On 5/10/11 12:23 PM, Jonathan Corbet wrote:
I'm not talking about this week. The X11 problem was reported
three weeks
ago, and the "don't run Rawhide" advice given to me came rather before
that. Rawhide has been an unusually painful place to be for some time
now, and a lot of people, I believe, have opted out of it.
Yum has pattern excludes, man. You have the tools.
I'm getting close to doing the same, despite having run Rawhide
on my
desktop for a *long* time.
Your response really just reinforces my concern. What value does the
distribution get from a "dumping ground"?
Sigh.
Catching this instance of the problem is not entirely trivial. I have a
plan, and I'll even take patches for it, but it's simply not a priority.
Releases matter more.
I suspect they matter more for a lot of people, in fact.
Which is an entire set of problems that simply did not exist before
no-frozen-rawhide was a thing. Before that, the problem we had was five
hundred builds all hitting rawhide at once and literally nothing working
for weeks _after_ the release was out. Here, instead, we've slipped that
broken stage to be parallel with tuning the release. That seems like a
win to me; once F15 is out, some F16hide problems have a chance of
having been fixed, and the rest will get done sooner (in the sense of
"by an earlier date", not "at a faster rate"), which gives more useful
time before F16 to do actual work.
---
What _I'm_ dismayed about is that literally zero people cared about this
problem enough to have the courage to look at fixing it. Not so much
that it leaves rawhide unusable, more that we've basically lost any
notion of collective ownership, and/or that nobody is fearless enough to
go read things they don't already know. I had to write a patch for
frickin' LLVM for F15, of all things, code I'd never read before in a
language I don't really know in a domain where I have zero expertise.
That was just the problem that was in my way.
X being broken in rawhide was in who knows how many peoples' ways, and
even though it's provenpackager+ and even though all it would have taken
was a mass driver rebuild, nobody even tried. Where _are_ you people?
Why do I bother to open the ACLs on my packages if nobody's going to
take advantage of it?
---
But to your question of "what value do we get from rawhide existing",
well, builds have to go somewhere. We may as well keep the compose tools
running on them all the time, so when things do break we see them. All
we had here was a case where one tool didn't catch a breakage because
the other bit of the tools weren't complete enough yet. I don't think
that's sufficient cause to question rawhide's existence. I suspect,
instead, that most other consumability problems with rawhide are
basically the same pattern: nobody tries to fix anything outside their
scope.
If one is not concerned with the whole being greater than the sum of its
parts, why exactly would one work on a distribution?
- ajax