Unregister ros1 subscribers?

[lucasw]

Is there a current method for shutting down subscribers fully? So far I'm having luck unregistering publishers, if I make sure they go out of scope in their tokio tasks before the process exits then they don't appear in rosnode info for the dead node, but subscribers linger and a rosnode cleanup is needed.

I see unregister_subscriber() in https://github.com/Carter12s/roslibrust/blob/master/roslibrust/src/ros1/master_client.rs#L224 - but it's not exposed through a node handle, nothing calls it?

Beyond that what about interrupting my_subscriber.next().await- I have several of those running inside tokio tasks- if unregister_subscriber were exposed would/could it cause that next() to return an error, like how a channel errors if the other end of it (sender or receiver) goes away?

[lucasw]

I have a manual unregister_subscriber working in https://github.com/lucasw/roslibrust/tree/get_rx_msg_def lucasw@5fbbd2e - but so far I haven't been able to get drop to do it automatically.

I'll rebase the minimal version of it onto master and make at least a draft PR.

[Carter12s]

Nice!

I had the chance to look at your code this morning, and then while I was on a flight today decided to implement this:
#204

It isn't a full answer to your question and we should make Subscriptions clean up after themselves, but I added a Node level cleanup of any ROS references that (at least in my testing) means we don't leave any junk in rosnode info / rosnode list.

[Carter12s]

I'd love to see a draft PR of having Subscribers cleanup after themselves. I think the best paradigm for how to do this right is what ServiceServer does. I think having each of Publisher, Subscriber, and ServiceServer hold a weak_clone() of NodeHandle and then to call a non-async clean-up function on NodeHandle in their Drop impl will end up working well and being pretty readable.