--- Begin Message ---
Package: release.debian.org
Severity: normal
User: release.debian.org@packages.debian.org
Usertags: unblock
Please unblock package openmpi to fix RC bug #848574
Openmpi 2.0.2 was released just as Stretch was being frozen. The package in testing,
2.0.2~git.20161225 was packaged to get relevant (2.0.2) changes into Stretch, but unfortuanatelt contained
a significant bug on mips64el release architecture that was not caught before transition.
debdiff too large to be useful.
unblock openmpi/2.0.2-2
-- System Information:
Debian Release: 9.0
APT prefers testing
APT policy: (500, 'testing')
Architecture: i386 (i686)
Kernel: Linux 4.9.0-1-686-pae (SMP w/1 CPU core)
Locale: LANG=en_IE.UTF-8, LC_CTYPE=en_IE.UTF-8 (charmap=UTF-8) (ignored: LC_ALL set to en_IE.UTF-8)
Shell: /bin/sh linked to /bin/dash
Init: systemd (via /run/systemd/system)
--- End Message ---
--- Begin Message ---
Hi,
On Fri, Feb 17, 2017 at 04:29:05PM +0000, Alastair McKinstry wrote:
> >> debdiff too large to be useful.
> > Yes, that's the issue:
> >
> > 4391 files changed, 1353847 insertions(+), 423120 deletions(-)
> ...
> > So perhaps you can get a filtered diff, saying what you have excluded (and why),
> > and explain what changes there are in this release and why we should accept it,
> > rather than a targeted fix for #848574?
> >
> I agree about the problem, the trouble is that 2.0.2~git.20161225 turns
> out not to be a good baseline to be working against; it should not be in
> testing, either 2.0.1 or 2.0.2 should be. (It was in the middle of a
> large change; upstream do a lot of commits ; not only were 4391 files
> changed between Dec and now, about the same number of commits were
> involved. (I had believed that 2.0.2 would include a major SOVERSION
> change that in the end wasn't made, and wanted to get a version with
> fixes but no SOVERSION change into stretch).
It looks like we don't have any other option, as the version in testing is
clearly broken. I unblocked it.
> I had picked that git as a 'near release point', expecting it to be
> nearly stable. It wasn't. I have little confidence that if I cherry pick
> just the fix for #848574, that the resulting package would be fit for
> stretch.
> 2.0.2 on the other hand _has_ had extensive testing, (it has one of the
> best test infrastructures around) it just wasn't ready in time for the
> freeze date.
>
> The other alternative is removal, which is less desirable:
> $ apt-cache rdepends libopenmpi2 | wc -l
> 186
> directly.
Let's hope we don't have to resort to this, if it turns out the unblock was
the wrong call...
Cheers,
Ivo
--- End Message ---