Inconsistency with POINT(nan nan)

[Mike Taves]

Consider two points that use the "NaN" convention to try and create POINT EMPTY. This example uses shapely:

from shapely import wkt
from shapely.geometry import Point

# two different ways to create a point with NaN
point_nan_from_val = Point(float('nan'), float('nan'))
point_nan_from_wkt = wkt.loads('point(nan nan)')

point_nan_from_val.is_empty  # False
point_nan_from_wkt.is_empty  # True

point_nan_from_val.wkt  # POINT (nan nan)
point_nan_from_wkt.wkt  # POINT EMPTY

And a similar example with PyGEOS (master), which side-steps #1048 by ​correctly writing WKB for POINT EMPTY:

import pygeos

point_nan_from_val = pygeos.points(float('nan'), float('nan'))
point_nan_from_wkt = pygeos.Geometry('point(nan nan)')

pygeos.is_empty(point_nan_from_val)  # False
pygeos.is_empty(point_nan_from_wkt)  # True

# Different WKT
pygeos.to_wkt(point_nan_from_val)  # POINT (nan nan)
pygeos.to_wkt(point_nan_from_wkt)  # POINT EMPTY

# Same WKB
pygeos.to_wkb(point_nan_from_val) == pygeos.to_wkb(point_nan_from_wkt)  # True

The two points behave very differently, with different WKT and same WKB (when #1048 is resolved).

Seeing that POINT EMPTY is generally serialised in WKB as POINT(nan nan) (#1005), it could be expected that geometries created with nan coordinate values should be POINT EMPTY too, and not non-empty points with nan coordinates.

[Paul Ramsey <pramsey@…>]

In 693c1aa/git:

Correctly return isEmpty for hand-built empty points, closes #1049