[Debian-astro-maintainers] Bug#915207: gnudatalanguage: autopkgtests fail: test_bug_n000720, test_idlneturl, test_point_lun, test_save_restore, test_zip

[Ole Streicher]

On 03.12.18 10:02, Sebastiaan Couwenberg wrote:
> On 12/3/18 9:54 AM, Ole Streicher wrote:
>> On 03.12.18 09:41, Sebastiaan Couwenberg wrote:
>>> hdf5 is unable to migrate to testing because gnudatalanguage
>>> autopkgtests fail. Test failures indicate that the package is broken,
>>> and not suitable for release. It should be removed from testing until
>>> the test failures are fixed, which also allows hdf5 and its rdeps to
>>> migrate to testing.
>>
>> That is currently in the decision of the package maintainer. Neither the
>> policy nor the RC document you cited list CI test failures or migration
>> delays as a possible cause for severity "serious".
>>
>>> As announced the autopkgtest delay is increased exponentially which will
>>> block packages from getting into buster before the final freeze. We do
>>> not want that to happen, but your package is not allowing hdf5 and its
>>> rdeps to migrate even though there is nothing else preventing this part
>>> of the transition.
>>
>>> Fix your failing autopkgtests or remove them. Not fixing failing
>>> autopkgtests in your package that prevent its dependencies from (timely)
>>> testing migration is very unfriendly (to phrase it very mildly).
>>
>> I totally agree, I am affected as well, and I am working on it. You
>> shouldn't think that an "important" bug does not get attention. But this
>> also does not reason the severity "serious".
> 
> Yes, it is. Your package is broken and needs to be removed from testing
> as soon as possible, as it negatively affects its dependencies.

The package works well, with a few things that do not work. This is what
we call "normal" or "important" bugs. "negative effects" is not
"breaks". Migration delays are not breakages.

> important severity does not trigger autoremoval, serious does.

Sure, so what?

> The RC policy hasn't been updated for autopkgtests yet, but I expect
> failing tests to become an official reason for RC severity as that
> removal from testing of that package will be the only way to let its
> dependencies migrate to testing (without release team hints).

Unless this is the case you can't cite that as an argument for being RC.
Also not the policy.

> If a fix is forthcoming, why not let gnudatalanguage be autoremoved from
> testing, and let it migrate again when it has been fixed?

> Do you think that keeping your package in testing is more important than
> all the other ~100 packages involved in the hdf5 transitions? According
> to popcon it's not.

About 30-40 of those are mine, BTW. And I am still unsure whether hdf5
is not really the cause of the problem yet, since it is the HDF5 test
case in gnudatalanguage is the only one where I don't see a reason for
the failure yet. Just "let us migrate" is the wrong way here -- the CI
migration delays are to find out where the problem is before.

Best regards

Ole