crash in renepay plugin about lightning HOT 10 CLOSED

Actually, I think I don't need any more log messages. I seems like the flow destructor is called twice. I'm looking into it.

from lightning.

@rustyrussell just PR this #6523 to avoid crashing the node.

from lightning.

We should not make renepay marked as important, I do not think it has enough tests to do it?

Maybe this underflow is easier for @Lagrang3 to debug? assigning to it till I will be away from my desk

from lightning.

First thing I notice is that invoice string comes before the payment parameters otherwise the parser will not let you continue.

Can you upload more details of the logs, to simplify maybe just the lines containing renepay?

from lightning.

Sorry, I did get the BOLT11 in before the parameters in the actual test. The report is updated to reflect that.

That was all the log output I received - there was no traceback available. Will see if I can up the logging level and get some more details.

from lightning.

We should not make renepay marked as important, I do not think it has enough tests to do it?

Maybe this underflow is easier for @Lagrang3 to debug? assigning to it till I will be away from my desk

As a builtin plugin, the bool important flag is true by default. I am not sure if I can flip it at will. There are conflicting definitions of struct plugin in lightningd/plugin.h and plugins/libplugin.c, the first contains a bool important, the second does not.

from lightning.

If there is no possibility to do it, we should add it. Till now, all the plugins built-in are important (except grpc one), but we should have the possibility to say, "it is fine if this plugin dies"

from lightning.

@rustyrussell there is also the fix to the underflow that triggered this issue #6505

from lightning.

My fault I was assuming that this was already merged

from lightning.

I believe this should be fixed by #6538

from lightning.