17:41 <rixed> Is it intended that the tcpip stack doesn't implement a loopback and instead emmit an ARP requests when trying to connect to itself?

17:50 <yomimono> rixed: see https://github.com/mirage/mirage-tcpip/issues/133

17:51 <yomimono> we had some discussion on this that landed on "it's weird to do this in the context of a unikernel, and would take some work to support, so we won't do it for now"

17:54 <yomimono> if you have a use case that needs some different behavior, we could think about it some more though

17:59 <rixed> yomimono: It's painfull to implement this at the application layer because it's hard to inject the bouncing frame into the receiving callback that's already blocked in a read. :-/ I guess that's a similar issue that you faced from within the kernel

18:00 <rixed> yomimono: I don't really buy the unikernel exemption argument though. You may have independant services running in the same unikernel and sometime one might want to communicate with another exactly as in Unix.

18:00 <rixed> yomimono: indeed that's my situation. So now I'm going to implement a net device and a routing stage above the kernel, which looks a bit weird

18:01 <rixed> Hopefuly that's temporary only, will reopen that bug and see :)

18:02 <rixed> BTW, my use case is a monitoring service that monitors itself, and a registration service that register itself... So I've hit this limit twice already :)

18:04 <yomimono> rixed: any interest in bringing this up on mirageos-devel?

18:05 <yomimono> oh nm, I see you commented on the issue

18:05 <rixed> yomimono: That's what I did a few days ago but the ML does not work for me (can receive the messages but cannot post aparently)

18:05 <rixed> yomimono: (that, or I'm greylisted for a looong time)

18:06 <yomimono> rixed: ouch :( I'm asking around to see who the ML admin is these days; I have a few guesses but I know it's not me

18:07 <yomimono> I know it's moderated but not sure who's doing the scutwork

18:08 <rixed> yomimono: Then my message will appear in the ML eventualy. Meanwhile I'm going to hack a workaround. It's the first behavioral difference I can spot with the unix version of my pet project