Skip to content

A programming framework for agentic AI

Fix GraphFlow cycle detection to properly clean up recursion state (#7026)

Copilot14809f51dd55
diff --git a/python/packages/autogen-agentchat/src/autogen_agentchat/teams/_group_chat/_graph/_digraph_group_chat.py b/python/packages/autogen-agentchat/src/autogen_agentchat/teams/_group_chat/_graph/_digraph_group_chat.py
index b9b607b22..d77b42dd1 100644
--- a/python/packages/autogen-agentchat/src/autogen_agentchat/teams/_group_chat/_graph/_digraph_group_chat.py
+++ b/python/packages/autogen-agentchat/src/autogen_agentchat/teams/_group_chat/_graph/_digraph_group_chat.py
@@ -165,12 +165,13 @@ class DiGraph(BaseModel):
visited.add(node_name)
rec_stack.add(node_name)
path.append(node_name)
+ cycle = False
for edge in self.nodes[node_name].edges:
target = edge.target
if target not in visited:
if dfs(target):
- return True
+ cycle = True
elif target in rec_stack:
# Found a cycle → extract the cycle
cycle_start_index = path.index(target)
@@ -182,11 +183,11 @@ class DiGraph(BaseModel):
raise ValueError(
f"Cycle detected without exit condition: {' -> '.join(cycle_nodes + cycle_nodes[:1])}"
)
- return True # Found cycle, but it has an exit condition
+ cycle = True # Found cycle, but it has an exit condition
rec_stack.remove(node_name)
path.pop()
- return False
+ return cycle
has_cycle = False
for node in self.nodes:
diff --git a/python/packages/autogen-agentchat/tests/test_group_chat_graph.py b/python/packages/autogen-agentchat/tests/test_group_chat_graph.py
index 98cc8ca66..b3a82e9ee 100644
--- a/python/packages/autogen-agentchat/tests/test_group_chat_graph.py
+++ b/python/packages/autogen-agentchat/tests/test_group_chat_graph.py
@@ -247,6 +247,51 @@ def test_cycle_detection_without_exit_condition() -> None:
graph.has_cycles_with_exit()
+def test_cycle_detection_cleanup_bug() -> None:
+ """Test that cycle detection properly cleans up recursion state.
+
+ This test reproduces the bug where the DFS algorithm in has_cycles_with_exit
+ didn't properly clean up rec_stack and path when returning early upon finding
+ a cycle with valid exit conditions. The bug could cause incorrect behavior
+ when processing subsequent unvisited nodes in graphs with multiple components.
+ """
+
+ # Create a graph that exposes the cleanup bug:
+ # A -> B -> C -> A (cycle with condition)
+ # A -> D (separate branch that could be affected by stale rec_stack/path)
+ # E (disconnected component that could be affected by cleanup issues)
+ graph = DiGraph(
+ nodes={
+ "A": DiGraphNode(name="A", edges=[DiGraphEdge(target="B"), DiGraphEdge(target="D")]),
+ "B": DiGraphNode(name="B", edges=[DiGraphEdge(target="C")]),
+ "C": DiGraphNode(name="C", edges=[DiGraphEdge(target="A", condition="loop")]),
+ "D": DiGraphNode(name="D", edges=[]),
+ "E": DiGraphNode(name="E", edges=[]), # Disconnected component
+ }
+ )
+
+ # This should work correctly with the fix - proper cleanup ensures
+ # that processing node E is not affected by stale recursion state
+ # from processing the A->B->C->A cycle
+ result = graph.has_cycles_with_exit()
+ assert result is True # Has valid cycles
+
+ # Test with multiple cycles to ensure thorough cleanup
+ multi_cycle_graph = DiGraph(
+ nodes={
+ "A": DiGraphNode(name="A", edges=[DiGraphEdge(target="B")]),
+ "B": DiGraphNode(name="B", edges=[DiGraphEdge(target="A", condition="cycle1")]),
+ "C": DiGraphNode(name="C", edges=[DiGraphEdge(target="D")]),
+ "D": DiGraphNode(name="D", edges=[DiGraphEdge(target="C", condition="cycle2")]),
+ "E": DiGraphNode(name="E", edges=[DiGraphEdge(target="F")]),
+ "F": DiGraphNode(name="F", edges=[]),
+ }
+ )
+
+ result = multi_cycle_graph.has_cycles_with_exit()
+ assert result is True # Has valid cycles
+
+
def test_different_activation_groups_detection() -> None:
"""Test different activation groups."""
graph = DiGraph(