Hi,
On 2020-08-11 13:52:00 -0400, Tom Lane wrote:
> 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.
Don't think that necessarily implies having a compatible xmin / xmax
around. Certainly not a 32bit one. I guess an AM could always return
InvalidOid, but that doesn't seem particularly helpful either.
I think it'd be better if we actually tried to provide a way to do
whatever xmin is being used properly. I've seen many uses of xmin/xmax
and many of them didn't work at all, and most of remaining ones only
worked in common cases.
Greetings,
Andres Freund