decrease tol for camera test due to MacOS

[akaszynski]

We recently had a test failure in this run.

=================================== FAILURES ===================================
________________________________ test_distance _________________________________
camera = <Camera at 0x1348840c0>
    def test_distance(camera):
        focal_point = np.random.random(3)
        position = np.random.random(3)
        camera.position = position
        camera.focal_point = focal_point
        assert np.isclose(camera.distance, np.linalg.norm(focal_point - position, ord=2), rtol=1e-8)
        distance = np.random.random()
        camera.distance = distance
>       assert np.isclose(camera.distance, distance)
E       assert False
E        +  where False = <function isclose at 0x118bb1200>(0.0002, 0.00016663273706207793)
E        +    where <function isclose at 0x118bb1200> = np.isclose
E        +    and   0.0002 = <Camera at 0x1348840c0>.distance
tests/test_camera.py:59: AssertionError

This PR simply increases the tolerance for this.

[codecov]

Codecov Report

Merging #2836 (0bcd1e8) into main (5f3da83) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##             main    #2836   +/-   ##
=======================================
  Coverage   94.03%   94.03%           
=======================================
  Files          76       76           
  Lines       16399    16399           
=======================================
  Hits        15420    15420           
  Misses        979      979           

[MatthewFlamm]

I'm not quite sure whether this is a MacOS thing or some pathological combination of random numbers. If the latter, we could use specified values for the test.

[adeak]

There's some really creepy thing going on. Notice that the error comes from this assert:

camera.distance = distance
assert np.isclose(camera.distance, distance)

In any sane setup I'd expect those two be very close! Instead this is what I see, on Linux:

>>> import pyvista as pv
>>> cam = pv.Camera()
>>> cam.distance
1.0
>>> cam.distance = 0.00016663273706207793
>>> cam.distance
0.0002
>>> type(cam.distance)
<class 'float'>
>>> cam.distance - 0.00016663273706207793
3.336726293792208e-05

What's up with this?? This can only be VTK's fault, right? Considering

pyvista/pyvista/plotting/camera.py

Lines 225 to 228 in 36ab8b5

	@distance.setter
	def distance(self, distance):
	"""Set the distance from the camera position to the focal point."""
	self.SetDistance(distance)

[adeak]

VTK seems to set a hard lower bound...

>>> cam.distance = 0.001
>>> cam.distance
0.001
>>> cam.distance = 0.0001
>>> cam.distance
0.0002
>>> cam.distance = 0.00001
>>> cam.distance
0.0002

[MatthewFlamm]

Good catch. It indeed has a limit:

https://github.com/Kitware/VTK/blob/5f855ff8f1237cbb5e5fa55a5ace48149237006e/Rendering/Core/vtkCamera.cxx#L563-L577

This confirms that it is a pathological random number. We should pick a number greater than this limit.

[akaszynski]

Good catch. It indeed has a limit:

https://github.com/Kitware/VTK/blob/5f855ff8f1237cbb5e5fa55a5ace48149237006e/Rendering/Core/vtkCamera.cxx#L563-L577

This confirms that it is a pathological random number. We should pick a number greater than this limit.

This works:

>>> np.isclose(0.0002, 0, atol=0.0002)
True

Since the smallest possible value for distance is 0 and VTK's smallest is 0.0002.

[adeak]

This works:

>>> np.isclose(0.0002, 0, atol=0.0002)
True

Since the smallest possible value for distance is 0 and VTK's smallest is 0.0002.

Yeah, I think this can work.

[adeak]

LGTM!

[adeak]

Let's see if re-approving adds the label.

Edit: nevermind, the labelling action is not merged yet, right?

[akaszynski]

nevermind, the labelling action is not merged yet, right?

It is, but it needs an additional PR.