Also, simplify the spec with:
* Dropping armhv7l and i686 conditions
* Dropping old test skips that seem to have been fixed
* Use pyproject macro's `-l` license option
* Drop patch for old versioneer, as the new one is available
Well, this is awful, but I can't find anything better. As the
tests are part of the library, we can't really run them out of
the source tree but against the installed library as we'd like
to, and as we usually do for sane projects. We can't really
remove them from the installed library and run the tests out of
the source tree because we have to set PYTHONPATH to do that
and it means we're not testing the installed copy of the lib,
plus it causes other issues.
So put them back in, but do an ugly hack to remove a single test
file when running the tests on s390x (as all the tests in it
fail, they are not big-endian safe), but put it back in the
installed package so we don't have an inconsistency between the
noarch package as built on s390x and the noarch package as built
on other arches. Sigh.
This updates to the latest upstream release, drops several patches
that have been merged or that are no longer necessary since we
have newer versions of various deps in Fedora 40 and Rawhide, and
backports https://github.com/dask/dask/pull/11185 to fix a couple
of issues with Python 3.13.
In the next major update `pandas` will drop support for i686. This
package depends on `pandas`, thus it will need to stop building for i686
as well.
More information is available in Bugzilla:
https://bugzilla.redhat.com/show_bug.cgi?id=2263999
This is important because the python-pandas package will now install
pandas.tests only with python3-pandas+tests, and Dask’s tests import
from Pandas’s test suite.
- Conditionalize fastavro BR, which is not available on 32-bit
- Run tests on all arches to reliably track arch-dependent bugs
- Add necessary arch-dependent skips for failing tests
- Drop the “diagnostics” extras metapackage because no recent enough
Bokeh version is available.
- Switch to the PyPI sdist as source (with workarounds)
- Ensure NUMPY_LICENSE.txt is packaged
- Add Provides: bundled(numpy)