https://bugzilla.redhat.com/show_bug.cgi?id=2176933
--- Comment #9 from Jerry James loganjerry@gmail.com --- (In reply to Miro Hrončok from comment #5)
I see. IMHO this is a bug in upstream packaging. The setup.py script has:
from a11y_pygments.utils.utils import find_all_themes_packages
Hence before we can execute setup.py, we already need to have install_requires installed.
A possible upstream fix (if you intend to persuade it) is to introduce a pyproject.toml file with pygments listed in [build-system].requires. Something like:
[build-system] requires = ["setuptools", "pygments >= 1.5"] build-backend = "setuptools.build_meta"
A better fix is not to import the package from setup.py, but that might be tricky.
(This might deserve a comment near the manual BR.)
I will speak with upstream about this. Thanks for the diagnosis.
The intended usage is:
%{py3_test_envvars} %{python3} test/run_tests.py
I had no idea that such variables are not considered "environment" when introduced the macro (and nobody who discussed or reviewed it had that idea neither). The macro will be documented in the guidelines. Do you think it needs to be renamed?
Ah, I misunderstood. When used like that, they are environment variables. If the documentation says to use the macro that way, then the name is fine.
(In reply to Miro Hrončok from comment #7)
test/run_tests.py has no output. Is that expected?
The test regenerates the output in the test/results directory. We could save the original output, then compare the two. However, they are not the same, because we have a newer version of pygments than was used to generate the results packaged by upstream. So all we are testing is that this package can be imported and used to generate output. Whether the output is actually correct is not tested. I will make a comment about that in the spec file.
I have updated the URLs with the version I intend to import. Thank you for the review!