kernels in the packaging universe

[Dave Jones]

On Wed, Dec 20, 2006 at 02:06:39PM -0500, Greg Dekoenigsberg wrote:

 > To the best of my knowledge, the problem you have with kmods/alternate 
 > kernels is that people complain when they break, and they fill bugzilla 
 > with bugs that don't make sense -- because people don't understand that 
 > they're running funky kernels.
 > 
 > Right?  Are there any other reasons not to package these alternate 
 > kernels?
 > 
 > Because that's a valid reason.  But it also gives us something to shoot 
 > for: better reporting tools.

The bugzilla issue is the #1 reason.
I don't want to do another round-trip in bugzilla where I have to ask..

"Now try and repeat this issue without kmod-blah loaded".

This has always been a problem for users who choose to build their own addons,
but offering prebuilt binaries makes it easier, and hence more of a problem.

It's as big a problem as the nvidia & ati modules. In fact worse, because
we actually endorse these by their prescence in extras.  There's usually
good reason that code isn't accepted upstream.  Quite often that reason is
"driver xyz is a steaming shitpile". As a result, I refuse to look into any
bug until out-of-tree drivers are removed from the equation. And if users
refuse to try without that driver, I'll close it CANTFIX.

Kernel bugs in Fedora are _out of control_.  And yet it seems we're determined
to do whatever we can to make the situation even more unmanagable.

If these drivers are important enough to be in Fedora, they're important
enough to get beaten into shape to survive an upstream code review and
subsequent inclusion there.  Packaging them into RPMS doesn't achieve this. At all.
Any comments about "increased testing" are pointless, because they typically
find superficial bugs that are usually easily fixed. Fundamental problems
are typically only found by code review.

		Dave

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