Fix issue where sectors showed a straight line instead of a curved #266
Conversation
|
I added 5102055 because I have a hard time running the tests locally. There is always something with the fonts being slightly different. And it's hard to know why a test failed in CI if you cannot get the comparisons. |
| @@ -38,3 +38,10 @@ jobs: | |||
| run: | | |||
| pytest -s -rxs -vv -Werror tests/ --mpl --mpl-generate-summary=html \ | |||
| --mpl-results-path="windrose_test_output-${{ matrix.os }}-${{ matrix.python-version }}" | |||
| - name: Store mpl-results | |||
| uses: actions/upload-artifact@v4 | |||
| @@ -642,6 +642,8 @@ def bar(self, direction, var, **kwargs): | |||
| zorder=zorder, | |||
| **kwargs, | |||
| ) | |||
| # needed so the the line of the rectangle becomes curved | |||
| patch.get_path()._interpolation_steps = 100 | |||
There was a problem hiding this comment.
Ultimately, if I understand this correctly, this is a hack b/ it is using a private method to please our eyes with something that did not looked curved due to low resolution but was "correct." With that said, I do prefer this behavior even if it is prone to break with future changes in mpl.
There was a problem hiding this comment.
I 100% agree on that being not optimal and originally private, but since it's even in the docs I think that's the only way to get there?
I still do think a straight line is technically incorrect in a polar projection.
There was a problem hiding this comment.
I still do think a straight line is technically incorrect in a polar projection.
For our eyes? Yes. I agree. But in theory the "data" is correct otherwise the higher interpolation steps would still produce a straight line, right?
Anyway, I did not mean that this was the problem as for a graph what we see is the important part. I think that mpl is probably choosing non-ideals defaults for us in this case.
resolves #137
I think I found the solution by looking at what the usual
.bar()does. We need to set._interpolation_stepsto a value > 1 (https://matplotlib.org/stable/api/projections/polar.html#matplotlib.projections.polar.PolarTransform).I updated the baseline images affected, since we were (sometimes) initially testing the "wrong" behavior - there was also some test pollution going on making some plots curved others straight.
I think the only correct behavior are curved lines because in a polar coordinate system straight lines would mean varying values depending on where you read the value from.
Happy to hear your thoughts on this