On Fri, Mar 20, 2015 at 9:04 AM, Robert Haas <robertmh...@gmail.com> wrote:
> On Fri, Mar 20, 2015 at 10:54 AM, David Christensen <da...@endpoint.com> > wrote: > > In that case, the other thought I had here is that we change the > function signature of current_setting() to be a two-arg form where the > second argument is a boolean "throw_error", with a default argument of true > to preserve existing semantics, and returning NULL if that argument is > false. However, I'm not sure if there are some issues with changing the > signature of an existing function (e.g., with pg_upgrade, etc.). > > > > My *impression* is that since pg_upgrade rebuilds the system tables for > a new install it shouldn't be an issue, but not sure if having the same > pg_proc OID with different values or an alternate pg_proc OID would cause > issues down the line; anyone know if this is a dead-end? > > I think if the second argument is defaulted it would be OK. However > it might make sense to instead add a new two-argument function and > leave the existing one-argument function alone, because setting > default arguments for functions defined in pg_proc.h is kind of a > chore. > Isn't there some other update along this whole error-vs-null choice going around where a separate name was chosen for the new null-returning function instead of adding a boolean switch argument? David J.
