Re: Rethinking representation of partial-aggregate steps

On 24 June 2016 at 05:55, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> #define AGGOP_COMBINESTATES   0x1
> #define AGGOP_SERIALIZESTATES  0x2
> #define AGGOP_DESERIALIZESTATES  0x4
> #define AGGOP_FINALIZEAGGS 0x8
>
> typedef enum AggPartialMode
> {
>   AGGPARTIAL_SIMPLE = AGGOP_FINALIZEAGGS,
>   AGGPARTIAL_PARTIAL = AGGOP_SERIALIZESTATES,
>   AGGPARTIAL_FINAL = AGGOP_COMBINESTATES | AGGOP_DESERIALIZESTATES | AGGOP_FINALIZEAGGS
> } AggPartialMode;
>
> #define DO_AGGPARTIAL_COMBINE(apm)  (((apm) & AGGOP_COMBINESTATES) != 0)
> #define DO_AGGPARTIAL_SERIALIZE(apm)  (((apm) & AGGOP_SERIALIZESTATES) != 0)
> #define DO_AGGPARTIAL_DESERIALIZE(apm)  (((apm) & AGGOP_DESERIALIZESTATES) != 0)
> #define DO_AGGPARTIAL_FINALIZE(apm)  (((apm) & AGGOP_FINALIZEAGGS) != 0)

The attached implements this, with the exception that I didn't really
think AggPartialMode was the best name for the enum. I've named this
AggregateMode instead, as the aggregate is only partial in some cases.
I also appended _SERIAL and _DESERIAL to the end of the partial and
final modes for now, as likely in 10.0 we'll be performing some
partial non-serialized aggregation and we'll likely want to keep those
other names for that instead.

--
 David Rowley                   http://www.2ndQuadrant.com/
 PostgreSQL Development, 24x7 Support, Training & Services