[fab] Architecture Policy.

[Dave Jones]

On Thu, Nov 16, 2006 at 01:27:55PM +0800, David Woodhouse wrote:

 > > Looking at the current bug FC6 kernel lists, there's only a half dozen
 > > obvious "ppc only" bugs out of 213, but that considering it's used
 > > a lot less than x86/x86-64, the ratio of bugs/users is probably the
 > > pretty close.  The only difference is that the x86/x86-64 bugs are getting
 > > attention (slowly but surely).
 > 
 > To confirm your meaning above: You're saying that the number of PowerPC
 > bugs _isn't_ out of proportion -- that the 6/213 you see is what you
 > might reasonable expect from the statistics? It isn't out of proportion?

Maybe it's because I don't have direct exposure as a user of ppc that it
seems to be in more of a dire state than it is, but 99% of bugs
that get filed for ppc are "this doesn't boot/install", which is a showstopper.
Compared that to things like "my sound doesn't work" "I got an oops"
"nfs doesn't work" or anything else that pops up on the more common archs.
Some bugs get reported a lot more than others, like when say, CIFS is broken.
This sort of thing never gets reported on PPC, even though it's equally
as affected.  Now, that may imply that PPC users don't care about CIFS, NFS
and sound, or it could imply that there's not a sufficient userbase period.
Or it could be just "it didn't even install, so I didn't get that far".
We've no real metrics other than the types of bugs that /do/ get filed,
and the download stats, which have been dropping off every release from
what I hear. (unsurprising given the only people that previously cared
with either a) mac users, who are now a dying breed due to the x86 migration
or b) folks like IBM who used Fedora as a testbed until a ppc RHEL beta came
along).

 > It is my perception that the PowerPC kernel _does_ get attention -- also
 > slowly, but surely. Our hardware support is better now than it was with
 > the FC4 release, and hardware manufacturers have been working with us to
 > include new hardware while we've also had improvements in support for
 > older machines.

I don't recall there being as many "doesn't boot" complaints in FC4 or FC5
as what we have open in FC6.

 > If the amount/speed of bug handling is considered to be a problem, then
 > I'm very sure we can improve on that. Fedora doesn't have to be all
 > @redhat.com people, and we can _certainly_ get assistance from others if
 > it's known to be necessary.

Right, and imo, it shouldn't be just @redhat.com people.
If we can convince community folks to help, I'm all for it.

 > My concern is that life gets somewhat fraught around release time, and
 > there will be no time available, while we're in the middle of the
 > scramble on x86/x86_64, to work through a _separate_ process for
 > releasing on PowerPC. And thus that _however_ hard the PowerPC folks
 > work, we might not be able to release FC7/PPC in sync with the FC7
 > release on other architectures because we'd be blocked on approval and
 > process issues.

For FC6, towards the end, PPC got very little love, and that's continuing
post-release.   Unless something changes, we shouldn't pretend to care about
it equally as much as x86/x86-64 whilst simultaneously letting it become
the red headed stepchild for FC7.

 > Thus the suggestion -- and I know I'm repeating myself here -- that we
 > don't use PowerPC as a guinea pig for the new process. Use it first to
 > bring Aurora and/or AlphaCore into the fold, rather than breaking an
 > architecture which is working already today.

Sparc/Alpha is an unknown commodity at this point wrt to "do things get fixed
when users report them broken" (at least from my point of view, I've no idea
how those projects even track bugs if they do at all), so I don't think
it's really an apples/oranges comparison to PPC.

If you really believe that PPC can remain as a primary answer, the solution
isn't persuading me, or the f-a-b, or really anyone else to change policies.
The solution is to improve the situation so that there's something in place
so that users don't get left dangling when they report bugs.  Dennis mentioned
the idea of having per-arch mailing lists for such things that could get
CC'd/assigned bugs.  If there was sufficient interest from folks, then I'd
agree that'd be a great start.

		Dave

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