ksedgwic/clboss v0.17.3-rc1

on GitHub

Release candidate 1 for 0.17.3. Changes since 0.17.2 are listed below. Two notes on swaps first:

• Swaps on mainnet are not available at present. Boltz, the swap provider compiled into CLBOSS, suspended all swaps on 2026-08-03, and its host names no longer resolve. A mainnet node on this release gets no quotation and makes no swap; nothing is paid. The swap changes in this release were validated on signet.
• The swap invoice check (#329) is the fix that would have mattered during the eMudhra certificate authority key compromise of October 2026. An impersonated provider endpoint could have had CLBOSS pay an invoice of its choosing; the check removes that path, whatever the host's trust store contains.

Upgrading from 0.17.2

• Finish or wait out any swap in flight before upgrading, and before
  downgrading again. The swap table gains ten columns on first
  use, and a swap created by one version cannot be claimed by the
  other, so it ends at its timeout and the payment fails back to the
  node. The Boltz::Service log lines show one in flight. Earlier
  releases still run on the upgraded database.

Changed

• Boltz swaps use the provider's API v2 with taproot lockups: the swap
  is created at /v2/swap/reverse, polled at /v2/swap/{id}, quoted
  from the v2 pairs, and claimed cooperatively with one MuSig2 key-path
  signature, or through the claim leaf with our own signature and the
  preimage when the provider does not co-sign. Before paying, the setup
  writes both tapscript leaves itself and accepts the provider's only
  when they match byte for byte. The claim waits until our own node has
  the lockup three blocks deep, is recorded before the preimage is sent,
  is replaced when the going feerate rises above its own, and is tracked
  in our own blocks until the lockup is spent, whatever the provider
  reports; a spend that does not pay us is logged as the loss it is.
  The API v1 code is gone; the v1 table column redeemScript stays,
  empty, beside the new claimScript, refundScript, refundPubkey,
  createdHeight, lockupSearchedHeight, spendSearchedHeight,
  preimageRevealed, lockupTx, claimTx and claimFeerate (#254).
• The vendored libsecp256k1 moves from 0.3.2 to 0.6.0, which carries
  the MuSig2, extrakeys and schnorrsig modules that taproot swaps
  need.

Fixed

• A Boltz endpoint answering with something other than JSON, such
  as an HTML error page or an empty body, surfaced as an unhandled
  "Parse error near character '<'" with a backtrace and no hint of
  which request failed, and a feerates response without the
  opening estimate failed the same way. The connection now reports
  a parse failure or a missing result as an API error naming the
  endpoint, the handlers log that text with the response, and the
  fee lookup warns and falls back to 2000 perkw (#269).
• C++ exceptions are unwound by libgcc's unwinder on Debian and
  Ubuntu builds. Those distributions build libunwind with its own
  exception unwinder, and with -lunwind ahead of libstdc++ on the
  link line it took over every throw in CLBOSS and its tests, which
  made make check fail under valgrind. The build now links
  libgcc_s explicitly ahead of libunwind.
• A swap whose newaddr or swap-provider conversation died
  mid-flight could wedge SwapManager for the process lifetime:
  the flag serializing each processing loop was only cleared when
  the loop's queue drained, and the response that would have
  advanced it sometimes never arrives. The periodic tick now
  restarts a loop whose current request has had no answer for
  30 minutes; the previous remedy was restarting Core Lightning.
• A swap invoice from the provider was paid as received. The
  payer now checks the decoded invoice against the swap: the
  payment hash must be of the preimage CLBOSS generated and the
  amount must be the one the provider was asked for; any other
  invoice is logged and left unpaid. Before this check, anyone
  able to impersonate the provider's HTTPS endpoint, for example
  with a certificate issued under a compromised certificate
  authority, could have had CLBOSS pay an invoice of their
  choosing for the swap amount (#329).
• With two or more channels to one peer, the fee modifier by
  balance read only the first live channel, applied the multiplier
  it derived to all of them and recorded that one channel's balance
  as the peer's. It now sums our balance and the capacity over
  the peer's live channels, so the multiplier reflects the whole
  position and the balance samples PeerTrackRecord and the
  contrib tools read are the peer's (#352).
• With two or more channels to one peer, the active prober sized
  its probe from the first live channel listed, which could be the
  smaller one. It now probes through the channel with the most to
  spend; the probe names that channel, so its amount stays within
  one channel (#352).
• clboss-auto-close on a peer with more than one channel issued
  close id=<node>, which lightningd refuses ("Peer has multiple
  channels"); the error was logged and the close retried every
  cycle. It now closes each closable channel by channel id
  (#352).
• contrib/clboss-channel-sizing reads a peer's balance history
  recorded before 0.17.3 against the one channel it describes. Those
  samples carry the first channel's balance and capacity, not the
  peer's sum, and the tool apportioned them over all of the peer's
  channels, so a peer that opened a 5M channel beside a 1M one showed
  the 1M channel at several times its real turns. When the newest
  sample's total is not the peer's summed capacity, the channel whose
  capacity matches reads the samples and the peer's other channels
  read as young (#352).
• A new signet node can find peers again. Signet has no DNS seed, so
  the hardcoded seeds are the only way in, and all four of them had
  stopped accepting connections; they are replaced with seven nodes
  that accept them.

Credits

Thanks to the contributors to this release:

• The contributor who sent the API v2 draft: the first version of
  the Boltz API v2 code, with taproot reverse swaps and MuSig2
  claims (#254).
• @Mohil-Ahuja: the three fixes for several channels to one peer
  (#368, #371, #369), the balance-sample fix in
  contrib/clboss-channel-sizing (#378) and the replacement signet
  seeds (#379).
• @Amperstrand: reported that a swap invoice was paid without a
  check of its payment hash (#329).
• @tsjk: runs the signet swap backend that the swap changes were
  tested against.