test_load_zip_with_local_header_extra_field failure on s390x

[emollier]

Greetings,

While integrating trx-python 0.4.0 in Debian sid, our continuous integration caught a regression on s390x (maybe also on other big endian platforms, but that's the only one tested):

https://ci.debian.net/packages/p/python-trx-python/testing/s390x/69895533/

The interesting part of the test log looks like:

 84s             trx = tmm.load_from_zip(trx_path)
 84s >           np.testing.assert_array_almost_equal(trx.streamlines._data, positions)
 84s E           AssertionError: 
 84s E           Arrays are not almost equal to 6 decimals
 84s E           
 84s E           Mismatched elements: 6 / 6 (100%)
 84s E           Max absolute difference among violations: 6.
 84s E           Max relative difference among violations: 1.
 84s E            ACTUAL: memmap([[4.6006e-41, 8.9683e-44, 2.3049e-41],
 84s E                   [4.6007e-41, 5.7487e-41, 6.8966e-41]], dtype='<f4')
 84s E            DESIRED: array([[1., 2., 3.],
 84s E                  [4., 5., 6.]], dtype=float32)
 84s 
 84s trx/test_memmap.py:609: AssertionError

I'm not sure exactly where the issue lies. It could be a need to update the reference data for big endian platform, but the looks of the result feels more like a need to adjust the logic somewhere for big endian platforms appropriately.

I'm not expecting a lot of our users to run big endian platforms and intend to make this issue non-blocking for a migration to testing, so no pressure. But if you feel this could be symptomatic of a more problematic issue, please don't hesitate to say so.

Have a nice day, :)
Étienne.

[skoudoro]

Hi @emollier,

Thank you for the feedback, we will look into it.

Not sure about the timeline yet