Deprecation fix mpl 3.10 and beyond

[cvanelteren]

There are a lot of deprecation warnings from MPL and or Ultraplot existing in the current build. The majority of these fixes can be addressed by modifying the tests. Some of the deprecation warnings ensures keywords of usage to be compatible with mpl 3.10 which will result in errors in later versions.

There are a few errors I have not addressed that warrants some discussion.

• The first one

  ultraplot/tests/test_1dplots.py::test_boxplot_vectors
  /home/casper/micromamba/envs/ultraplot-dev/lib/python3.11/site-packages/pytest_mpl/plugin.py:125: UltraPlotWarning: Passing lists to fillalpha was removed in v0.9. Please specify different opacities using the property cycle colors instead.
      store.return_value[test_name] = obj(*args, **kwargs)

• The second one

  ultraplot/tests/test_subplots.py::test_axis_sharing[labels]
  /home/casper/micromamba/envs/ultraplot-dev/lib/python3.11/site-packages/pytest_mpl/plugin.py:125: UltraPlotWarning: Calling arbitrary axes methods from SubplotGrid was deprecated in v0.8 and will be removed in a future release. Please index the grid or loop over the grid instead.
  store.return_value[test_name] = obj(*args, **kwargs)

For both of these, I don't agree with the deprecation warning. The first issue indicates that we should move towards a cycler object. However, adjusting the subplot to match the target baseline cannot be readily achieved through the cycler object. It requires setting the patches to a different alpha and leaving the lines with alpha = 1. This would (I think) require an extension of what the cycler object can handle which is to set all potential properties of all the artists. I don't think it can currently do this.

The second error implies the removal of multi-axis formatting. Personally, I would be against removing this feature as it is one of the most powerful features of UltraPlot.

[cvanelteren]

Making this a draft as we have not updated to 3.10 yet.

[cvanelteren]

Note I had to rename the warning UltraplotWarning to UltraPlotWarning due to some weird pytest-xdist issue.

[cvanelteren]

@beckermr

I had some spare time so pushed this along. There is one warning I could use your input for. There is a warning for the boxplot that keyword fillalphas is removed in 0.9 for proplot. I cannot determined directly why this was deprecated from the source but I have a hunch.

The warning implies that properties have to be fed through the Cycle object. However, the source is currently geared towards dealing with line properties only, setting color, alpha and marker style accordingly. This creates a stress point as the deprecation warning is not compatible with the current implementation of the source.

The major problem is that it is not trivial to determine from a singular entry point how to cycle general objects. Line objects have different properties from boxplots, and contours. As such, setting all the properties through a general cycler won't be possible unless we write the code for this. There are some properties that can be leveraged for this in ultraplot.internals.__init__.py that categorize the properties for the major artist types.

In short, we either
a) allow for the fillalphas to remain
b) extend the cycler object to allow cycling for different parts of the plot. For example we could have an internal map to line, patch, fliers, medians etc, and cycle the properties accordingly.

For b however, retooling would be necessary that will touch much of the codebase and I a don't know what will be gained tbh. I could be missing the point of the warnings and Luke's intentions, hence this msg to get some input on this issue. Happy to hear your thoughts.

P.S. This warning can be found by running ultraplot/tests/test_1dplots.py::test_boxplot_vectors (for example).

[beckermr]

I left some comments.

[cvanelteren]

Hi @beckermr thanks for the comment but they don't address the things I outlined.

[beckermr]

Yes I know, sorry. I don't fully understand why things were deprecated either. I think not deprecating it fow now is fine.

See comment

[cvanelteren]

Tests look as they should and fail on newly added once or correct existing errors.

[cvanelteren]

Can we go ahead an merge this @beckermr?

[beckermr]

I had a few more comments, plus there are test failures in the plots that do not look like we expect them.

[beckermr]

Also the docs build is failing.

[cvanelteren]

Will fix this and perhaps split this PR up for clarity.

[beckermr]

This continue statement needs more discussion.

[beckermr]

Code looks good, though I do not understand all of the changes.

The test failures need fixing, and then I think we can merge.

[cvanelteren]

Ok merging this as the newer plots more correct. We may need to revert the draw in fig.show as it caused the image sizes to differ.

[beckermr]

I was about to comment that the newer plots were more correct as well.

If you want code review, please wait until the reviewer approves before merging.

[cvanelteren]

Ah ok I gathered it was ok. Will wait next time.