Justifed state in LMD GHOST

[djrtwo]

Issue

justified_head_state which is used to get the active validator set for lmd_ghost is currently defined as "the resulting BeaconState object from processing the chain up to the justified_head".

This is correct in the "happy" case in which no slots were skipped between the justified_head and the epoch boundary that was justified, but in the case that slots were skipped between the head and epoch boundary slot (% epoch_length == 0), the justified head is actually the block from a prior slot but at the state at the epoch boundary slot. For example, if a block at slot 63 (epoch 0) is justified for epoch 1 (epoch boundary @ slot 64) in the case that there was no block for slot 64, then the state associated with this (justified_block, justified_epoch) tuple is actually the justified block from slot 63 transitioned through an additional slot to slot 64.

Solution

Define justified_head_state as "the resulting BeaconState object from processing the chain up to the justified_head through the start slot of justified_epoch"

[JustinDrake]

We probably also want to make the fork choice rule spec executable.

[mratsim]

That would be great to produce test vectors.

The only existing reference code in Python is unfortunately outdated:

• https://github.com/ethereum/research/blob/master/clock_disparity/lmd_node.py (uses notaries instead of validators and heights instead of slots)
• https://github.com/ethereum/research/blob/master/ghost/ghost.py seems to be bitwise LMD Ghost

[JustinDrake]

This was addressed in 1198