Electrum

merge part of "lnpeer: harden _check_unfulfilled_htlc"

extracted from https://github.com/spesmilo/electrum/pull/10899

Merge pull request #10909 from accumulator/lnpeer_fix_connection_wedges

lnpeer: fix wedged peer connections

lnpeer: fix wedged peer connections.

On android, when the app is suspended in the background long enough that LN peer sockets die,
electrum failed to recover these wedged connections. On desktop, the same issue occurs when suspending
the system.

- add timeout on ping messages. avoid wait forever on a dead socket
- add connection monitor task on Peer taskgroup, similar to Interface.

Merge pull request #10911 from SomberNight/202608_transaction_var_int

transaction: de-dupe write_compact_size and var_int

Merge pull request #10904 from SomberNight/202608_tests_lnpeer_htlc_tracker

tests: lnpeer: factor out "SenderHtlcResolvedTracker"

Merge pull request #10900 from Alwoch/unskip_chan_reserve_tests

tests: Unskip, repair and make discoverable tests under TestChanReserve

merge most of "lnchannel: extract_preimage: fix len(witness) check"

extracted from https://github.com/spesmilo/electrum/pull/10906

transaction: fix return value of already_has_some_signatures()

harmless in current usage.

type hints: (trivial) small fixes all over

channel: ensure htlcs get resolved after force close

Minor refactor to ensure:

* htlcs_fulfilled gets called even if we already know the preimage:
If we e.g. sent a mpp on multiple channels and got the preimage
somewhere else, `extract_preimage_from_htlc_txin` wouldn't call
htlc_fulfilled anymore, but the htlc needs to be resolved once.

* htlc_fulfilled doesn't trigger `payment_succeeded` callbacks
if the payment is already `PR_PAID`, similar dedup to
`htlc_failed`, once is enough.

* htlc_failed doesn't set a payment `PR_UNPAID` if we already know
the preimage. Maybe `extract_preimage_from_htlc_txin` tries to
fail the htlc after a revocation tx got confirmed, but if we
already have the preimage the payment is not neccessarily failed.

lnwatcher: catch exc during ctx output sweeping

Catch exceptions occuring during the sweep of a single
ctx output and keep going sweeping the other outputs so
we sweep at least those that do not crash and don't leave
them on the table.

test: regtest: make breach test wait for REDEEMED

Make the breach regtest wait for the channel to get set into
REDEEMED state to catch regressions preventing the channel state
from advancing into REDEEMED.

tests: add unittests for lnchannel onchain preimage extraction

Cover Channel.extract_preimage_from_htlc_txin() txin htlc preimage
extraction in unittests.

type hints: (trivial) add missing "Optional" qualifiers

In many cases with our type hinting, we do not explicitly specify that variables are allowed to be None when they have a default value of None.
This PR now adds the missing explicit "Optional" markers.

-----

Consider:
```
class Foo(NamedTuple):
field: str = None
```
This type-hinting is technically incorrect, we should type hint it as either one of these options:
- `field: Optional[str] = None`
- `field: str | None = None`

PyCharm has had soft-warnings about these lines for quite a while,
however when doing type-inference, it used to internally "fix" the type by allowing None.
Recently however it seems it is not doing the internal inference-fixing anymore, but assume that `path` has type `str` (as we specified) and then make "incorrect" inferences:

```
class Foo(NamedTuple):
field: str = None

def func(foo: Foo):
assert foo.field is None
print("function body") # <<< PyCharm now says this code is unreachable
```
I have whole function bodies grayed out and marked as dead code due to this.

Example: https://github.com/spesmilo/electrum/blob/c82a0d12ef1e03a96d6180839f215c05efe9efe3/electrum/lnpeer.py#L2320, where htlc_set.parent_set_key is type-hinted as str, so the assert "must" always fail

lnutil: rm unused "name" param from make_htlc_tx_with_open_channel()

Merge pull request #10901 from SomberNight/202608_trampoline_drawings

trampoline: add links to t-bast's drawings

lnpeer._check_unfulfilled_htlc: check that config allows forwarding

if we receive an htlc that need to be forwarded,
check that config allows forwarding.

lnworker: add sanity check to trampoline forwarding

These new asserts should not be triggered unless there are bugs in the code.
Which there are.
(still the asserts are not intended as primary checks, just defense-in-depth)

merge prep commits from "fix anchor channel backup"

extracted from https://github.com/spesmilo/electrum/pull/10852

Merge pull request #10897 from SomberNight/202608_refactor_lnchan_can_update_ctx

lnpeer: split can_update_ctx checks

lnchan: (rename-only) rename can_update_ctx

transformations:
- can_update_ctx(proposer=LOCAL) -> can_send_ctx_updates()
- can_update_ctx(proposer=REMOTE) -> can_progress_ctx()

Merge pull request #10894 from f321x/fix_qml_use_after_free

qt/qml: handle RuntimeError when accessing deleted QObject

Merge pull request #10898 from SomberNight/202608_comment_swapserver_learns_ips

swaps: add comment that swapserver colluding with relay can learn IP

swaps: add comment that swapserver colluding with relay can learn IP

lnpeer: on illegal ctx_update, disconnect instead of ignoring msg

I was the one who wrote the existing code that simply drops the illegal message,
but I cannot remember the reasoning. Thinking about it now whether we should
drop the message or disconnect should depend on the message.
- "channel_ready" and "channel_reestablish" we can simply drop
- but "update_", "commitment_signed", "revoke_and_ack" we cannot just ignore...
- it seems saner to disconnect, otherwise the channel could get into a weird state, some kind of consensus failure between the peers