For what it's worth, I think all the systems I have seen hostname
trouble on have multiple interfaces.  I also tend to have 'too many'
interfaces, which causes sun rpc to choke sometimes since SIOCGIFCONF
needs more than the 1kbytes that freebsd libc gives it by default and
the fallback/expand code seems imperfect.  I wonder if there is
something in the rpc2 code that does something similar.

My server is now ipv6-ready (netbsd+kame) and I also have some
ipv6-ready clients (freebsd 41-stable) - I haven't had any problems on
these.  Both are single-homed and have no ipv6 addresses configured -
just the automatic link-local.  (I don't have any Linux clients,
unfortunately - the only one handy is RH sparc, and too important to
muck with.)

My rough take on things is that the host table being right doesn't
matter, and that one has to have DNS working ok.  But this is a very
fuzzy hunch from trying things - I have never gotten this at all
nailed down.

        Greg Troxel <gdt_at_ir.bbn.com>