Re: default partition and concurrent attach partition

On 2020-Sep-07, Alvaro Herrera wrote:

> Ah, it looks like we can get away with initializing the RRI to 0, and
> then explicitly handle that case in ExecPartitionCheckEmitError, as in
> the attached (which means reindenting, but I left it alone to make it
> easy to read).

Well, that was silly -- this seems fixable by mapping the tuple columns
prior to ExecPartitionCheck, which is achievable with something like

if (partidx == partdesc->boundinfo->default_index)
{
TupleTableSlot *tempslot;

if (dispatch->tupmap != NULL)
{
tempslot = MakeSingleTupleTableSlot(RelationGetDescr(rri->ri_RelationDesc),
&TTSOpsHeapTuple);
tempslot = execute_attr_map_slot(dispatch->tupmap, slot, tempslot);
}
else
tempslot = slot;
ExecPartitionCheck(rri, tempslot, estate, true);
if (dispatch->tupmap != NULL)
ExecDropSingleTupleTableSlot(tempslot);
}

though this exact incantation, apart from being pretty horrible, doesn't
actually work (other than to fix this specific test case -- it crashes
elsewhere.)

--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services