Fix RXFilter.node_ids() dropping selected nodes with include flags

src/tracksdata/graph/_rustworkx_graph.py

@@ -199,26 +199,34 @@ def _current_node_ids(self) -> list[int]:
 
     @cache_method
     def node_ids(self) -> list[int]:
-        # if there are no edge filters, we can return the current node ids
+        # if there are no edge filters nor include flags, we can return the current node ids
         if not self._edge_attr_comps and (not self._include_targets and not self._include_sources):
             return self._current_node_ids()
 
-        # find nodes that are connected to edges that pass the edge filters
+        edges_df = self._edge_attrs()
+        node_filtered = self._node_ids is not None or bool(self._node_attr_comps)
+
         node_ids = []
-        edge_node_ids = (
-            self._edge_attrs()
-            .select(
-                DEFAULT_ATTR_KEYS.EDGE_SOURCE,
-                DEFAULT_ATTR_KEYS.EDGE_TARGET,
+        if node_filtered or not self._edge_attr_comps:
+            # nodes selected by `node_ids`/node filters (or all nodes when
+            # unfiltered) are always kept; the include flags only extend the
+            # selection with the respective edge endpoints, matching SQLFilter.
+            node_ids.append(np.asarray(self._current_node_ids(), dtype=int))
+        else:
+            # only edge filters: nodes are the endpoints of the matching edges
+            node_ids.append(
+                edges_df.select(
+                    DEFAULT_ATTR_KEYS.EDGE_SOURCE,
+                    DEFAULT_ATTR_KEYS.EDGE_TARGET,
+                )
+                .to_numpy()
+                .ravel()
             )
-            .to_numpy()
-            .ravel()
-        )
-        node_ids.append(edge_node_ids)
 
-        if self._node_attr_comps:
-            # if there are node filters, we need to add the nodes that pass the node filters
-            node_ids.append(self._current_node_ids())
+        if self._include_sources:
+            node_ids.append(edges_df[DEFAULT_ATTR_KEYS.EDGE_SOURCE].to_numpy())
+        if self._include_targets:
+            node_ids.append(edges_df[DEFAULT_ATTR_KEYS.EDGE_TARGET].to_numpy())
 
         node_ids = [v for v in node_ids if len(v) > 0]
 

src/tracksdata/graph/_test/test_graph_backends.py

@@ -726,6 +726,29 @@ def test_edge_attrs_include_targets(graph_backend: BaseGraph) -> None:
     assert single_exclusive_edge_ids == expected_single_exclusive, msg
 
 
+def test_filter_node_ids_with_include_flags(graph_backend: BaseGraph) -> None:
+    """Selected nodes must be kept when include flags extend the selection."""
+    graph_backend.add_edge_attr_key("weight", dtype=pl.Float64)
+
+    node0 = graph_backend.add_node({"t": 0})
+    node1 = graph_backend.add_node({"t": 1})
+    isolated = graph_backend.add_node({"t": 0})
+
+    graph_backend.add_edge(node0, node1, attrs={"weight": 0.5})
+
+    # explicitly selected nodes without edges must not be dropped
+    node_ids = graph_backend.filter(node_ids=[node0, isolated], include_targets=True).node_ids()
+    assert sorted(node_ids) == sorted([node0, node1, isolated])
+
+    # node attribute filters behave the same way
+    node_ids = graph_backend.filter(NodeAttr("t") == 0, include_targets=True).node_ids()
+    assert sorted(node_ids) == sorted([node0, node1, isolated])
+
+    # include_sources only extends with edge sources, not targets
+    node_ids = graph_backend.filter(node_ids=[node0, isolated], include_sources=True).node_ids()
+    assert sorted(node_ids) == sorted([node0, isolated])
+
+
 def test_from_ctc(
     ctc_data_dir: Path,
     graph_backend: BaseGraph,