"Merlin Moncure" <mmoncure@xxxxxxxxx> writes: > select f(x) from t where id = 1 order by n; > can cause f to execute for the entire table even if id is unique. Really? I'd consider it a bug if so. Compare select 1/x from t where x > 0 If the presence of zeroes in t can make this throw a zero-divide error, the database is broken. In my mind the SQL spec is perfectly clear that WHERE filtering occurs before evaluation of the SELECT targetlist. (Sorting, however, occurs afterward --- so there are certainly potential gotchas of this ilk. But the specific example you give is bogus.) regards, tom lane
