remove plt.close() from animations

[JoelLucaAdams]

• Removed closing of figure in animations
• Updated documentation to close figures explicitly

Closing the figure early so that it would not display the first frame in a jupyter notebook or on the documentation is bad practice and I shouldn't have let it through in the first place. This causes a bug when using Tkinter as seen below.

To reproduce this bug using the dataarray.epoch.animate:

import matplotlib
# Force the Tkinter interactive backend
matplotlib.use('TkAgg') 

ds = sdfxr.open_mfdataset("docs/tutorial_dataset_2d")
anim = ds["Derived_Number_Density_Electron"].epoch.animate()

Alternatively if you're more interested in the general bug then you can reproduce this using this simple script:

import matplotlib
# Force the Tkinter interactive backend
matplotlib.use('TkAgg') 
import matplotlib.pyplot as plt
import numpy as np

def trigger_colorbar_crash():
    print("Creating figure...")
    fig, ax = plt.subplots()

    print("Creating dummy 2D data...")
    data = np.random.rand(10, 10)
    plot = ax.pcolormesh(data)

    print("Closing figure prematurely (simulating the bug)...")
    plt.close(fig)

    print("Attempting to add colorbar to the closed figure...")
    # Matplotlib tries to resize the axes for the colorbar,
    # which tells the Tkinter toolbar to update its "Undo/Back" history state.
    # Since plt.close() destroyed the toolbar, it throws a TclError.
    fig.colorbar(plot, ax=ax)

trigger_colorbar_crash()

[LiamPattinson]

Looks okay, but just to be sure, has this been fully tested in Jupyter notebooks and from the terminal? Closing a fig but then returning something that uses it feels a bit sketchy.

[JoelLucaAdams]

Changed it now. We could actually simplify this even more on the documentation by simply calling plt.close() which will close all figs. This would remove the plt.subplots() line and the ax=ax from the animation calls but I think it's better to be explicit here. Thoughts on this @LiamPattinson ?

[LiamPattinson]

Looks good, even if I'm still not 100% clear on what's going on. Is it that Jupyter notebooks keep inserting an implicit plt.show() (or equivalent), which messes up the rendering in the docs?

[LiamPattinson]

Neat, today I learned something

[JoelLucaAdams]

Yay!! I helped :D

[JoelLucaAdams]

Looks good, even if I'm still not 100% clear on what's going on. Is it that Jupyter notebooks keep inserting an implicit plt.show() (or equivalent), which messes up the rendering in the docs?

So from my understanding what happens is that Jupyter will always attempt to render a figure at the end of each code cell but using certain backends such as notebook stops this from happening until you exclusively call plt.show(). https://matplotlib.org/stable/users/explain/figure/figure_intro.html#notebooks-and-ides

[LiamPattinson]

Good to know! I've seen %matplotlib inline so many times, I had no idea there were any alternatives.