Bug summary

Trying to create an arc path of 0/360 degrees sometimes results in one of 360/0 degrees due to floating point weirdness.

Code for reproduction

from matplotlib import patches, path, pyplot

fig, ax = pyplot.subplots(ncols=3)
ax[0].add_patch(patches.PathPatch(path.Path.arc(theta1=-90 - 1e-14, theta2=270)))
ax[0].set_title("arc(-90-1e-14, 270), should be a circle")
ax[1].add_patch(patches.PathPatch(path.Path.arc(theta1=-90, theta2=270)))
ax[1].set_title("arc(-90, 270), is a circle")
ax[2].add_patch(patches.PathPatch(path.Path.arc(theta1=-90, theta2=-90 - 1e-14)))
ax[2].set_title("arc(-90, -90-1e-14), should not be a circle")
for a in ax:
    a.set_xlim(-1, 1)
    a.set_ylim(-1, 1)
    a.set_aspect("equal")
pyplot.show()

Actual outcome

Expected outcome

I expect the first plot to be a circle, identical to the second.
I expect the third plot to be empty

Additional information

The location of the bug is not hard to find:

At runtime the first example becomes (-90 <= -90 - 1e-14) == False, which therefore skips the +360 bump.

What is much harder is to understand is what the intention is here.
If you can help me to understand what behaviour you'd like, I'd be happy to put together a PR for it.

I think theta2 < theta1 always gives an unexpected result:

• The docstring does not clarify what should happen, therefore my assumption is that it should return an anti-clockwise arc
• The np.floor will mean it becomes the angle range [theta1, theta2+360], which gives the third plot which I believe should not be a circle
• The special casing only treats eta2 += 360, not eta2 -= 360

It would be great if either the docstring could be clarified or a ValueError raised on unexpected inputs here.

Operating system

No response

Matplotlib Version

3.10.0

Matplotlib Backend

No response

Python version

No response

Jupyter version

No response

Installation

pip