Zeugswetter Andreas IZ5 wrote:
>
> I think the best compromise would be to have ctid in the where clause.
And we told about this a days ago, but no one did it.
> This would need to always be considered as best path by the optimizer.
> Then the situation is similar to a primary key foreign key scenario.
>
> The referenced table does not need an index, since we know the physical
> position of the row we want (where ctid='(5,255)').
>
> What we need second is an update trigger for the referenced table that
> updates old.ctid to new.ctid in the referencing table. For this to be
Triggers are not fired by vacuum...
Vadim
