Markus Schaber <schabios@logi-track.com> writes:
> Tom Lane schrieb:
>> given the improved support in 8.0 for anonymous record types, we could
>> in theory have the backend invent a record type on-the-fly to match
>> whatever list of OUT parameters a particular function has.)
> It would not be necessarily on the fly, at least in the first step we
> possibly get away with declaraing the returned tuples at creation time
> and implicitly creating those tuple types. The declaration could be like
> "returns (touchedrows int, somethingelse datetime), setof (article int,
> description text)" for a function/method that has two resultsets, one of
> those with always one row.
The advantage of not explicitly creating the rowtypes is that we don't
need to worry about choosing nonconflicting names for them. So I think
I'd go down the anonymous-rowtype path even in the first cut.
regards, tom lane