3974806e5a added this:
CI: Try not running group 2/10 UBSAN in parallel.
It's being killed with signal 143, which means docker isn't happy; too much memory consumption?
But since we're now at 12 groups, that probably doesn't apply (it
might not have even before, in the two years since that commit since
so may things have been added). And it caused this shard to take over
2 hours and timed out.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Under Postgres, this actually takes more than 2 seconds, so w2
really has timed out already:
```
time.sleep(2) # total 2
assert not w1.done()
> assert not w2.done()
E assert not True
E + where True = done()
E + where done = <Future at 0x7fe14e54fee0 state=finished raised RpcError>.done
tests/test_invoices.py:420: AssertionError
```
So space the timeouts out more, and sleep one second too short; the
.result() (which sleeps) will catch up if we were extremely slow.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This means that we won't complain to peers which gossip about our
channels, but it does mean that our channel graph (like other nodes on
the network) will show two channels, not one, for the duration.
For this reason, we need askrene to omit local dying channels.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We restart the nodeL if the coin_movements.py plugin hasn't processed the
notification yet, it will be incorrect:
```
> assert account_balance(l2, chanid_1) == 100001001
E AssertionError: assert 150_001_001msat == 100_001_001
E + where 150001001msat = account_balance(<fixtures.LightningNode object at 0x7f0634e1eb00>, '39ac52c818c5304cf0664940ff236c4e3f8f4ceb8993cb1491347142d61b62bc')
```
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
1. It was flaky, probably because it didn't wait for the remote update_channel.
2. Rusty applied a fix in 5f664dac77, not clear if it worked.
3. Christian disabled it altogether in 23ce9a947d.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Since this was written, we now test if remote side would get into this situation and stop
it from happening, so the test doesn't work any more.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We can still get a warning:
lightningd-1 2025-12-10T01:11:07.232Z DEBUG 022d223620a359a47ff7f7ac447c85c46c923da53389221a0054c11c1e3ca31d59-connectd: Received WIRE_WARNING: WARNING: channel_announcement: no unspent txout 109x1x1
This has nothing to do with l1 talking about the original channel
(which would be 103x1x): it's because l2's gossipd (being the node
which does the splice) immediately forgets the pre-splice id. If l1
sends some gossip, it will get a warning message.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
We used to just run these without valgrind, but we already run them in
CI (which sets SLOW_MACHINE) without valgrind, so this just doubles
up.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
```
common/test/run-param.c:381:8: runtime error: applying zero offset to null pointer
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior common/test/run-param.c:381:8
```
Probably because CI now on 24.04, so more recent clang. But the test really does
want to see what happens when the callback is NULL, so workaround.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
When paying through a direct channel, direct_pay_override() creates a
route bypassing the normal routing path, which skips the CLTV budget
check in payment_getroute(). This allows payments to succeed even when
maxdelay is set below the required min_final_cltv_expiry.
Add a check in direct_pay_override() to verify the required CLTV
doesn't exceed cltv_budget before using the direct channel shortcut.
If it exceeds, skip the direct channel and let normal routing handle
the failure with a proper error message.
Fixes: #8609
Changelog-Fixed: pay: `maxdelay` parameter now enforced for direct channel payments
gossipd no longer makes gossip messages, and hasn't since v24.02, so it
doesn't actually need to talk to the hsm daemon.
Also, various comments were out of date, so fix those too.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
- ReadMe API v2 now requires category.uri instead of category.id.
- ReadMe v2 page responses include a renderable field indicating whether the document is MDX-compatible and can be rendered successfully. The script now checks this field and prints detailed compilation errors if rendering fails.
We had the test backwards, so we moved it *all the time*. This bloats our gossip store, as well as
not moving it in the case where we need to.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Changelog-Fixed: gossipd: we would occasionally not show a node announcement in listnodes().
Remove anyhow as a dependency from the tlv module. This allows for a
cleaner error handling
Changelog-None
Signed-off-by: Peter Neuroth <pet.v.ne@gmail.com>
The core module shouldn't depend on anything from the cln_adapters
module. This ensures that we can switch out interfaces in the future
(e.g a different database than cln's datastore).
This change made it necessary to refactor the service plugin which
allowed us to clean it up on the go
Signed-off-by: Peter Neuroth <pet.v.ne@gmail.com>
Separate concerns of the lsps2 modules into their respective layers:
core for tlv, as it is lightning related, not specifically
core-lightning. cln_adapters for core-lightning related types.
Signed-off-by: Peter Neuroth <pet.v.ne@gmail.com>
This commit separates the business logic from core-lighting behaviour
and moves the core-lightning related implementations into the
cln_adapters module.
Splits up tests as well
Signed-off-by: Peter Neuroth <pet.v.ne@gmail.com>
This now only makes it easier to test, but also more modular. It allows
us to use any kind of storage, not only the datastore.
Signed-off-by: Peter Neuroth <pet.v.ne@gmail.com>
The ClnApi trait is way overloaded. We break things down to make it more
modular on the service side.
Signed-off-by: Peter Neuroth <pet.v.ne@gmail.com>
