[ATrpms-users] FESCo Meeting

Jarod Wilson lists at wilsonet.com
Fri Sep 7 04:14:23 CEST 2007


On Sep 05, 2007, at 17:36, Axel Thimm wrote:

> On Wed, Sep 05, 2007 at 12:05:53PM -0600, Richi Plana wrote:
>> On Wed, 2007-09-05 at 19:55 +0200, Axel Thimm wrote:
>>> We had an irc meeting a year ago when I tried to convince Fedora  
>>> that
>>> kmods are broken and that kmdls should be introduced. It was mainly
>>> torpedoed by Thorsten Leemhouis, and I don't have the nerve and time
>>> to waste part of my life again. :(
>>
>> Perhaps he'll have a change of heart. Or other engineers will see  
>> your
>> point now.
>
> Thorsten made me leave EPEL and also caused the split in rpmfusion vs
> rpmrepo. I'd doubt that there is a change of heart.

Wow have I had my head in the sand for a while... I hadn't heard a  
lick about rpmfusion in quite a while, didn't know things had  
actually fallen apart there. Sucky.


>> With regards to kmdl, that'll only be true for some modules and  
>> modules
>> that aren't in the kernel tree.
>>
>> *sigh*
>
> Well, you're free to enter the fesco discussion arguing in favour of
> kmdls, I know I won't as even if I had the energy left I'm just
> packing my stuff to relocate and won't be online for 1 week+.
>
> The point is that kernel modules outside the kernel package are not
> seen in favour, not becasue the packaging sucks, but because the
> kernel masters at Red Hat/Fedora lose control over what kernel space
> bits run and are afraid of bug reports that were induced by kernel
> parts that they didn't package.
>
> There is some truth in this, and the bad kmod implementation didn
> t help persuade them to accept this kind of packaging, but OTOH we're
> talking about Linux, an open system which is not to be controlled by
> vendors, but by users, so an all-in-one-kernel approach is
> semantically wrong to start with.
>
> But still whatever the kernel rpm will ship, if there is need to ship
> extrenal stuff the kmdls will not be affected by this decision. Of
> course it would had been nice if kmlds had been adopted by Fedora, but
> they haven't and with people at the edge of banning any kmdl-like
> setup pushing them off the clip, one won't manage to get it into  
> Fedora.

Not necessarily. There are multiple camps within Red Hat/Fedora, and  
plenty of us do NOT want to see externally packaged kernel modules  
deep-sixed. There's even good business justification for it (see: Red  
Hat Enterprise Linux...).

The push on the Fedora side is to not have the Fedora project itself  
pushing out any kernel module packages, and I think most folks are  
down with that. Instead, we put the bits into the kernel package  
itself and work on getting them into the upstream kernel itself  
(which I'm doing w/lirc right now). The Fedora kernel, especially in  
rawhide, gets updated so often, its just much easier to maintain  
these bits in the kernel itself than to spin kernel module packages  
for every new kernel. It also does more to drive upstream adoption.

Outside of this particular debate, is whether or not to publish an  
"official" kernel module packaging standard that the Fedora project  
suggests 3rd-party repos use to package any kernel module bits they  
may have. This is where kmdl2 could still make some headway.  
Regardless of what the Fedora side of the house thinks, externally  
packaged kernel modules and a consistent standard for putting them  
together are and will continue to be an important detail for Red Hat  
Enterprise Linux.


-- 
Jarod Wilson
jarod at wilsonet.com






More information about the atrpms-users mailing list