[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: need repositories for keras subprojects



On Sat, Jun 9, 2018 at 1:25 PM, Lumin <cdluminate@gmail.com> wrote:
> Hi Stephen,
>
>> I am updating the keras package and the newest release of keras 2.2.0
>> now depends on some modules that have been broken off upstream into
>> separate tarballs, namely "keras-applications" and
>> "keras-preprocessing".
>>
>> So I am making corresponding packages for these.  Would someone be
>> able to create repositories for these on salsa?  Spelled exactly as
>> above; the upstream repositories can be found at:
>> https://github.com/keras-team
>>
>> I shall file the appropriate ITP shortly.
>
> Thank you for your effort of maintaining the keras package. I've
> created several repos for you as requested, and grant you master
> access to these repos:
>
> [1] https://salsa.debian.org/science-team/keras-applications
> [2] https://salsa.debian.org/science-team/keras-preprocessing
>
> Next time please provide your salsa account name, so the other's
> won't need to try searching your name on salsa :-)

Thank you very much!  Yes, sorry I forgot that credentials were
needed.  I'll try not to forget next time.

>> One issue regarding autopkgtest for these subprojects: I have found
>> that it is impossible to run any tests on them without keras installed
>> ("import keras_applications" fails if "import keras" was not executed
>> first.)  However, keras depends on them.  Therefore it seems
>> impossible to include any autopkgtest tests for these packages since
>> for this purpose it is bidirectional dependency.  Is it ok to skip the
>> autopkgtest in this case and therefore ignore the corresponding
>> lintian warning?
>
> It sounds strange. Would adding all the related packages to Depends: field
> in autopkgtest control file work?

I'll try it, I'm not sure since the keras package needs these ones,
but I suppose if they are uploaded at the same time then test-time
dependencies should not fail.

Daniel Stender suggested to me in a separate mail that perhaps I
should just bundle them into the keras package but it seemed simpler
to do it this way.  I am not sure how to do multi-tarball upstreams,
but I suppose it is possible, I may look into it, but I thought that
arranging the packages to mirror upstream's divisions would be the
best initial approach.  Suggestions are welcome, I will give it a try
and post here for a review when it's ready.

regards,
Steve


Reply to: