FIX: return points rather than path to fix regression

[bsipocz]

Regression in an astropy plotting example (astropy/astropy#8792) is due to #11407

Locally this PR fixes the issue and the test seem to be still working, but I didn't run the whole suite locally.

Please advise what kind of test is preferred to be added for this.

[jklymak]

Can you reproduce the error in matplotlib somehow? I can’t really tell what the issue is from the astropy example.

[bsipocz]

The problem is that we wanted to plot points rather than paths. This example from the issue in fact doesn't use astropy:

import numpy as np
import matplotlib.pyplot as plt

plt.figure(figsize=(8, 4))
plt.subplot(111, projection="aitoff")

x = [0, np.pi/4, np.pi/2]
y = [0, np.pi/4, 3*np.pi/8]

plt.scatter(x, y, color='tab:orange')

plt.plot(x, y,
         marker='o', linestyle='none', markersize=2, alpha=0.3)

[jklymak]

Great - I guess this is fine - if @anntzer had a reason for changing paths to points, I'm sure he'd have written a test to make sure we didn't re-break it on him 😛 But I actually suspect this was just a typo.

I'm not 100% clear why this requires a projection to show up. A test would be very useful - if there was a way to make it not be an image test, even better, but sometimes those are unavoidable.

[anntzer]

I clearly typoed in #11407, sorry about that.

[dstansby]

Is it possible to get the Line2D objects returned by plot() and do some checks there? Otherwise I think this might need an image test (it might be possible to modify an existing custom projection image test to exercise this though).

[tacaswell]

Merged to get the back ports going for 3.1.1, will open an issue to add tests.

[tacaswell]

Thank you @bsipocz !

I am very happy that there is cross-project communication / collaboration going on (which to be fair has mostly been astropy telling Matplotlib we broke you ....).

[bsipocz]

@tacaswell - I'm happy about the widening communication channels, too. I suppose it's a side effect of being upstream that we mostly report about breakage, but at least we start to run into them while testing against the dev version, even though this instance it wasn't the case.

Also, I was looking into to add a test here, too, just haven't yet got to a point to identify what I could use in Line2D that is different. If I don't open a PR by the end of the weekend about it, then I will have given up on it.