Time Nick Message 03:23 quin Hi, I'm the Lead Deveoper for Red Eclipse. I have a heads-up about some merged pull requests that may fall under your project's AI policy. Who would be the best person to send the details to privately? 03:26 MTDiscord one must wonder what Red Eclipse has to do with luanti 03:28 MTDiscord anyways, dm'd you an email address 03:30 qreeves (y) 16:35 Sheriff_U3 PlayerRef:get_eye_offset() could use some better documentation, especially since it returns three(?) vectors. 16:37 rubenwardy_ obviously it's incase you have 3 eyes 20:41 sfan5 i think we need a consensus on replacing the protocol and move that forward 21:08 sfan5 actually for many issues it would help if we had a consensus. because if there's one (+ a rough design) someone else could implement it 21:10 Krock nailing down a dependency would be a good start, I think? 21:14 Krock and the server might have to listen on two ports: one for the current mtp, and the other for library X 21:14 Krock because shoving a peer from one handler to the other might be somewhat difficult 21:15 sfan5 yeah that's clear 21:15 Krock let's have a meeting to discuss a few suggestions? 21:16 sfan5 pinning down a library is where the problem starts. which protocol do we even want? should it have encryption? if yes, how do we handle certificates? 21:17 sfan5 the QUIC PR caused a lot of discussion so I'm almost thinking it could be better to first switch to a different transport layer and add encryption later 21:18 sfan5 (wasting the opportunity to do both at once...) 21:18 Krock I assumed encryption is the main concern? In either case - doing some research on established libraries might be a good idea. some might already depend on the system-wide ca-certificates. 21:19 sfan5 IMO the main concern is that our current protocol code is garbage. it's slow and falls apart if you have even slightly unrealiable network (there are open issues for this) 21:20 Krock I'm afraid adding (proper) encryption as a later thought could prove difficult. Hence, using a library that supports both would be favourable 21:22 sfan5 if we only replace the protocol layer first then the encryption later would have to be another protocol layer (e.g. DTLS) 21:24 sfan5 doing both at once means we can directly switch to a protocol that does both (such as QUIC) - but be more complex 21:27 sfan5 I have the impression that big, complex tasks that can't be broken down into parts are doomed to be done "effectively never" in Luanti 21:31 Krock "system-wide ca-certificates" <-- sorry. mislead thoughts. The servers would have to request a certificate like websites to be secure. 21:32 Krock issuing custom asymmetric pairs would be possible too, depending on how high the security standards should be 21:33 sfan5 we could also say that the primary goal is transport encryption and ignore certificated verification for now 21:33 Krock right. It'll certainly result in a >1000 LoC PR, which is already difficult to merge in most cases 21:34 sfan5 (or more precisely: protection against a passive attacker) 21:35 Krock yes. That's the route I'd go with encryption. 21:37 Krock As for the library or features, the common route of "what do other games do?" could also help 21:38 sfan5 there's enet which was kinda ruled out I think 21:40 Krock (such as 0 A.D., OpenTTD, Freeciv, Warzone 2100, to name a few) 21:47 MTDiscord does anything other than quic make sense if eventual core web support is wanted? otherwise anything else would require relay and central ip problems again. unless you add nodebalancer/proxy forwarded ip 23:39 rubenwardy_ We should change both at once and we should do it using that big thing that is happening soon 23:41 rubenwardy_ https://github.com/orgs/luanti-org/discussions/160#discussioncomment-16296652