Re: An out-of-date comment in nodeIndexonlyscan.c

On Mon, 14 Jun 2021 at 10:03, Thomas Munro <thomas(dot)munro(at)gmail(dot)com> wrote:
>
> On Mon, Jun 14, 2021 at 12:54 AM David Rowley <dgrowleyml(at)gmail(dot)com> wrote:
> > I think a more optimal and nicer way of doing that would be setting
> > bits in a Bitmapset then checking bms_num_members is equal to
> > n_scan_keys.
>
> Shouldn't it be compared with indnkeyatts? Yes, much nicer, thanks!

Oh yeah, I did mean that. Thanks for the correction.

Have you also thought about deferrable unique / primary key constraints?

It's possible to the uniqueness temporarily violated during a
transaction when the unique constraint is deferred,

For example:
create table t (id int primary key deferrable initially deferred);
begin;
insert into t values(1),(1);
select * from t;
id
----
1
1
(2 rows)

I think you'd just need to add a check to ensure that indimmediate is
true around where you're checking the indisunique flag.

David