On Tue, Jan 02, 2007 at 10:15:59AM -0500, der Mouse wrote:
> >> If we [remove routes to noncontiguous netmasks], I think we either
> >> should forbid noncontiguous netmasks on interfaces or add
> >> corresponding code somewhere to do the splitting you refer to when
> >> adding or removing interface routes for interfaces with
> >> noncontiguous netmasks.
> > I want to kill the "feature" completely.
> 
> Another reason to be glad I'mn not tracking -current, I guess.  (I'd
> much rather just pay a performance penalty if noncontiguous netmasks
> are used.  But I don't expect my opinion to make much difference.)

It is not so much about the performance penalty, but the code
complexity. I would have to keep track of the original routes, check any
new route for possible affecting a non-contiguous netmask route, resolve
that conflict, possibly readd the partial route etc. That's a lot of
complexity for a functionality almost noone is using. I expect to find
it in pre-CIDR networks if it is still used at all. The real question
for those users is whether or not they have to equivalent route setup
already for other operating systems like Linux. I guess that is the
case. I'm asking this now as I want to deprecate it for NetBSD 4 first.

> 
> > I don't want the kernel to handle it at all -- I don't think any
> > non-local network still depends on it, so admins should be able to
> > convert the networks accordingly if they still have such rules.
> 
> You can't imagine any reason to use them, so no such reason exists?

I can image reasons to use them, but I don't think they outweight the
complexity.

Joerg