TheVerge: spec draft

[gballet]

This PR adds the execution witness to the ExecutionPayload container, and also describes the fork.

TODO:

• Check that the fork description is correct
• Add a document about p2p, if needed
• Address the FIXME in fork.md for historical_summaries

Co-Authored-By: Dankrad Feist [email protected]

[hwwhww]

Hi @gballet, given that we merged #3248, could you rebase and move specs/verge to specs/_features/verge? 🙏

[dapplion]

Should this be rebased on top of deneb? CC @terencechain

[djrtwo]

Let's swap this feature name/dir to reflect the EIP number -- https://eips.ethereum.org/EIPS/eip-6800

[etan-status]

These should either be updated for EIP-7495 StableContainer / Profile (on SuffixStateDiff), or unwrapped so that e.g. a zero value is used instead of None. Would like to withdraw EIP-6475 style Optional as the functionality is covered by the more advanced EIP-7495 as well.

SSZ library implementation progress here: https://stabilitynow.box

[hwwhww]

Oh right.

@gballet
The current Optional in pyspec is the Python typing.Optional. Our SSZ lib Remerkleable only implemented SSZ Union, so technically we can use Union[None, Bytes32] here in the specs. However, other SSZ libs didn't implement Union and may go implement the newer SSZ types as @etan-status suggested.

Is it possible to use Bytes32(), the empty 0x00...00, to represent null in this case so that we don't introduce new SSZ typing here?

[gballet]

no it is not possible to use Bytes32() and I was initially using Union[None, Byte32] and was asked to replace it with Optional. Precisely because Union didn't exist (except in the wonderful gballet/ssz.zig, of course 😇).

[etan-status]

If you want to give StableContainer a shot, this is how it would look there:

class StableSuffixStateDiff(StableContainer[8]):  # or [4] if this is not planned to grow in the future
    suffix: Optional[Bytes1]
    # Null means not currently present
    current_value: Optional[Bytes32]
    # Null means value not updated
    new_value: Optional[Bytes32]

class SuffixStateDiff(Profile[StableSuffixStateDiff]):
    suffix: Bytes1
    # Null means not currently present
    current_value: Optional[Bytes32]
    # Null means value not updated
    new_value: Optional[Bytes32]

Serialization of the Profile looks like this:

1. 1 byte optional_fields, 0b1/0 for current_value presence/absence, 0b1/0 for new_value presence/absence
2. 1 byte suffix
3. 32 bytes current_value if present, or nothing otherwise
4. 32 bytes new_value if present, or nothing otherwise

Notably, no variable length offsets, this is 7-9 bytes more compact than the current EIP-6475 based impl, and requires one fewer hash as well.

Merkleization would look like this, with htr == hash_tree_root and based on 8 capacity. can lower to 4 if sufficient, should reflect theoretical design space, not actual space used today, for future extensibility without breaking merkleization of existing EIP-4788 based verifiers:

• htr
  • htr
    • htr
      • htr
        • htr(suffix)
        • htr(current_value) if current_value is not None else htr(0)
      • htr
        • htr(new_value) if new_value is not None else htr(0)
        • htr(0)
    • htr
      • htr
        • htr(0)
        • htr(0)
      • htr
        • htr(0)
        • htr(0)
  • active_fields (suffix always 0b1, then 0b0/0b1 for current_value, then 0b0/0b1 for new_value, rest 0)

[etan-status]

Needs custom remerkleable from here:

#3777

[hwwhww]

lgtm as the initial version. 👍

It now passes the linter but we haven't enabled the tests yet.