On Mon, 2013-07-01 at 19:27 -0400, Tom Lane wrote:
> I think probably we'd have just not compiled the dependent code, and
> would have found out about it only when somebody complained that peer
> auth didn't work on OpenBSD. Not sure that's really a more attractive
> behavior :-(
That might be the case, but that's more a problem of how this particular
feature is implemented, not how the configure test works.
