Support sjoin on small or empty dataframes

Author: mrocklin

geopandas/__init__.py

@@ -9,6 +9,8 @@
 from geopandas.tools import sjoin
 from geopandas.tools import overlay
 
+from shapely.geometry.base import CAP_STYLE, JOIN_STYLE
+
 import geopandas.datasets
 
 # make the interactive namespace easier to use

geopandas/algos.c

@@ -22,7 +22,7 @@ void strtree_query_callback(void *item, void *vec)
 GEOSSTRtree *create_index(GEOSContextHandle_t handle, GEOSGeometry **geoms, size_t n)
 {
     int i;
-    GEOSSTRtree* tree = GEOSSTRtree_create_r(handle, n);
+    GEOSSTRtree* tree = GEOSSTRtree_create_r(handle, n > 1 ? n : 2);
 
     for (i = 0; i < n ; i++)
     {
@@ -62,6 +62,8 @@ size_vector sjoin(GEOSContextHandle_t handle,
     kv_init(vec);
 
     // begin = clock();
+    if (nright == 0)
+        return out;
 
     GEOSSTRtree* tree = create_index(handle, right, nright);
 

geopandas/tools/tests/test_sjoin.py

@@ -103,3 +103,19 @@ def test_errors():
 
     assert "both" in str(info.value)
     assert "inner" in str(info.value)
+
+
+@pytest.mark.parametrize('l', [0, 1, 5])
+@pytest.mark.parametrize('r', [0, 1, 5])
+def test_small(l, r):
+    left = gpd.GeoDataFrame({'geometry': triangles,
+                             'x': np.random.random(len(triangles)),
+                             'y': np.random.random(len(triangles))},
+                             index=np.arange(len(triangles)) * 2).iloc[:l]
+
+    right = gpd.GeoDataFrame({'geometry': points,
+                              'x': np.random.random(len(points)),
+                              'z': np.random.random(len(points))},
+                              index=list(string.ascii_lowercase[:len(points)])).iloc[:r]
+
+    result = sjoin(left, right)