On Fri, Nov 21, 2025 at 01:52:07PM -0500, Tom Lane wrote: > But the conclusion that "where t1.a = t2.a and t1.b = 0" means that > t1's rowcount is 1 only applies if the join is implemented as an inner > indexscan. If we choose some other method, say a hash join based on > a seqscan of t1, having forced that rowcount to 1 would produce > completely false estimates.
But wouldn't knowing that for an inner indexscan the cardinality is one then drive the optimizer to choose inner indexscan over other methods? Nico --
