test_can_restrict_endianness fails on big-endian s390x

[ginggs]

Hi!

test_can_restrict_endianness fails on s390x, the only big-endian architecture being tested in Ubuntu.
Links to full tests logs can be found on the following page:
http://autopkgtest.ubuntu.com/packages/p/python-hypothesis/bionic/s390x

Is this by design?

The test fails when dt=dtype('int16') and dt.byteorder == '='

=================================== FAILURES ===================================
_________________________ test_can_restrict_endianness _________________________

    @given(nps.integer_dtypes(endianness='>'))
>   def test_can_restrict_endianness(dt):

tests/numpy/test_gen_data.py:189: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/usr/lib/python3/dist-packages/hypothesis/core.py:579: in execute
    result = self.test_runner(data, run)
/usr/lib/python3/dist-packages/hypothesis/executors.py:58: in default_new_style_executor
    return function(data)
/usr/lib/python3/dist-packages/hypothesis/core.py:571: in run
    return test(*args, **kwargs)
tests/numpy/test_gen_data.py:189: in test_can_restrict_endianness
    def test_can_restrict_endianness(dt):
/usr/lib/python3/dist-packages/hypothesis/core.py:518: in test
    result = self.test(*args, **kwargs)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

dt = dtype('int16')

    @given(nps.integer_dtypes(endianness='>'))
    def test_can_restrict_endianness(dt):
        if dt.itemsize == 1:
            assert dt.byteorder == '|'
        else:
>           assert dt.byteorder == '>'
E           AssertionError

tests/numpy/test_gen_data.py:193: AssertionError
---------------------------------- Hypothesis ----------------------------------
Falsifying example: test_can_restrict_endianness(dt=dtype('int16'))
======== 1 failed, 2156 passed, 4 skipped, 1 xfailed in 278.16 seconds =========

[Zac-HD]

Thanks for the report @ginggs - it looks like our behaviour in this case is correct, but the test is wrong! Given a > (big-endian) argument to the dtype strategy we do always generate big-endian dtypes, but native byte-order is always reported as =.