Tests failing on main

[emlys]

Tests are failing on macOS, all python versions. Probably due to some dependency update.

FAILED tests/test_coastal_vulnerability.py::CoastalVulnerabilityTests::test_aoi_invalid_geometry - RuntimeWarning: invalid value encountered in intersection
FAILED tests/test_ndr.py::NDRTests::test_masking_invalid_geometry - AssertionError: 
Arrays are not equal

Mismatched elements: 1 / 1 (100%)
Max absolute difference: 1
Max relative difference: inf
 x: array([[1]])
 y: array([[0]], dtype=uint8)
FAILED tests/test_scenic_quality.py::ViewshedTests::test_primitive_peak - AssertionError: 
Arrays are not equal

Mismatched elements: 4 / 64 (6.25%)
Max absolute difference: 1.
Max relative difference: 1.
 x: array([[1, 1, 1, 1, 1, 1, 1, 1],
       [1, 1, 1, 1, 1, 1, 1, 1],
       [1, 1, 1, 1, 1, 1, 1, 1],...
 y: array([[1., 1., 1., 1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1., 1., 1., 1.],...

[emlys]

This appears to be related to a change in the github actions runner. macos-latest is now macos-14-arm64. Since the difference has to do with ARM64 and/or ARM64 versions of packages, I can't reproduce it locally.

[emlys]

Coastal Vulnerability test failure

This was failing due to a RuntimeWarning coming from shapely. This looks like a M1 specific bug with shapely or geos? I asked about it here: shapely/shapely#2045
We can add an exception to our pytest configuration so that this specific warning doesn't fail the test.

Scenic Quality test failure

@phargogh was able to reproduce this one locally on his M1 mac. It comes down to a difference in floating point error between Intel and M1. The relevant code is:

invest/src/natcap/invest/scenic_quality/viewshed.pyx

Lines 834 to 841 in 55804f9

	if (adjusted_dem_height >= z and
	target_distance < max_visible_radius and
	target_dem_height != nodata):
	visibility_managed_raster.set(ix_target, iy_target, 1)
	aux_managed_raster.set(ix_target, iy_target, adjusted_dem_height)
	else:
	visibility_managed_raster.set(ix_target, iy_target, 0)
	aux_managed_raster.set(ix_target, iy_target, z)

We have a situation where adjusted_dem_height and z should be equal, but on M1, z has floating point error that makes it greater than adjusted_dem_height, and so the conditional evaluates differently. This is like how need isclose for equality comparison, we need a similar thing for greater/lesser-than-or-equal comparison. Probably worth scanning the rest of our code base for this situation.

NDR test failure

The failing test uses an invalid geometry, which is questionable, but I reproduced the issue with a valid geometry too. The issue comes down to different results from gdal.RasterizeLayer (reported here: OSGeo/gdal#9822). Ultimately, seems to be an upstream issue and we don't need to specifically test this edge case. We could test with an invalid geometry that more obviously overlaps the centerpoint, or remove this test case altogether - we don't need to support invalid geometries anyway.