"Eduardo Naschenweng" <eduardo.naschenweng@digitro.com.br> writes:
> bxs=3D# EXPLAIN ANALYZE SELECT dt_inicial, identidadea FROM cham_chamada cc=
> ;=0D
> [ is faster than ]
> bxs=3D# EXPLAIN ANALYZE SELECT * FROM teste();=0D
nodeFunctionscan.c insists on cramming the results of the function into
a tuplestore and then reading them back. This is fairly expensive for a
large result set. I complained about this back when the implementation
was first proposed, but we set the problem aside for the time being, and
it hasn't been revisited. One reason for it is that a function is by
nature not very transparent, so it's hard to tell whether the storage
overhead is necessary or not.
regards, tom lane
