Arik Schimmel <arik.schimmel@wiz.io> writes:
> Is there a reason partial index predicates aren't reordered by cost?
It hasn't come up AFAIR. I'm dubious that it'd be worth the trouble,
because order_qual_clauses is really quite crude when dealing with
simple expressions. We don't have accurate costing data for most
functions/operators --- they're all just labeled with procost 1 ---
so that the "cost-based ordering" reduces to just counting the
functions. That gets the right answer in your example, but only
accidentally IMO; it has no idea that jsonb_extract_path_text()
is particularly expensive. order_qual_clauses exists mostly to
ensure that subplans get pushed to the end, and that's not relevant
in this context because we don't support those in indexes.
regards, tom lane
