I refreshed and re-enabled Andreas' mark_tests_working_on_intel.patch
 which was disabled in 0.22.0-1.
I additionally marked test_sum_nanops_timedelta and
test_timedelta_ops_with_missing_values with pytest.mark.intel which were
failing at least on arm64.
To address a new failure on big-endian architectures, I added a patch
 and forwarded it upstream.
I uploaded 0.22.0-3 including these changes, and as of this writing,
builds have succeeded on s390x, ppc64el, amd64 and i386.
debian-science-maintainers mailing list