Josh Boyer wrote:
On Mon, Sep 15, 2008 at 08:58:09PM -0400, Chris Snook wrote:
> Jesse Keating wrote:
>> Back in December, I had made a change that blocked kernel-devel packages
>> from winding up in the install media for the Fedora spin. I don't
>> recall getting any push back at the time, but I've gotten at least one
>> angry comment since then. So I'm putting it out for more discussion.
>> Do we feel that the kernel-devel (5~megs) should be in the install
> Yes please. There's always new hardware we don't support yet, so some
> people will need to build drivers just to get online and access the
> repos. If we ship it on the install media, it's much easier to
> distribute code that you're reasonably confident will work on a new
> install. If people have to hunt down matching kernel and kernel-devel
> rpms, it's a moving target for people working on these device drivers.
> I'm not saying we should bend over backwards for out-of-tree drivers, but
> this is precisely the scenario that determines the first impression for
> someone trying this "Linux" thing on their shiny new bleeding-edge box,
> and it's pretty easy to accomodate.
Erm... if this is the first time someone is trying a shiny new "Linux" thing
on a bleeding edge box, and they have to grab a kernel-devel package and
build drivers _themselves_, then they are obviously smart enough to run
'yum install kernel-devel'. Somehow I think your example is slightly off.
I don't know many Linux newbies that know 1) that they need to build a
driver, 2) what driver to build, and 3) what packages they need to build it
all without knowing how to install anything.
These are not the people you are targeting. They know enough to not require
kernel-devel on the install media. You can move on to finding some other
use case that makes sense.
How are they going to 'yum install kernel-devel' with no internet connection?
And how does the maintainer of the not-yet-merged driver ensure that his code
always works with whatever kernel Fedora has rebased to today? If you can be
reasonably confident that every Fedora 10 user can at least get a certain
matching kernel and kernel-devel, it's a lot easier to maintain drivers. In
theory you could build packages, but when you're doing this for Fedora, Debian,
Ubuntu, CentOS, etc. it isn't really feasible.
I know this is a niche use case, but it's a very large niche, and it's the niche
that's most likely to become avid Linux users if we don't push them away the
first time. I know because I had this experience when I first started
maintaining the atl2 network driver. Tens of thousands of EeePC users, many of
them technically savvy but Linux novices, wanted to replace the hacked up
Xandros that ships on the Eee with something a bit more flexible, but they
needed atl2 to make it work, and it wasn't merged in the distros yet. When
fast-moving distros like Fedora rebased their kernels to 2.6.24, my code broke,
but the users who had kernel-devel or equivalent on their install media were
fine, while the others either had to dig around to find a matching kernel and
development headers, since only the latest were in the repos, or wait for me to
The situation is much worse in the wireless world, where many of the drivers are
reverse engineered and break on every new hardware rev, forcing users to install
experimental stuff that isn't ready for merging in order to get their shiny new
I'm not saying that kernel-devel is an absolutely critical package to have on
the install media, but it's extremely convenient at a time when users are most
frustrated. If we're going to trim the install media, which I'm not
fundamentally opposed to, there are much more frivolous things that should go first.