Time Nick Message 08:30 nerzhul question, is the next version a breaking version or not ? 08:30 nerzhul regarding irrlicht i mean it was intended 08:30 nerzhul i was thinking about again good entities with less performance usage server side 08:31 nerzhul ie. having good mobs, less low level packets sent through AOMessages 08:32 nerzhul i mean AOM ENTITY_MOVETO or a thing like this which involve pathfinder on client side to have accurate movements, instead of manually sending all the movement things from server :) 08:32 nerzhul and it's not a replacement, it's a refinement which can help many entites to use less server, and reduce lua code which requires low level actions 08:35 sfan5 there are no breaking changes done or planned 08:36 sfan5 maybe I should work on experimantal SSCSM, because ultimately client-side programmable mob movement prediction is better 09:42 AntumDeluge Is this the right place for modding questions, or should that go in #minetest? 09:50 sfan5 latter 09:50 AntumDeluge Thanks sfan5. I will ask there. 13:02 nerzhul sfan5 you don't need SSCSM for mob predicution if the server says, move to x,y,z with speed x 13:03 nerzhul we can directly call the server & client path finder, and regularly send the move_to order fro mthe server with the current server position to fix laggy clients 13:03 nerzhul in multiple online games it's how it's done, the path find is both sides 13:06 MTDiscord sfan5 mind uh allowing us to enable entities client side entirely for things like visual effects 13:20 sfan5 sure you don't need SSCSM but it's more flexible and at the same time covers the client-side pathfinding usecase 13:20 sfan5 also is the pathfinder even deterministic? you'd run into problems doing this if not 14:10 MTDiscord i mean "entities" 14:11 MTDiscord as in client side models that work similar but more like HUD/formspec 14:57 nerzhul Jordash, entiters won't never be fully client side, server always has something to calculate to verify context :) 14:58 nerzhul sfan5, i think it is, if we use the same algorithm on both sides, or very close, the errors margin will be low 14:59 nerzhul also if there is some tiny changesi n the decision, the regular packet sending should fix the original position and setup the new path finding 14:59 rubenwardy entirely client-side entities would be useful for some visual effects. It wouldn't be network synced 14:59 nerzhul also i'm not sure we need a breaking for this, only protocol breakage, because old versions cannot understand this new way to move, and backward compat will just generate hell code on server side 15:00 nerzhul rubenwardy: you mean which example ? just to see if there is no side effect :) 15:00 rubenwardy \o/ 15:01 rubenwardy no specifics, I can see it being useful 15:03 MTDiscord simple example, client side entity in front of the eyes to prevent seeing, currently you can do this with hud, but can be bypassed by turning hud off 15:04 MTDiscord prevent seeing/limit vision 15:04 rubenwardy well, you could use it for selection boxes but that could also be a per-player entity rather than a client-side one 15:05 MTDiscord another example is if you want tracers only for the person that shot 15:38 sfan5 nerzhul: if it doesn't replace the regular position updates, what advantage would it have? 15:38 nerzhul you say to an entity: go there 15:39 nerzhul you don't say: move with acceleration, turn , move , (jump maybe too possible, who know), etc 15:39 nerzhul and path finder does the job 15:39 nerzhul it's like this in majority of the MMORPG/MOBA nowadays 15:40 nerzhul you don't let the server send you all mini orders, else on big map with huge number of players you just let your server exponentially ddie :D 15:40 nerzhul the regular current position sending is just to ensure server & client agree about where is the mob for next pathfinder calculation on both sides 17:07 Pexin re clientside entities: fireworks mod was bogging down my group's server to the point chat took 20 literal minutes to get through 19:44 lhofhansl Any comments on this design/patch/mod: https://github.com/EvidenceBKidscode/ffiopt ? Perhaps we can aim integrating that into MT proper. 19:45 lhofhansl ( I guess us LuaJit only, so likely not) 19:47 MTDiscord Why not? 19:48 MTDiscord If LuaJIT is available, such a patch seems reasonable 19:48 lhofhansl Fair enough actually. 19:50 sfan5 IIRC pgimeno showed a similar patch recently 19:51 sfan5 though actually, they're not that similar except for using FFI 19:52 pgimeno https://forum.minetest.net/viewtopic.php?f=9&t=22377&p=346105 19:52 sfan5 is replacing a table with a dummy table + metatable actually faster? I can see it saving memory but not so much performance 19:52 pgimeno I wrote mine trying to tighten the security a bit, but it's not really secure anyway 19:53 pgimeno sfan5: the trick is that the access is to FFI structures 19:54 sfan5 and that can be faster than accessing a lua table? 19:58 MTDiscord well you save the Lua table generation 20:02 pgimeno indeed 20:02 pgimeno especially when LuaJIT compiles, because it flattens function calls, so the metamethod call gets eliminated 20:07 pgimeno anyway, it's just a hack, and a faster method would be to give direct access to the structures as in #6863 20:07 ShadowBot https://github.com/minetest/minetest/issues/6863 -- Add LuaJIT FFI-friendly memory-intensive functions 20:08 pgimeno (now https://notabug.org/pgimeno/minetest/pulls/2 )