Closed taldcroft closed 1 year ago
It took me a while to realize that the issue was with the python setup.py --version
call...
I verified that this builds in the build workflow.
It took me a while to realize that the issue was with the python setup.py --version call...
Me too. It came to me out of nowhere when I wasn't even really thinking about this. Brain.
Sorry, I guess I just need to write things out more. As in my slack comment "From the setup.py it is still supposed to return version without the cython calls, so the failure on OSX to end up with version (‘chandra_time-None-py38_0.tar.bz2’) still seems odd." should have said "Our build process does a preliminary step where "python setup.py --version" is called, and the setup.py has an if statement to do that without needing Cython, but it doesn't appear to be working now on OSX or Windows". I still don't know why this stopped working but removing the lambda seems great.
Oh I get, "python setup.py --version" started to fail because the lambda was incompatible with the language_level=3 piece that was the change.
Description
Fixes (hopefully) the problem discussed on Slack "we are getting this error when building chandra_time on Windows" in
#aspect_team
. The original strategy of makingcythonize
into a lambda function that returnsNone
was silly.I also noticed when building that the
MANIFEST.in
file referenced non-existent files and was generating warnings when building. But the distribution was still working so I don't think we need the manifest. Evenpython setup.py sdist
seems to be fine, see testing below.Interface impacts
None.
Testing
Unit tests
No code changes.
Functional tests
Install test
I did the usual of removing the
build
dir and installing to a dev ska and running tests. All good.Version
MANIFEST.in
The
.pyx
and.h
files are automatically included even without a manifest.