rbitcoin v0.8.0

rbitcoin v0.8.0

Linux musl x86_64 is the operator binary (statically linked).
Windows is CRT-static PE (no IoRing). Darwin aarch64 is ad-hoc
codesigned, not notarized (xattr -d com.apple.quarantine).

Highlights

• Schema 26 opens a 0.7 datadir in place. Schema 24 and 25 rewrite
  header.body from 96 B to 88 B and meta to 26. A store older than schema
  22, including occupied 0.6.x, refuses with one wipe-and-IBD line.
• Mainnet script skip is path-anchored. The default milestone still names
  height 840000 and block
  0000000000000000000320283a032748cef8227873ff4872689bf23f1cda83a5. Scripts
  skip only on that header path once chain work meets Core
  nMinimumChainWork. Signet’s default milestone is 0 (every script). An
  explicit --milestone HEIGHT stays height-only.
• Initial sync keeps moving on ordinary connections. One peer serves
  headers. The others download blocks, at most 64 blocks or 16 MiB in
  flight, sized from recent bodies. A peer holding the tip block is
  replaced only after 5 seconds, and only when a free peer would finish
  it in half the time. A full serve queue waits out the rest of a
  getdata instead of dropping it.
• Consensus checks match Bitcoin Core where they used to disagree.
  BIP30 is off only when the header at BIP34 height is that network’s
  BIP34 block, and it is always on from height 1,983,702. A confirm batch
  cannot spend an output created by a later block in that batch. The
  genesis coinbase is not a spendable coin. An empty signature in legacy
  CHECKMULTISIG deletes OP_0 from scriptCode. A bad copy of a valid
  block is requested again and is not cached as invalid.
• Peers and RPC fail closed under load. Counts above 50,000 in inv,
  getdata, and notfound fail decode. Compact-block and block
  transaction counts are rejected before allocation. The RPC body cap is
  2 MiB, and a full work queue is HTTP 503 before the handler. A session
  pauses once its outbound queue passes 4 MiB. An unpaged scripthash join
  above --max-sh-creates (default 10000) is refused.
• Wallets, Lightning nodes, and explorers have a setup guide.
  docs/wallets.md and docs/lightning.md. Electrum and Esplora serve
  Electrum, Sparrow, Liana, Specter, Cake Wallet, BlueWallet, BitBox App,
  Envoy, and BDK (including Alby Hub). Wasabi’s full-node path uses
  getblockfilter and --rpc-cookie-file (HTTP Basic on TCP).
  getnetworkinfo.version stays 190000 (Core 0.19’s client integer); the
  semver stays in subversion. CLN (bcli and sauron), ldk-node
  (Esplora, Electrum, or REST), and LND with bitcoind.rpcpolling use
  this node as the chain source. Esplora /internal/* on the unix socket
  is the mempool.space electrs drop-in. /fee-estimates keeps the
  estimator’s precision. /fees/recommended and /ws are 404.
• Feerate uses sigop-adjusted size. Admission, RBF, eviction, estimates,
  and -blockmintxfee use max(weight, sigop_cost × bytes_per_sigop)
  (default 20, --bytes-per-sigop, 0 disables). A transaction whose sigop
  cost reaches the template budget is rejected. Template selection applies
  the mintxfee floor per chunk and keeps the block sigop cap.
• Fee quotes follow a log grid and a fullness blend. About 100 steps per
  decade from min relay (0.1 sat/vB) through 1000 sat/vB, plus one open
  bucket above that. The published rate is (1−α)·history + α·live flow.
  α is decayed admitted weight over about 1.44M WU (150s half-life). History
  is filled from txstat (at most 1 GiB; each block’s vsize-weighted p10
  of individual rates). estimatesmartfee returns {feerate, blocks}, or
  {errors, blocks} with no feerate when it has no estimate, and
  feerate is at least mempoolminfee.
• Block filters and silent-payment tweaks build during IBD, and after
  it when they are behind. With the index on from the start, each
  confirmed block is built on the script pool and appended on the confirm
  write. A restart gap of at most 2,016 blocks is sealed before confirm.
  A larger gap, including an index enabled on an already synced datadir,
  is built afterward by rbtc-idx-wb (index: build) and does not hold
  the node in initial download. NODE_COMPACT_FILTERS is advertised once
  filters reach the tip. With --datadir-cold, blockfilter and
  sp_tweaks must sit on the cold store.
• Spend annotations recover after a crash past the tip seal. A missing
  spent slot stays unspent until open replays above the spend_durable
  marker. rbtc-spend-sync checkpoints about every ten minutes, and once
  more on shutdown.

Thanks to @otaliptus for the security review, and to @dergoegge, @Rob1Ham, and @1440000bytes for earlier findings.

Thanks to @benthecarman, @average-gary, @bobberb, @Hero-Gamer, @bkeroack, @xstoicunicornx, and @RandyMcMillan for changes in this release.

Full notes: CHANGELOG.md ## [0.8.0].