On Mon, 28 Apr 2025 at 10:42, Álvaro Herrera <alvherre@kurilemu.de> wrote:
> On 2025-Apr-24, Japin Li wrote:
>
>> On Thu, 17 Apr 2025 at 17:18, Álvaro Herrera <alvherre@kurilemu.de> wrote:
>> > On 2025-Apr-17, Japin Li wrote:
>> >
>> >> It seems PG 16 does not support exclusion constraints on
>> >> partitioned tables.
>> >
>> > Yeah, my recollection is that they were purposefully disallowed
>> > (mainly because I didn't want to research how to fully make them
>> > work when adding local partitioned indexes), and that we needed to
>> > do more work if we wanted to let them through. I suspect commit
>> > 8c852ba9a4 was mistaken to allow that case without looking for
>> > further implications.
>>
>> Sorry, I’m unclear on “more work.” Can you explain further?
>
> Well, there are no tests in the patch. 8c852ba9a434 added some, but
> it's now clear that something was overlooked. I think this patch should
> make more of an effort to cover all interesting cases in regression
> tests if there are holes in coverage; and also add something to verify
> that pg_dump and pg_upgrade work correctly for these constraints.
>
My understanding, based on the src/bin/pg_dump tests, is that they don't
involve a genuine restore of the dumped data to a database. Instead, it
dumps to a file using pg_restore. Is that correct?
I doubt whether I can add a test to pg_dump that would cover this issue.
--
Regrads,
Japin Li
