Fix test

I think this failure starts in Python3.8 due to changes in unraiseable
exception behaviour, or possibly in how pytest
handles them. See `sys.unraisablehook`.

tests/test_aioca.py

@@ -11,6 +11,7 @@
 from typing import Callable, List, Tuple, Union
 
 import pytest
+from _pytest.unraisableexception import catch_unraisable_exception
 from epicscorelibs.ca import cadef, dbr
 
 from aioca import (
@@ -530,41 +531,46 @@ async def monitor_for_a_bit(callback: Callable, ioc) -> Subscription:
     return m
 
 
-def test_closing_event_loop(ioc: subprocess.Popen, capsys) -> None:
-    def closed_messages(text):
-        return [x for x in text.splitlines() if x.endswith("Event loop is closed")]
-
-    q: queue.Queue = queue.Queue()
-    m = run(monitor_for_a_bit(q.put, ioc))
-    # Most of we get the initial value 0, then an update of 1
-    # On travis sometimes the IOC is too late to startup and we
-    # get 1 and 2
-    updates = (q.get_nowait(), q.get_nowait())
-    assert updates in [(0, 1), (1, 2)]
-    assert q.qsize() == 0
-    assert len(m.pending_values) == 0
+def test_closing_event_loop(
+    ioc: subprocess.Popen,
+) -> None:
+    with catch_unraisable_exception() as cm:
+        q: queue.Queue = queue.Queue()
+        m = run(monitor_for_a_bit(q.put, ioc))
+        # Most of we get the initial value 0, then an update of 1
+        # On travis sometimes the IOC is too late to startup and we
+        # get 1 and 2
+        updates = (q.get_nowait(), q.get_nowait())
+        assert updates in [(0, 1), (1, 2)]
+        assert q.qsize() == 0
+        assert len(m.pending_values) == 0
 
-    time.sleep(2.0)
-    m.close()
-    time.sleep(1.0)
+        time.sleep(2.0)
+        m.close()
+        time.sleep(1.0)
 
-    # We should have 2 more updates that didn't make it to the queue
-    # because loop closed
-    assert q.qsize() == 0
-    assert len(m.pending_values) == 2
+        # We should have 2 more updates that didn't make it to the queue
+        # because loop closed
+        assert q.qsize() == 0
+        assert len(m.pending_values) == 2
 
-    # Check that there are no more updates
-    time.sleep(2.0)
-    assert len(m.pending_values) == 2
+        # Check that there are no more updates
+        time.sleep(2.0)
+        assert len(m.pending_values) == 2
 
-    ioc.communicate("exit")
-    time.sleep(0.5)
-    # There should be no more updates, but one error from the close message
-    assert q.qsize() == 0
-    assert len(m.pending_values) == 2
-    captured = capsys.readouterr()
-    assert captured.out == ""
-    assert len(closed_messages(captured.err)) == 1, captured.err
+        ioc.communicate("exit")
+        time.sleep(0.5)
+        # There should be no more updates, but one error from the close message
+        assert q.qsize() == 0
+        assert len(m.pending_values) == 2
+
+        # We expect there to be an unraiseable exception, caused by trying dispatch a
+        # function to a closed event loop.
+        # Detecting this is a bit hacky - pytest has code to catch unraiseable
+        # exceptions, but no publicly declared API to examine them nor configuration
+        # settings to do anything but ignore them.
+        # So we use an internal API to examine it.
+        assert "Event loop is closed" in cm.unraisable.exc_value.args
 
 
 @pytest.mark.asyncio