swarm-derive/: Add where clause of behaviour to generated out event

[mxinden]

Description

When generating the OutEvent for a NetworkBehaviour, add the where clause of the
NetworkBehaviour struct to the generated enum.

Links to any relevant issues

Follow up to #2792

Open Questions

Change checklist

• I have performed a self-review of my own code
• I have made corresponding changes to the documentation
• I have added tests that prove my fix is effective or that my feature works
• A changelog entry has been made in the appropriate crates

[elenaf9]

Won't this cause an error in case of

    #[derive(NetworkBehaviour)]
    struct Foo<T>
    where
        T: Copy
    {
        ping: libp2p::ping::Ping,
        #[behaviour(ignore)]
        bar: T,
    }

which will generate

pub enum FooEvent<T> // Error: unused Parameter
where
    T: Copy
{
    Ping(<libp2p::ping::Ping as NetworkBehaviour>::OutEvent)
}

Am I missing something?

[mxinden]

@elenaf9 that is correct. Good catch. Though why would someone use the OutEvent mechanism in combination with #[behaviour(ignore)]?

Once we remove NetworkBehaviourEventProcess (#2751) I don't think there is any use for poll_method nor ignore at all. See #2751 (comment).

With that in mind, I would leave this as is. Do you agree @elenaf9?

[elenaf9]

Though why would someone use the OutEvent mechanism in combination with #[behaviour(ignore)]?

Once we remove NetworkBehaviourEventProcess (#2751) I don't think there is any use for poll_method nor ignore at all. See #2751 (comment).

I am a bit confused. Why would we not need #[behaviour(ignore)] anymore? Isn't it still needed to signal the derive macro that this field is not a network behaviour? E.g. when implementing the methods of the NetworkBehaviour trait and we pass the event to our inner behaviours?
And with OutEvent I would think we stilll need it to mark that this field won't generate an event:

• If the OutEvent is provided by the user, we should not need the trait bound From<<FIELD as NetworkBehaviour>::OutEvent> for this field
• If the OutEvent is generated, we should not generated an enum variant for this field. See above code example.

@mxinden what am I missing here?

[mxinden]

I am a bit confused. Why would we not need #[behaviour(ignore)] anymore?

Let me rephrase:

With NetworkBehaviourEventProcess users sometimes needed an events: VecDeque which they could push events into from NetworkBehaviourEventProcess and pop events from within their custom poll method. events would not implement NetworkBehaviour and thus needed the #[behaviour(ignore)].

With the removal of NetworkBehaviourEventProcess there is no more need for fields on the root NetworkBehaviour that do not themselves implement NetworkBehaviour, at least I can not think of a use-case.

Can you think of a use case for #[behaviour(ignore)] and a custom poll method once NetworkBehaviourEventProcess is removed @elenaf9?

[elenaf9]

Thanks for elaborating @mxinden! Makes sense now.

Can you think of a use case for #[behaviour(ignore)] and a custom poll method once NetworkBehaviourEventProcess is removed @elenaf9?

I guess the only use case is if the wrapping NetworkBehaviour wants to implement some additional logic on top of the inner behaviours and / or hide some of the implementation details about them. But this can also be implemented on top of the Swarm, so I agree that we don't need #[behaviour(ignore)] or #[behaviour(poll = ..)] anymore.

With that in mind, I would leave this as is. Do you agree @elenaf9?

Yep :)