On Feb24, 2014, at 17:50 , Dean Rasheed <dean.a.rasheed@gmail.com> wrote:
> On 20 February 2014 01:48, Florian Pflug <fgp@phlo.org> wrote:
>> On Jan29, 2014, at 13:45 , Florian Pflug <fgp@phlo.org> wrote:
>>> In fact, I'm
>>> currently leaning towards just forbidding non-strict forward transition
>>> function with strict inverses, and adding non-NULL counters to the
>>> aggregates that then require them. It's really only the SUM() aggregates
>>> that are affected by this, I think.
>>
>> I finally got around to doing that, and the results aren't too bad. The
>> attached patches required that the strictness settings of the forward and
>> reverse transition functions agree, and employ exactly the same NULL-skipping
>> logic we always had.
>>
>> The only aggregates seriously affected by that change were SUM(int2) and
>> SUM(int4).
>
> I haven't looked at this in any detail yet, but that seems much neater
> to me. It seems perfectly sensible that the forward and inverse
> transition functions should have the same strictness settings, and
> enforcing that keeps the logic simple, as well as hopefully making it
> easier to document.
Good to hear that you agree! I'll try to find some time to update the docs.
best regards,
Florian Pflug