On Wed, Apr 8, 2009 at 3:49 PM, Bruce Momjian <bruce@momjian.us> wrote:
> We already had a huge discussion over 'S' and I think we did as good as
> we can. I think we risk overcomplicating the API by adding U, but we
> can revisit this in 8.5 once we get more feedback from users.
I think we'll need to take stock before 8.4 actually. Tom's pointed
out a whole pile of problems with the current approach and I'm
becoming convinced he's right. I know I was one of the proponents of
the change but I didn't realize how bad the problems were.
As I understand his proposal is that \df with no pattern could list
all user functions but \df <pattern> should always follow the
search_path and show the same functions that would actually be called.
One possibility for reducing clutter would be moving a whole slew of
the system functions which are never intended for users to call
explicitly to a different schema which isn't implicitly added to
search_path. That would at least get all the RI functions, bt procs,
maybe even the operator functions out of the way.
--
greg