Re: Initial libnl support committed

[Robert Love <rml novell com>]

On Fri, 2005-10-28 at 12:22 -0400, Dan Williams wrote:

> > Is this solid code?  Is anything else using it?  I have never even heard
> > of it.
> 
> Here's my argument.  I was going to have to talk to netlink anyway.  You
> evidently can't do stuff like remove duplicate IP addresses from
> interfaces with ioctl().  Routing table manipulation code is insane with
> ioctl().  Netlink is obviously the way to go, I don't think anyone would
> disagree.

Agreed.

> AFAIK, there are no existing netlink convenience/shim libraries other
> than libnl.  We _could_ suck in a private copy of libnl, but nobody
> likes to do that, nobody thinks doing this is a good idea that _I_ know
> of, and it's pretty much the same thing as using it externally.

There is something called libnlink.  I know nothing about it.
Description says "libnetlink provides a higher level interface to
rtnetlink(7)."

I have never heard of either libnl or libnlink until googling today.

> I've mentioned for about 4 or 5 months that moving to libnl is the
> forward looking plan, were you simply looking for more lead-time before
> the change happened? 

No, I'm not concerned about lead time -- just the choice itself.  I like
moving fast.  ;-)

But I never heard of libnl or the intended move; I must have missed it.

> If you'd  like some configure-time magic
> to use either an external or an internal version of the lib so you don't
> have to actually package libnl, I'd entertain that idea and take a patch
> for it.

Nah, making it an option seems silly.  I would be happy if YOU also
wanted to use it internal-only, I would go for that, but making it an
option just so I don't have to make a new package is dumb.  I want to
diverge less, not more.

So, here is my thinking:  If future NM needs are going to use more and
more netlink (as we agree) and the netlink code utilized is complicated
and not worth reimplementing, then using libnl seems sane.  But if we
are talking about more of the same (saving a mere 30 lines), then I am
not for it.  Since you imply that we are going to have greater and
greater dependence on netlink, then I suspect the former is true and
let's move forward.  I will package libnl for SUSE.  I would like to see
more usage sooner rather than later, though, to validate libnl's use.

Best,

	Robert Love