Pandas and dependencies used to be keyworded for ppc64 but is no longer. I built dev-python/pandas-2.2.2-r1 and all the dependencies listed by nattka on power9 with FEATURES=test and they all work. I built pandas with "full-support" and "debug" USE flags to ensure that all the optional features are tested and that assertions are enabled.
N.B. I had problems with dev-python/netcdf4 giving the below error, which turns out to be triggered by compiling with debug info (`-g` in CFLAGS). Removing the `-g` fixed the problem. ImportError: /var/tmp/portage/dev-python/netcdf4-1.7.1-r1/work/netcdf4-1.7.1-python3_12/install/usr/lib/python3.12/site-packages/netCDF4/_netCDF4.cpython-312-powerpc64le-linux-gnu.so: undefined symbol: pfnc_inq_vardimid
Helpful but not strictly required: #937755 for a properly open source implementation of szip.
N.B. This is all with numpy 1.26.4 because numpy 2.0.0 fails to build on power9 - see #937757
Created attachment 906407 [details] xarray build and test with netcdf4 roundtrip failures Now that I have found a workaround for #937757 I rebuilt these packages with against numpy 2.0.0 All the tests pass in all packages if I keep dev-python/xarray to version 2024.6.0-r1 which built successfully with numpy v1. However the latest dev-python/xarray-2024.9.0 fails two tests in the NetCDF4 backend.
dev-python/xarray-2024.11.0 also fails src_test with the same two errors in NetCDF4 FAILED xarray/tests/test_backends.py::TestNetCDF4Data::test_roundtrip_mask_and_scale[dtype0-create_unsigned_false_masked_scaled_data-create_encoded_unsigned_false_masked_scaled_data] - AssertionError: Left and right Dataset objects are not close Differing data variables: L x (t) float64 32B -1.0 10.1 22.7 nan R x (t) float64 32B 10.0 10.1 22.7 nan FAILED xarray/tests/test_backends.py::TestNetCDF4Data::test_roundtrip_mask_and_scale[dtype1-create_unsigned_false_masked_scaled_data-create_encoded_unsigned_false_masked_scaled_data] - AssertionError: Left and right Dataset objects are not close Differing data variables: L x (t) float32 16B -1.0 10.1 22.7 nan R x (t) float32 16B 10.0 10.1 22.7 nan
