Luanti logo

IRC log for #minetest-dev, 2018-01-12

| Channels | #minetest-dev index | Today | | Google Search | Plaintext

All times shown according to UTC.

Enable nick filtering
Time Nick Message
00:22 EvergreenTree joined #minetest-dev
01:25 paramat will merge trivial #6906 later with #6838
01:25 ShadowBot https://github.com/minetest/minetest/issues/6906 -- Various settings and docs improvements by paramat
01:25 ShadowBot https://github.com/minetest/minetest/issues/6838 -- Fix ambient occlusion and dark lines at mapblock borders by numberZero
5 more elements. Show/hide.
04:44 paramat !tell lhofhansl see https://github.com/minetest/minetest/issues/6907
04:44 ShadowBot paramat: O.K.
3 more elements. Show/hide.
05:41 paramat merging #6838 #6906
05:41 ShadowBot https://github.com/minetest/minetest/issues/6838 -- Fix ambient occlusion and dark lines at mapblock borders by numberZero
05:41 ShadowBot https://github.com/minetest/minetest/issues/6906 -- Various settings and docs improvements by paramat
06:01 paramat merged
07:35 nrzkt joined #minetest-dev
07:35 nrzkt hello, pushing a tiny regression fix for #6897
07:35 ShadowBot https://github.com/minetest/minetest/issues/6897 -- Build fails without freetype
07:56 nrzkt i just pushed #6908, there is an interesting Calinou question here
07:56 ShadowBot https://github.com/minetest/minetest/issues/6908 -- Add a build step to test non freetype builds by nerzhul
5 more elements. Show/hide.
12:55 nerzhul hey sfan5 are you around ?
12:56 sfan5 yes
12:57 nerzhul can you take a little amount of time to review #6908
12:57 ShadowBot https://github.com/minetest/minetest/issues/6908 -- Add a build step to test non freetype builds by nerzhul
12:57 nerzhul pleaes
12:57 sfan5 sure
12:57 nerzhul please*
12:58 sfan5 nerzhul: wasn't there some problem with clang-5.0 builds or is that fixed already?
12:59 nerzhul sfan5, what do you mean ? clang build works, is there any runtime issue ? i hope it's not the case clang 5 is used everywhere in the world
12:59 sfan5 no not at runtime
13:00 nerzhul generally i do my own tests with clang on my machine and i never see any issue with that
13:00 sfan5 https://travis-ci.org/minetest/minetest/jobs/326540973#L490
13:01 nerzhul sfan5, oh it's a provider issue
13:02 nerzhul travis is not very stable these days
13:02 nerzhul i think it's due to the meltdown problem and some AWS issues as travis runs on heroku, then it runs on AWS :)
13:04 sfan5 possibly
13:05 nerzhul sfan5, thanks i fixed the -n will merge after travis build pass
13:06 sfan5 you forgot the quotes
13:06 sfan5 they're not required I know...
13:06 nerzhul you are right :)
13:06 nerzhul i missed them and i agree it's better
4 more elements. Show/hide.
14:36 nerzhul merging #6908 (macosx build doesn't want to start) thanks sfan5 for the review
14:36 ShadowBot https://github.com/minetest/minetest/issues/6908 -- Add a build step to test non freetype builds by nerzhul
11 more elements. Show/hide.
18:15 lhofhansl I'm going to revert 9c66901. See #6907
18:15 ShadowBot https://github.com/minetest/minetest/issues/6907 -- Vehicle acceleration and turn are slow and jerky
18:16 lhofhansl A shame really since active objects are *very* heavy weight, and currently there's no way to reduce that load and keep responsiveness where needed.
18:18 lhofhansl They can be heavy weight (such as in NSSM)
18:29 Krock so there's no workaround?
2 more elements. Show/hide.
18:39 lhofhansl Krock: None that worked satisfactory. Without that change AsyncRunStep is essentially run at full speed - only slowed by retrieving packets from clients. In mob heavy setups (like nssm) that quickly renders the server useless... Put out a few a ice-salamanders and the world stops rendering since all the time is spent in the active objects. That's what this addressed. But now of course there *is* a time budget, and it increase lat
18:39 lhofhansl cy.
18:39 lhofhansl I tried handling both players and attached objects separately but that lead to other forms of jitter when a player is attached.
18:40 rubenwardy can you prioritise by player proximity? Sorry, I haven't actually looked at the patches yet
18:41 rubenwardy I guess that replies on the entity storage system being less shit and O(N^2)
18:41 lhofhansl rubenwardy: won't help. Even when I prioritize all players and attached objects (and have a single player only) I see the jitter.
18:42 rubenwardy is that rendering or server step jitter?
18:42 lhofhansl Perhaps a less drastic way is to lower the active object internal to 0.01s (< 1/60s). That defeats some of the advantages but it will slow AOs down when they become too CPU intensive (and only then). Actually... Lemme try that.
18:42 lhofhansl server step jitter
18:46 pgimeno_ hmmm... does world generation only happen during idle time?
18:46 lhofhansl nssm has a lot of Lua map gen hooks, those get slowed down to the point of being useless.
18:47 lhofhansl perhaps I'm optimizing too much for nssm :) It just happens to be my test mod for mob heavy scenarios.
18:49 rubenwardy does NSSM use the heightmap to find good locations, or does it scan for particular nodes?
18:49 rubenwardy if the later, LVM or get_node?
5 more elements. Show/hide.
19:31 lhofhansl rubenwardy: I am not entirely sure actually.
19:32 lhofhansl As expected reducing active_object_interval to 0.01 negates the aim of that change. World rendering just stops with a bunch of mobs around. But responses for attached objects are fast again. Perhaps that would be the simplest change right now.
19:43 lhofhansl OK... Best solution: active_object_interval to 0.01 and reducing moving average smoothing in ServerEnvironment... I'll create a PR.
4 more elements. Show/hide.
22:33 lhofhansl reverting is best for now.
22:34 lhofhansl Do folks want PR for that or should I just go ahead and revert?
22:34 lhofhansl re #6907
22:34 ShadowBot https://github.com/minetest/minetest/issues/6907 -- Vehicle acceleration and turn are slow and jerky
3 more elements. Show/hide.
23:11 lhofhansl rubenwardy: Looks like NSSM is using mobs_redo's spawn_specific, and all calls pass in neighbors, is that what you meant?
23:13 lhofhansl But these register ABMs... So I guess I do not quite understand why exactly putting a time budget on active_objects helped anything.
23:18 EvergreenTree joined #minetest-dev

| Channels | #minetest-dev index | Today | | Google Search | Plaintext