Re: Index only scan for cube and seg

On Fri, Oct 27, 2017 at 9:54 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> > On Thu, Oct 26, 2017 at 12:22 PM, Andrey Borodin <x4mmm(at)yandex-team(dot)ru>
> wrote:
> >> For cube there is new default opclass.
>
> > I seem to recall that changing the default opclass causes unsolvable
> > problems with upgrades. You might want to check the archives for
> > previous discussions of this issue; unfortunately, I don't recall the
> > details off-hand.
>
> Quite aside from that, replacing the opclass with a new one creates
> user-visible headaches that I don't think are justified, i.e. having to
> reconstruct indexes in order to get the benefit.
>
> Maybe I'm missing something, but ISTM you could just drop the compress
> function and call it good. This would mean that an IOS scan would
> sometimes hand back a toast-compressed value, but what's the problem
> with that?
>

+1,
I think in this case replacing default opclass or even duplicating opclass
isn't justified.

(The only reason for making a decompress function that just detoasts
> is that your other support functions then do not have to consider
> the possibility that they're handed a toast-compressed value. I have
> not checked whether that really matters for cube.)
>

As I can see, cube GiST code always uses DatumGetNDBOX() macro to transform
Datum to (NDBOX *).

#define DatumGetNDBOX(x) ((NDBOX *) PG_DETOAST_DATUM(x))

Thus, it should be safe to just remove both compress/decompress methods
from existing opclass.

------
Alexander Korotkov
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company