@timhoffm Can you add a test so this doesn't break in the future? Could probably just check that the call doesn't error...

@timhoffm Does this make it through to plot, and can a test be added to test_axes.py?

plot(np.array([]), np.array([]), drawstyle="steps")

Yes it does. I've intentionally only added unit tests for the pts_to_*step functions. These were the ones failing. We shouldn't add full integration tests for all possible inputs. This really slows the test suite down.

Ok I’m not sure what the testing philosophy is/should be. To me if it broke once it’ll break again and having a test will stop that. If you can make the test a non-image test I don’t think it takes very long.

The things that was broken, were the pts_to_*step functions. And they are tested now, so they can't break again.

Unfortunately, every axes instantiation is quite slow #6664. For a single plot, this doesn't take too long, but it really adds up for several thousand tests.

Just to carry on this discussion: to me, its the user-facing features that need to keep working. Someone could decide to do away with pts_to_pre etc, and implement steps= in a different way, but the user-experience in plot shouldn't change (or at least we should know it breaks a test). But I'm pretty new here, so please don't take that as our policy!

I think the test is fine.
Thanks!