>>>>> "Ekta" == Ekta Khanna <ekhanna@pivotal.io> writes:
Ekta> Hello PGSQL Hackers,
Ekta> We have come across the following issue on Postgres
Ekta> REL_10_STABLE. Below is the repro:
[...]
Ekta> In the plan, we see that planner merges the quals from FROM
Ekta> clause and the WHERE clause in the same RESTRICTINFO. Is this the
Ekta> expected behavior?
Yes, it's entirely expected. You CANNOT make assumptions about the order
of evaluation of quals; the planner will rearrange them freely, even
across subquery boundaries (where the semantics allow).
You can do this:
WHERE CASE WHEN length(b) = 8
THEN to_date(b, 'YYYYMMDD') > '2018-05-04'
ELSE false END
since one of the few guarantees about execution order is that a CASE
will evaluate its condition tests before any non-constant subexpressions
in the corresponding THEN clause.
(Another method is to put an OFFSET 0 in the subquery, but that's more
of a hack)
--
Andrew (irc:RhodiumToad)