The test_ecs_intake_ek80_BB_complex test currently fails on the Python 3.12/3.13 CI jobs but the Python 3.11 job passes. The failure happens in the broadband ECS interpolation check, where interpolated calibration parameters are compared using .identical(). The Python 3.12/3.13 jobs resolve newer dependencies than Python 3.11 (NumPy 2.4.6 vs. 2.4.3, SciPy 1.18.0 vs. 1.17.1, pandas 3.0.3 vs. 3.0.2) but its unclear what changes.. The pb is angle_offset_alongship. The arrays are numerically equivalent:
np.allclose(...) == Truexr.testing.assert_allclose(...) passes- max absolute difference:
5.551115123125783e-17
- diff /
np.finfo(float).eps: 0.25
attributes, coordinates, dtype, indexes, and encoding are identical, so this appears to be a tiny floating point difference. I'll open a small PR replacing .identical() assertion with xr.testing.assert_allclose() for this comparison!
The
test_ecs_intake_ek80_BB_complextest currently fails on the Python 3.12/3.13 CI jobs but the Python 3.11 job passes. The failure happens in the broadband ECS interpolation check, where interpolated calibration parameters are compared using.identical(). The Python 3.12/3.13 jobs resolve newer dependencies than Python 3.11 (NumPy 2.4.6 vs. 2.4.3, SciPy 1.18.0 vs. 1.17.1, pandas 3.0.3 vs. 3.0.2) but its unclear what changes.. The pb isangle_offset_alongship. The arrays are numerically equivalent:np.allclose(...) == Truexr.testing.assert_allclose(...)passes5.551115123125783e-17np.finfo(float).eps:0.25attributes, coordinates, dtype, indexes, and encoding are identical, so this appears to be a tiny floating point difference. I'll open a small PR replacing
.identical()assertion withxr.testing.assert_allclose()for this comparison!