached. I don't see
how the copied state will help in other frame types, but is it rather
than that? We assume the frame starting and ending boundary never goes
back.
Regards,
--
Hitoshi Harada
