Thomas Munro <thomas.munro@gmail.com> writes:
> On Tue, Jul 9, 2019 at 2:19 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Given the purposes of this test, I think it'd be reasonable to force
>> both enable_hashjoin = on and enable_mergejoin = off at the very top
>> of the join_hash script, or the corresponding place in join.sql in
>> v11. Thomas, was there a specific reason for forcing enable_mergejoin
>> = off for only some of these tests?
> Based on a suggestion from Andres (if I recall correctly), I wrapped
> each individual test in savepoint/rollback, and then set just the GUCs
> needed to get the plan shape and execution code path I wanted to
> exercise, and I guess I found that I only needed to disable merge
> joins for some of them. The idea was that the individual tests could
> be understood independently.
But per this discussion, they can only be "understood independently"
if you make some assumptions about the prevailing values of the
planner GUCs.
regards, tom lane