[Ksummit-2012-discuss] [ATTEND] linux-next and process
linux at roeck-us.net
Tue Jun 19 01:59:37 UTC 2012
On Mon, Jun 18, 2012 at 02:31:19PM -0700, Andrew Morton wrote:
> On Mon, 18 Jun 2012 17:17:52 -0400
> Dave Jones <davej at redhat.com> wrote:
> > On Mon, Jun 18, 2012 at 01:08:12PM -0700, Andrew Morton wrote:
> > > On Sun, 17 Jun 2012 11:28:12 +1000
> > > Stephen Rothwell <sfr at canb.auug.org.au> wrote:
> > >
> > > > Also, I am wondering if we should talk a little about the process of
> > > > being maintainers (I hear groans already :-) - I don't really want to
> > > > teach you all how to suck eggs) and whether people have any (hopefully
> > > > positive) criticism of the linux-next process itself.
> > >
> > > I'd like to hear about the code which is turning up in mainline but
> > > wasn't in linux-next, or wasn't in linux-next in the same form/version.
> > > How common is this, why is it happening, do we have a problem and if
> > > so, what do we do about it.
> > This has been puzzling me for a while. Sasha Levin runs my fuzzing tool
> > on -next almost daily, and keeps finding bugs.
> That's a useful thing to do.
> > Then when Linus merges it all,
> > we see a ton of new bugs appear that neither of us hit before.
> > I've observed this for the last 2-3 releases, but it's probably been happening
> > for longer. This isn't driver specific either, it's core code that everyone runs.
> Oh my, I didn't know that. So we have regressions going into mainline
> which would have been detected beforehand, only they're not, because
> code isn't getting a trial run in -next? If so, that is a massive fail.
> It would be interesting to bisect some of these regressions, then do a
> bit of investigation into how the offending patches managed to get into
> mainline. Perhaps a pattern will emerge...
Someone published statistics a couple of months ago - I think it was for 3.2 or
3.3. My memory may be wrong, but at as far as I recall the percentage of commits
in the commit window which did not go through -next was quite high, something
in the range of 20 or even 30%.
More information about the Ksummit-2012-discuss