Re: Add starelid, attnum to pg_stats and leverage this in pg_dump

On Tue, Mar 17, 2026 at 06:49:17AM +0900, Michael Paquier wrote:
> On Mon, Mar 16, 2026 at 03:15:14PM -0500, Nathan Bossart wrote:
>> On Mon, Mar 16, 2026 at 04:04:23PM -0400, Corey Huinker wrote:
>>> expr_attnum was something that Michael Paquier had lamented that the view
>>> didn't have. There is obviously no present need for it, as pg_dump isn't
>>> being modified for extended stats at all.
>>
>> Okay. I think I'll continue to leave this one out for now.
>
> Lamenting is the right term. The expressions stored in an extended
> stats object have attnumbers computed by the backend, starting from -1
> and decremented, and we don't expose this information at all in any
> system view. It could have helped in enforcing a stronger ordering of
> the items dumps for the extstats restore functions. I still think
> that it could provide an extra layer of safety. Now, we don't
> critically require it either based on how we pass down the input
> arrays, how we dump the data from the catalogs, and how we treat the
> order of the items given as function args.

FTR I'm not mortally opposed to the idea. I just want to get the easier
stuff out of the way first so we can commit the pg_dump change. Then we
can give expr_attnum our undivided attention.

--
nathan