Re: pg_restore handles extended statistics inconsistently with statistics data

> On Jun 16, 2026, at 14:59, Michael Paquier <michael(at)paquier(dot)xyz> wrote:
>
> On Tue, Jun 16, 2026 at 02:36:21PM +0800, Chao Li wrote:
>> I think the dependency-based matching is better, but what happens if
>> the archive is created with --statistics-only? In that case, the
>> statistics entries still have dependencies, but the referenced
>> parent TOC entries are not present in the archive. See the attached
>> test script. It creates the archive with --statistics-only, with the
>> new patch, neither table nor index stats are restored:
>
> I can see your argument in that HEAD can improve the detection in what
> gets restored if the dumps only include the stats data, but I don't
> really buy that this is a use case interesting to support: one can
> also dump the data with the index definition and then filter back the
> contents when restoring using --index. So at the end having a tighter
> check at the restore is more appealing to me, because it leads to a
> more predictible result, like in the scenario I have posted upthread
> when a schema has the idea to include tables and indexes with the same
> names (unlikely, okay, still). Or in short, don't do that.
>
> For the case of upgrades one or the other does not really matter,
> pg_upgrade treats all the objects in full.
> --
> Michael

Make sense. Then the new patch looks good to me.

Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/