fix: assign receive-loop attributes before spawning recv task

Author: SAY-5

src/acp/connection.py

@@ -86,6 +86,8 @@ def __init__(
         self._closed = False
         self._disconnected = False
         self._sender = (sender_factory or self._default_sender_factory)(self._writer, self._tasks)
+        self._observers: list[StreamObserver] = list(observers or [])
+        self._receive_timeout = receive_timeout
         if listening:
             self._recv_task = self._tasks.create(
                 self._receive_loop(),
@@ -103,8 +105,6 @@ def __init__(
             self._run_notification,
         )
         self._dispatcher.start()
-        self._observers: list[StreamObserver] = list(observers or [])
-        self._receive_timeout = receive_timeout
 
     async def close(self) -> None:
         """Stop the receive loop and cancel any in-flight handler tasks."""

tests/test_rpc.py

@@ -676,3 +676,28 @@ async def test_spawn_agent_process_roundtrip(tmp_path):
         assert test_client.notifications
 
     assert process.returncode is not None
+
+
+@pytest.mark.asyncio
+async def test_connection_init_under_eager_task_factory(server):
+    # Regression: under asyncio.eager_task_factory the receive loop runs synchronously
+    # up to its first await inside Connection.__init__, so every attribute it reads
+    # (e.g. _receive_timeout) must be assigned before _tasks.create(_receive_loop()).
+    loop = asyncio.get_running_loop()
+    previous_factory = loop.get_task_factory()
+    loop.set_task_factory(asyncio.eager_task_factory)
+    try:
+        conn = Connection(
+            lambda method, params, is_notification: None,
+            server.client_writer,
+            server.client_reader,
+            receive_timeout=0.5,
+        )
+    finally:
+        loop.set_task_factory(previous_factory)
+
+    assert conn._receive_timeout == 0.5
+    # Let the loop tick once so any deferred receive-task crash would land.
+    await asyncio.sleep(0)
+    assert conn._disconnected is False
+    await conn.close()