Carrying alternate kernel builds in Extras

[Dave Jones]

On Tue, Mar 28, 2006 at 08:55:07PM -0600, Jason L Tibbitts III wrote:
 > I have an SMP VIA C3 machine that is not supported by FC5 in SMP mode
 > because it doesn't have PAE and thus won't run the i686 SMP kernel.
 > Compiling a kernel without CONFIG_HIGHMEM64G=y works fine.
 > 
 > Core doesn't really want to carry yet another kernel permutation.
 > Would it be reasonable to consider carrying alternate kernel builds in
 > Extras?

This has come up before. (Though I believe it was 'suspend2' and the like
last time).

There's a few things that jump to mind that make this a problem.
- Errata kernels
  extras kernels will always lag, leaving people stuck with a problem
  until their -extra kernel rebases.  Trawl bugzilla some time, and
  you'll find a bunch of people stuck on some ancient kernel that
  refuse to update (probably because they have some out of tree module
  that only works on that one version). Keeping people on the latest
  versions is vital to keeping bugzilla under control.

- More flavours that all out-of-tree modules shipped in extras need to
  be built against.

- What happens when bugs get found?
  I'm way overloaded with the bugs in the kernels shipped by core,
  and have no time to look at more flavours, so the first thing I'd
  be doing would be to ask reporters to try a -core kernel.
  (And you can guarantee there'd be people who refuse to give up
   their favorite feature than go test a -core kernel for a while,
   leaving another source of festering 'never to be fixed' bugs).

I'm sure there were other reasons too that came up last time,
but my memory is hazy right now.

		Dave

-- 
http://www.codemonkey.org.uk