Time  Nick      Message
18:25 MTDiscord <herowl> @luatic re param3 out of band: we surely need a better implementation for the drawtype "submerged" present in a PR
20:17 lhofhansl Re: My question fro yesterday... I added some simple code to track the number of meshes and the the numbers of buffers. In my case there are no object/entities, and all block meshes have a single buffer (all the meshes of all blocks touched in ClientMap::touchMapBlocks), and yet there are many more hw buffers then mesh buffers, which does not seem to make sense.
20:18 lhofhansl Is anyone aware of a reason why that might be the case - before I spent more time on it?
20:19 sfan5     transparency sorting
20:21 lhofhansl ACK
20:23 lhofhansl I already have transparency_sorting_distance set to 0, though.
20:24 sfan5     dunno then
20:26 sfan5     https://github.com/luanti-org/luanti/blob/73aba2e1f058868ad03340086a4429a00af3d43e/src/client/mapblock_mesh.cpp#L887-L905 but actually that can still create multiple buffers
20:26 sfan5     also mesh_buffer_min_vertices will create lots of buffers on the fly
20:27 lhofhansl I find that hw-buffers is suspiciously close to the 2 x the number of loaded buffers.
20:27 lhofhansl I have mesh_buffer_min_vertices set to 0 as well (it seemed to hurt performance on my machine)
20:27 sfan5     vertex and index buffers count as separate
20:28 lhofhansl Ah. That would explain it.
20:35 lhofhansl Any thoughts on #16945? Brings back a pretty good optimization that I had revert... Based on the observation that "close" events should never be considered "low priority". (That leaves the server to finish the current send-blocks loop instead of interrupting it.)
20:35 ShadowBot https://github.com/luanti-org/luanti/issues/16945 -- Bring back low priority block events in a safe way. by lhofhansl