On Mon, 2007-01-15 at 15:05 -0500, Tom Lane wrote:
> maybe we should just do the constant for starters and see how many
> people really want to write C-code estimators ...
+1
BTW, your proposal would still pushdown all qualifiers, right?
Hellerstein's xfunc work discusses situations in which it makes sense to
pullup expensive qualifiers above joins, for example, in order to reduce
the number of tuples the qualifier is applied to. Unfortunately, this
would probably increase the optimizer's search space by a fairly
significant degree, so it might need to be controlled by a GUC variable,
or only applied when the estimated cost of applying a qualifier is
particularly large relative to the total estimated cost of the plan.
-Neil