Tests fail on ppc64 and s390x

[susilehtola]

I tried updating the Fedora packages of pybind11 to version 2.2.2. However, the tests fail on the ppc64 and s390x architectures, see https://koji.fedoraproject.org/koji/taskinfo?taskID=25049665 .

ppc64 build log https://kojipkgs.fedoraproject.org//work/tasks/9682/25049682/build.log
s390x build log https://kojipkgs.fedoraproject.org//work/tasks/9684/25049684/build.log

It appears both architectures have the same errors

=================================== FAILURES ===================================
__________________________________ test_dtype __________________________________
simple_dtype = dtype({'names':['bool_','uint_','float_','ldbl_'], 'formats':['?','>u4','>f4','>f16'], 'offsets':[0,4,8,16], 'itemsize':32})
    def test_dtype(simple_dtype):
        from sys import byteorder
        e = '<' if byteorder == 'little' else '>'
    
>       assert m.print_dtypes() == [
            simple_dtype_fmt(),
            packed_dtype_fmt(),
            "[('a', {}), ('b', {})]".format(simple_dtype_fmt(), packed_dtype_fmt()),
            partial_dtype_fmt(),
            partial_nested_fmt(),
            "[('a', 'S3'), ('b', 'S3')]",
            ("{{'names':['a','b','c','d'], " +
             "'formats':[('S4', (3,)),('<i4', (2,)),('u1', (3,)),('<f4', (4, 2))], " +
             "'offsets':[0,12,20,24], 'itemsize':56}}").format(e=e),
            "[('e1', '" + e + "i8'), ('e2', 'u1')]",
            "[('x', 'i1'), ('y', '" + e + "u8')]",
            "[('cflt', '" + e + "c8'), ('cdbl', '" + e + "c16')]"
        ]
E       assert ["{'names':['... 'S3')]", ...] == ["{'names':['b... 'S3')]", ...]
E         At index 6 diff: u"{'names':['a','b','c','d'], 'formats':[('S4', (3,)),('>i4', (2,)),('u1', (3,)),('>f4', (4, 2))], 'offsets':[0,12,20,24], 'itemsize':56}" != "{'names':['a','b','c','d'], 'formats':[('S4', (3,)),('<i4', (2,)),('u1', (3,)),('<f4', (4, 2))], 'offsets':[0,12,20,24], 'itemsize':56}"
E         Use -v to get the full diff
test_numpy_dtypes.py:98: AssertionError
_______________________________ test_array_array _______________________________
    def test_array_array():
        from sys import byteorder
        e = '<' if byteorder == 'little' else '>'
    
        arr = m.create_array_array(3)
>       assert str(arr.dtype) == (
            "{{'names':['a','b','c','d'], " +
            "'formats':[('S4', (3,)),('<i4', (2,)),('u1', (3,)),('{e}f4', (4, 2))], " +
            "'offsets':[0,12,20,24], 'itemsize':56}}").format(e=e)
E       assert "{'names':['a...itemsize':56}" == "{'names':['a'...itemsize':56}"
E         Skipping 44 identical leading characters in diff, use -v to show
E         Skipping 71 identical trailing characters in diff, use -v to show
E         - , (3,)),('>i4', (2,)
E         ?           ^
E         + , (3,)),('<i4', (2,)
E         ?           ^
test_numpy_dtypes.py:216: AssertionError
=============== 2 failed, 284 passed, 6 skipped in 7.07 seconds ================

[jagerman]

Thanks for the report. It looks like a false positive from a little-endianism that crept in since the last time we fixed this in #699/#694.

#1287 should solve it: can you give that patch a try and confirm that it fixes things on those (big-endian) arches?

[susilehtola]

Yes, confirmed. Thanks for the speedy fix.