Attestation committee refactor

[vbuterin]

Mostly cosmetic and enabling some functionality needed in phase 1. A few substantive changes to Attestation data structure.

• Remove get_crosslink_committees_at_slot (that function's ugly man...)
• Make the "base" that everything works off instead be get_crosslink_committee (simpler and more generic)
• Attestations store epoch, start shard and shard, no longer slot (slot can be calculated from the other three)
• store inclusion_distance in PendingAttestation rather than inclusion_slot for more direct use in subsequent calculations.
• Retaining start shard in attestations allows get_attesting_indices to peek much further back into the past, making it useful for slashings (Phase 1)
• Some two-layer-deep nested loops become one-layer-deep loops
• alter get_winning_crosslink_and_attesting_indices function signature to mirror that of get_crosslink_committee
• is_double_vote and is_surround_vote replaced with single is_slashable_attestation_data

[djrtwo]

I'm working on some edits based on the above feedback

[djrtwo]

and now working on some bugs :)

[djrtwo]

@JustinDrake if we remove get_state_root helper, then I'd like to do it in a separate PR with an explanation rather than it just disappearing at some point

[vbuterin]

I still am not fully comfortable with compute_committee being removed. It's definitely the right choice if you look at phase 0 in a vacuum, but the reason I separated out that function initially was that get_persistent_committee in phase 1 also uses it, so we're making both kinds of committees from the same basic building block.

Here is the usage in phase 1: https://github.com/ethereum/eth2.0-specs/blob/dev/specs/core/1_shard-data-chains.md#get_period_committee

One possible compromise I see would be to define a function:

def get_indices_of_chunk(list_size: int, chunks: int, chunk: int) -> List[int]:
    start = (len(list_size) * chunk) // chunks
    end = (len(list_size) * (chunk + 1)) // chunks
    return list(range(start, end))

Then the PR would be, in get_crosslink_committee:

    return [
        active_indices[get_shuffled_index(i, len(active_indices), seed)]
        for i in get_indices_of_chunk(len(active_indices), committee_count, committee_index)
    ]

And the same in get_period_committee.

[JustinDrake]

@vbuterin: I've exposed get_committee which abstracts away common logic for both get_crosslink_committee (now a 2-liner) and get_period_committee (now a 1-liner).

[djrtwo]

I think the previous function signature of compute_committee is more straight forward and readable and also is much clearer to the client what the explicit inputs are which could be used to cache results.

I vote going back to that function signature or something similar

[vbuterin]

I think the previous function signature of compute_committee is more straight forward and readable and also is much clearer to the client what the explicit inputs are which could be used to cache results.

Agreed.

[JustinDrake]

compute_committee restored! :)