Re: wip: functions median and percentile

Hello

2010/9/22 Hitoshi Harada <umi.tanuki@gmail.com>:
> 2010/9/22 Pavel Stehule <pavel.stehule@gmail.com>:
>> Hello
>>
>> I found probably hard problem in cooperation with window functions :(
>>
>> tuplesort_begin_datum creates child context inside aggcontext. This
>> context is used for tuplestore. But when this function is called from
>> WindowAgg_Aggregates context - someone drops all child context every
>> iteration, and then tuplestore state is invalid. For this moment we
>> can block using a median function as window function, but it should be
>> solved better - if it is possible - I don't see inside window function
>> implementation.
>
> Does it happen when the window frame starts from not UNBOUNDED
> PRECEDING? In those cases, nodeWindowAgg tries to discard all
> aggregate contexts and to initialize the aggregate state. AFAIK the
> memory context is brand-new although it was reset and its children
> deleted, so if the function is well-formed and initializes its state
> on NULL input, it doesn't cause a problem.

maybe I was confused. I found a other possible problems.

The problem with median function is probably inside a final function
implementation. Actually we request possibility of repetitive call of
final function. But final function call tuplesort_end function and
tuplesort_performsort. These function changes a state of tuplesort.
The most basic question is "who has to call tuplesort_end function and
when?

Regards

Pavel Stehule

>
>
> Regards,
>
>
> --
> Hitoshi Harada
>