importing napari-matplotlib modifies how plots are shown in jupyter notebooks

[haesleinhuepf]

... and it should not.

The visualization of plots in matplotlib change after napari_matplotlib has been imported. Global settings e.g. here should not be modified by a napari-plugin. Consider programming napari-plugins side-effect free.

To reproduce:

import matplotlib.pyplot as plt
import numpy as np

image = np.random.random((3,3))
plt.imshow(image)

import napari_matplotlib
plt.imshow(image)

The same can be observed when not importing napari_matplotlib but instead opening a viewer where napari-stress is installed, because starting the viewer imports napari_stress and this imports napari_matplotlib in the background. A napari-plugin should not modify how things are displayed in jupyter notebooks. It's kind of evil usage of global variables.

CC @jo-mueller

[haesleinhuepf]

We had a similar issue earlier in another plugin and this is how we solved it:

https://github.com/BiAPoL/napari-clusters-plotter/blob/b917c1c7e01acf28b24fdbe64913e20588bed642/napari_clusters_plotter/_Qt_code.py#L257-L274

I hope this helps! :-)

[jo-mueller]

Hi @haesleinhuepf ,

I think a similar issue has been discussed here and I thought it was also seeb as not desirable to change the global matplotlib behavior. I've seen the same thing happening for other plots (napari-style color scheme) so the global variables must have been set somewhere else, anyway.

Another quick fix that I use a lot would be:

import matplotlib as mpl
mpl.style.use('default')

[haesleinhuepf]

That's a great workaround. Thanks, @jo-mueller .

so the global variables must have been set somewhere else, anyway.

Can you give an example and create a GitHub issue on the affected project?

[jo-mueller]

@haesleinhuepf

so the global variables must have been set somewhere else, anyway.

I mean more that the matplotlib global variables must have been set somewhere else within the scope of napari-matplotlib. I think here. A workaround would be to set the ax parameters (facecolors, etc) where the canvas.axes objects are created, e.g. here and here.

@dstansby would it be acceptable for you to define the display parameters in the __init__ of the respective "end-user" widget definitions? Doing so would probably also close #36.