On Sat, Dec 18, 1999 at 10:10:02PM -0800, Michael Graff wrote:
> matthew green <mrg@eterna.com.au> writes:
> 
> > i would like for it to be a sysctl; i don't care about the default.
> 
> I tend to think the default is what the RFC says -- we should have it
> on by default, and put a comment in inetd.conf that we ship, and on
> the inet6 and inet man pages.

Well, standards are standards but I don't like to have it on by
default as to me it doesn't follow the rule `of least suprise';
It makes a bind on INET6 get fuzzy semantics, as application
programmer you may think you're accepting only real INET6 connections
but you have no way of telling whether you get IPv4 connections in
disguise too. So, as itojun implied: if you cannot rely on the feature
when porting an app it really is of no use anymore (you have to deal
with the case that it is switched off anyway). Therfore, I think it is
one of these things that will become obsolete in the standard some
day because it is not used in practice. (Note also, that for special
applications like faithd(8) one *needs* to know the address family
of the connection. See also my send-PR on faithd).

But to conform to the RFC it needs to be there, but in addition to a
sysctl I'd like a kernel option to specify the default for that 
sysctl. Just to reduce the potential for suprise a little bit on
systems where that is  really needed.

Feico.