Test failure in test_frac.py for function float_to_twelths_frac

[mhucka]

With Python 3.10.0 and 3.12.5 on macOS 14.7, running pytest produces a failure in test_frac.py:

    def test_float_to_twelths_frac() -> None:
        with pytest.raises(ValueError):
            u.float_to_twelths_frac(1.0 / 24)
        with pytest.raises(ValueError):
            u.float_to_twelths_frac(1.0 / 7)
        with pytest.raises(ValueError):
            u.float_to_twelths_frac(1.0 / 5)
        with pytest.raises(ValueError):
            u.float_to_twelths_frac(1.0 / 11)
        with pytest.raises(ValueError):
            u.float_to_twelths_frac(1.0 / 13)

        assert u.float_to_twelths_frac(0) == frac(0)
        assert u.float_to_twelths_frac(502) == frac(502)
        assert u.float_to_twelths_frac(1.0 / 12) == frac(denom=12)
        assert u.float_to_twelths_frac(-1.0 / 12) == frac(-1, 12)
        assert u.float_to_twelths_frac(501.0 / 3) == frac(167)
        assert u.float_to_twelths_frac(502.0 / 3) == frac(502, 3)

        # Precision.
        assert u.float_to_twelths_frac((1 << 55) + 1) == frac((1 << 55) + 1)
>       assert u.float_to_twelths_frac(float(1 << 55) / 3.0) == frac(1 << 55, 3)

test/cython/test_frac.py:128:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
tunits/core/cython/frac.pyx:83: in tunits_core.float_to_twelths_frac
    cpdef frac float_to_twelths_frac(a) except *:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

>   raise ValueError("Not a twelfth.")
E   ValueError: Not a twelfth.

tunits/core/cython/frac.pyx:91: ValueError

This is using a fresh pyenv virtual environment, then installing dependencies using

pip install -r dev_tools/dev.env.txt

followed by running pytest. Stopping in the test function and inspecting the values, it appears that the difference between values for the case involving float(1 << 55 / 3.0) is indeed greater than the absolute tolerance specified.

I don't experience this on Linux.