Add test_genesis.py and cleanup genesis trigger

[hwwhww]

1. Fix is_genesis_trigger fails to verify deposit merkle branch #1199

• modify to initialize_beacon_state_from_eth
• add is_valid_genesis_state

2. Add test_libs/pyspec/eth2spec/test/genesis/ tests
3. Update deposit contract (per Make deposit root the root of an SSZ list #1229)

[JustinDrake]

@hwwhww What do you think of a3f8f50?

[hwwhww]

@JustinDrake

In practice, a client, they would probably:

1. Build deposit_data_list: List[DepositData] from deposit contract log.
2. Generate proof against the latest deposit_root (can be gotten from the contract, or just merkleize the deposit_data_list).
3. Build the deposits: List[Deposit] with these proof.
4. Check if it triggers the genesis starting point.

So it's unlikely they would need to regenerate deposit_root after they have created the Deposits, in practice. That's why I'd say we don't need to regenerate the deposit_root again.

Another way to describe the logic is changing the function to def is_genesis_trigger(deposit_data_list: List[DepositData], timestamp: uint64) -> bool, and generate the proof in this function. I hope to balance between (i) making the spec describe the protocol in details and (ii) making it too far from practice. 😄

What do you think about it? 😸

[JustinDrake]

Right. In practice I expect implementers to have a cache (e.g. a Store mapping deposits to beacon states) and an incremental version of is_genesis_trigger. This would avoid recomputing things from genesis, and handle Eth1 reorgs. I'd favour spec readability over efficiency, but don't feel too strongly either way :)

[protolambda]

• fixed the merge from Justin, it was broken:
  • introduced back an invalid deposit test, we don't have deposit.index anymore
  • syntax problem, decorator was over empty line
  • confusing big merge, few merge conflicts, but know not to pick the old(!!!) diff parts, when they changed intentionally...
• Moved genesis tests out of sanity. The idea is that the shape of each test is the same within one file.
• Annotated the yielded data, so that it actually gets into a dict of key-value for a future test generator
• Fixed the genesis-trigger-true tests partially. Date was the same as the false case...
• Introduced local SECONDS_PER_DAY constant. You can make it a real global, but if it's for an "example" part of the code, we shouldn't include it in the main constants imho.
• Fixed problems with python syntax, map was used wrong, and generator (... for ... in ...) results were used as if it were lists (they are types.GeneratorType)
• The Deposit-root for the truee-trigger is still wrong somehow, but something for you to figure out, may be the expected root that is wrong, or some other funky reason more familiar to the PR authors

[protolambda]

• Added signed=True args to make it work with BLS
• Added a few remarks to avoid confusion in genesis/deposit verification

Ready to merge imho. We can add a few additional tests after spec-freeze. Possibly even improve the efficiency of genesis construction. As clients are (very) likely not implementing it like this anyway, and state output will be the same. This description and pseudocode will at least set a baseline, and set the interface to be compatible with SSZ partials in the future.

[djrtwo]

this cleaned up nicely. Great work everyone!