[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: kqemu is now GPLv2



On Wed, Feb 07, 2007 at 09:28:14AM -0500, Jesse Keating wrote:
> On Wednesday 07 February 2007 03:15, Andreas Bierfert wrote:
> > Maybe we should just follow a different approach for kmods then? Why not do
> > something like a module manager (I heard some other distros have that ;) )?
> > With it people could easily build their modules themselves but have them
> > integrated via rpm so their filesystems don't go into nirvana after a
> > couple of system upgrades. If you make it easy (and graphical) enough I
> > suspect that people would be ok with it for external module stuff. That
> > would solve the problems of repo inconsistency but still give users what
> > they want...
> 
> And when they don't rebuild cleanly?  Then what?  The user is left holding the 
> bag of a broken system should they ever reboot to that new kernel.
> 
> I am entirely unconvinced that out of tree kernel modules adds any value over 
> the long run.  It may work for a kernel or two, but it will lag, it will 
> break, and somebody will get blamed for it, more often than not, it will be 
> us for moving the kernel too fast, or not caring enough about external 
> modules to hold back updates, or, or, or...  Out of tree modules _will_ lead 
> to poor user experiences and I do _not_ want the Fedora name attached to 
> _any_ of them.

#1: I agree, I really want all kernel modules in the Fedora kernel.
No question.  For this I push very hard, and where I have influence
(or can hack the driver code myself), it gets into upstream.  We're
not backing down from that stance.

#2: For the subset of kernel modules where #1 isn't (yet) reality, we
 need to strive for #1.  In the mean time, having
 not-yet-merged-upstream-but-in-progress kernel modules available can
 demonstrate both the value and the bug-freeness of the new code.
 (e.g. 1000 people have been running it for a month, and no new bug
 reports have been filed in the last 3 weeks).  Here, tools like kmods
 and DKMS may provide some benefit.  It's the same as with
 updates-testing.  It needs to exist, even if we don't necessarily
 want all mainstream users using it.

-Matt


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]