Commit Graph
1 Commits
Author SHA1 Message Date
Kgothatso NgakoandClaude Opus 5.5 d8d8d038f9 test(lightning): the group's wallet end to end on regtest, and what it found
`testing/regtest/` builds bitcoind, electrs, an eclair v0.14.4 LSP set up as
a Phoenix LSP, and a payer. `GroupLightningRegtestJvmTest` (run with
BENNU_REGTEST=1) drives four devices through it with a real lightning-kmp
node: open, receive on a channel opened on the fly, pay through the LSP's
trampoline, close mutually, then force-close a second channel and sweep its
delayed output with a FROST script-path spend. It passes from a fresh
network in about thirty seconds.

The run found four bugs in the operator's node, fixed here:

- It learned the chain's height only at the next block, and until then
  could answer no invoice request and take no payment. It now asks as it
  starts.
- It turned down every on-the-fly channel, because lightning-kmp's default
  liquidity policy caps fees at 2,000 sats and the statement's signed
  `maxLiquidityFeeSat` never reached it. The node's policy is now the
  statement's.
- A channel could ask for round one before the pool held it. The node now
  waits for a minimum pool before connecting, and otherwise refuses and
  retries.
- A force close waited forever for feerates, which lightning-kmp fetches
  only once. The node now refreshes them, and takes a flat rate on regtest.

Public eclair cannot route a Bolt11 payment to a wallet by its peer-id
route hint, so the run pays the group's Bolt12 offer. The offer is now part
of the wallet: every snapshot carries it, every member keeps it only if
its own copy of the seed derives it, and the receive screen shows it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-23 22:30:25 +02:00