Anyone interested in moving this forwards? I think it is good to go...

I am not seeing why the test image should have changed.

I am not seeing why the test image should have changed.

Fair comment. This is the result I get when running the tests with anything since the released v1.3.1. Do you get the original image? I notice that our default tolerance is 13 - that is the RMS pixel difference is 13. It is feasible that the two images are different to within this rms value.

On master I get the new image, have not tried to track it back to where it changed.

I am not sure if I think it is worth figuring out what changed (no one has complained), but it does point out a weakness in the image comparison. Maybe we should also add a check on the number of pixel different by greater than a threshold ie assert sum(np.abs(im1 - im2) > threshold) < N.

Maybe we should also add a check on the number of pixel different by greater than a threshold

I've been thinking exactly the same thing in the last couple of days. I think that would be a sensible complimentary approach. Perhaps not in this PR though 😉

No, not in this PR. I'll push the big green button and then get a PR for the new checks going (and I now see why getting the compare.py pep8 in was a good thing).