On Wed, May 28, 2014 at 08:39:32PM +1200, David Rowley wrote:
> The attached patch allows join removals for both sub queries with left
> joins and also semi joins where a foreign key can prove the existence of
> the record.
When a snapshot can see modifications that queued referential integrity
triggers for some FK constraint, that constraint is not guaranteed to hold
within the snapshot until those triggers have fired. For example, a query
running within a VOLATILE function f() in a statement like "UPDATE t SET c =
f(c)" may read data that contradicts FK constraints involving table "t".
Deferred UNIQUE constraints, which we also do not yet use for deductions in
the planner, have the same problem; see commit 0f39d50. This project will
need a design accounting for that hazard.
As a point of procedure, I recommend separating the semijoin support into its
own patch. Your patch is already not small; delaying non-essential parts will
make the essential parts more accessible to reviewers.
Thanks,
nm
--
Noah Misch
EnterpriseDB http://www.enterprisedb.com