TODO: add screenshot-based tests

[tlambert03]

a testing tool that would be particularly important is to setup a framework for screenshot based testing. both vispy and pygfx have examples of doing this (recommend looking through pygfx's though). Basic idea is that we commit, somewhere, a folder of "expectations" for how something should look. Something we have manually checked at one point, and then tests confirm that things continue to look the same.

[gselzer]

Hmm, out of curiosity, what do you think the benefit is here over explicit comparison with the backend values?

In other words, I presume that since pygfx/vispy do this, we can be reasonably certain that if e.g. a pygfx object has a transformation matrix M, it will always look the same way on the canvas/screenshot. So why can't we just check that if our pygfx-backed scenex Image node always has a transformation matrix N, then associated pygfx object always has a transformation matrix M? What are we missing out on?

[tlambert03]

Visual confirmation is the end all be all, and requires no programmatic backend specific check.

For example, if you wanted to make sure that adding a node to a graph does actually show the object on the screen, you would either need to write backend specific test runners that translate the intention of the test (is visible, is red, is over here, etc) to a backend-specific, often private api.

If, however, you write a scenex api test, check the image and say "yep that's correct", then screenshot comparisons are excellent ways to confirm that expectations remain met, even after refactoring, without backend specific logic

[tlambert03]

(There are tons of these sorts of tests in napari too... we ultimately found it very necessary)