feat(geospatial): accept geopandas GDFs in memtable

ibis/backends/tests/test_generic.py

@@ -1294,6 +1294,19 @@ def test_memtable_column_naming_mismatch(con, monkeypatch, df, columns):
         ibis.memtable(df, columns=columns)
 
 
+@pytest.mark.notyet(
+    ["mssql", "mysql", "exasol", "impala"], reason="various syntax errors reported"
+)
+def test_memtable_from_geopandas_dataframe(con, data_dir):
+    gpd = pytest.importorskip("geopandas")
+    gdf = gpd.read_file(data_dir / "geojson" / "zones.geojson")[:5]
+
+    # Read in memtable
+    t = ibis.memtable(gdf)
+    # Execute a few rows to force ingestion
+    res = con.to_pandas(t.limit(2).select("geometry"))
+
+
 @pytest.mark.notimpl(["oracle", "exasol"], raises=com.OperationNotDefinedError)
 @pytest.mark.notimpl(["druid"], raises=AssertionError)
 @pytest.mark.notyet(

ibis/expr/api.py

@@ -552,6 +552,23 @@ def _memtable_from_polars_dataframe(
     ).to_expr()
 
 
+@_memtable.register("geopandas.geodataframe.GeoDataFrame")
+def _memtable_from_geopandas_geodataframe(
+    data: pl.DataFrame,
+    *,
+    name: str | None = None,
+    schema: SchemaLike | None = None,
+    columns: Iterable[str] | None = None,
+):
+    # The Pandas data proxy and the `to_arrow` method on it can't handle
+    # geopandas geometry columns. But if we first make the geometry columns WKB,
+    # then the geo column gets treated (correctly) as just a binary blob, and
+    # DuckDB can cast it to a proper geometry column after import.
+    wkb_df = data.to_wkb()
+
+    return _memtable(wkb_df, name=name, schema=schema, columns=columns)
+
+
 def _deferred_method_call(expr, method_name, **kwargs):
     method = operator.methodcaller(method_name, **kwargs)
     if isinstance(expr, str):