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

Tinyarray packaging



Hello,

As mentioned [1], Steffen Möller and Andreas Tille helped to finish the
packaging of Tinyarray [2].  Here, I would like to continue off-list
discussions we had.

Andreas Tille wrote:

> The cythonized files are not really needed for Debian packaging.  Its
> mandatory to re-build them in the package build process anyway.
> Otherwise this is considered as the usage of foreign binary code.
> Thus it makes sense to use a tarball without cython results.

Actually, tinyarray is a pure C++ Python extension.  There’s no Cython
in there.  My remark on upstream tarballs containing cythonized files
was more general, extending to the other packages that I intent to
package, like Kwant.

Kwant contains substantial amounts of Cython code, and since the
beginning we have been following the advice of Cython documentation [3]:

“It is strongly recommended that you distribute the generated .c files
as well as your Cython sources, so that users can install your module
without needing to have Cython available.”

But for Debian it is of course possible to re-cythonize.  However, this
does not change the fact that the upstream tarballs contain the
cythonized files.  And I always believed that Debian is very careful
(not to say pedantic) about upstream tarballs and keeps them as
immutable artifacts.

> If you are tagging releases properly in your
> 
>     https://gitlab.kwant-project.org/kwant/kwant
> 
> I would prefer this over PyPI.  But using PyPI is fine was well.

Sure, the tags are correctly signed and we keep them immutable.  We’ve
been using PyPI in debian/watch, since it is likely *even* more
persistent ;-) than our own gitlab instance.

However, the only officially released tarballs are those to be found on
PyPI and on https://downloads.kwant-project.org/.

Is it OK use the gitlab-generated tarballs as upstream for Debian?
I guess that’s fine, other than that the upstream tarball for a given
version will then differ from the one used by everyone else.

> I just sponsored tinyarray.  In principle I would love if Christoph
> would use
> 
>      https://wiki.debian.org/DebianPureBlends/SoB
> 
> It perfectly fits since the its scientific software.  May be even
> tinyarray would fit in some of the Debian Science Blend?  The final
> target will fit in any case and I'd strongly recommend to use
> a repository under science-team for its packaging.

I’m happy to follow any guidelines, I’m just a bit confused with what
“DebianPureBlends” has to do with packaging tinyarray for Debian.
I thought that Debian Pure Blends is an umbrella-term for specialized
sub-distributions of Debian.

I’m also OK with moving tinyarray to debian-science if this is useful
and possible.

Thanks
Christoph

[1] https://lists.debian.org/debian-science/2020/09/msg00043.html
[2] https://salsa.debian.org/python-team/modules/python-tinyarray
[3] https://cython.readthedocs.io/en/latest/src/userguide/source_files_and_compilation.html#distributing-cython-modules

Attachment: signature.asc
Description: PGP signature


Reply to: