getNumGeometries with empty geometries

[mwtoews]

GEOS' getNumGeometries is inconsistent with empty geometry types. Investigating via Shapely:

import shapely
import numpy as np

shapely.geos_version_string  # 3.12.0
geom_types = [
    "POINT", "LINESTRING", "POLYGON", "LINEARRING",
    "GEOMETRYCOLLECTION", "MULTIPOINT", "MULTILINESTRING", "MULTIPOLYGON"]
empties = shapely.from_wkt(np.char.add(geom_types, " EMPTY"))
dict(zip(geom_types, shapely.get_num_geometries(empties)))
shapely.get_num_geometries(empties)  # array([1, 1, 1, 1, 0, 0, 0, 0], dtype=int32)
dict(zip(geom_types, shapely.get_num_geometries(empties)))
# shows
{'POINT': 1,
 'LINESTRING': 1,
 'POLYGON': 1,
 'LINEARRING': 1,
 'GEOMETRYCOLLECTION': 0,
 'MULTIPOINT': 0,
 'MULTILINESTRING': 0,
 'MULTIPOLYGON': 0}

PostGIS's documentation for ST_NumGeometries does not document the behaviour with EMPTY geometries, but these counts are all zero:

SELECT ST_AsText(postgis_typmod_type(i) || ' EMPTY') AS empty_geom,
    ST_NumGeometries((postgis_typmod_type(i) || ' EMPTY')::geometry)
FROM generate_series(4, 63, 4) AS i;

        empty_geom        | st_numgeometries 
--------------------------+------------------
 POINT EMPTY              |                0
 LINESTRING EMPTY         |                0
 POLYGON EMPTY            |                0
 MULTIPOINT EMPTY         |                0
 MULTILINESTRING EMPTY    |                0
 MULTIPOLYGON EMPTY       |                0
 GEOMETRYCOLLECTION EMPTY |                0
 CIRCULARSTRING EMPTY     |                0
 COMPOUNDCURVE EMPTY      |                0
 CURVEPOLYGON EMPTY       |                0
 MULTICURVE EMPTY         |                0
 MULTISURFACE EMPTY       |                0
 POLYHEDRALSURFACE EMPTY  |                0
 TRIANGLE EMPTY           |                0
 TIN EMPTY                |                0
(15 rows)

[dr-jts]

Why do you say this is inconsistent? An empty atomic geometry is still a geometry, so can argue that getNumGeometries == 1 is correct.

I'm not sure there is a "right" answer to this. It just has to be documented behaviour.

[mwtoews]

Agree the best fix is to document the differences. I forgot there is a similar discussion for PostGIS ~2 years ago https://trac.osgeo.org/postgis/ticket/4736

More importantly, the number of geometries needs to be correct when they are unpacked, which is OK with GEOS.

See also https://trac.osgeo.org/postgis/ticket/5483 to improve docs for PostGIS.

[mwtoews]

An analogy that I can think of is that you can count one empty bottle, or zero bottles in an empty crate. But I'll probably document it differently than that.