Phragmen should ignore "empty" voters

[kianenigma]

We build the voters as such:

substrate/primitives/npos-elections/src/lib.rs

Lines 385 to 405 in 0c8329c

	voters.extend(initial_voters.into_iter().map(|(who, voter_stake, votes)| {
	let mut edges: Vec<Edge<AccountId>> = Vec::with_capacity(votes.len());
	for v in votes {
	if edges.iter().any(|e| e.who == v) {
	// duplicate edge.
	continue;
	}
	if let Some(idx) = c_idx_cache.get(&v) {
	// This candidate is valid + already cached.
	candidates[*idx].approval_stake = candidates[*idx].approval_stake
	.saturating_add(voter_stake.into());
	edges.push(Edge { who: v.clone(), candidate_index: *idx, ..Default::default() });
	} // else {} would be wrong votes. We don't really care about it.
	}
	Voter {
	who,
	edges: edges,
	budget: voter_stake.into(),
	load: Rational128::zero(),
	}
	}));

If by the end of the map(), edges.len() is empty, then this voter should not be even included. Essentially, we have to change the map() to filter_map.

Ideally there should be a test added for this, else we need to make sure the change is reflected in #6685 as well.

This (as far as I know) does not cause any issues. It is just extra iterations here and there that will make the execution a bit slower.

[jaxter03]

Hey @kianenigma , is this issue still open? ... I see #6685 is already merged.

[kianenigma]

ahh yes, thanks.

[kianenigma]

Well.. actually it seems like the answer is actually no. If you see the setup_inputs you will see that we still have this issue.

[jaxter03]

Okay Thanks @kianenigma