Re: pyensembl autopkgtest

[Andreas Tille <andreas@an3as.eu>]

Hi,

I'm CCing debian-med list.  Please be so kind to always write this kind
of mails to the list and not to any maintainers private mailbox.

Am Fri, Aug 11, 2023 at 11:09:16PM +0700 schrieb Lin Qigang:
> I have been working on the autopkgtest for pyensembl, and I discovered that
> pyensembl is missing a dependency for tinytimer. It fails to run with a
> simple `pyensembl -h` after an apt install of the package. As far as I know,
> tinytimer is not in Debian. Do you think I should create an ITP for it?

Thanks a lot for verifying pyensembl and finding this problem.  Yes.  if
`pyensembl -h` needs tinytimer to run pyensembl sensibly it needs to be
packaged.  I get

$ pyensembl -h
Traceback (most recent call last):
  File "/usr/bin/pyensembl", line 33, in <module>
    sys.exit(load_entry_point('pyensembl==2.2.8', 'console_scripts', 'pyensembl')())
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
  File "/usr/lib/python3/dist-packages/pkg_resources/__init__.py", line 868, in _resolve_dist
    raise DistributionNotFound(req, requirers)
pkg_resources.DistributionNotFound: The 'tinytimer' distribution was not found and is required by pyensembl


So yes, we need python3-tinytimer.  Do you have time to package this -
probably in Debian Python Team since this seems to be a generic package
and the DPT prefers to have control over generic Python packages.  In
case you are no member of this team I might package it in the next
couple of days.
 
> I am using the tests created upstream for the autopkgtest. Currently, 19 of
> 36 tests are failing due to data that needs to be imported with pyensembl.

We have several options in this case.  The most simple option is to
simply skip the affected tests.  If we really care for all tests we
should check whether the missing data files are sensibly small we can
provide these files in debian/tests/data.   If the data files are large
we should create a multi-source tarball (see Debian Med policy how this
works).
 
> Also, 14 of 25 test files are being ignored during testing. These tests need
> files that are not present in the tarball from Pypi, but these files are
> present in the Github source.

Usually its better to stick to the Github source instead of Pypi.  I
need to check this situation if I have more time (if nobody will beat
me in doing so ;-) ).

> If the package version was released on Github,
> I would switch the watch file to that, but the last release and tag on
> Github was in 2017.

This might be the reason why the watch file is pointing to Pypi.  May
be you can file a Github issue asking for proper tags?

> I am ignoring these tests for now, but the other testing
> problems could be fixed with the missing dependency.
> 
> What do you think is the best solution for this package?

Please start with asking upstream for a proper tag.  Hopefully upstream
responds quickly to this.  We should decide who will package tinytimer
an than we see what might be sensible further.

Thanks a lot for working on this
    Andreas.

-- 
http://fam-tille.de