ST_Covers and ST_Contains output does not match DuckDB or Shapely

[utkarshkukreti]

Hi,

I was porting some DuckDB and Shapely code to SedonaDB to see if I could get better performance. While it does run faster, I found that the output of ST_Covers was not identical to PostGIS, DuckDB and Shapely in some cases.

I narrowed it down to the case when the child geometry is identical to a hole of the parent geometry.

Here's the smallest reproduction I could create:

# /// script
# requires-python = ">=3.14"
# dependencies = [
#     "apache-sedona[db]>=1.8.1",
#     "duckdb>=1.5.0",
#     "shapely>=2.1.2",
# ]
# ///

import duckdb
import sedona.db
import shapely

parent_wkt = "POLYGON ((0 0, 3 0, 3 3, 0 3, 0 0), (1 1, 1 2, 2 1, 1 1))"
child_wkt = "POLYGON ((1 1, 1 2, 2 1, 1 1))"

duckdb_connection = duckdb.connect()
duckdb_connection.execute("install spatial")
duckdb_connection.execute("load spatial")

sedona_connection = sedona.db.connect()

parent_geometry = shapely.from_wkt(parent_wkt)
child_geometry = shapely.from_wkt(child_wkt)

for predicate, shapely_value in [
    ("ST_Intersects", shapely.intersects(parent_geometry, child_geometry)),
    ("ST_Covers", shapely.covers(parent_geometry, child_geometry)),
    ("ST_Contains", shapely.contains(parent_geometry, child_geometry)),
]:
    duckdb_value = duckdb_connection.execute(
        f"select {predicate}(ST_GeomFromText('{parent_wkt}'), ST_GeomFromText('{child_wkt}'))"
    ).fetchone()[0]
    sedona_value = sedona_connection.sql(
        f"select {predicate}(ST_GeomFromWKT('{parent_wkt}'), ST_GeomFromWKT('{child_wkt}')) as value"
    ).to_pandas()["value"][0]
    print(
        f"{predicate} duckdb={duckdb_value} sedona={sedona_value} shapely={shapely_value}"
    )

Output:

ST_Intersects duckdb=True sedona=True shapely=True
ST_Covers duckdb=False sedona=True shapely=False
ST_Contains duckdb=False sedona=True shapely=False

Postgis:

postgres=# SELECT
  ST_Intersects(parent, child) AS intersects,
  ST_Covers(parent, child) AS covers,
  ST_Contains(parent, child) AS contains
FROM (
  SELECT
    'POLYGON ((0 0, 3 0, 3 3, 0 3, 0 0), (1 1, 1 2, 2 1, 1 1))'::geometry AS parent,
    'POLYGON ((1 1, 1 2, 2 1, 1 1))'::geometry AS child
) t;
 intersects | covers | contains
------------+--------+----------
 t          | f      | f
(1 row)

ST_Covers and ST_Contains return true with SedonaDB but false with the other 3.

[paleolimbot]

Thanks for opening! There's one other case we know about too: Contains(GEOMETRYCOLLECTION (POINT (0 0), LINESTRING (0 0, 0 1)), POINT (0 0)) should evaluate to false, but actually evaluates to true ( tidwall/tg#20 ).

We use tg to evaluate predicates because it's faster and because it has a thread-safe prepared geometry (the same reason why DuckDB's DWithin is faster). I'll add this one to that issue and see if we can fix upstream.