On Fri, 2019-08-09 at 15:14 +0100, Daniel P. Berrangé wrote:
On Fri, Aug 09, 2019 at 03:13:07PM +0200, Martin Kolman wrote:
> On Fri, 2019-08-09 at 14:00 +0100, Daniel P. Berrangé wrote:
> > On Fri, Aug 09, 2019 at 02:28:55PM +0200, Jens-Ulrik Petersen wrote:
> > > On Fri, Aug 9, 2019 at 9:27 AM Igor Gnatenko <
> > > ignatenkobrain(a)fedoraproject.org> wrote:
> > >
> > > > Well, it was retired because it did not built since F30 mass
rebuild…
> > > >
> > >
> > > I went ahead and built it with the testsuite disabled for now: I suppose
> > > any Proven Packager could also have done, but yeah normally it should be
> > > done by the maintainers.
> > > I admit the ball was dropped on this by various people (myself included),
> > > and sorry about that. [1]
> > > The new major upstream release was also a long time coming...
> > >
> > > But to me the deeper question is still "why are we proactively
breaking the
> > > distro" in this way with package retirements by non-maintainers?
> > >
> > > Sure FTBFS is bad but there is no need to proactively remove core
packages
> > > which are still working okay.
> > > I really really wish could stop this... causing more busy work and
stress.
> >
> > When a FTBFS hits, we don't know whether the package is still working
> > ok or not as there are many possible reasons for the failure.
> Maybe this is something gating tests could help with ? If a package is FTBFS
> and has reasonable gating test coverage, you will know it is working.
The gating CI is a pretty low bar right now. As CI is made stronger,
it could well actually make FTBFS *more* common, as CI is introducing
more scope for things to be classed as a failure. So be careful what
you wish for :-)
> > Filing
> > the FTBFS BZs informs the maintainer(s) & allows them to investigate,
> > figure what has gone wrong & decide what changes are needed. Missing
> > on 2 mass rebuilds means the package is still build with F29 toolchain,
> > and thus lacking desired improvements Fedora is introducing, so this
> > has a cost for the rest of the distro. Somewhere there's a balance
> > between cost for the maintainer in work & cost for the distro in the
> > package being outdated.
> >
> > There was no acknowlegement on the BZ that anyone was actively working
> > on fixing it in 6 months. This is true for so many of the FTBFS BZs that
> > get filed. If the packages don't get orphaned after 6+ months of being
> > ignored, when would they ever be fixed ?
> >
> > Having said that, I think in the case of packages which are deps of
> > so much of the distro, it could have been useful to have a warning of
> > imminent orphaning on fedora-devel. There was a warning that orphaning
> > was starting, but no list of affected packages included.
> Maybe another job for automated tests/CI ?
>
> Before dropping a batch of packages, do a test compose without them and postpone
> the drop if the compose run crashes and burns.
>
> Sounds really like something doable which could save a lot of everyones time once in
place.
We can't carry on postponing things indefinitely though - at some point we
have to say enough, and expect a maintainer to actually do some maintaining.
Sure
& I totally agree with that.
I'm just trying to find ways that can sound the alarm bells &
prevent everyone impacting Rawhide breakage before it's too late
and things need to be fixed post-mortem. An this case really looks
like something that an automated check should be able to catch soon
enough. :)
Regards,
Daniel