In the subgraph, when there is a branch point to __end__(or __start__ point to ...), it will lose part of the edges

[gbaian10]

Checked other resources

• I added a very descriptive title to this issue.
• I searched the LangGraph/LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.
• I am sure that this is a bug in LangGraph/LangChain rather than my code.
• I am sure this is better as an issue rather than a GitHub discussion, since this is a LangGraph bug and not a design question.

Example Code

import secrets

from langgraph.graph import END, START, MessagesState, StateGraph
from langgraph.graph.state import CompiledStateGraph
from rich import get_console


def foo(_: MessagesState) -> None:
    return


def branch() -> bool:
    return secrets.choice([True, False])


def sub() -> CompiledStateGraph:
    workflow = StateGraph(MessagesState)
    for node in ["A", "B", "Z"]:
        workflow.add_node(node, foo)

    workflow.add_conditional_edges(START, branch, {True: "A", False: END})
    workflow.add_edge("A", "B")
    workflow.add_conditional_edges("B", branch, {True: "Z", False: END})
    workflow.add_edge("Z", END)
    return workflow.compile()


def main() -> CompiledStateGraph:
    workflow = StateGraph(MessagesState)
    workflow.add_node("main_entry", foo)
    workflow.add_node("sub_1", sub())
    workflow.add_node("main_exit", foo)

    workflow.add_edge(START, "main_entry")
    workflow.add_edge("main_entry", "sub_1")
    workflow.add_edge("sub_1", "main_exit")
    workflow.add_edge("main_exit", END)
    return workflow.compile()


if __name__ == "__main__":
    sub_graph = sub()
    get_console().print(sub_graph.get_graph())
    sub_graph.get_graph().draw_mermaid_png(output_file_path="_only_sub.png")

    graph = main()
    get_console().print(graph.get_graph())
    graph.get_graph(xray=1).draw_mermaid_png(output_file_path="_with_sub.png")

Error Message and Stack Trace (if applicable)

Graph(
    nodes={
        '__start__': Node(id='__start__', name='__start__', data=<class 'pydantic.v1.main.LangGraphInput'>, metadata=None),
        'A': Node(id='A', name='A', data=A(func_accepts_config=False, afunc_accepts_config=False, recurse=True), metadata=None),
        'B': Node(id='B', name='B', data=B(func_accepts_config=False, afunc_accepts_config=False, recurse=True), metadata=None),
        'Z': Node(id='Z', name='Z', data=Z(func_accepts_config=False, afunc_accepts_config=False, recurse=True), metadata=None),
        '__end__': Node(id='__end__', name='__end__', data=<class 'pydantic.v1.main.LangGraphOutput'>, metadata=None)
    },
    edges=[
        Edge(source='A', target='B', data=None, conditional=False),
        Edge(source='Z', target='__end__', data=None, conditional=False),
        Edge(source='__start__', target='A', data=True, conditional=True),
        Edge(source='__start__', target='__end__', data=False, conditional=True),
        Edge(source='B', target='Z', data=True, conditional=True),
        Edge(source='B', target='__end__', data=False, conditional=True)
    ]
)
Graph(
    nodes={
        '__start__': Node(id='__start__', name='__start__', data=<class 'pydantic.v1.main.LangGraphInput'>, metadata=None),
        'main_entry': Node(id='main_entry', name='main_entry', data=main_entry(func_accepts_config=False, afunc_accepts_config=False, recurse=True), metadata=None),
        'sub_1': Node(id='sub_1', name='sub_1', data=<langgraph.graph.state.CompiledStateGraph object at 0x000001D7541179B0>, metadata=None),
        'main_exit': Node(id='main_exit', name='main_exit', data=main_exit(func_accepts_config=False, afunc_accepts_config=False, recurse=True), metadata=None),
        '__end__': Node(id='__end__', name='__end__', data=<class 'pydantic.v1.main.LangGraphOutput'>, metadata=None)
    },
    edges=[
        Edge(source='__start__', target='main_entry', data=None, conditional=False),
        Edge(source='main_entry', target='sub_1', data=None, conditional=False),
        Edge(source='main_exit', target='__end__', data=None, conditional=False),
        Edge(source='sub_1', target='main_exit', data=None, conditional=False)
    ]
)

Description

When there is an edge pointing from start or towards end, and another edge points to a node, part of the edges will disappear.

System Info

langgraph==0.2.19
langchain-core==0.2.38

[gbaian10]

import secrets

from langgraph.graph import END, START, MessagesState, StateGraph
from langgraph.graph.state import CompiledStateGraph


def foo(_: MessagesState) -> None:
    return


def branch() -> bool:
    return secrets.choice([True, False])


def sub() -> CompiledStateGraph:
    workflow = StateGraph(MessagesState)
    for node in ["A", "B", "M", "Y", "Z"]:
        workflow.add_node(node, foo)

    workflow.add_conditional_edges(START, branch, {True: "A", False: "B"})
    workflow.add_edge("A", "M")
    workflow.add_edge("B", "M")
    workflow.add_conditional_edges("M", branch, {True: "Y", False: "Z"})
    workflow.add_edge("Y", END)
    workflow.add_edge("Z", END)
    return workflow.compile()


def main() -> CompiledStateGraph:
    workflow = StateGraph(MessagesState)
    workflow.add_node("main_entry", foo)
    workflow.add_node("sub_1", sub())
    workflow.add_node("main_exit", foo)

    workflow.add_edge(START, "main_entry")
    workflow.add_edge("main_entry", "sub_1")
    workflow.add_edge("sub_1", "main_exit")
    workflow.add_edge("main_exit", END)
    return workflow.compile()


if __name__ == "__main__":
    sub_graph = sub()
    sub_graph.get_graph().draw_mermaid_png(output_file_path="_only_sub.png")

    graph = main()
    graph.get_graph(xray=1).draw_mermaid_png(output_file_path="_with_sub.png")

But in this example, edges didn't disappear; instead, start and end appeared.
I'm not sure if they should appear.

Because in a simple single-line subgraph, start and end will be eliminated.

import secrets

from langgraph.graph import END, START, MessagesState, StateGraph
from langgraph.graph.state import CompiledStateGraph


def foo(_: MessagesState) -> None:
    return


def branch() -> bool:
    return secrets.choice([True, False])


def sub() -> CompiledStateGraph:
    workflow = StateGraph(MessagesState)
    for node in ["A", "B", "C"]:
        workflow.add_node(node, foo)

    workflow.add_edge(START, "A")
    workflow.add_edge("A", "B")
    workflow.add_edge("B", "C")
    workflow.add_edge("C", END)
    return workflow.compile()


def main() -> CompiledStateGraph:
    workflow = StateGraph(MessagesState)
    workflow.add_node("main_entry", foo)
    workflow.add_node("sub_1", sub())
    workflow.add_node("main_exit", foo)

    workflow.add_edge(START, "main_entry")
    workflow.add_edge("main_entry", "sub_1")
    workflow.add_edge("sub_1", "main_exit")
    workflow.add_edge("main_exit", END)
    return workflow.compile()


if __name__ == "__main__":
    sub_graph = sub()
    sub_graph.get_graph().draw_mermaid_png(output_file_path="_only_sub.png")

    graph = main()
    graph.get_graph(xray=1).draw_mermaid_png(output_file_path="_with_sub.png")

[gbaian10]

But when we try to direct the three edges towards end, everything seems to become normal again.

import secrets

from langgraph.graph import END, START, MessagesState, StateGraph
from langgraph.graph.state import CompiledStateGraph


def foo(_: MessagesState) -> None:
    return


def branch() -> bool:
    return secrets.choice([True, False])


def sub() -> CompiledStateGraph:
    workflow = StateGraph(MessagesState)
    for node in ["A", "B", "M", "N", "Z"]:
        workflow.add_node(node, foo)

    workflow.add_conditional_edges(START, branch, {True: "A", False: "B"})
    workflow.add_edge("A", "M")
    workflow.add_edge("B", "N")

    workflow.add_conditional_edges("M", branch, {True: "Z", False: END})
    workflow.add_edge("Z", END)
    workflow.add_edge("N", END)
    return workflow.compile()


def main() -> CompiledStateGraph:
    workflow = StateGraph(MessagesState)
    workflow.add_node("main_entry", foo)
    workflow.add_node("sub_1", sub())
    workflow.add_node("main_exit", foo)

    workflow.add_edge(START, "main_entry")
    workflow.add_edge("main_entry", "sub_1")
    workflow.add_edge("sub_1", "main_exit")
    workflow.add_edge("main_exit", END)
    return workflow.compile()


if __name__ == "__main__":
    sub_graph = sub()
    sub_graph.get_graph().draw_mermaid_png(output_file_path="_only_sub.png")

    graph = main()
    graph.get_graph(xray=1).draw_mermaid_png(output_file_path="_with_sub.png")