Failing set_precision collapse test with GEOS 3.10.1

[jorisvandenbossche]

From the conda-forge build (conda-forge/pygeos-feedstock#48) rebuilding pygeos with GEOS 3.10.1, one test is failing:

_______ test_set_precision_collapse[geometry8-keep_collapsed-expected8] ________

geometry = <pygeos.Geometry LINEARRING (0 0, 0.1 0, 0.1 0.1, 0 0.1, 0 0)>
mode = 'keep_collapsed'
expected = <pygeos.Geometry LINEARRING (0 0, 0 0, 0 0, 0 0, 0 0)>

....

    def test_set_precision_collapse(geometry, mode, expected):
        """Lines and polygons collapse to empty geometries if vertices are too close"""
        actual = pygeos.set_precision(geometry, 1, mode=mode)
        if pygeos.geos_version < (3, 9, 0):
            # pre GEOS 3.9 has difficulty comparing empty geometries exactly
            # normalize and compare by WKT instead
            assert pygeos.to_wkt(pygeos.normalize(actual)) == pygeos.to_wkt(
                pygeos.normalize(expected)
            )
        else:
            # force to 2D because GEOS 3.10 yields 3D geometries when they are empty.
>           assert_geometries_equal(pygeos.force_2d(actual), expected)
E           AssertionError: 
E           Geometries are not equal:
E           
E           Not equal to tolerance 1e-07
E            x: array(<pygeos.Geometry LINESTRING (0 0, 0 0, 0 0)>, dtype=object)
E            y: array(<pygeos.Geometry LINEARRING (0 0, 0 0, 0 0, 0 0, 0 0)>, dtype=object)

[jorisvandenbossche]

cc @pramsey there is one related change in GEoS 3.10.1: libgeos/geos@14f0a69, although I don't see how that would cause a LinearRing to become a LineString instead

[caspervdw]

This is expected behaviour from GEOS. The test in pygeos has been fixed in #419; I guess we just need to push the next release.

[caspervdw]

Closing as the next release is tracked by #429

[jorisvandenbossche]

Sorry, I actually checked if it was already changed on master, not sure how I missed this :)