shard_committees_at_slots missing when processing previous epoch attestations

[arnetheduck]

The scenario is the following:

At slot 6, we included an attestation for slot 1. We advance through time all the way to slot 127. The next slot will be an epoch update. We start the slot processing by increasing state.slot to 128. Then we proceed through the functions until we pick out previous_epoch_attestations = [a for a in state.latest_attestations if state.slot - 2 * EPOCH_LENGTH <= a.slot < state.slot - EPOCH_LENGTH] - our attestation from slot 1 fits this criteria (0 <= 1 < 64).

We call get_attestation_participants which calls get_shard_committees_at_slot(state, 1) which computes:
earliest_slot_in_array = state.slot - (state.slot % EPOCH_LENGTH) - EPOCH_LENGTH or 128 - 0 - 64 = 64.

subsequently, assert earliest_slot_in_array <= slot < earliest_slot_in_array + EPOCH_LENGTH * 2 fires.

related to #348, #92, #323

c8efdc7

[djrtwo]

I think using the previous slot in the calculations fixes this. Note as is (and with the fix) this helper can only safely be called in the context of the state transition.

prev_slot = state.slot - 1
earliest_slot_in_array = prev_slot - (prev_slot % EPOCH_LENGTH) - EPOCH_LENGTH

[arnetheduck]

I'd love to, but that causes another error with get_beacon_proposer_index(state, state.slot) - at the epoch boundary there's no longer coverage for the proposer!

[djrtwo]

yeah, yeah, subtle quirks in shuffling and helpers.
on it!

[JustinDrake]

Is this addressed @djrtwo?