Time Nick Message 04:00 MTDiscord Im pleased to report that minetest on android works better than it has in many years with the release candidate. The bug where minetest would close (though remain in the app drawer) as soon as it was opened is fixed. The bug where minetest being minimized to the app drawer caused a loss of server connection is fixed. Bloom and post-processing seem to work. Shadows still do not work. 15:45 kilbith_ https://github.com/minetest/minetest/pull/13296 20:58 rubenwardy should #13200 be on the 5.7 milestone? 20:58 ShadowBot https://github.com/minetest/minetest/issues/13200 -- Fix missing particlespawner values by velartrill 21:42 lhofhansl Let's think about #13250 and #13225. 21:42 ShadowBot https://github.com/minetest/minetest/issues/13250 -- Scale culler steps proportionally to the mesh sizes by x2048 21:42 ShadowBot https://github.com/minetest/minetest/issues/13225 -- Fix infinite viewing_range by lhofhansl 21:43 lhofhansl I'm happy to finish 13225, I'd like 13250 to go in first. 21:47 rubenwardy why do you need to fix infinite view range twice? 21:48 rubenwardy for good measure? 21:54 sfan5 the new culler perform bad over long distances 21:54 sfan5 the kind of distances range_all is used at 22:00 lhofhansl We can remove the infinite part of 13250... But it is needed for performance at larger (non-infinite) viewing ranges. 22:01 lhofhansl 13225 pulls parts of the old culler back. We have since learned how to make it work with the client mesh chunking. 22:03 lhofhansl The new culler performs bad when only a small fraction of the potentially visible is loaded at the client, because it has traverse the unloaded blocks as well, while the old culler starts from the loaded blocks (but does that much slower) 22:03 lhofhansl badly 22:09 rubenwardy how bad does it perform without #13250? How much risk is that PR? 22:09 ShadowBot https://github.com/minetest/minetest/issues/13250 -- Scale culler steps proportionally to the mesh sizes by x2048 22:11 rubenwardy I guess pretty badly 22:11 lhofhansl For viewing_range of 1000 before that PR the culler can go up 70ms or more. I.e. every 0.2s you see a 70ms blip. The new culler comfortably stays below 5ms. 22:11 lhofhansl That part I'd want. :) 22:11 lhofhansl I've seen it go to 100ms even. 22:12 lhofhansl (Relatively fast Laptop) 22:12 rubenwardy The reason I questioned it is that performance PRs aren't usually allowed in feature freezes, but this seems close enough to a bug fix 22:13 lhofhansl I could argue this from both sides :) The old culler was even slower at that viewing_range, so it's already better. 22:14 lhofhansl (I mean the old loops culler that we removed, and I would partially bring back for the infinite vewing_range) 22:16 lhofhansl Or I could say that the culler fix is really needed for the client mesh chunking, and should be considered part of that. 22:28 lhofhansl I'll say one more thing: The current culler is bad enough at larg'ish viewing_range that I went through the trouble of #13207. With this PR that is completely unneeded. 22:28 ShadowBot https://github.com/minetest/minetest/issues/13207 -- Spread updating the drawlist over multiple frames. by lhofhansl