[Bug]: Figure manager is gone - AttributeError: 'NoneType' object has no attribute 'canvas' about matplotlib-inline HOT 6 CLOSED

Sorry, that return should have been a return manager, I think. Not in front of my computer right now so feel free to make a pr out of the suggestion, or I'll do one when I can.

from matplotlib-inline.

Thx @1kastner for the fix! I just pushed 0.1.6 with this fix to PyPI, should be up on conda-forge soon.

Sorry @sryza for the hassles and thx for the prompt report.

from matplotlib-inline.

No, there aren't - yesterday I was very pressed for time and basically had to decide "do I spend the window I have adding some CI/tests to this, or do I finish this release quickly so the manual fix is out?" And I went for the latter, but given all my preaching about good testing, I felt bad about it :)

This package is tiny and moves very rarely, so in the past it hasn't been too terrible, but I 100% agree we should. Even putting in a few high-level tests in a notebook would be useful.

I'll try to find the time for that now that 0.1.6 is out and we're back where we needed to, but feel free to beat me to it!

from matplotlib-inline.

I completely feel you! Some aspects can be tested better, such as "does this throw an exception in situations it should not", and some worse, such as "does it look beautiful". I have seen projects where they started a pixel-wise comparison of the expected and intended output. I believe humans are better in that task and it might be over the top for this project. But it is one important step to make the decision where to draw the line. I guess unittesting might be helpful but time-consuming as well.

If there is some testing infrastructure in place that the leading maintainers are all happy with, I could add one or two automated tests as well - probably some nbconvert invocations or something similar.

from matplotlib-inline.

Thanks a lot, I have done as you have proposed!

from matplotlib-inline.

@fperez thanks for the fast reaction and no worries. I just checked the repository and I was wondering whether there are any automated checks in place for the code? For example, unittests or Python type hints in combination with code linting might have helped to spot such an issue before release. I am just saying this because many, many people rely on this package.

from matplotlib-inline.