test_hdx_rgb_roundtrip fail son osx with python 3.8

[step21]

Description

osx python 3.8
test_hdx_rgb_roundtrip

Way to reproduce

run tests on osx with python 3.8

Version information

python 3.8
conda forge scikit-image version

[step21]

_____________________ TestColorconv.test_hdx_rgb_roundtrip _____________________

self = <skimage.color.tests.test_colorconv.TestColorconv testMethod=test_hdx_rgb_roundtrip>

    def test_hdx_rgb_roundtrip(self):
        from skimage.color.colorconv import hdx_from_rgb, rgb_from_hdx
        img_rgb = self.img_rgb
        conv = combine_stains(separate_stains(img_rgb, hdx_from_rgb),
                              rgb_from_hdx)
>       assert_equal(img_as_ubyte(conv), img_rgb)
E       AssertionError: 
E       Arrays are not equal
E       
E       Mismatched elements: 420 / 411810 (0.102%)
E       Max absolute difference: 253
E       Max relative difference: 9.66666667
E        x: array([[[0, 0, 0],
E               [0, 0, 0],
E               [0, 0, 0],...
E        y: array([[[0, 0, 0],
E               [0, 0, 0],
E               [0, 0, 0],...

../_test_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_p/lib/python3.8/site-packages/skimage/color/tests/test_colorconv.py:203: AssertionError

https://dev.azure.com/conda-forge/feedstock-builds/_build/results?buildId=169451&view=logs&j=a8dee1cf-d17f-5cb0-b53f-4a93dfdbfb0c&t=d9dbf137-9a7f-5666-c60c-9e1b2fc31445&l=6527

[rfezzani]

Hi @step21 and sorry for the toooooooooooo late answer, this issue passed under my radar 😕.
Do you still face the problem? I have no mac to test it by my self....

[step21]

Hi. No worries, this was mainly while testing/building for conda-forge. The relevant test(s) are still commented out: https://github.com/conda-forge/scikit-image-feedstock/blob/master/recipe/meta.yaml
Though I do not know if it still happens. For testing, I could offer to make a PR in the scikit-image-feedstock repo, with the relevant test(s) not skipped. But not sure how helpful that would be for debugging, as you couldn't test fixes. I could also test fixes locally, but this might also be frustrating for both of us :) Soooo I am assuming it still happens, but what would be the best way to work on fixing it?

[grlee77]

It wonder if this was fixed by #5172 (which added clipping of any negative values: https://github.com/scikit-image/scikit-image/pull/5172/files#diff-a5b79edf6268ee850675946c603a0c03fac862574ab9ac2a661031049ea628e3R1454). I don't think that PR was backported to the 0.18.x branch, though.