nimbus-eth2/beacon_chain/spec/helpers.nim

655 lines
27 KiB
Nim

# beacon_chain
# Copyright (c) 2018-2022 Status Research & Development GmbH
# Licensed and distributed under either of
# * MIT license (license terms in the root directory or at https://opensource.org/licenses/MIT).
# * Apache v2 license (license terms in the root directory or at https://www.apache.org/licenses/LICENSE-2.0).
# at your option. This file may not be copied, modified, or distributed except according to those terms.
# Uncategorized helper functions from the spec
{.push raises: [Defect].}
# References to `vFuture` refer to the pre-release proposal of the libp2p based
# light client sync protocol. Conflicting release versions are not in use.
# https://github.com/ethereum/consensus-specs/pull/2802
import
# Standard lib
std/[algorithm, math, sequtils, sets, tables],
# Status libraries
stew/[bitops2, byteutils, endians2, objects],
chronicles,
# Internal
./datatypes/[phase0, altair, bellatrix],
"."/[eth2_merkleization, forks, ssz_codec]
# TODO although eth2_merkleization already exports ssz_codec, *sometimes* code
# fails to compile if the export is not done here also
export
forks, eth2_merkleization, ssz_codec
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/phase0/beacon-chain.md#integer_squareroot
func integer_squareroot*(n: SomeInteger): SomeInteger =
## Return the largest integer ``x`` such that ``x**2 <= n``.
doAssert n >= 0'u64
var
x = n
y = (x + 1) div 2
while y < x:
x = y
y = (x + n div x) div 2
x
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/ssz/merkle-proofs.md#generalized_index_sibling
template generalized_index_sibling*(
index: GeneralizedIndex): GeneralizedIndex =
index xor 1.GeneralizedIndex
template generalized_index_sibling_left(
index: GeneralizedIndex): GeneralizedIndex =
index and not 1.GeneralizedIndex
template generalized_index_sibling_right(
index: GeneralizedIndex): GeneralizedIndex =
index or 1.GeneralizedIndex
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/ssz/merkle-proofs.md#generalized_index_parent
template generalized_index_parent*(
index: GeneralizedIndex): GeneralizedIndex =
index shr 1
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/ssz/merkle-proofs.md#merkle-multiproofs
iterator get_branch_indices*(
tree_index: GeneralizedIndex): GeneralizedIndex =
## Get the generalized indices of the sister chunks along the path
## from the chunk with the given tree index to the root.
var index = tree_index
while index > 1.GeneralizedIndex:
yield generalized_index_sibling(index)
index = generalized_index_parent(index)
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/ssz/merkle-proofs.md#merkle-multiproofs
iterator get_path_indices*(
tree_index: GeneralizedIndex): GeneralizedIndex =
## Get the generalized indices of the chunks along the path
## from the chunk with the given tree index to the root.
var index = tree_index
while index > 1.GeneralizedIndex:
yield index
index = generalized_index_parent(index)
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/ssz/merkle-proofs.md#merkle-multiproofs
func get_helper_indices*(
indices: openArray[GeneralizedIndex]): seq[GeneralizedIndex] =
## Get the generalized indices of all "extra" chunks in the tree needed
## to prove the chunks with the given generalized indices. Note that the
## decreasing order is chosen deliberately to ensure equivalence to the order
## of hashes in a regular single-item Merkle proof in the single-item case.
var all_helper_indices = initHashSet[GeneralizedIndex]()
for index in indices:
for idx in get_branch_indices(index):
all_helper_indices.incl idx
for index in indices:
for idx in get_path_indices(index):
all_helper_indices.excl idx
var res = newSeqOfCap[GeneralizedIndex](all_helper_indices.len)
for idx in all_helper_indices:
res.add idx
res.sort(SortOrder.Descending)
res
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/ssz/merkle-proofs.md#merkle-multiproofs
func check_multiproof_acceptable*(
indices: openArray[GeneralizedIndex]): Result[void, string] =
# Check that proof verification won't allocate excessive amounts of memory.
const max_multiproof_complexity = nextPowerOfTwo(256)
if indices.len > max_multiproof_complexity:
trace "Max multiproof complexity exceeded",
num_indices=indices.len, max_multiproof_complexity
return err("Unsupported multiproof complexity (" & $indices.len & ")")
if indices.len == 0:
return err("No indices specified")
if indices.anyIt(it == 0.GeneralizedIndex):
return err("Invalid index specified")
ok()
func calculate_multi_merkle_root_impl(
leaves: openArray[Eth2Digest],
proof: openArray[Eth2Digest],
indices: openArray[GeneralizedIndex],
helper_indices: openArray[GeneralizedIndex]): Result[Eth2Digest, string] =
# All callers have already verified the checks in check_multiproof_acceptable,
# as well as whether lengths of leaves/indices and proof/helper_indices match.
# Helper to retrieve a value from a table that is statically known to exist.
template getExisting[A, B](t: var Table[A, B], key: A): var B =
try: t[key]
except KeyError: raiseAssert "Unreachable"
# Populate data structure with all leaves.
# This data structure only scales with the number of `leaves`,
# in contrast to the spec one that also scales with the number of `proof`
# items and the number of all intermediate roots, potentially the entire tree.
let capacity = nextPowerOfTwo(leaves.len)
var objects = initTable[GeneralizedIndex, Eth2Digest](capacity)
for i, index in indices:
if objects.mgetOrPut(index, leaves[i]) != leaves[i]:
return err("Conflicting roots for same index")
# Create list with keys of all active nodes that need to be visited.
# This list is sorted in descending order, same as `helper_indices`.
# Pulling from `objects` instead of from `indices` deduplicates the list.
var keys = newSeqOfCap[GeneralizedIndex](objects.len)
for index in objects.keys:
if index > 1.GeneralizedIndex: # For the root, no work needs to be done.
keys.add index
keys.sort(SortOrder.Descending)
# The merkle tree is processed from bottom to top, pulling in helper
# indices from `proof` as needed. During processing, the `keys` list
# may temporarily end up being split into two parts, sorted individually.
# An additional index tracks the current maximum element of the list.
var
completed = 0 # All key indices before this are fully processed.
maxIndex = completed # Index of the list's largest key.
helper = 0 # Helper index from `proof` to be pulled next.
# Processing is done when there are no more keys to process.
while completed < keys.len:
let
k = keys[maxIndex]
sibling = generalized_index_sibling(k)
left = generalized_index_sibling_left(k)
right = generalized_index_sibling_right(k)
parent = generalized_index_parent(k)
parentRight = generalized_index_sibling_right(parent)
# Keys need to be processed in descending order to ensure that intermediate
# roots remain available until they are no longer needed. This ensures that
# conflicting roots are detected in all cases.
keys[maxIndex] =
if not objects.hasKey(k):
# A previous computation did already merge this key with its sibling.
0.GeneralizedIndex
else:
# Compute expected root for parent. This deletes child roots.
# Because the list is sorted in descending order, they are not needed.
let root = withEth2Hash:
if helper < helper_indices.len and helper_indices[helper] == sibling:
# The next proof item is required to form the parent hash.
if sibling == left:
h.update proof[helper].data
h.update objects.getExisting(right).data; objects.del right
else:
h.update objects.getExisting(left).data; objects.del left
h.update proof[helper].data
inc helper
else:
# Both siblings are already known.
h.update objects.getExisting(left).data; objects.del left
h.update objects.getExisting(right).data; objects.del right
# Store parent root, and replace the current list entry with its parent.
if objects.hasKeyOrPut(parent, root):
if objects.getExisting(parent) != root:
return err("Conflicting roots for same index")
0.GeneralizedIndex
elif parent > 1.GeneralizedIndex:
# Note that the list may contain further nodes that are on a layer
# beneath the parent, so this may break the strictly descending order
# of the list. For example, given [12, 9], this will lead to [6, 9].
# This will resolve itself after the additional nodes are processed,
# i.e., [6, 9] -> [6, 4] -> [3, 4] -> [3, 2] -> [1].
parent
else:
0.GeneralizedIndex
if keys[maxIndex] != 0.GeneralizedIndex:
# The list may have been temporarily split up into two parts that are
# individually sorted in descending order. Have to first process further
# nodes until the list is sorted once more.
inc maxIndex
# Determine whether descending sort order has been restored.
let isSorted =
if maxIndex == completed: true
else:
while maxIndex < keys.len and keys[maxIndex] == 0.GeneralizedIndex:
inc maxIndex
maxIndex >= keys.len or keys[maxIndex] <= parentRight
if isSorted:
# List is sorted once more. Reset `maxIndex` to its start.
while completed < keys.len and keys[completed] == 0.GeneralizedIndex:
inc completed
maxIndex = completed
# Proof is guaranteed to provide all info needed to reach the root.
doAssert helper == helper_indices.len
doAssert objects.len == 1
ok(objects.getExisting(1.GeneralizedIndex))
func calculate_multi_merkle_root*(
leaves: openArray[Eth2Digest],
proof: openArray[Eth2Digest],
indices: openArray[GeneralizedIndex],
helper_indices: openArray[GeneralizedIndex]): Result[Eth2Digest, string] =
doAssert proof.len == helper_indices.len
if leaves.len != indices.len:
return err("Length mismatch for leaves and indices")
? check_multiproof_acceptable(indices)
calculate_multi_merkle_root_impl(
leaves, proof, indices, helper_indices)
func calculate_multi_merkle_root*(
leaves: openArray[Eth2Digest],
proof: openArray[Eth2Digest],
indices: openArray[GeneralizedIndex]): Result[Eth2Digest, string] =
if leaves.len != indices.len:
return err("Length mismatch for leaves and indices")
? check_multiproof_acceptable(indices)
calculate_multi_merkle_root_impl(
leaves, proof, indices, get_helper_indices(indices))
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/ssz/merkle-proofs.md#merkle-multiproofs
func verify_merkle_multiproof*(
leaves: openArray[Eth2Digest],
proof: openArray[Eth2Digest],
indices: openArray[GeneralizedIndex],
helper_indices: openArray[GeneralizedIndex],
root: Eth2Digest): bool =
let calc = calculate_multi_merkle_root(leaves, proof, indices, helper_indices)
if calc.isErr: return false
calc.get == root
func verify_merkle_multiproof*(
leaves: openArray[Eth2Digest],
proof: openArray[Eth2Digest],
indices: openArray[GeneralizedIndex],
root: Eth2Digest): bool =
let calc = calculate_multi_merkle_root(leaves, proof, indices)
if calc.isErr: return false
calc.get == root
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/phase0/beacon-chain.md#is_valid_merkle_branch
func is_valid_merkle_branch*(leaf: Eth2Digest, branch: openArray[Eth2Digest],
depth: int, index: uint64,
root: Eth2Digest): bool =
## Check if ``leaf`` at ``index`` verifies against the Merkle ``root`` and
## ``branch``.
var
value = leaf
buf: array[64, byte]
for i in 0 ..< depth:
if (index div (1'u64 shl i)) mod 2 != 0:
buf[0..31] = branch[i].data
buf[32..63] = value.data
else:
buf[0..31] = value.data
buf[32..63] = branch[i].data
value = eth2digest(buf)
value == root
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/tests/core/pyspec/eth2spec/test/helpers/merkle.py#L4-L21
func build_proof_impl(anchor: object, leaf_index: uint64,
proof: var openArray[Eth2Digest]) =
let
bottom_length = nextPow2(typeof(anchor).totalSerializedFields.uint64)
tree_depth = log2trunc(bottom_length)
parent_index =
if leaf_index < bottom_length shl 1:
0'u64
else:
var i = leaf_index
while i >= bottom_length shl 1:
i = i shr 1
i
var
prefix_len = 0
proof_len = log2trunc(leaf_index)
cache = newSeq[Eth2Digest](bottom_length shl 1)
block:
var i = bottom_length
anchor.enumInstanceSerializedFields(fieldNameVar, fieldVar):
if i == parent_index:
when fieldVar is object:
prefix_len = log2trunc(leaf_index) - tree_depth
proof_len -= prefix_len
let
bottom_bits = leaf_index and not (uint64.high shl prefix_len)
prefix_leaf_index = (1'u64 shl prefix_len) + bottom_bits
build_proof_impl(fieldVar, prefix_leaf_index, proof)
else: raiseAssert "Invalid leaf_index"
cache[i] = hash_tree_root(fieldVar)
i += 1
for i in countdown(bottom_length - 1, 1):
cache[i] = withEth2Hash:
h.update cache[i shl 1].data
h.update cache[i shl 1 + 1].data
var i = if parent_index != 0: parent_index
else: leaf_index
doAssert i > 0 and i < bottom_length shl 1
for proof_index in prefix_len ..< prefix_len + proof_len:
let b = (i and 1) != 0
i = i shr 1
proof[proof_index] = if b: cache[i shl 1]
else: cache[i shl 1 + 1]
func build_proof*(anchor: object, leaf_index: uint64,
proof: var openArray[Eth2Digest]) =
doAssert leaf_index > 0
doAssert proof.len == log2trunc(leaf_index)
build_proof_impl(anchor, leaf_index, proof)
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/phase0/beacon-chain.md#is_active_validator
func is_active_validator*(validator: Validator, epoch: Epoch): bool =
## Check if ``validator`` is active
validator.activation_epoch <= epoch and epoch < validator.exit_epoch
func is_exited_validator*(validator: Validator, epoch: Epoch): bool =
## Check if ``validator`` is exited
validator.exit_epoch <= epoch
func is_withdrawable_validator*(validator: Validator, epoch: Epoch): bool =
epoch >= validator.withdrawable_epoch
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/phase0/beacon-chain.md#get_active_validator_indices
iterator get_active_validator_indices*(state: ForkyBeaconState, epoch: Epoch):
ValidatorIndex =
for vidx in state.validators.vindices:
if is_active_validator(state.validators[vidx], epoch):
yield vidx
func get_active_validator_indices*(state: ForkyBeaconState, epoch: Epoch):
seq[ValidatorIndex] =
## Return the sequence of active validator indices at ``epoch``.
var res = newSeqOfCap[ValidatorIndex](state.validators.len)
for vidx in get_active_validator_indices(state, epoch):
res.add vidx
res
func get_active_validator_indices_len*(state: ForkyBeaconState, epoch: Epoch):
uint64 =
for vidx in state.validators.vindices:
if is_active_validator(state.validators.item(vidx), epoch):
inc result
func get_active_validator_indices_len*(
state: ForkedHashedBeaconState; epoch: Epoch): uint64 =
withState(state):
get_active_validator_indices_len(state.data, epoch)
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/phase0/beacon-chain.md#get_current_epoch
func get_current_epoch*(state: ForkyBeaconState): Epoch =
## Return the current epoch.
state.slot.epoch
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/phase0/beacon-chain.md#get_current_epoch
func get_current_epoch*(state: ForkedHashedBeaconState): Epoch =
## Return the current epoch.
withState(state): get_current_epoch(state.data)
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/phase0/beacon-chain.md#get_randao_mix
func get_randao_mix*(state: ForkyBeaconState, epoch: Epoch): Eth2Digest =
## Returns the randao mix at a recent ``epoch``.
state.randao_mixes[epoch mod EPOCHS_PER_HISTORICAL_VECTOR]
func bytes_to_uint64*(data: openArray[byte]): uint64 =
doAssert data.len == 8
# Little-endian data representation
uint64.fromBytesLE(data)
func uint_to_bytes*(x: uint64): array[8, byte] = toBytesLE(x)
func uint_to_bytes*(x: uint32): array[4, byte] = toBytesLE(x)
func uint_to_bytes*(x: uint16): array[2, byte] = toBytesLE(x)
func uint_to_bytes*(x: uint8): array[1, byte] = toBytesLE(x)
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/phase0/beacon-chain.md#compute_domain
func compute_domain*(
domain_type: DomainType,
fork_version: Version,
genesis_validators_root: Eth2Digest = ZERO_HASH): Eth2Domain =
## Return the domain for the ``domain_type`` and ``fork_version``.
let fork_data_root =
compute_fork_data_root(fork_version, genesis_validators_root)
result[0..3] = domain_type.data
result[4..31] = fork_data_root.data.toOpenArray(0, 27)
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/phase0/beacon-chain.md#get_domain
func get_domain*(
fork: Fork,
domain_type: DomainType,
epoch: Epoch,
genesis_validators_root: Eth2Digest): Eth2Domain =
## Return the signature domain (fork version concatenated with domain type)
## of a message.
let fork_version =
if epoch < fork.epoch:
fork.previous_version
else:
fork.current_version
compute_domain(domain_type, fork_version, genesis_validators_root)
func get_domain*(
state: ForkyBeaconState, domain_type: DomainType, epoch: Epoch): Eth2Domain =
## Return the signature domain (fork version concatenated with domain type)
## of a message.
get_domain(state.fork, domain_type, epoch, state.genesis_validators_root)
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/phase0/beacon-chain.md#compute_signing_root
func compute_signing_root*(ssz_object: auto, domain: Eth2Domain): Eth2Digest =
## Return the signing root of an object by calculating the root of the
## object-domain tree.
let domain_wrapped_object = SigningData(
object_root: hash_tree_root(ssz_object),
domain: domain
)
hash_tree_root(domain_wrapped_object)
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/phase0/beacon-chain.md#get_seed
func get_seed*(state: ForkyBeaconState, epoch: Epoch, domain_type: DomainType):
Eth2Digest =
## Return the seed at ``epoch``.
var seed_input : array[4+8+32, byte]
# Detect potential underflow
static:
doAssert EPOCHS_PER_HISTORICAL_VECTOR > MIN_SEED_LOOKAHEAD
seed_input[0..3] = domain_type.data
seed_input[4..11] = uint_to_bytes(epoch.uint64)
seed_input[12..43] =
get_randao_mix(state, # Avoid underflow
epoch + EPOCHS_PER_HISTORICAL_VECTOR - MIN_SEED_LOOKAHEAD - 1).data
eth2digest(seed_input)
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/altair/beacon-chain.md#add_flag
func add_flag*(flags: ParticipationFlags, flag_index: int): ParticipationFlags =
let flag = ParticipationFlags(1'u8 shl flag_index)
flags or flag
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/altair/beacon-chain.md#has_flag
func has_flag*(flags: ParticipationFlags, flag_index: int): bool =
let flag = ParticipationFlags(1'u8 shl flag_index)
(flags and flag) == flag
# https://github.com/ethereum/consensus-specs/blob/vFuture/specs/altair/sync-protocol.md#is_sync_committee_update
template is_sync_committee_update*(update: SomeLightClientUpdate): bool =
when update is SomeLightClientUpdateWithSyncCommittee:
not isZeroMemory(update.next_sync_committee_branch)
else:
false
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/altair/sync-protocol.md#get_active_header
template is_finality_update*(update: SomeLightClientUpdate): bool =
when update is SomeLightClientUpdateWithFinality:
not isZeroMemory(update.finality_branch)
else:
false
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/altair/sync-protocol.md#get_subtree_index
func get_subtree_index*(idx: GeneralizedIndex): uint64 =
doAssert idx > 0
uint64(idx mod (type(idx)(1) shl log2trunc(idx)))
# https://github.com/ethereum/consensus-specs/blob/vFuture/specs/altair/sync-protocol.md#is_next_sync_committee_known
template is_next_sync_committee_known*(store: LightClientStore): bool =
not isZeroMemory(store.next_sync_committee)
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/altair/sync-protocol.md#get_safety_threshold
func get_safety_threshold*(store: LightClientStore): uint64 =
max(
store.previous_max_active_participants,
store.current_max_active_participants
) div 2
# https://github.com/ethereum/consensus-specs/blob/vFuture/specs/altair/sync-protocol.md#is_better_update
type LightClientUpdateMetadata* = object
attested_slot*, finalized_slot*, signature_slot*: Slot
has_sync_committee*, has_finality*: bool
num_active_participants*: uint64
func toMeta*(update: SomeLightClientUpdate): LightClientUpdateMetadata =
var meta {.noinit.}: LightClientUpdateMetadata
meta.attested_slot =
update.attested_header.slot
meta.finalized_slot =
when update is SomeLightClientUpdateWithFinality:
update.finalized_header.slot
else:
GENESIS_SLOT
meta.signature_slot =
update.signature_slot
meta.has_sync_committee =
when update is SomeLightClientUpdateWithSyncCommittee:
not update.next_sync_committee_branch.isZeroMemory
else:
false
meta.has_finality =
when update is SomeLightClientUpdateWithFinality:
not update.finality_branch.isZeroMemory
else:
false
meta.num_active_participants =
countOnes(update.sync_aggregate.sync_committee_bits).uint64
meta
func is_better_data*(new_meta, old_meta: LightClientUpdateMetadata): bool =
# Compare supermajority (> 2/3) sync committee participation
const max_active_participants = SYNC_COMMITTEE_SIZE.uint64
let
new_has_supermajority =
new_meta.num_active_participants * 3 >= max_active_participants * 2
old_has_supermajority =
old_meta.num_active_participants * 3 >= max_active_participants * 2
if new_has_supermajority != old_has_supermajority:
return new_has_supermajority > old_has_supermajority
if not new_has_supermajority:
if new_meta.num_active_participants != old_meta.num_active_participants:
return new_meta.num_active_participants > old_meta.num_active_participants
# Compare presence of relevant sync committee
let
new_has_relevant_sync_committee = new_meta.has_sync_committee and
new_meta.attested_slot.sync_committee_period ==
new_meta.signature_slot.sync_committee_period
old_has_relevant_sync_committee = old_meta.has_sync_committee and
old_meta.attested_slot.sync_committee_period ==
old_meta.signature_slot.sync_committee_period
if new_has_relevant_sync_committee != old_has_relevant_sync_committee:
return new_has_relevant_sync_committee > old_has_relevant_sync_committee
# Compare indication of any finality
if new_meta.has_finality != old_meta.has_finality:
return new_meta.has_finality > old_meta.has_finality
# Compare sync committee finality
if new_meta.has_finality:
let
new_has_sync_committee_finality =
new_meta.finalized_slot.sync_committee_period ==
new_meta.attested_slot.sync_committee_period
old_has_sync_committee_finality =
old_meta.finalized_slot.sync_committee_period ==
old_meta.attested_slot.sync_committee_period
if new_has_sync_committee_finality != old_has_sync_committee_finality:
return new_has_sync_committee_finality > old_has_sync_committee_finality
# Tiebreaker 1: Sync committee participation beyond supermajority
if new_meta.num_active_participants != old_meta.num_active_participants:
return new_meta.num_active_participants > old_meta.num_active_participants
# Tiebreaker 2: Prefer older data (fewer changes to best data)
new_meta.attested_slot < old_meta.attested_slot
template is_better_update*[A, B: SomeLightClientUpdate](
new_update: A, old_update: B): bool =
is_better_data(toMeta(new_update), toMeta(old_update))
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/bellatrix/beacon-chain.md#is_merge_transition_complete
func is_merge_transition_complete*(state: bellatrix.BeaconState): bool =
const defaultExecutionPayloadHeader = default(ExecutionPayloadHeader)
state.latest_execution_payload_header != defaultExecutionPayloadHeader
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/sync/optimistic.md#helpers
func is_execution_block*(
body: bellatrix.BeaconBlockBody | bellatrix.TrustedBeaconBlockBody |
bellatrix.SigVerifiedBeaconBlockBody): bool =
const defaultBellatrixExecutionPayload = default(bellatrix.ExecutionPayload)
body.execution_payload != defaultBellatrixExecutionPayload
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/bellatrix/beacon-chain.md#is_merge_transition_block
func is_merge_transition_block(
state: bellatrix.BeaconState,
body: bellatrix.BeaconBlockBody | bellatrix.TrustedBeaconBlockBody |
bellatrix.SigVerifiedBeaconBlockBody): bool =
const defaultBellatrixExecutionPayload = default(bellatrix.ExecutionPayload)
not is_merge_transition_complete(state) and
body.execution_payload != defaultBellatrixExecutionPayload
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/bellatrix/beacon-chain.md#is_execution_enabled
func is_execution_enabled*(
state: bellatrix.BeaconState,
body: bellatrix.BeaconBlockBody | bellatrix.TrustedBeaconBlockBody |
bellatrix.SigVerifiedBeaconBlockBody): bool =
is_merge_transition_block(state, body) or is_merge_transition_complete(state)
# https://github.com/ethereum/consensus-specs/blob/v1.2.0-rc.1/specs/bellatrix/beacon-chain.md#compute_timestamp_at_slot
func compute_timestamp_at_slot*(state: ForkyBeaconState, slot: Slot): uint64 =
# Note: This function is unsafe with respect to overflows and underflows.
let slots_since_genesis = slot - GENESIS_SLOT
state.genesis_time + slots_since_genesis * SECONDS_PER_SLOT
# https://github.com/ethereum/consensus-specs/blob/v1.1.10/tests/core/pyspec/eth2spec/test/helpers/execution_payload.py#L1-L31
func build_empty_execution_payload*(state: bellatrix.BeaconState): ExecutionPayload =
## Assuming a pre-state of the same slot, build a valid ExecutionPayload
## without any transactions.
let
latest = state.latest_execution_payload_header
timestamp = compute_timestamp_at_slot(state, state.slot)
randao_mix = get_randao_mix(state, get_current_epoch(state))
var payload = ExecutionPayload(
parent_hash: latest.block_hash,
state_root: latest.state_root, # no changes to the state
receipts_root: static(Eth2Digest.fromHex(
"0x56e81f171bcc55a6ff8345e692c0f86e5b48e01b996cadc001622fb5e363b421")),
block_number: latest.block_number + 1,
prev_randao: randao_mix,
gas_limit: latest.gas_limit, # retain same limit
gas_used: 0, # empty block, 0 gas
timestamp: timestamp,
base_fee_per_gas: latest.base_fee_per_gas) # retain same base_fee
payload.block_hash = withEth2Hash:
h.update payload.hash_tree_root().data
h.update cast[array[13, uint8]]("FAKE RLP HASH")
payload