unit test for plot2D based on image hashes

[oskooi]

plot2D seems to no longer work correctly for cylindrical coordinates. As an example, python/examples/zone_plate.py of Tutorial/Focusing Properties of a Binary-Phase Zone Plate produces the wrong output:

This particular example was most recently corrected in #2136:

From the images, the problem seems to be that the $R$ axis is not properly defined: it should be the interval (0,R) rather than (-R/2,+R/2).

[oskooi]

As a second check, I rebuilt from the master branch (b86b476) and then reran the example above. The output is correct. Not sure why the output was incorrect previously. There is no bug in plot2D after all.

Anyhow, it would be useful to add a set of unit tests for plot2D. It seems there are unit tests for plot2D in test_visualization.py but they do not actually do anything:

meep/python/tests/test_visualization.py

Lines 187 to 213 in b86b476

	def test_plot2D(self):
	# Check plotting of geometry with several sources, monitors, and PMLs
	f = plt.figure()
	ax = f.gca()
	sim = setup_sim()
	ax = sim.plot2D(ax=ax)
	if mp.am_master():
	hash_figure(f)
	# self.assertAlmostEqual(hash_figure(f),10231488)
	# Check plotting of fields after timestepping
	f = plt.figure()
	ax = f.gca()
	sim.run(until=200)
	ax = sim.plot2D(ax=ax, fields=mp.Ez)
	if mp.am_master():
	hash_figure(f)
	# self.assertAlmostEqual(hash_figure(f),79786722)
	# Check output_plane feature
	f = plt.figure()
	ax = f.gca()
	vol = mp.Volume(center=mp.Vector3(), size=mp.Vector3(2, 2))
	ax = sim.plot2D(ax=ax, fields=mp.Ez, output_plane=vol)
	if mp.am_master():
	hash_figure(f)
	# self.assertAlmostEqual(hash_figure(f),68926258)

Note that the assertAlmostEqual statements are all commented out.

@smartalecH, do we just need to update the hash values in this test in order to reenable it?

[smartalecH]

IIRC, the hash value checks were rather flaky. Essentially the same plots were producing different hashes depending on the CI build machine etc (back in the days of Travis).

We could once again look into how matplotlib does CI (which is how we originally came up with the hash comparisons). I'm sure some new techniques have emerged in the last five years. Maybe even some native to matplotlib itself.

[oskooi]

We could once again look into how matplotlib does CI (which is how we originally came up with the hash comparisons). I'm sure some new techniques have emerged in the last five years. Maybe even some native to matplotlib itself.

Seems there is indeed support within matplotlib for comparing images as part of a unit test:

https://matplotlib.org/stable/devel/testing.html#writing-an-image-comparison-test