"Florian G. Pflug" <fgp@phlo.org> writes:
> Image a complex, autogenerated query with looks something like this
> select ....
> from t1
> join t2 on ...
> join t3 on ...
> join t4 on ...
> ...
> ...
> where
> <big, complicated expression derived from some user input>.
> This big, complicated expression looks different for every query - and
> currently, postgres often vastly overestimates the selectivity of this
> expression.
This is a straw man. There is no way that your application can throw in
a chosen-at-random selectivity value for a join condition that it
doesn't understand and have that be more likely to be right than the
planner's guess.
regards, tom lane