Fork choice EF consensus tests

[mratsim]

Consensus spec v1.1.1 finally added fork choice tests, this integrates them.

[github-actions]

Unit Test Results

     12 files  ±  0     756 suites  +8   52m 29s ⏱️ + 22m 23s
1 515 tests +48  1 477 ✔️ +12  38 💤 +36  0 ❌ ±0 
9 032 runs  +96  8 952 ✔️ +24  80 💤 +72  0 ❌ ±0 

Results for commit 608964f. ± Comparison against base commit cc0dbd5.

♻️ This comment has been updated with latest results.

[mratsim]

Error is an IO Error on windows, maybe path is too long?

https://github.com/status-im/nimbus-eth2/runs/4061397307?check_suite_focus=true#step:20:1199

D:\a\nimbus-eth2\nimbus-eth2\nimbus-eth2\vendor\nim-eth2-scenarios\tests-v1.1.3\mainnet\phase0\fork_choice\on_block\pyspec_tests\new_finalized_slot_is_not_justified_checkpoint_ancestor\block_0xee90f2a441552708e67e14230750d3503b0b3219b829d6e839900e428c1fa2de.ssz_snappy is 269 characters

[tersec]

Why are these skipped?

[mratsim]

Details are given here https://github.com/status-im/nimbus-eth2/pull/3041/files#diff-993d81740d3beb56996cc3af0c4b741fa075eb06a7b0eaea23003b7ec27ed26fR308-R324

Some are skipped inherently due to the difference between ProtoArray and the spec reference implementations.

For the one highlighted in particular "protoArray changes fork as expected and directly returns the head of the new chain instead of the root of the fork

[tersec]

Arguably an issue in test case construction, to the extent it's testing implementation details rather than correct results.

[tersec]

Error is an IO Error on windows, maybe path is too long?

https://github.com/status-im/nimbus-eth2/runs/4061397307?check_suite_focus=true#step:20:1199

D:\a\nimbus-eth2\nimbus-eth2\nimbus-eth2\vendor\nim-eth2-scenarios\tests-v1.1.3\mainnet\phase0\fork_choice\on_block\pyspec_tests\new_finalized_slot_is_not_justified_checkpoint_ancestor\block_0xee90f2a441552708e67e14230750d3503b0b3219b829d6e839900e428c1fa2de.ssz_snappy is 269 characters

Seems likely -- but it's also possible that whether Git was handling it properly before or not, and might have provided the proximate error cause, the test runner could still have issues opening it.

[tersec]

It might suffice to eliminate those long test names by updating past v1.1.3, to v1.1.5:

[arnetheduck]

this is still broken - it needs to use less stack space on some tests (see #3055)

[mratsim]

It might suffice to eliminate those long test names by updating past v1.1.3, to v1.1.5:

Mmmh so the tests we skipped on mainnet were removed but not for minimal. We might want to skip minimal for now.

[tersec]

Mmmh so the tests we skipped on mainnet were removed but not for minimal. We might want to skip minimal for now.

Could also skip (as in, not even attempt to use) the long-named tests in minimal, and use the short-named tests. Either that or starting out by skipping minimal seems reasonable.