voter: only primary voter proposes and only once

[andresilva]

No description provided.

[rphmeier]

We discussed this in the original PR and I was against including the VoterId in the voter at all. We haven't done it for any other messages either

[andresilva]

Sorry I missed that discussion. What if the local id is exposed from the Environment? I don't really want to spam consensus gossip with useless messages, and we either filter here or on substrate (which means the logic for primary election leaks there).

[marcio-diaz]

I think @rphmeier wanted to filter them at Substrate level. Except for that lgtm.

[rphmeier]

Yeah, although it does seem like a somewhat debatable point. I think we either go all-the-way with VoterId in finality-grandpa or not at all. So if Id is None we don't send messages at all, for instance.

[rphmeier]

@andresilva

we either filter here or on substrate (which means the logic for primary election leaks there).

The idea is that with #55 we will just call voter_state.round_data(round).primary == self.id for filtering -- the logic for primary election doesn't need to leak.

[andresilva]

I agree that doing the filtering on substrate is more inline with what we have now, especially with #55. I'll update this PR shortly to remove this.

[andresilva]

After discussing this with @rphmeier we've decided to go "all-in" on using the voter_id and move the logic for filtering messages to this crate. I've updated the PR so that no votes are sent out if we're not a current voter in the voter set. The local voter id is given per-round through Environment::round_data. The protocol state machine still transitions as if the votes were cast, the only difference being that we don't push the messages to the sink. Commits are still pushed out and should be filtered on substrate side (although it may be beneficial in some cases for observers to broadcast commits).