On Wed, Feb 22, 2006 at 02:28:46AM +0000, Thomas E. Spanjaard wrote:
> Sean Davis wrote:
> >Some people (myself included) have no IPv6 connectivity. Nor do I want or
> >need it. We don't insist on SCSI in kernels with support for hard disks, do
> >we? No, if I don't have SCSI in a machine, I can leave it out. I think the
> >same argument is valid for IPv6.
> 
> It might make sense to merge the code paths as far as they're similar, 
> even though you don't need the other version. You can separate the 
> shared part, and have the rest of each reside under their respective 
> options.

I agree.

> >We support IDE-and-no-SCSI, as well as SCSI-and-no-IDE. Kernels can have
> >both even if the machine only has one of the two, but why add extra bloat?
> 
> Well, for example ATAPI devices are basically SCSI devices on an IDE 
> connection (simplified). This means that to support ATAPI devices like 
> CD-ROMs or tape drives and such, you need some common code between SCSI 
> and IDE(ATAPI). I think this is a similar case to the IP versions in 
> this respect.

I left ATAPI out in my (imperfect) analogy, because it was the best I could
think of at the time of that mail for a one-or-the-other-but-not-both
example that we currently support. Yes, I realize this makes my analogy
flawed; I just don't see any practical reason to force everyone to have IPv6
in their kernel when I'd hazard a guess more than half of the NetBSD
community doesn't use IPv6.

> >This isn't even going into the fact that NetBSD insists on trying IPv6 
> >first
> >on everything, even if there is no IPv6 connectivity, which is a very good
> >reason to leave it out of a kernel, IMO.
> 
> That's a bug indeed, just as having gethostbyname(3) with a query for 
> NULL not responding with both :: and 0.0.0.0 when net.inet6.ip6.v6only=1 
> (though this isn't the kernel's fault)(assuming you have IPv6 compiled 
> in and working, ofcourse).

Aye.

-Sean