setchannelfee: fix crash when channel is not in valid state.

[rustyrussell]

You can't fail a cmd when you've already started streaming
a successful response:

lightningd: ccan/ccan/json_out/json_out.c:343: json_out_finished: Assertion `tal_count(jout->wrapping) == 0' failed.
lightningd: FATAL SIGNAL 6 (version v0.9.2-119-gf7cdf1d)
0x1847d1 send_backtrace
common/daemon.c:38
0x184877 crashdump
common/daemon.c:51
0x5bda03f ???
???:0
0x5bd9fb7 ???
???:0
0x5bdb920 ???
???:0
0x5bcb489 ???
???:0
0x5bcb501 ???
???:0
0x1e07a8 json_out_finished
ccan/ccan/json_out/json_out.c:343
0x18db0a json_stream_double_cr
common/json_stream.c:95
0x18dbf3 json_stream_close
common/json_stream.c:117
0x12fd98 command_raw_complete
lightningd/jsonrpc.c:459
0x12fec9 command_failed
lightningd/jsonrpc.c:488
0x12ffb9 command_fail
lightningd/jsonrpc.c:503
0x14dc20 json_setchannelfee
lightningd/peer_control.c:2052

Signed-off-by: Rusty Russell [email protected]
Changelog-Fixed: JSONRPC: setchannelfee would fail an assertion if channel wasn't in normal state.

[m-schmoock]

ACK 93e2b19

That is correct.
I wrote a testcase that checks for this. Feel free to add it to tests/test_pay.py.
It also checks that channel lookup by channel_id is possible, which was uncovered before.

def test_setchannelfee_closed(node_factory, bitcoind):
    l1, l2 = node_factory.line_graph(2)
    # We use the `channel_id` (not scid) by intention
    # to also test channel lookup using `channel_id`.
    cid = l1.get_channel_id(l2)
    l2.stop()
    l1.rpc.close(cid, 1)
    with pytest.raises(RpcError, match="Channel is in state AWAITING_UNILATERAL"):
        l1.rpc.setchannelfee(cid, 12, 34)

[darosior]

ACK 93e2b19