After checkpoint sync, blocks between the finalized checkpoint and head
are already in fork choice but their envelopes aren't in the store.
When range sync downloads these blocks, they get skipped as
DuplicateFullyImported without persisting the envelope. The next new
block then fails load_parent because it can't find its parent's envelope.
Fix: persist the envelope in the DuplicateFullyImported path, same as
the existing WouldRevertFinalizedSlot path.
Allow for the vc to submit its proposer preferences to the network
Co-Authored-By: Eitan Seri-Levi <eserilev@ucsc.edu>
Co-Authored-By: Jimmy Chen <jchen.tc@gmail.com>
- pending_column.rs: split try_to_sidecar into is_complete-checked to_sidecar
with typed PendingColumnError so 'incomplete column' is no longer conflated
with VariableList size-bound failures (jimmygchen, dapplion).
- pending_components.rs: get_cached_data_columns filters by is_complete first,
then logs an error if a complete column fails to assemble (dknopik's sanity
check on filter_map silent drops).
- data_column_verification.rs: add the missing column.slot == bid.slot
consistency check in validate_data_column_sidecar_for_gossip_gloas, using
the previously-defined-but-unused BlockSlotMismatch error variant
(jimmygchen).
Brings the FORK_NAME=gloas beacon_chain test suite from 31 failures to green:
- v1 KZG batch verifier couldn't verify Gloas columns. Added
verify_columns_against_block helper that picks commitments per fork
(Fulu: inline on column; Gloas: signed_execution_payload_bid).
- BeaconChainHarness::process_envelope didn't persist columns. Now mirrors
what production does in import_available_execution_payload_envelope.
- get_or_reconstruct_blobs returned an error for Gloas. Now short-circuits to
Ok(None); WSS test copies columns from source to dest directly.
- update_data_column_signed_header (block_verification tests) only handled
Fulu shape. Added a Gloas branch that re-keys to canonical_root.
- BlockError::EnvelopeBlockRootUnknown changed to tuple variant.
- Removed duplicate process_payload_envelope_availability.
Brings the FORK_NAME=gloas beacon_chain test suite from 31 failures to green:
- v1 KZG batch verifier couldn't verify Gloas columns. Added
verify_columns_against_block helper that picks commitments per fork
(Fulu: inline on column; Gloas: signed_execution_payload_bid).
- BeaconChainHarness::process_envelope didn't persist columns. Now mirrors
what production does in import_available_execution_payload_envelope.
- get_or_reconstruct_blobs returned an error for Gloas. Now short-circuits to
Ok(None); WSS test copies columns from source to dest directly.
- update_data_column_signed_header (block_verification tests) only handled
Fulu shape. Added a Gloas branch that re-keys to canonical_root.
- BlockError::EnvelopeBlockRootUnknown changed to tuple variant.
- Removed duplicate process_payload_envelope_availability.
- Replace the `PendingPayloadBid` projection (slot + blob_kzg_commitments)
with `Arc<SignedExecutionPayloadBid<E>>`. Cloning becomes a cheap Arc
bump and the bid carries enough context for future bid<->envelope
cross-checks. Add a `signed_payload_bid_from_block` helper.
- `PendingColumn` switches from a pre-sized `Vec<Option<_>>` to a sparse
`HashMap<usize, _>`; the `new_with_capacity(num_blobs)` constructor is
gone since callers no longer need to know the blob count up front.
- `PendingComponents::merge_data_columns` takes a slice instead of an
owning iterator (it only borrows + clones cells).
- Store `block_root` in `PendingComponents` so `make_available` and
`get_cached_data_columns` no longer require it as an argument (the
arg was misnamed `block_hash` in `make_available`).
- Rename `PendingComponents::empty` -> `new`; it is the only constructor.
Tests were switched from put_kzg_verified_custody_data_columns (no
filtering) to put_rpc_custody_columns (filters to sampling columns) but
assertions still assumed all provided columns would be stored. Account
for the sampling filter in each test's setup and assertions.
We yolo'd to alpha 7. We're just changing the proposer preference to include dependent root, instead of checkpoint root. This way we can actually construct it within the VC without needing a view of fork choice.
Co-Authored-By: Eitan Seri-Levi <eserilev@ucsc.edu>
