ST_ClusterKMeans: handle effective K=0 when all the inputs are EMPTY

Closes #4101
Closes #254



git-svn-id: http://svn.osgeo.org/postgis/trunk@16605 b70326c6-7e19-0410-871a-916f4a2858ee

liblwgeom/lwkmeans.c

@@ -135,20 +135,8 @@ kmeans_init(POINT2D** objs, int* clusters, uint32_t n, POINT2D** centers, POINT2
 	double max_dst = -1;
 	double dst_p1, dst_p2;
 
-	assert(k > 0);
-
-	/* k = 1: first non-null is ok, and input check guarantees there's one */
-	if (k == 1)
-	{
-		for (i = 0; i < n; i++)
-		{
-			if (!objs[i]) continue;
-			centers_raw[0] = *((POINT2D *)objs[i]);
-			centers[0] = &(centers_raw[0]);
-			return;
-		}
-		assert(0);
-	}
+	/* k=0, k=1: "clustering" is just input validation */
+	assert(k > 1);
 
 	/* k >= 2: find two distant points greedily */
 	for (i = 1; i < n; i++)
@@ -333,9 +321,24 @@ lwgeom_cluster_2d_kmeans(const LWGEOM** geoms, uint32_t n, uint32_t k)
 		k = num_non_empty;
 	}
 
-	kmeans_init(objs, clusters, n, centers, centers_raw, k);
-
-	result = kmeans(objs, clusters, n, centers, k);
+	if (k > 1)
+	{
+		kmeans_init(objs, clusters, n, centers, centers_raw, k);
+		result = kmeans(objs, clusters, n, centers, k);
+	}
+	else
+	{
+		/* k=0: everythong is unclusterable
+		 * k=1: mark up NULL and non-NULL */
+		for (i = 0; i < n; i++)
+		{
+			if (k == 0 || !objs[i])
+				clusters[i] = KMEANS_NULL_CLUSTER;
+			else
+				clusters[i] = 0;
+		}
+		result = LW_TRUE;
+	}
 
 	/* Before error handling, might as well clean up all the inputs */
 	lwfree(objs);

regress/cluster.sql

@@ -50,3 +50,13 @@ select '#4100a', count(distinct result) from (SELECT ST_ClusterKMeans(foo1.the_g
 	( ST_GeomFromEWKT('SRID=4326;POLYGON((-71.1261 42.2703 1,-71.1257 42.2703 1,-71.1257 42.2701 1,-71.126 42.2701 1,-71.1261 42.2702 1,-71.1261 42.2703 1))') ) ) As g(geom) CROSS JOIN generate_series(1,3) As i GROUP BY i )) As foo1 LIMIT 10) kmeans;
 
 select '#4100b', count(distinct cid) from (select ST_ClusterKMeans(geom,2) over () as cid from (values ('POINT(0 0)'::geometry), ('POINT(0 0)')) g(geom)) kmeans;
+
+
+select '#4101a', count(distinct result) from (SELECT ST_ClusterKMeans(foo1.the_geom, 3) OVER()  As result
+							FROM ((SELECT ST_GeomFromText('POINT EMPTY',4326) As the_geom
+			UNION ALL SELECT ST_GeomFromText('MULTIPOINT EMPTY',4326) As the_geom
+			UNION ALL SELECT ST_GeomFromText('MULTIPOLYGON EMPTY',4326) As the_geom
+			UNION ALL SELECT ST_GeomFromText('LINESTRING EMPTY',4326) As the_geom
+			UNION ALL SELECT ST_GeomFromText('MULTILINESTRING EMPTY',4326) As the_geom ) ) As foo1 LIMIT 10) kmeans;
+
+select '#4101b', count(distinct cid) from (select ST_ClusterKMeans(geom,2) over () as cid from (values ('POINT EMPTY'::geometry), ('POINT EMPTY')) g(geom)) kmeans;

regress/cluster_expected

@@ -34,3 +34,7 @@ NOTICE:  kmeans_init: there are at least 3 duplicate inputs, number of output cl
 #4100a|1
 NOTICE:  kmeans_init: there are at least 2 duplicate inputs, number of output clusters may be less than you requested
 #4100b|1
+NOTICE:  lwgeom_cluster_2d_kmeans: number of non-empty geometries is less than the number of clusters requested, not all clusters will get data
+#4101a|1
+NOTICE:  lwgeom_cluster_2d_kmeans: number of non-empty geometries is less than the number of clusters requested, not all clusters will get data
+#4101b|1