Hi Chiara,
On 2022-04-01 05:47, Chiara Marmo wrote:
> I have investigated the scikit-learn bug #1008369 [1].
> I thought it could be fixed by an update of numpydoc [2], but I was
> wrong (sorry Christian... I believe you are in this list too so I'm not
> cross-posting on debian-python).
Thanks for trying, anyway!
> Indeed, the failures are related to docstring issues in scikit-learn
> that have been fixed in the current main version on github.
> I have created a number of patches already but new failures pop out... :(
> I am quite confident that 1.1 will soon be out (~end of April).
>
> Could it be acceptable to wait for 1.1 rather than create a number of
> patches on the current version?
If the offending tests all contain the string "docstring" (as is the
case in #1008369), I'd just skip those for now.
This could be as simple as adding "not doctest" to "EXCLUDE_TESTS" in
debian/rules and to "exclude_tests" in debian/tests/python3.
I think in the past, even larger parts of the test suite have been
omitted temporarily to easily deal with transient stuff.
Otherwise, I think waiting until 1.1 wouldn't be too bad.
I don't think the effort of anything more than pulling a patch or two
from 1.1 are worth it, if the issue will indeed be resolved shortly with
the new upstream release.
> Thanks for your understanding.
>
> Best regards,
>
> Chiara
>
> [1] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1008369
> <https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1008369>
> [2] https://lists.debian.org/debian-python/2022/03/msg00048.html
> <https://lists.debian.org/debian-python/2022/03/msg00048.html>
Best,
Christian
