On 2 November 2015 at 10:59, Marko Tiikkaja <marko@joh.to> wrote:
> On 11/2/15 9:32 AM, Dean Rasheed wrote:
>>
>> On 28 October 2015 at 16:50, Marko Tiikkaja <marko@joh.to> wrote:
>>>
>>> Here's a patch for the aggregate function outlined by Corey Huinker in
>>> CADkLM=foA_oC_Ri23F9PbfLnfwXFbC3Lt8bBzRu3=CB77G9_qw@mail.gmail.com .
>>
>>
>> +1. I've wanted something like this a few times. Of the names
>> suggested so far, I think I prefer "single_value", and yes I think it
>> should work with NULLs.
>
>
> This was actually a last-minute design change I made before submitting the
> patch. The two times I've previously written this aggregate both accepted
> NULLs and only enforced that there must not be more than one non-NULL value,
> but that's only because I was thinking about the "poor man's FILTER" case,
> which is obsolete since version 9.4. The reason I changed in this version
> is that accepting NULLs can also hide bugs, and it's (usually) easy to
> filter them out with FILTER.
>
> Did you have some specific use case in mind where accepting NULLs would be
> beneficial?
>
I'm not sure what you mean when you say accepting NULLs can hide bugs.
I think that if the input values to the aggregate were
1,1,1,NULL,1,1,1 then it should raise an error. ITSM that that is more
likely to reveal problems with your underlying data or the query. If
you want to ignore NULLs, you can always add a FILTER(WHERE val IS NOT
NULL) clause.
Regards,
Dean
