On Mon, 2020-09-07 at 11:42 -0300, Alvaro Herrera wrote:
> > This patch would provide a more convenient way to do that.
> > Again, I am not sure if that justifies the effort.
>
> I have to admit I've seen cases where it'd be useful to have included
> columns in primary keys.
>
> TBH I think if we really wanted the feature of primary keys with
> included columns, we'd have to add it to the PRIMARY KEY syntax rather
> than having an ad-hoc ALTER TABLE ALTER CONSTRAINT USING INDEX command
> to replace the index underneath. Then things like pg_dump would work
> normally.
>
> (I have an answer for the information_schema question Tom posed; I'd
> like to know what's yours.)
Gah, now I see my mistake. I was under the impression that a
primary key can have an INCLUDE clause today, which is not true.
So this would introduce that feature in a weird way.
I agree that that is undesirable.
We should at least have
ALTER TABLE ... ADD PRIMARY KEY (id) INCLUDE (val);
or something before we consider this patch.
As to the information_schema, that could pretend that the INCLUDE
columns just don't exist.
Yours,
Laurenz Albe
