On Wed, Feb 14, 2018 at 2:50 PM, Etsuro Fujita <fujita.ets...@lab.ntt.co.jp> wrote: > > I'd vote for #1, but ISTM that that's more like a feature, not a fix. > Pushing down ConvertRowtypeExprs to the remote seems to me to be the same > problem as pushing down PHVs to the remote, which is definitely a feature > development. I think a fix for this would be to just give up pushing down a > child join in foreign_join_ok or somewhere else if the reltarget of the > child-join relation contains any ConvertRowtypeExprs.
That means we won't be able to push down any scans under a DML on a partitioned table. That will be too restrictive. I think the case of PHV and ConvertRowtypeExprs are different. The the former we need a subquery to push PHVs, whereas to deparse ConvertRowtypeExpr on the nullable side we can use the same strategy as whole-row expression deparsing. -- Best Wishes, Ashutosh Bapat EnterpriseDB Corporation The Postgres Database Company