Turns out the to_asyncio. stuff wasn't robust enough to handle this; we weren't propagating errors from an IPC context embedded task opened via .open_channel_from()..
To start this adds 2 more tests:
[x] one where the far end asyncio task errors after a short sleep but this error is not propagated and the test times out.
[x] a fix will come for this in a follow up commit
[x] one test that does pass where the parent actor that also opens the IPC context cancels the actor using the portal api.
Turns out the
to_asyncio.stuff wasn't robust enough to handle this; we weren't propagating errors from an IPC context embedded task opened via.open_channel_from()..To start this adds 2 more tests:
asynciotask errors after a short sleep but this error is not propagated and the test times out.