Fix fork choice on_block tests and update test format

[hwwhww]

1. Update test format of checks: add checkpoint epoch number and rephrase them in dict. Example:

- checks:
    time: 192
    head: {slot: 32, root: '0xdaa1d49d57594ced0c35688a6da133abb086d191a2ebdfd736fad95299325aeb'}
    justified_checkpoint: {epoch: 3, root: '0xc25faab4acab38d3560864ca01e4d5cc4dc2cd473da053fbc03c2669143a2de4'}
    finalized_checkpoint: {epoch: 2, root: '0x40d32d6283ec11c53317a46808bc88f55657d93b95a1af920403187accf48f4f'}
    best_justified_checkpoint: {epoch: 3, root: '0xc25faab4acab38d3560864ca01e4d5cc4dc2cd473da053fbc03c2669143a2de4'}

2. Fix test_new_justified_is_later_than_store_justified: there was a missing tick step.
   • Thanks @ajsutton for pointing it out!
3. Update test_new_finalized_slot_is_justified_checkpoint_ancestor: remove the useless tick step.
4. Rewrite test_on_block_finalized_skip_slots_not_in_skip_chain and test_on_block_finalized_skip_slots:
   • test_on_block_finalized_skip_slots_not_in_skip_chain used non-genesis anchor state and block to initialize fork choice store, which was hacky and having potential issues (Issues with fork choice and non-genesis anchors #2566 explained it)
   • I rewrote it to build a store from genesis. The test scenario is similar to Properly handle skip slots in fork choice #1579.

[hwwhww]

@ajsutton @tuyennhv

If you'd like to try the updated test vectors, I dumped new fork choice test vectors to: https://github.com/hwwhww/consensus-spec-tests/tree/fork-choice-pr2577 (b23ed05 version)

p.s. It doesn't fix the test_new_finalized_slot_is_justified_checkpoint_ancestor issue.

[hwwhww]

Update: @tuyennhv confirmed that Lodestar passed all fork choice tests against this new branch. 🙌

[ajsutton]

Sorry got distracted from this. I've never worked out how to build the actual test vectors from the source (only got as far as running them against the spec implementation), do you have instructions for that or is there a tarball somewhere?

[hwwhww]

@ajsutton

fork_choice tests only tarballs:

• minimal.tar.gz
• mainnet.tar.gz

---

If you'd like to download from source, it required Git LFS installed.

git lfs install

If LFS has been installed correctly, git clone should download the large files by default:

git clone git@github.com:hwwhww/consensus-spec-tests.git -b fork-choice-pr2577 --single-branch
cd consensus-spec-tests

Or you can trigger it by

git lfs pull

The test vectors will be under consensus-spec-tests/tests folder.

[djrtwo]

i got a little hung up on what was actually being tested in the valid case, but figured it out!

Looks correct to me

[djrtwo]

This block is valid wrt the finalized epoch (including the skip slots), but this block isn't the head, right?

apply_next_epoch_with_attestations for epoch 3 and 4 will have created blocks of much higher weight and win out

So -- Does adding this block change anything? or does it catch a regression in the prior fork choice implementations?

[djrtwo]

and if block is supposed to be the head? can we check that it is against a dynamic call to get_head()?

[djrtwo]

Ah, I see. It's just checking in on_block can even be run anymore which would yield valid: False if it didn't work

[hwwhww]

Ah, I see. It's just checking in on_block can even be run anymore which would yield valid: False if it didn't work

Right, it is for comparing with test_on_block_finalized_skip_slots_not_in_skip_chain (valid: False) test case.