Core Lightning

chaintopology: don't use txfilter, rely on wallet_extract_owned_outputs.

txfilter.c keeps a hash with interface:
* INSERT: txfilter_add_scriptpubkey &txfilter_add_derkey
* SEARCH: txfilter_match & txfilter_scriptpubkey_matches. It simply returns true/false.

wallet.c keeps an our_addresses hash:
* INSERT: on-demand when asked based on max keyidx (plus gap)
* SEARCH: wallet_can_spend (and thus its caller wallet_extract_owned_outputs).

Here are the places which populate the txfilter:

lightningd/channel.c:
- When we generate our own address final address
=> this is already covered when we call wallet_get_newindex() to get the final_key_idx
lightningd/lightningd.c:
- init_txfilter initializes all the prior addresses
=> wallet_can_spend populates this on first call.
wallet/reservation.c:
- finish_psbt and json_addpsbtoutput for change output
=> this is already covered when get call wallet_get_newindex() in the same funciton.
wallet/wallet.c:
- got_utxo for unconfirmed outputs
=> This is called when we have already determined we can spend the output (wallet_extract_owned_outputs), or on the fixup migration where the addresses are populated from a keyindex we've already created.
wallet/walletrpc.c:
- newaddr_inner
=> Already covered by wallet_get_newindex()

The result: we can just rely on the wallet to find our addresses, and
we don't need the txfilter at all. Just make chaintopology ask the
wallet directly.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

gossip: Don’t reset channel on stale annoncement

Splicing can create stale announcements where the original funding output gets spent creating spruious “channel_announcement: no unspent txout” warnings.

This happens particuarlly often with agressive splice tests in CI, so we downgrade this from a warning to a log message.

splice: Fix weight calculations & use opening feerate

Here we update `psbt_input_get_weight` to allow the caller to specify what kind of input size estimation they would like. Before it was just “zero witness bytes” which we now move to the default behavior and add an assumption option for P2WSH -> 2of2 multisig which is what we typically see in splicing.

At the same time we move channeld over to using the opening feerate estimation instead of the less useful “max feerate.” We make splice script use this same feerate.

After auditing the feerates for these two places against each other and the actually created transaction, we implement some fixes making them all match.

While we’re here, we add relevant debug log messages.

lightningd: unreserve UTXOs from withheld funding PSBT when dropping channel

When a channel using a withheld funding PSBT is dropped, we need to
unreserve the UTXOs from the PSBT. Otherwise those funds will remain
blocked until the reservation expires, delaying their availability for
other transactions.

Changelog-Fixed: lightningd: unreserve UTXOs from withheld funding PSBT
on channel close.

reestablish: Send announcement sigs when asked

A new TLV in `channel_reestablish` lets our peer request announcement signatures. This updates CLN to (re)send them when requested.

As per splice spec PR https://github.com/lightning/bolts/pull/1160/changes#diff-ed04ca2c673fd6aabde69389511fa9ee60cb44d6b2ef6c88b549ffaa753d6afeR3303-R3308

Changelog-None

schemas: fix ping requirements documentation

Changelog-None

CI: use nproc + 1 for pytest parallelism..

More future proof, better for self-hosted runners.

Example run times before:

4s CI completion
2m 36s Check we can downgrade the node (compile-gcc, sqlite3, liquid-regtest)
2m 52s Check we can downgrade the node (compile-gcc, postgres, regtest)
3m 20s Check we can downgrade the node (compile-gcc, sqlite3, regtest, 1)
3m 20s Check we can downgrade the node (compile-gcc, sqlite3, regtest, 1)
8m 1s Run fuzz regression tests
11m 46s Run unit tests (compile-clang-sanitizers, 0)
12m 59s Compile CLN compile-gcc
13m 42s Compile CLN compile-clang
14m 10s Compile CLN compile-gcc-O3
14m 15s Compile CLN compile-clang-sanitizers
15m 23s Pre-build checks
19m 37s Test CLN dual-fund
26m 49s Run unit tests (compile-gcc, 1)
43m 19s Valgrind Test CLN Valgrind (06/10)
43m 21s Valgrind Test CLN Valgrind (08/10)
43m 40s Valgrind Test CLN Valgrind (05/10)
43m 59s Valgrind Test CLN Valgrind (03/10)
44m 4s Valgrind Test CLN Valgrind (07/10)
47m 28s Valgrind Test CLN Valgrind (09/10)
47m 37s Valgrind Test CLN Valgrind (02/10)
47m 45s Valgrind Test CLN Valgrind (01/10)
52m 51s Valgrind Test CLN Valgrind (04/10)
54m 6s Valgrind Test CLN Valgrind (10/10)
1h 14m 13s Sanitizers Test CLN (ASan/UBSan (03/12), 3, --test-group=3 --test-group-count=12)
1h 14m 6s Sanitizers Test CLN (ASan/UBSan (06/12), 6, --test-group=6 --test-group-count=12)
1h 15m 23s Sanitizers Test CLN (ASan/UBSan (01/12), 1, --test-group=1 --test-group-count=12)
1h 20m 30s Sanitizers Test CLN (ASan/UBSan (08/12), 8, --test-group=8 --test-group-count=12)
1h 22m 35s Sanitizers Test CLN (ASan/UBSan (12/12), 12, --test-group=12 --test-group-count=12)
1h 23m 49s Sanitizers Test CLN (ASan/UBSan (04/12), 4, --test-group=4 --test-group-count=12)
1h 26m 11s Sanitizers Test CLN (ASan/UBSan (07/12), 7, --test-group=7 --test-group-count=12)
1h 28m 44s Sanitizers Test CLN (ASan/UBSan (05/12), 5, --test-group=5 --test-group-count=12)
1h 29m 1s Sanitizers Test CLN (ASan/UBSan (02/12), 2, --test-group=2 --test-group-count=12)
1h 29m 4s Sanitizers Test CLN (ASan/UBSan (11/12), 11, --test-group=11 --test-group-count=12)
1h 31m 36s Sanitizers Test CLN (ASan/UBSan (09/12), 9, --test-group=9 --test-group-count=12)
1h 38m 54s Sanitizers Test CLN (ASan/UBSan (10/12), 10, --test-group=10 --test-group-count=12)
1h 39m 22s Test CLN liquid
1h 52m 23s Test minimum supported BTC v25.0 with clang
1h 52m 28s Test CLN gcc
1h 53m 57s Test CLN splicing
1h 57m 25s Test CLN postgres
1h 59m 25s Test CLN clang

2d 17h 56m 16s TOTAL

Example runtimes afterwards:



Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

CI: make downgrade and slow integration tests depend on check-compiled-source step.

As pointed out by Alex Myers: we don't want to waste time running the
complex steps if the generated files are not up-to-date.

But we still run the faster "integration" tests (not valgrind or
sanitizer ones), since they often reveal early failures.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Build: don't use HAVE_FUNCTION_SECTIONS for update-mocks.

I ran `uv run make update-mocks` and many mocks vanished. But that
broke MacOS (and maybe other archs) which doesn't support function
sections and no longer linked.

If this proves problematic, we might end up generating mocks properly:
1. Use #include "mocks-sphinx-xor_cipher_stream_gen.c"
2. Generate those separate files as necessary.
3. Don't commit them, build them locally.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

pytest: make global exclusion for "That's weird: Request X took Y msec" under valgrind.

Rather than playing whack-a-mole:

```
ERROR tests/test_misc.py::test_emergencyrecover - ValueError:
Node errors:
- lightningd-1: had BROKEN or That's weird messages
...
lightningd-1 2026-02-18T02:29:54.826Z UNUSUAL jsonrpc#76: That's weird: Request signpsbt took 7466 milliseconds
```

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

gossmap: add callback for gossipd to see dying messages.

gossmap doesn't care, so gossipd currently has to iterate through the
store to find them at startup. Create a callback for gossipd to use
instead.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

tools: delete gossip_store of needed for downgrade even if db hasn't changed.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

pytest: changed old_hsmsecret to default to false.

This changes various tests in minor ways:

1. The "l2" secret key in tests/plugins/channeld_fakenet.c is updated.
2. The decompressed gossip data node id needs changing.
3. The coinmoves order changes in bookkeeper for anchors.
4. Various harcoded gossip constants change.
5. Some hardcoded makesecret results change.
6. zeroconf tests which hardcoded node ids change.
7. Arbitrary rune strings change.
8. A log message which uses node ids changes.

pytest: make sure to use old_hsmsecret on downgrade tests.

We only support downgrade if you upgraded, so your hsm_secret will be old.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

wallet: conduct unit tests with both bip86 and bip32.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Makefile: update next and prev versions now we've released.

This means:
1. downgrade changes (we no longer fail due to node biases).
2. various deprecations no longer are

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

lightningd: use bip86 derivation for anchor spend change outputs

lightningd: use BIP86 derivation for P2TR in HTLC rebroadcast change outputs

hsmd: fix derivation for non-taproot addresses in modern mnemonic (25.12+) nodes.

This is a fix from https://github.com/rauaap who correctly diagnosed the problem:

```
Error broadcasting transaction: error code: -26\nerror message\nmempool-script-verify-flag-failed (Script failed an OP_EQUALVERIFY operation), input 0 of ...
```

The decision to use the changed derivation for all addresses, not just
taproot, came up during review. Unfortunately, the signing code
(here) was not changed to match the address generation code (in the
wallet).

Reported-by: https://github.com/postanissue
Fixes: https://github.com/ElementsProject/lightning/issues/8804
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Changelog-Fixed: lightningd: we now correctly sign for non-taproot addresses given by nodes created by v25.12 or newer.

wallet/hsmd: fix signmessagewithkey for BIP86 wallets

The signmessagewithkey RPC was failing for BIP86 (mnemonic-based)
wallets because:

1. The wallet RPC was iterating through BIP32-derived addresses only,
so it couldn't find BIP86-derived addresses.

2. The HSM's handle_bip137_sign_message always used bitcoin_key()
(BIP32 derivation) regardless of wallet type.

jsonrpc: fix recover command for BIP86 wallets

The recover command checks if a node has already issued bitcoin
addresses before allowing recovery. This check only looked at
bip32_max_index, but with BIP86 wallets, newaddr() increments
bip86_max_index instead.

Also, the recover test asserted on hex but now it's asserting on codex32 instead. We should probably go in and fix the end point. @rustyrussell what do you think?

lightningd: Fix penalty tx output derivation for BIP86 wallets

When using the new BIP39 mnemonic HSM secret format, the wallet uses
BIP86 derivation for taproot addresses. However, onchaind_tx_unsigned() was always using bip32_pubkey() to derive the final key for penalty transaction outputs.

pytest: don't record IO when we kill nodes

```
ERROR tests/test_connection.py::test_channel_persistence - json.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0)
```

Similarly:

```
ERROR tests/test_opening.py::test_sendpsbt_crash - json.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0)
```

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

pytest: fix feerate check in test_peer_anchor_push

We didn't log when anchor transactions had short signatures,
which causes this test to not assert (did_short_sig):

```
total_feerate_perkw = total_fees / total_weight * 1000
> check_feerate([l3, l2], total_feerate_perkw, feerate)

tests/test_closing.py:4063:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

nodes = [<fixtures.LightningNode object at 0x7f0fb322bb20>, <fixtures.LightningNode object at 0x7f0fb1b5ead0>]
actual_feerate = 14006.105538595726, expected_feerate = 14000

def check_feerate(nodes, actual_feerate, expected_feerate):
# Feerate can't be lower.
assert actual_feerate > expected_feerate - 2
if actual_feerate >= expected_feerate + 2:
if any([did_short_sig(n) for n in nodes]):
return
# Use assert as it shows the actual values on failure
> assert actual_feerate < expected_feerate + 2
E AssertionError
```

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

pytest: don't get upset at slow multi-input signing under valgrind.

```
2026-01-06T07:46:35.5710043Z lightningd-1 2026-01-06T07:45:11.040Z UNUSUAL jsonrpc#68: That's weird: Request signpsbt took 5099 milliseconds
```

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>