On 2016-03-02 17:22, Ben Hutchings wrote:
On Wed, 2016-03-02 at 17:21 +0100, Emilio Pozuelo Monfort wrote:On 02/03/16 13:24, Ben Hutchings wrote: > On Wed, 2016-03-02 at 12:31 +0100, Andreas Beckmann wrote: > > Source: linux-tools > > Version: 4.4-4 > > Severity: important > > > > Hi, > > > > new major upstream releases (e.g. 4.4) of src:linux and src:linux-tools > > should migrate together to testing. Right now src:linux-tools 4.4-4 is > > already in testing, which doesn't look too useful to me without > > src:linux 4.4.x-y. > > This should have been prevented by the fact that it makes linux-perf > uninstallable in testing (it depends on linux-perf-4.3). Cc'ing the > release team to comment on why it happened anyway. Because linux-perf-4.3 is still in testing.Yet only on two architectures!
Yes, because...
It was smooth-updated. Not sure if that was expected, given the package is in section devel and not (old)libs.
Actually, I'm not sure that it was. The excuses for the run during which it migrated has:
<li>old binaries left on <a href="http://buildd.debian.org/status/logs.php?arch=amd64&pkg=linux-tools&ver=4.4-4"; target="_blank">amd64</a>: liblockdep4.3, linux-kbuild-4.3, linux-perf-4.3 (from <a href="http://buildd.debian.org/status/logs.php?arch=amd64&pkg=linux-tools&ver=4.3.1-2"; target="_blank">4.3.1-2</a>) (but ignoring cruft, so nevermind) <li>old binaries left on <a href="http://buildd.debian.org/status/logs.php?arch=i386&pkg=linux-tools&ver=4.4-4"; target="_blank">i386</a>: liblockdep4.3, linux-kbuild-4.3, linux-perf-4.3 (from <a href="http://buildd.debian.org/status/logs.php?arch=i386&pkg=linux-tools&ver=4.3.1-2"; target="_blank">4.3.1-2</a>) (but ignoring cruft, so nevermind) <li>old binaries left on <a href="http://buildd.debian.org/status/logs.php?arch=arm64&pkg=linux-tools&ver=4.4-4"; target="_blank">arm64</a>: liblockdep4.3, linux-kbuild-4.3, linux-perf-4.3 (from <a href="http://buildd.debian.org/status/logs.php?arch=arm64&pkg=linux-tools&ver=4.3.1-2"; target="_blank">4.3.1-2</a>) (but ignoring cruft, so nevermind)
and so on. So the new version was migrated at the start of the run, and then she tried to remove the cruft at the end (and is still trying, in fact).
Does that mean there was manual intervention? Or is that automatic behaviour? Does it ignore all->any dependencies on !x86?
It was all automatic. britney only requires that arch:all packages are installable on i386 and amd64. At the end of the britney run during which linux-tools migrated, the old binary packages from each of the other architectures were removed as doing so did not create any new uninstallability so far as britney was concerned.
Regards, Adam