atmel drivers in the main distro?

Dave Jones davej at redhat.com
Thu Mar 4 05:27:47 UTC 2004


On Wed, Mar 03, 2004 at 06:49:19PM +1100, msevior at physics.unimelb.edu.au wrote:

 > I also think these should be put into whatever Kernel Fedora is rolling
 > for FC 2. Since this is RedHat's perogative, I'll happily email the person
 > doing this if (s)he isn't reading this list or missed my post.

The problem with saying "sure, I'll merge this" is that once I've
set precident, I start getting dozens of requests "oh, you merged driver
A, how about driver B too?", and before I know it I have $DEITY knows
how many 'add-on' drivers in the Fedora kernel, which turns into a
maintainence nightmare.

>From my experience with maintaining the RHL7-9 kernels (which were
quite heavily patched with external drivers), and FC1 (ditto, although
to a somewhat lesser extent), I've already been through the hell of
having to run around finding out the latest versions of patches on
project webpages, digging mangling diffs so they fit against our
kernel, checking they aren't backing out fixes etc.. it's a nightmare.

Realworld example: I got a large number of requests to update the
Adaptec SCSI driver in RHL, so set about digging around and found
the latest version of the driver (which folks had claimed fixed their problems).
I merged the _700KB_ update after skimming through the diff
(with a diff of that size, with no broken-into-individual-changesets diffs,
 reviewing is a real headache, and mistakes do get overlooked).
I rolled this into an update, and handed off to our QA folks.
It didn't even boot on any of the Adaptec cards that they tested with.
And this was just 1 single driver. With a dozen or so drivers, with
lots of users mailing you/bugzilla'ing "new version upstream supports
my xxx card" might look like a 10 minute merge-job, but it quickly
spirals out of control. That's why we're taking the attitude
"get it upstream" with Fedora. It's not just a cheap way out and
a chance for me to slack off, it's about actually trying to get a
product out there which has a chance in hell of being maintained
effectively after release.

Remember that "we'll stay as close to upstream as possible" mantra? 
For FC2, with a low patch count, we actually have a damn good shot 
at pulling that off.
Doing it for FC1 turned out to be a miserable experience due to the
large amount of change still going on in mainline continually conflicting
with the large number of patches we applied.  If we ship FC2 with
say a 2.6.5 kernel, it's not outside the realms of possiblity
that through its maintainence cycle, it continually tracks upstream
when 2.6.6, 2.6.7 etc get released, rebasing at each point release.

There must be a reason that these folks haven't got their drivers
in the upstream kernel yet. Frankly, if they aren't ready for mainline,
they likely aren't ready for Fedora either.

		Dave





More information about the fedora-devel-list mailing list