Re: nd6 change and rc.d/network_ipv6 -> rc.d/netif integration

Hiroki Sato wrote:
> Hi,
> 
>  I would like your comments about merging the network_ipv6 -> netif
>  integration to stable/8.  

I maintain my objection to MFC'ing this prior to the 8.0-RELEASE. As
stated previously my objections are as follows (in decreasing order of
general importance):

1. It is a fairly significant change happening too late in the release
cycle. IMO that is reason enough to not allow the change.
2. Although 8.0 seems to be getting more beta/rc testing than previous
.0 releases, the overall number of users testing it is still a small
percentage of the userbase.
3. A dramatically smaller percentage of those users who are actually
doing the testing is also using IPv6.
4. There are still rough edges to the changes.
5. I personally disagree with some of the choices you've made and
would like to see more discussion about them. (More about 4 and 5 below.)

The rough edges I've noticed have to do with the various problems
people have reported to the lists, including what seems to be a lack
of testing without IPv6 in the kernel, continuing evolution of how to
deal with the afnet tests, and personally I've noticed the following
on my console, although I haven't had time to research yet whether
it's definitely coming from your changes:

in6_ifattach_linklocal: failed to add a link-local addr to wpi0

In terms of design decisions you've made, I am still confused about
why you insist on deprecating ipv6_enable. Recent discussion on the
lists indicates to me that I'm not alone in thinking that this is a
valuable mechanism and that there is not only no reason to deprecate
it, to do so is not desirable.

I'd also like to explore further the idea that I suggested in a
previous thread that it should not be necessary to specify
ifconfig_IF_ipv6 at all. The vast majority of users will be using RA
for the next couple of years at least, so in my mind it makes sense to
default to using ipv6_network_interfaces=$network_interfaces and RA by
default. If the user has a need to configure something explicitly then
you've provided the mechanism for them to do that, but they shouldn't
be forced to use it. This is another reason that I think ipv6_enable
should be the "master" knob. I like the idea of the ipv6_prefer knob,
but I do not like the idea of overloading it with the function of
ipv6_enable too.

I can certainly understand why you are eager to get these changes into
8.0, however if we do a proper job of maintaining backwards
compatibility (which I think we should do anyway) I don't see any
reason that they cannot be merged after 8.0, and more importantly
after they have had a proper opportunity to shake out in HEAD.


Doug

-- 

    This .signature sanitized for your protection