Re: posgres 12 bug (partitioned table)

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> On Tue, Aug 11, 2020 at 12:02 PM Pavel Biryukov <79166341370(at)yandex(dot)ru> wrote:
>> I just want to point that Npgsql provider for .Net Core builds queries like that (RETURNING xmin) to keep track for concurrency.
>> This bug stops us from moving to partitioned tables in Postgres 12 with Npgsql.

> That's certainly a good reason to try to make it work. And we can make
> it work, if we're willing to assume that everything's a heap table.
> But at some point, that hopefully won't be true any more, and then
> this whole idea becomes pretty dubious. I think we shouldn't wait
> until it happens to start thinking about that problem.

For xmin in particular, you don't have to assume "everything's a heap".
What you have to assume is "everything uses MVCC", which seems a more
defensible position. It'll still fall down for foreign tables that are
partitions, though.

I echo Andres' nearby question about exactly why npgsql has such a
hard dependency on xmin. Maybe what we need is to try to abstract
that a little, and see if we could require all partition members
to support some unified concept of it.

regards, tom lane