Re: Table AM modifications to accept column projection lists

On Thu, Jul 1, 2021 at 7:42 AM Jacob Champion <pchampion(at)vmware(dot)com> wrote:
>
> Hi all,
>
> Ashwin, Deep, and I were discussing this patch today. We agree that
> it's fairly difficult to review in its current state, and the lack of a
> concrete implementation of the new API isn't helping. (A big chunk of
> the context for the patch exists in the zedstore megathread, which
> isn't exactly light reading.)
>
> We'd like to improve that, but with current time constraints, we won't
> be able to get to it for the July commitfest. So I'll mark this patch
> Withdrawn for now, to reduce the review load. (239 Needs Review and
> counting!) We hope to revisit in the September timeframe.
>
> Thanks for all the reviews and input!
>
> --Jacob

Since this thread is mentioned again, I want to share some thoughts on the lazy
material part[1]. If we go with that direction, we may not need the AM for pass
projection to heapam. we just need something like AM like

Datum fetch_column_value_for_column(.., rowid, colid);

I worked in this direction with a small step and stopped quickly because of some
other reasons. I would like to share my work so far here [2], However
lazy material
is not always good.

[1] https://stratos.seas.harvard.edu/files/stratos/files/columnstoresfntdbs.pdf
[2] https://github.com/zhihuiFan/postgres/tree/lazy_material_v2

--
Best Regards
Andy Fan (https://www.aliyun.com/)