convex_hull gives incorrect polygon near poles, maybe

[spenczar]

I am seeing strange behavior with polygon point containment near the poles.

Here's an example. Build a convex hull from points which swing around the north pole at latitude 80:

positions = [radec_to_vector(x, 80) for x in range(360)]
c_hull = SphericalPolygon.convex_hull(positions)

Does the hull contain the north pole? It should...

>>> c_hull.contains_point((0, 0, 1))
False

This looks incorrect to me. Minor perturbations of the north pole sometimes work, and sometimes don't. It seems very inconsistent.

[mcara]

Thank you for reporting this. Indeed, after taking a look at the code, this is a limitation of the algorithm. If you modify positions in this way (as an example):

positions = [radec_to_vector(x, 80 + x / 360) for x in range(360)]

the code will work correctly. With your original example, all points are at the same "distance" (in degrees) from the pole and this is where the algorithm fails. You could try applying a rotation to the x, y, z coordinates of your points so that the North pole ((0, 0, 1)) is not inside the convex hull.