Richard Guo <guofenglinux@gmail.com> writes:
> I agree that the new test case for v15 does not seem to be worth
> including in v16+. It seems to me that it would be better if we can
> have another new test case to verify that we've included child rel's
> em_relids even for appendrel child relations with pseudoconstant
> translated variables, i.e. to verify that the change in equivclass.c
> takes effect. Maybe with a query like below:
> explain (costs off)
> select * from tenk1 t1
> left join lateral
> (select t1.unique1 as t1u, 0 as c
> union all
> select t1.unique1 as t1u, 1 as c) s on true
> where t1.unique1 = s.c;
> Without the change in equivclass.c, this query would trigger the new
> added assert in get_baserel_parampathinfo for v16, and give a wrong plan
> for v15. What do you think?
I didn't add such a test because there are already several cases
(in foreign_data.sql, IIRC) that trigger the assert, which is
how come I found the problem in the first place. Admittedly,
those depend on potentially-changeable details of an
information_schema view, so maybe it'd be better to have a
bespoke test.
regards, tom lane