This document describes the process of the Capella upgrade.
Name | Value |
---|---|
CAPELLA_FORK_VERSION |
Version('0x03000000') |
CAPELLA_FORK_EPOCH |
Epoch(194048) (April 12, 2023, 10:27:35pm UTC) |
def compute_fork_version(epoch: Epoch) -> Version:
"""
Return the fork version at the given ``epoch``.
"""
if epoch >= CAPELLA_FORK_EPOCH:
return CAPELLA_FORK_VERSION
if epoch >= BELLATRIX_FORK_EPOCH:
return BELLATRIX_FORK_VERSION
if epoch >= ALTAIR_FORK_EPOCH:
return ALTAIR_FORK_VERSION
return GENESIS_FORK_VERSION
The fork is triggered at epoch CAPELLA_FORK_EPOCH
.
Note that for the pure Capella networks, we don't apply upgrade_to_capella
since it starts with Capella version logic.
If state.slot % SLOTS_PER_EPOCH == 0
and compute_epoch_at_slot(state.slot) == CAPELLA_FORK_EPOCH
,
an irregular state change is made to upgrade to Capella.
The upgrade occurs after the completion of the inner loop of process_slots
that sets state.slot
equal to CAPELLA_FORK_EPOCH * SLOTS_PER_EPOCH
.
Care must be taken when transitioning through the fork boundary as implementations will need a modified state transition function that deviates from the Phase 0 document.
In particular, the outer state_transition
function defined in the Phase 0 document will not expose the precise fork slot to execute the upgrade in the presence of skipped slots at the fork boundary. Instead, the logic must be within process_slots
.
def upgrade_to_capella(pre: bellatrix.BeaconState) -> BeaconState:
epoch = bellatrix.get_current_epoch(pre)
latest_execution_payload_header = ExecutionPayloadHeader(
parent_hash=pre.latest_execution_payload_header.parent_hash,
fee_recipient=pre.latest_execution_payload_header.fee_recipient,
state_root=pre.latest_execution_payload_header.state_root,
receipts_root=pre.latest_execution_payload_header.receipts_root,
logs_bloom=pre.latest_execution_payload_header.logs_bloom,
prev_randao=pre.latest_execution_payload_header.prev_randao,
block_number=pre.latest_execution_payload_header.block_number,
gas_limit=pre.latest_execution_payload_header.gas_limit,
gas_used=pre.latest_execution_payload_header.gas_used,
timestamp=pre.latest_execution_payload_header.timestamp,
extra_data=pre.latest_execution_payload_header.extra_data,
base_fee_per_gas=pre.latest_execution_payload_header.base_fee_per_gas,
block_hash=pre.latest_execution_payload_header.block_hash,
transactions_root=pre.latest_execution_payload_header.transactions_root,
withdrawals_root=Root(), # [New in Capella]
)
post = BeaconState(
# Versioning
genesis_time=pre.genesis_time,
genesis_validators_root=pre.genesis_validators_root,
slot=pre.slot,
fork=Fork(
previous_version=pre.fork.current_version,
current_version=CAPELLA_FORK_VERSION,
epoch=epoch,
),
# History
latest_block_header=pre.latest_block_header,
block_roots=pre.block_roots,
state_roots=pre.state_roots,
historical_roots=pre.historical_roots,
# Eth1
eth1_data=pre.eth1_data,
eth1_data_votes=pre.eth1_data_votes,
eth1_deposit_index=pre.eth1_deposit_index,
# Registry
validators=pre.validators,
balances=pre.balances,
# Randomness
randao_mixes=pre.randao_mixes,
# Slashings
slashings=pre.slashings,
# Participation
previous_epoch_participation=pre.previous_epoch_participation,
current_epoch_participation=pre.current_epoch_participation,
# Finality
justification_bits=pre.justification_bits,
previous_justified_checkpoint=pre.previous_justified_checkpoint,
current_justified_checkpoint=pre.current_justified_checkpoint,
finalized_checkpoint=pre.finalized_checkpoint,
# Inactivity
inactivity_scores=pre.inactivity_scores,
# Sync
current_sync_committee=pre.current_sync_committee,
next_sync_committee=pre.next_sync_committee,
# Execution-layer
latest_execution_payload_header=latest_execution_payload_header,
# Withdrawals
next_withdrawal_index=WithdrawalIndex(0), # [New in Capella]
next_withdrawal_validator_index=ValidatorIndex(0), # [New in Capella]
# Deep history valid from Capella onwards
historical_summaries=List[HistoricalSummary, HISTORICAL_ROOTS_LIMIT]([]), # [New in Capella]
)
return post