Initialize anchor PTC vote arrays in get_forkchoice_store

[0xsamalt]

Description

get_forkchoice_store initializes the store's PTC vote arrays (payload_timeliness_vote and payload_data_availability_vote) for every block added via on_block, but leaves the anchor root out. Since payload_timeliness, payload_data_availability, and on_payload_attestation_message require every known block root to be present in these arrays, the anchor breaks the invariant: get_head fails with an AssertionError once the anchor is a previous-slot payload decision with its payload delivered, because should_extend_payload consults payload_timeliness on the anchor root.

Initialize both arrays for the anchor root with no votes cast ([None] * PTC_SIZE), matching on_block. test_genesis previously codified the missing arrays as expected behavior; it now asserts that the anchor's vote arrays are initialized with no votes recorded, and a regression test covers get_head with the anchor's payload delivered.

Checklist

• Update documentation: not applicable, the fix is in the spec itself
• Add tests for new functionality: added test_get_head_with_anchor_payload_delivered and updated test_genesis
• Run make lint to check formatting
• Run make test to check tests

Relations

None

[jtraglia]

LGTM, thanks for this @0xsamalt!

[jtraglia]

Sorry, I don't think I like this fix. Seems like a hack. Also the annotation (modified in gloas) is over the wrong thing now. I believe @jihoonsong is looking at this. Let's see what he thinks.

[jihoonsong]

The issue is related to Heze genesis, not here. I'll make a commit, which will exclude this change.

[jihoonsong]

After looking into this, it was not about Heze genesis either. See the other comment.

[jihoonsong]

After reviewing this PR, I've concluded that the proposed spec change is correct. #5135 removed wrongly seeding the PTC arrays with True. Later, #5180 introduced None votes and it seems this spec change should've included there.

However, the new test case is unreachable. I think there are two cases relevant to this change: genesis block and checkpoint sync. The genesis block is always EMPTY post-Gloas (see #5172 for more context) and during checkpoint syncing, the anchor block cannot be from the previous slot. Hence, the test is removed in 745433f.

While I think it's unreachable that the PTC arrays being accessed by anchor root, I believe having this code makes the invariant hold without an exception as the author described.

[nflaig]

LGTM, this matches what lodestar does, but this code should not be reachable as @jihoonsong pointed out as we only evaluate ptc votes near the tip, ie. for payload status of a block from current_slot - 1 but the finalized anchor during checkpoint sync is at least 2 epochs behind the current epoch