[fab] Alternative kernels?

Jeremy Katz katzj at redhat.com
Wed Sep 20 19:38:33 UTC 2006


On Wed, 2006-09-20 at 15:27 -0400, Christopher Blizzard wrote:
> Jeremy Katz wrote:
> > If we allow arbitrary kernels that are maintained in Extras, how do we
> > make sure that there's actually a consistent set of features provided?
> > And that's ignoring the questions of currency and handling of security
> > errata, which is already hard enough.
> 
> In the case of the planet kernel, does it matter?  It's for a very 
> specific use.  We're not always talking about the general use case.  Be 
> it the planet kernel or the OLPC kernel, there's more than one use case 
> here.  Sometimes it makes a lot of sense to have a different kind of kernel.

For the PlanetCCRMA case, actually, it _does_ matter.  If we add a
wireless driver to the main kernel package and it doesn't end up in the
Extra CCRMA kernel package, then people using that get less
functionality and it's "choose A or B".  Which sucks.

OLPC being a slimmed down kernel targeted at a specific piece of
hardware is an entirely different discussion.

> > Additionally, more kernels ==> more pain for kernel module packagers.
> > Now they need to know about even _more_ variants and be able to build
> > against them.
> 
> Does it?  You're assuming that everyone pays for the pain for everyone 
> else's work.  I assume that some of that work will spill over to the 
> kernel packagers that we have today, but it's not going to be all of it. 

Read what I said again -- this is about third party module packagers,
not davej.  Having to guess at what kernels are available to build
modules for makes the complexity of the packaging of kernel modules
multiply far beyond the insanity it already is.

>   I think that enabling various uses is probably worth the pain, and the 
> mechanisms will work themselves out.  Maybe Dave will go a little crazy, 
> but maybe this will help him as well.  Maybe other people will be able 
> to step up and be able to actually help him with that work.  But we 
> can't even begin to tell that without taking the first step and saying 
> it's OK for other people to enter that space.

Frankly, "might"s and "maybe"s aren't really enough here for me.  Having
a separate kernel doesn't really do anything to enable people to help
him... so if they're not doing it now, why are they going to be _more_
willing to do so when we basically say "here, we endorse you having your
own playground"?

We went down this path for a while with the Xen kernel stuff and the
amount of pain was way way way huge.

Jeremy




More information about the fedora-advisory-board mailing list