zakura-core/zakura v1.8.5

on GitHub

Changed

• Relayed blocks are gossiped as soon as they are committed. Previously every tip change
  waited about 1 ms for a possible locally submitted block, which added to each relay hop
  (#2087).
• Require the shared mined-submission registry when constructing node RPC services. Update
  the state, consensus, and RPC crate major versions for the Rust interface changes, and the
  utilities crate dependency version. JSON-RPC methods and stored data formats are unchanged.
• Announce successfully submitted local blocks directly from committed state, including
  side-chain blocks, without waiting for the submitting RPC task to resume
  (#2093).
• Return state services and update receivers in a named InitializedState result
  so Rust library callers can select the fields they need.
• Change the Rust ParentInputs::Unspent response to retain a bounded subset of
  outputs. Library callers must look up omitted outputs rather than assume the
  map is complete; the state and consensus crates receive major version bumps
  (#2116).
• Skip the duplicate transparent-spend scan for blocks without non-coinbase
  transparent inputs, reusing the existing coinbase-validation pass
  (#2117).
• Updated Common to 5.3.0 to pick up Orchard and Halo 2 verification optimizations
  (#2133).
• Classify the compatible relay, verification, and dependency improvements as a
  patch node release; library crates retain their independent version
  requirements (#2148).

Fixed

• A peer that abandons a request by stopping its response stream now ends the
  request's service call and returns the request's output reservation at once,
  instead of at the first response write
  (#2064).
• Fixed the discovery candidate dialer so that candidates it can't dial, such
  as those on an IP at the per-IP cap or with more addresses than the free
  connection slots, no longer take a round's dial slots from other candidates,
  and so that finished dials no longer start rounds faster than the one-second
  dial interval
  (#2065).
• Discovery now confirms only the record addresses this node authenticated. A
  peer's own Hello confirms the record addresses on its connection's remote
  IP, and a dial confirms only the exact address it reached. A record that
  replaces an expired one without a higher sequence starts with no confirmed
  address. A record becomes eligible for peer samples once one of its
  addresses is confirmed. On each IP, the discovery dialer tries the address
  a successful dial reached first, then the record's other confirmed
  addresses (#2066).
• Replaced rolling block-advertisement windows with one size-dependent delay.
  Peers outside the initial group no longer wait for other peers to request or
  finish uploading a block. Restored the initial-group defaults to 32/64 peers
  and preserved explicitly configured bounds. New advert_delay_min and
  advert_delay_max settings default to zero, so all peers receive announcements
  immediately by default, including for blocks larger than 100 KiB
  (#2082).
• Notify block-upload tracking when a connection fails, before closing its writer.
  Retry failed announcement snapshots when replacement peers arrive, without
  restarting the announcement delay. Retain bounded recent-success history so
  timeout fallbacks can skip peers whose earlier announcement succeeded.
• Complete singleton block and transaction fetches over Zakura compatibility
  streams as soon as the item or missing-item record arrives, without waiting
  for the peer to finish the response stream
  (#2086).
• Coordinate mining broadcasts with state admission, so a peer's committed copy is relayed
  promptly even while the same local submission is still verifying. Admitted local submissions
  suppress ordinary committed-tip gossip for at most 90 seconds from their first deferral,
  including across tip changes, while retaining verification capacity. This bound covers
  submission coordination, not network delivery
  (#2090).
• Suppress delayed optimistic block announcements after state commits the submission,
  while preserving committed-broadcast retries.
• A peer that header sync faulted becomes eligible for a header session again
  as soon as its readmission cooldown ends. Previously the node could keep
  refusing its sessions for up to about another cooldown. On an inbound
  connection the peer still has to re-offer the session, which a Zakura peer
  does within a few seconds
  (#2104).
• Reduced block-processing overhead for indexed transparent addresses with long
  retained payment histories by sharing unchanged groups of output heights
  (#2112).
• Avoid repeating every parent-output lookup when transparent scripts cross the
  retention budget, keeping the existing limit on retained script bytes
  (#2116).
• Reject blocks whose transactions spend the same transparent output before
  performing parent-output reads and expensive transaction verification
  (#2117).