Actually there are lots of things that can be done with this sort of theorem proving. And NULL is a plenty good answer for a filter, just not for a check constraint. Amongst them INSERT through UNION ALL for symmetric views which can be handy for FDW partitioned tables.
One such implementation an be found here: https://www.google.com/patents/US6728952 (apparently expired) Cheers Serge Salesforce.com > On Dec 5, 2016, at 7:28 AM, Pavel Stehule <pavel.steh...@gmail.com> wrote: > > > > 2016-12-05 16:24 GMT+01:00 Tom Lane <t...@sss.pgh.pa.us > <mailto:t...@sss.pgh.pa.us>>: > Pavel Stehule <pavel.steh...@gmail.com <mailto:pavel.steh...@gmail.com>> > writes: > > I found some crazy queries in one customer application. These queries are > > stupid, but it was surprise for me so there are not some simple optimization > > > create table foo(a int); > > insert into foo select generate_series(1,100000); > > analyze foo; > > explain select * from foo where a <> a; > > > It does full scan of foo, although it should be replaced by false in > > planner time. > > > Same issue is a expression a = a .. can be replaced by true > > Wrong; those expressions yield NULL for NULL input. You could perhaps > optimize them slightly into some form of is-null test, but it hardly > seems worth the planner cycles to check for. > > understand > > > If you write something like "1 <> 1", it will be folded. > > it works, but a <> a not > > Regards > > Pavel > > regards, tom lane >
