v1.0.0: rpm build fails on fedora rawhide

[junghans]

From https://koji.fedoraproject.org/koji/taskinfo?taskID=56777342

I think the error is:

__init__.py:126: in <module>
    matplotlib.use('agg', warn=False)
E   TypeError: use() got an unexpected keyword argument 'warn'
___________ ERROR collecting MDAnalysisTests/analysis/test_align.py ____________

This is with python3-numpy-1.19.4-1.fc34, do we need a newer version on fedora?

Build log here

[IAlibay]

@junghans this seems to be related to #2938

The current solution is to pin to an older version of matplotlib, but we are hoping to release a 1.0.1 version very soon that should fix this (I think our own hurdle left is to re-enable CI since we've just migrated away from Travis).

edit: I'm glad @lilyminium isn't also awake, we might end up with three comments saying the exact same thing within a few seconds of each other 🙃

[RMeli]

Hi @junghans , unfortunately this is a known issue: #2938

It should be fixed for release 1.0.1. For the time being, the workaround is to pin the version of matplotlib to <3.3.

[junghans]

Ok, I will patch #2763 into the rpm.

[junghans]

Now it fails with a different error (see https://koji.fedoraproject.org/koji/taskinfo?taskID=56779633):

INTERNALERROR> AssertionError: ('MDAnalysisTests/analysis/test_data.py::test_all_exports', <WorkerController gw1>)
INTERNALERROR> assert not 'MDAnalysisTests/analysis/test_data.py::test_all_exports'

Build log here

[IAlibay]

😕 I'm not sure we've properly tested with 3.9 (@jbarnoud I think did?), any reason it's rc1 and not the full release?

hopefully @orbeckst might have a better handle on if this is in any related to #2763

[junghans]

It is rc1 of 3.9.1.

[junghans]

F33 has 3.9.0, let's test that: https://koji.fedoraproject.org/koji/taskinfo?taskID=56781133

[junghans]

Error stays the same:

INTERNALERROR> AssertionError: ('MDAnalysisTests/analysis/test_data.py::test_all_exports', <WorkerController gw39>)
INTERNALERROR> assert not 'MDAnalysisTests/analysis/test_data.py::test_all_exports'

End of build log here

[lilyminium]

I can't reproduce on the current version of develop in 3.9.0 with numpy 1.19.4 (osx, clang). @junghans are you using the 1.0 release or the development package? Although I can't actually install the 1.0.0 mdanalysistests package on 3.9. It's a weird error because that particular test just checks that all files in mdanalysistests.data.filenames that should be in __all__, are. Maybe it's related to #2939?

[junghans]

This was the 1.0.0 tarball, I can make a dockerfile to reproduce this.

[junghans]

Here it is:

Dockerfile
FROM registry.fedoraproject.org/fedora:rawhide
RUN dnf -y install fedpkg
RUN git clone -b v1.0.0 https://src.fedoraproject.org/forks/junghans/rpms/python-MDAnalysis.git
WORKDIR python-MDAnalysis
RUN dnf -y builddep python-MDAnalysis.spec
RUN spectool -g python-MDAnalysis.spec
RUN fedpkg --release f34 local

[junghans]

Now https://bugzilla.redhat.com/show_bug.cgi?id=1903956

[IAlibay]

@junghans sorry we kinda lapsed on looking into this - is this still an active issue?