Minetest logo

IRC log for #minetest-dev, 2020-12-03

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

All times shown according to UTC.

Time Nick Message
00:16 T4im joined #minetest-dev
00:41 kilbith joined #minetest-dev
00:57 lisac #10240 is ready for review :S
00:58 ShadowBot https://github.com/minetest/minetest/issues/10240 -- Return to main menu if shaders fail to compile, and display an error message. by lisacvuk
01:06 Taoki joined #minetest-dev
02:06 hecks what's the biggest known minetest world size in GB?
02:08 hecks a rough calculation gives me about 100 GB for a world's whole explored surface, 40 mapblocks thick, with some margin
04:31 MTDiscord <e​xe_virus> I should not be a core dev, but I should start helping review PR's that's something I think I can do.
04:33 MTDiscord <e​xe_virus> Also, kilbith, minetest is a game that comes and goes for most people. We get excited, do a lot of work, get bored/burnt out. and revolve back. I would expect the same from core devs. Only the spurts get longer and longer as MT develops for me haha
04:55 Seirdy joined #minetest-dev
05:00 MTDiscord joined #minetest-dev
06:13 jonadab joined #minetest-dev
07:25 m42uko joined #minetest-dev
08:00 ShadowNinja joined #minetest-dev
09:01 absurb joined #minetest-dev
09:28 calcul0n joined #minetest-dev
10:53 proller joined #minetest-dev
11:20 Fixer joined #minetest-dev
11:49 calcul0n_ joined #minetest-dev
12:38 pmp-p joined #minetest-dev
12:46 MTDiscord <A​pelta> Fair idea, more people reviewing PRs will help the core devs decide whether a change (especially larger ones) are stable and wanted by the community
12:47 MTDiscord <A​pelta> I encourage more people do it if they have time to kill lol, just write a useful review that outlines any problems you may have encountered as a direct result of the change or modification if you notice any
12:57 rubenwardy I did reviews long before I became a core dev
12:57 rubenwardy https://forum.minetest.net/viewtopic.php?t=19877
13:03 proller joined #minetest-dev
13:48 Wuzzy joined #minetest-dev
15:07 Taoki[mobile] joined #minetest-dev
15:15 lisac joined #minetest-dev
15:20 Wuzzy @sfan: Why is there an action/change needed  for #10690?
15:20 ShadowBot https://github.com/minetest/minetest/issues/10690 -- Fix misleading chat messages of /clearobjects by Wuzzy2
15:20 Wuzzy @sfan5 ↑
15:20 sfan5 I thought that was obvious from the discussion in #minetest
15:21 Wuzzy <celeron55> well if you want to clean up a map for sharing
15:21 sfan5 for example it clears all objects, but your change removes that from the description
15:21 Wuzzy oh right...
15:21 Wuzzy yeah, it technically does remove all objects, even if just time-deleay
15:50 Wuzzy ok i force-pushed to #10690 now
15:50 ShadowBot https://github.com/minetest/minetest/issues/10690 -- Fix misleading chat messages of /clearobjects by Wuzzy2
17:02 lhofhansl joined #minetest-dev
17:37 m42uko joined #minetest-dev
17:56 mizux joined #minetest-dev
18:22 proller joined #minetest-dev
18:31 fluxflux joined #minetest-dev
19:02 homthack joined #minetest-dev
19:51 fluxflux joined #minetest-dev
20:02 fluxflux joined #minetest-dev
20:05 rubenwardy It would be interesting to have Minetest report crashes to mod authors on ContentDB. The error dialog could include a report button when the mod name is detected and was installed from ContentDB, which sends the error message and a list of enabled mods. Although, attributing fault can be painful
20:05 rubenwardy in any case, supporting dependency installation is more important
20:06 Krock but only if the mod is up to date
20:07 rubenwardy it can send the contentdb release number
20:07 rubenwardy and ContentDB would ignore if it's not the latest
20:07 specing rubenwardy: you could host a redmine/bugzilla/etc instance and have the client automatically use its API to submit errors
20:09 specing SpringRTS has something half-assed: https://logs.springrts.com/ that lobby client[s] use to auto-report issues
20:10 rubenwardy nice
20:10 specing (half-assed because I'm not sure if there is any nice frontend to help developers)
20:13 specing I remember look into autoreporting briefly, and I couldn't find any system for preprocessing bug reports - removing duplicates, marking how many clients had this issue (with a graph over a time period), correlating bug reports to client-submitted identifiers (OS, version, hardware vendors...)...
20:33 proller joined #minetest-dev
20:39 Krock avoiding the 1000th open issue: #10692
20:39 ShadowBot https://github.com/minetest/minetest/issues/10692 -- Various documentation fixes by SmallJoker
20:40 lhofhansl Maybe we should all do a sweep through all open issues and close them as appropriate.
20:40 lhofhansl Or close any issue older than 1 year.
20:40 lhofhansl The oldest open issue is from 2011!
20:41 MTDiscord <J​onathon> i plan to open a PR to close my issue soon (just one of two)
20:48 Krock you could mark unpopular feature requests as [Possible close]
20:48 Krock which are then closed after the next iteration
20:54 specing lhofhansl: reminder that closing issues like that discourages further contributions
20:56 lhofhansl I think an issue that has been lingering for a year does not encourage contributions either.
20:56 lhofhansl or for 9 years :)
20:58 rubenwardy Wuzzy, Calinou, fancy looking through issues?
20:59 Wuzzy hi
20:59 Wuzzy i am currently making another attempt in translating builtin. so no (for today)
20:59 Wuzzy currently testing ...
20:59 specing lhofhansl: regardless, if someone invested their time opening it, then they deserve a good response (other than "closed after 1 year of inactivity")
21:01 lhofhansl I do not feel strongly. Most open source project automatically close old issues. Clearly nobody will pick up an issue now that is multiple years old, nor will the person who opened care after that time.
21:07 Wuzzy it is very frustrating tho if a PR dies due to ignorance
21:09 Wuzzy i am also against the idea of "closing old issues". issue age is not a justification. an ancient issue might still be relevant, for example, if its a crasher. bugs dont disappear just because you ignore them. 😉
21:10 Wuzzy a better justification is for example when there's too little information, hard to reproduce, obscure, its not clear if the claimed bug exists at all
21:12 Wuzzy i really, really hate bots that auto-close old issues. i have seen stupid bots close relevant, proven, well-documented but unfixed bug reports more than was healthy ...
21:23 twoelk joined #minetest-dev
21:32 twoelk maybe work in small steps from the oldest onwards in organized events? only closing the obvious, searching for the easiest to address, just to get the number down
21:46 MTDiscord <a​ppguru> rubenwardy: PLEASE
21:46 MTDiscord <a​ppguru> getting users to properly file bugs can be quite a hassle
21:47 rubenwardy auto reports in Android is nice for finding out how well your mod performs
22:36 Wuzzy My 2nd attempt of translating builtin is now ready for review: #10693
22:36 ShadowBot https://github.com/minetest/minetest/issues/10693 -- Builtin translate (2nd attempt) by Wuzzy2
22:44 sfan5 wasn't there a fundamental problem last time?
23:04 T4im joined #minetest-dev
23:28 hecks joined #minetest-dev
23:30 hecks Any netcode gurus around? I need help debugging a "zombie" client state
23:31 sfan5 hi
23:31 hecks hi =]
23:31 MTDiscord <o​neplustwo> is it the thing where you die but can't respawn?
23:31 hecks so I got this server and a lousy connection to it... say, roughly 200 ping
23:32 sfan5 minetest doesn't cope well with packet loss if you're about to say that too
23:32 hecks after a while, certain kinds of packets start... piling up? I don't know what's happening to them
23:33 hecks let's say that the part of the game that checks a selected item (client to server) and toggles hud (server to client) still works
23:33 hecks but the part that detects input and sets animation, attaches, detaches, moves - this one goes to hell
23:33 hecks pretty much until relogged
23:34 hecks I've seen it replay things I've done minutes before, and it never catches up
23:34 sfan5 yeah that's possible
23:34 hecks so is there anything special about those kinds of packets in particular?
23:34 hecks like does it rigidly enforce a certain queue for them and doesn't ever purge it
23:35 hecks because this sort of behavior might not be necessary for most of them
23:35 sfan5 some packets (this doesn't apply to movement) are guaranteed to deliver and only delivered in-order
23:36 sfan5 can't say what those are in your case without seeing them
23:36 hecks maybe not movement, but perhaps attachment and animations
23:36 hecks do they at least get separate queues?
23:36 sfan5 "they"?
23:37 hecks one kind of guaranteed packet and another
23:37 hecks or is there just one guaranteed queue and it's getting clogged?
23:37 sfan5 there's three channels and what the games does it somewhat distributed to them
23:37 sfan5 (with a fixed mapping)
23:38 hecks and where can I check which opcode is bound to which one?
23:38 hecks I'm starting to get why this is happening, maybe more channels are needed
23:39 sfan5 you can't just throw more channels at the problem, some packets need to be ordered relative to eachother
23:39 sfan5 rather the netcode needs to cope with packet loss better
23:39 sfan5 https://github.com/minetest/minetest/blob/master/src/network/serveropcodes.cpp#L124
23:39 sfan5 https://github.com/minetest/minetest/blob/master/src/network/clientopcodes.cpp#L140
23:39 hecks well I think that loosening up animation somehow might eliminate most of the problem
23:39 hecks because in reality, only the last issued animation range should matter
23:40 hecks and previous, lost packets can be safely dropped
23:40 sfan5 true
23:41 sfan5 but that currently doesn't exist
23:42 hecks oh, hud is a separate channel, that would explain its magical resistance to gremlins
23:43 hecks and the trailing boolean is reliable, gotcha
23:43 lhofhansl Did some tests this the zlib settings. The only change with an effect is the compression level. If I change it to 3 from the default of 6 (picked 3, because that is the highest compression level that uses deflate_fast internally), compression is 3x faster and the database is 34% larger.
23:44 hecks lhofhansl: would turning off compression make map saves less painful?
23:44 lhofhansl Generating a large new scene went from taking 120s to 80s (because the emerge threads are blocked during safe.
23:44 sfan5 of course
23:44 lhofhansl Yes. But turning it off is not a good option.
23:44 lhofhansl we might want to make 3 the default, or at least make it configurable.
23:44 hecks well I have a pretty big hard drive to spare, I might do this as a workaround
23:45 sfan5 you really don't want to disable it outright
23:45 lhofhansl I can test 3 again 0 and see what it does.
23:45 sfan5 any time saved you'll pay back tenfold due to limited IO bandwidth
23:45 lhofhansl or 6 against 0, rather
23:45 sfan5 lhofhansl: don't think the default should be changed, settings makes more sense to accomodate save-heavy usecases
23:46 sfan5 a setting*
23:46 hecks what if this could be autodetected?
23:46 lhofhansl going by what indicators?
23:46 lhofhansl (autodetection is better obviously)
23:46 hecks bench some r/w on all 9 levels in the world's directory
23:47 hecks pick the minimum
23:47 hecks I'd much rather wait an extra half a second while starting a world than fiddle with the setting manually and have to tune it for every new hard drive
23:47 lhofhansl end-to-end write performance improved 3x from level 6 to 3. That is the time the server "hangs" when exiting.
23:48 sfan5 so if you have an SSD the engine will automatically waste lots of disk space because it's fast
23:48 hecks maybe the setting can be a minimum then
23:48 lhofhansl 30-40% is not the end of the world. No compression would 10x, that'd be bad.
23:48 sfan5 really the "saving blocks takes too long" has never come up before so it should neither be autodetected nor should the default be changed
23:49 lhofhansl It's starting because I am testing with large viewing_ranges... 500 and more. There's a whole new set of bottlenecks. This is one.
23:49 sfan5 does making background saves more frequent help?
23:49 lhofhansl It has always bugged me that you cannot set the viewing_range to 1000. :)
23:50 lhofhansl It does not help in aggregate (just to state the obvious). So yeah, it helps with server hangs since it spreads them out more.
23:50 lhofhansl I also now always have my MAP_BLOCKSIZE set to 32.
23:51 sfan5 maybe logic could be added to "speed up" autosaving if the server detects its never gonna catch up at the current rate
23:52 lhofhansl yeah, but that complicated to get right. We simply picked zlibs default speed/compression trade-off. I'm suggesting that for MT it's not the right trade-off.
23:52 hecks an extra 30% world size is acceptable if it means doing something about the dreadful map save
23:53 hecks it's not just horrible on exit, map saves are capable of just stalling the whole game
23:53 sfan5 how do you know 6 wasn't picked intentionally? (I don't know either)
23:53 lhofhansl map save and map generation and active objects, etc (since that's blocking during save)
23:53 lhofhansl sfan5: Fair enough. :) I was assuming, since we use -1, which let's zlib pick its default, which is 6.
23:54 sfan5 yeah it *is* the default but I don't know if anyone has looked at changing it before
23:55 lhofhansl Now I looked at it :)
23:56 hecks regarding saving blocks never being a problem before: minetest_game didn't show it as much
23:56 hecks mtg has enough supporting client features to hide the lag, but the lag was always there
23:57 hecks the more you rely on the server step being responsive, the worse the map lock feels
23:57 lhofhansl Best would be to change the DB schema anyway - as I suggested in a github issue. No need to save the blob of mapnodes when only the block's TS changed.
23:57 hecks as for frequent saves: I tried that, and while it never got bad enough to throw a fit, it still resulted in uncomfortable spikes
23:58 lhofhansl and - for that matter - why do we need to persist the timestamp anyway?
23:58 hecks saving every 10 or 5 seconds would be a viable strategy if the saving process was just a bit leaner
23:59 sfan5 timestamps are used for clearobjects and probably nodetimers too

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