BUG: avoid invalid bbox for empty GeoDataFrame in to_parquet (#2761)

CHANGELOG.md

@@ -17,6 +17,8 @@ Bug fixes:
 - Ensure that GeoDataFrame created from DataFrame is a copy, not a view (#2667)
 - Fix mismatch between geometries and colors in ``plot()`` if an empty or missing
   geometry is present (#2224)
+- Fix `to_parquet`/`to_feather` to not write an invalid bbox (with NaNs) in the
+  metadata in case of an empty GeoDataFrame (#2653)
 
 Notes on (optional) dependencies:
 

geopandas/io/arrow.py

@@ -2,6 +2,7 @@ from packaging.version import Version
 import json
 import warnings
 
+import numpy as np
 from pandas import DataFrame, Series
 
 from geopandas._compat import import_optional_dependency
@@ -108,9 +109,13 @@ def _create_metadata(df, schema_version=None):
             "geometry_type": geometry_types[0]
             if len(geometry_types) == 1
             else geometry_types,
-            "bbox": series.total_bounds.tolist(),
         }
 
+        bbox = series.total_bounds.tolist()
+        if np.isfinite(bbox).all():
+            # don't add bbox with NaNs for empty / all-NA geometry column
+            column_metadata[col]["bbox"] = bbox
+
     return {
         "primary_column": df._geometry_column_name,
         "columns": column_metadata,

geopandas/io/tests/test_arrow.py

@@ -653,6 +653,20 @@ def test_write_read_feather_expand_user():
     os.remove(os.path.expanduser(test_file))
 
 
+@pytest.mark.parametrize("geometry", [[], [None]])
+def test_write_empty_bbox(tmpdir, geometry):
+    # empty dataframe or all missing geometries -> avoid bbox with NaNs
+    gdf = geopandas.GeoDataFrame({"col": [1] * len(geometry)}, geometry=geometry)
+    gdf.to_parquet(tmpdir / "test.parquet")
+
+    from pyarrow.parquet import read_table
+
+    table = read_table(tmpdir / "test.parquet")
+    metadata = json.loads(table.schema.metadata[b"geo"])
+    assert "encoding" in metadata["columns"]["geometry"]
+    assert "bbox" not in metadata["columns"]["geometry"]
+
+
 @pytest.mark.parametrize("format", ["feather", "parquet"])
 def test_write_read_default_crs(tmpdir, format):
     if format == "feather":