Image tests: set agg backend after rcdefaults

[rcomer]

I was following @tkknight's nice new instructions for installing and running tests, but got a bunch of errors and failures from assertBoundsTickLabels and assertPointsTickLabels on the TestGraphicStringCoord class. The errors look like this:

Traceback (most recent call last):
  File "[git-path]/iris/lib/iris/tests/unit/plot/test_plot.py", line 57, in test_xaxis_labels_with_axes
    self.assertBoundsTickLabels("xaxis", ax)
  File "[git-path]/iris/lib/iris/tests/unit/plot/__init__.py", line 45, in assertBoundsTickLabels
    actual = self.tick_loc_and_label(axis, axes)
  File "[git-path]/iris/lib/iris/tests/unit/plot/__init__.py", line 34, in tick_loc_and_label
    axes.figure.canvas.draw)
  File "[site-packages-path]/matplotlib/backends/backend_tkagg.py", line 10, in draw
    _backend_tk.blit(self._tkphoto, self.renderer._renderer, (0, 1, 2, 3))
  File "[site-packages-path]/matplotlib/backends/_backend_tk.py", line 75, in blit
    photoimage.blank()
  File "[py3-path]/tkinter/__init__.py", line 3548, in blank
    self.tk.call(self.name, 'blank')
_tkinter.TclError: invalid command name "pyimage353"

so it seemed the test was attempting to use TkAgg, even though the intent in tests/__init__.py is clearly to use Agg. Switching the order of the backend setting and rcdefaults call fixed the problem.

I'm not clear why this hasn't come up for anyone else. I do have a matplotlibrc file, which probably most devs don't - this file was the reason the rcdefaults call was originally added (#2746).

[trexfeathers]

This is being investigated in #3821 🙂

[trexfeathers]

Sorry shouldn't have closed it, read it as an issue! @stephenworsley does this help fix #3821?

[rcomer]

Thanks @trexfeathers, if I'd noticed that issue I might have got here quicker...!

[stephenworsley]

I think this probably does help fix #3821. Thanks @rcomer!

[bjlittle]

@rcomer Awesome spot 👀

Works for me, tested it locally 👍

Would you mind just adding a comment above matplotlib.use("agg") to ensure that a future developer doesn't move this statement back to being before matplotlib.rcdefaults(), thanks 😀

You could even quote this PR, just to add some connective tissue...

[bjlittle]

Otherwise LGTM 🥳

[bjlittle]

closes #3821

[rcomer]

Thanks @bjlittle, I've added that comment. Also a comment about why the rcdefaults is there for good measure.

[bjlittle]

@rcomer Awesome, thanks 👍