Time  Nick        Message
00:00 erlehmann   i should have probably read the commit messages before voicing that suspicion
00:07 erlehmann   just for future reference, you do not always control who else makes a texture and conveniently leaves some raw bitmaps lying around. in fact, almost all of the mcl_maps code i have written will deal with textures where this assumption is false.
00:09 erlehmann   and even if you had the raw bitmaps lying around, it's more code and uses more storage and … i really don't see an upside unless editing operations are *rare*. in which case you can use texture modifiers.
00:46 wsor        hey v-rob, when you have a chance could you rebase https://github.com/minetest/minetest/pull/10265 ?
08:02 proller     https://github.com/minetest/minetest/pull/11843
08:03 proller     https://github.com/minetest/minetest/pull/11910
08:05 proller     2.3 months is enough to merge?
08:25 erlehmann   proller what is your upgrade plan anyway?
08:26 erlehmann   like, assume a server on 5.5, how to enlarge map?
08:28 erlehmann   proller, you should know that people are often wrong about reserved identifiers, see “what we think we reserve” page 3 http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2409.pdf
08:28 erlehmann   > In effect, these identifiers are reserved for all uses in C regardless of what header files (if any) are included
08:28 erlehmann   > Identifiers starting with uint or int and ending with _t, or UINT or INT and ending with _MAX, _MIN, or _C
08:42 proller     erlehmann, just compile with USE_POS32=1  and run, but clients without this enabled options cant connect to your server
08:42 erlehmann   proller you can fix that
08:42 erlehmann   and arguably, shoud
08:42 erlehmann   should
08:43 proller     erlehmann, new type names can be changed to any other in 5 minutes
08:43 proller     erlehmann, not in this pr
08:43 erlehmann   i was just pointing out that the type name thing is not POSIX thing, but rather C or C++ thing in general
08:43 erlehmann   and many people do not know about it
08:44 erlehmann   for example, i believe you can not make a type starting with letter E
08:44 erlehmann   EAGAIN hehe
08:45 proller     you can name types as you want if they not match already existed standard types
08:45 erlehmann   proller i do not understand why you could not keep network protocol compatibility for old clients. you could have old position in the packet for old client and not load nodes past old mapgen limit.
08:46 erlehmann   then tell people haha world is bigger
08:46 erlehmann   proller other question what happens with stuff that people have built out of bounds? it just becomes visible, right?
08:46 proller     because network protocol is streamed and uses fixed number of bits for coords, it vas 16*3 and now it 32*3
08:47 erlehmann   i think you misunderstand me. keep the old bits. add some new bits somewhere else.
08:48 erlehmann   your idea of changing field size also breaks *every* software dealing with coordinates in network packets?
08:48 erlehmann   think of multiserver, mt-go library, the wireshark dissector …
08:48 proller     new client-server should use different serialization for old clients
08:48 proller     erlehmann, yes, breaks
08:48 erlehmann   it should not!
08:49 erlehmann   i mean, you can do this in a way that it breaks in a way that is more reasonable
08:50 proller     easiest way to send new big numbers at end of packets as new fields, but it requre rework all packets and add lot of if (version)... code
08:50 proller     it subject fot future prs
08:50 erlehmann   just changing the size of a struct in a network field is – to my knowledge – a big no no in terms reliable engineering
08:50 erlehmann   yes you have to change version code anyway
08:51 erlehmann   because this is new network protocol
08:51 erlehmann   proller have you talked to himbeerserver and anon55555 about it and what they would like to see?
08:52 proller     no
08:52 erlehmann   i mean they have written the most popular server proxies (i.e. they have parsing code for minetest network protocol the PR could break)
08:52 erlehmann   it is probably a good idea to talk to them just because they might give you ideas too
08:54 erlehmann   proller what is your opinion on the work of himbeerserver anyway? i find it always interesting how people who work on similar things have different approaches.
08:54 proller     this pr is just for rename types
08:55 erlehmann   i know you are doing it in stages
08:55 erlehmann   i think that is good btw
08:55 erlehmann   taking as small steps as is feasible, not biting off too much
08:55 proller     and it already have no move 2+ months
08:55 erlehmann   hahaha have you seen some simpler changes?
08:55 erlehmann   it takes ages to get something complex approved in any project sorry :/
08:55 erlehmann   look at the stuff that wuzzy made
08:55 erlehmann   it is a shame :(
08:56 proller     so for talking backward compatibility code we should wait... years?
08:56 erlehmann   no, not. talk now.
08:58 erlehmann   proller i believe the fundamental problem is that ultimately people like appgurueu who engage with your code can not give meaningful approval. i have largely stopped reviewing stuff and looking for bugs for that reason after i realized my opinion does not really count if i am not a coredev.
08:59 erlehmann   you have to get approval from coredevs
09:00 erlehmann   how to do? i know only repeatedly asking people.
09:01 proller     or fork 8)
09:01 erlehmann   you are not the first person to say this ^^
09:02 erlehmann   the problem is keeping critical mass. you either need to have a few very dedicated people or it becomes the same.
09:02 erlehmann   by the way i believe cleaning up types is a good idea, regardless of if you change map size later.
10:19 MTDiscord   <luatic> proller: I'm afraid you won't get a PR merged that breaks multiple things and then just say "fixing this is up to future PRs".
10:21 erlehmann   haha that too
10:21 erlehmann   but
10:21 erlehmann   does the typedef thing break anything?
10:23 ROllerozxa  Hmm, what would it take for Minetest sound functionality to be more modular? Like, right now it is pretty much hardcoded to only use Ogg Vorbis. The Lua API SimpleSoundSpec also assumes all sounds are Ogg Vorbis sounds with the .ogg file extension, requiring you to omit the file extension of any sounds you want to use.
10:26 erlehmann   ROllerozxa you could probably easily extend it to support opus, as that also uses the ogg container. but i assume this is about the tracker music support, right?
10:36 ROllerozxa  Heh, yeah. Either way if either Opus, MIDI or tracker music support were to be added, the code would still need to be refactored to support more sound formats. Would be nice if it was made to be similar to how Irrlicht's texture loaders are implemented.
10:43 erlehmann   i think opus does not make sense really. it's more efficient, but then you have mods that are not backwards compatible or have to encode everything twice.
10:43 erlehmann   vorbis is really good enough
10:43 erlehmann   source: i have written podcast publishing software, encoding stuff multiple times is a chore
10:44 erlehmann   midi or tracker music on the other hand would be cool
10:44 erlehmann   and give a completely different feel
10:44 erlehmann   additionally, stuff like bytebeat does not compress very well
10:52 sfan5       rubenwardy: do you have any idea how the 64-bit APKs ever worked correctly https://github.com/minetest/minetest/issues/12081#issuecomment-1045990017
10:53 sfan5       the context is https://github.com/minetest/minetest/pull/9614 which is a workaround that only applies correctly via CMake
10:57 ROllerozxa  The arm64 android APKs are quite honestly broken and shouldn't be made available, it only confuses people since most will assume 64-bit to be better than 32-bit and go for it if their device is 64-bit.
10:58 sfan5       they are not advertised on the website or forums
10:58 sfan5       the play store forces us to provide 64-bit builds
10:59 ROllerozxa  ...oh
11:03 erlehmann   i can test aarch64 as soon as my reform is back, if there is a general luajit issue (there seems to be?)
11:04 erlehmann   would using builtin lua change anything?
11:04 MTDiscord   <luatic> It would probably ruin performance
11:04 erlehmann   please elaborate
11:04 ROllerozxa  Minetest Android always uses the LuaJIT library it downloads
11:05 MTDiscord   <luatic> But yes, it should fix LuaJIT related bugs. PUC Lua is presumably a lot more platform-independent than LuaJIT.
11:05 MTDiscord   <luatic> erlehmann: LuaJIT usually provides a ~5x speedup from my experience. Android users in modded singleplayer absolutely need this.
11:05 MTDiscord   <luatic> (IDK if the speedup is as large on ARM though)
11:06 ROllerozxa  Although it certainly wouldn't hurt to write a PUC Lua dependency script for Android, if only for testing
11:06 erlehmann   oh, i never had such experiences
11:06 erlehmann   like, luajit is a bit better, but not that much
11:06 erlehmann   may be different on x86
11:06 MTDiscord   <luatic> LuaJIT is much better.
11:06 erlehmann   or i have the “wrong” workload to test it
11:06 erlehmann   i believe you, i just wonder why i did not notice
11:07 erlehmann   do you have some test mod that makes it obvious?
11:07 erlehmann   or some indication which operations are affected?
11:08 MTDiscord   <luatic> https://www.lua.org/gems/sample.pdf
11:09 MTDiscord   <luatic> I have observed this on a MT-unrelated algo repo and on a couple MT benchmarks I ran
11:10 erlehmann   yeah pls give the benchmarks
11:10 erlehmann   i'll read the pdf
11:11 erlehmann   wait i know this
11:11 MTDiscord   <luatic> The last page is where it mentions LuaJIT and the 5x speedup
11:11 erlehmann   > The drawbacks are that it runs only on x86 architectures and that you need a non-standard Lua interpreter (LuaJIT) to run your programs.
11:11 erlehmann   ah yes, this seems trustworthy
11:11 erlehmann   so then we should not use luajit on aarch64 at all
11:11 erlehmann   since it runs only on x86 architecture
11:11 erlehmann   … right? ;)
11:17 erlehmann   what change in the build process between minetest 5.4.1 and minetest 5.5.0 could be responsible for the latter pushing the system load to 4 on a 2 core system, making my computer smell weird and overheat when i build?
11:18 erlehmann   could be something else hmm
11:20 erlehmann   oh, firefox :(
11:20 ROllerozxa  browsers suck :(
11:21 erlehmann   well, firefox was good for a long time, but recently it often crashes and uses CPU while doing nothing (like chrome)
11:22 ROllerozxa  Anyways I'm writing an Android dependency build script for Lua but I can't get it to create the static library (liblua.a)
11:22 ROllerozxa  (by Lua I mean PUC Lua)
11:25 ROllerozxa  never mind, it was hiding in src/ along with object files, didn't get copied over for some reason
11:26 erlehmann   i really wonder if luajit on aarch64 is giving that speedup
13:28 ROllerozxa  ok so I managed to compile Minetest Android with PUC Lua... it crashes instantly.
13:28 ROllerozxa  how lovely
13:34 ROllerozxa  oh never mind, arm64-v8a crashes with PUC Lua but not armeabi-v7a... that's ironic
13:36 sfan5       I guarantee you you just compiled it wrong
13:36 erlehmann   yeah can you please show how you did it
13:38 sfan5       I didn't do anything but I know that it is possible
13:38 jingkaimori for the replacement of formspec, is it necessary for modders to layout elements manually by specifying position? or there must be a default layout
13:38 erlehmann   ROllerozxa so how did *you* do it?
13:39 erlehmann   where's the build script lewobski
13:41 ROllerozxa  Okay so at first I tried to do it by writing an Android dependency build script but I gave up on that idea and just edited the Android makefiles to point to the built-in lua source directory and commented out LuaJIT.
13:41 ROllerozxa  https://github.com/rollerozxa/minetest/commit/b4fa87dcdf157f58b17260acf55cf182810bdcf1
13:43 sfan5       that sounds close to correct
13:43 sfan5       so whats the error when starting MT?
13:47 ROllerozxa  here's the stacktrace: https://mister-muffin.de/p/p1i7.txt
13:49 erlehmann   > Build fingerprint: 'Nokia/DoctorDoom_00EEA/DRD_sprout:11/RKQ1.210303.002/00WW_2_19B:user/release-keys'
13:49 erlehmann   hehe
13:50 ROllerozxa  ...eh?
13:51 MTDiscord   <CartridgeZine> ….eh?
13:52 erlehmann   02-19 14:29:29.392  3683  3683 F DEBUG   : Abort message: 'FORTIFY: strchr: prevented read past end of buffer'
13:52 erlehmann   i guess this is legit and you could try to figure out what illegal thing strchr() wanted to do to that poor buffer
13:53 sfan5       the most interesting part of the stacktrace (frames 3 to 6) are unfortunately missing symbolsd
13:53 sfan5       -d
13:53 erlehmann   yeah, how come?
13:53 erlehmann   would a recompile with -g or so fix it?
13:54 sfan5       possibly
13:54 ROllerozxa  android debug builds are already done with -g right?
13:54 erlehmann   well no idea why we have *some* symbols but not others
13:57 erlehmann   ROllerozxa can you step through this using a debugger btw?
13:59 jingkaimori for the replacement of formspec, is it necessary for modders to layout
13:59 jingkaimori elements manually by specifying position? or there must be a default
13:59 jingkaimori layout
14:04 ROllerozxa  erlehmann: I don't know, am I able to? Only thing I know of is running adb logcat to get debug logs and whatnot
14:07 erlehmann   well i do not know about debugger on android sadly
14:08 ROllerozxa  also I am indeed compiling with -g, super weird how some symbols are still missing
14:12 sfan5       you can also try addr2line to manually resolve an address to the symbol
14:23 ROllerozxa  oh alright. addr2line says that #03-#06 are lua_yield... would that give any clues?
14:23 ShadowBot   https://github.com/minetest/minetest/issues/03 -- Furnace segfault
14:23 ShadowBot   https://github.com/minetest/minetest/issues/06 -- Apples on the trees can not be eaten
16:44 MTDiscord   <MisterE> If colored shadows are possible, is colored sunlight possible? For example, the sunlight could take on an orange tint at sunset
16:54 sfan5       sure
17:59 MTDiscord   <Benrob0329> I mean, sunlight is already colored, but only for day vs night
18:01 MTDiscord   <Benrob0329> (as well as artificial light vs sunlight)
18:11 rubenwardy  !tell sfan5 didn't the Lua error handling change in 5.5? I remember a PR on that
18:11 ShadowBot   rubenwardy: OK.
18:11 sfan5       .
18:11 sfan5       what do you mean
18:12 rubenwardy  huh
18:13 rubenwardy  why can't I see you in the user list
18:13 rubenwardy  client's borked
18:14 rubenwardy  wonder how long the user list was outdated for
18:18 rubenwardy  the one I was thinking of wasn't merged
18:19 rubenwardy  another possibility is that we've updated the dependencies since
18:20 rubenwardy  so the code/compile settings for LuaJIT may have changed
19:49 sfan5       possible
21:08 rubenwardy  Looks like GitHub actions is no longer making artifacts for Windows builds
21:09 rubenwardy  oh ffs, it would be nice if people could fix stuff rather than just disabling it
21:10 erlehmann   is it githubs fault or did some minetest dev disable it?
21:11 rubenwardy  mt de
21:12 rubenwardy  I imagine they intended to renable them after a while, but there's no issue to track it
21:13 rubenwardy  https://github.com/minetest/minetest/issues/12087
21:13 MTDiscord   <Jonathon> its been disabled for a while now
21:13 rubenwardy  !title
21:13 ShadowBot   Fix Visual Studio/MSVC CI builds, and renable them · Issue #12087 · minetest/minetest · GitHub
21:18 MTDiscord   <LandarVargan> Also note #11176, which I need to fix at some point
21:18 ShadowBot   https://github.com/minetest/minetest/issues/11176 -- Fix Windows Visual Studio actions by LoneWolfHT
21:20 rubenwardy  oh right
21:20 rubenwardy  I'll close #12088 then
21:20 ShadowBot   https://github.com/minetest/minetest/issues/12088 -- Fix MSVC build on GitHub Actions by rubenwardy
21:24 rubenwardy  LandarVargan: please may you rebase your branch
21:24 rubenwardy  or I can do it
21:29 MTDiscord   <LandarVargan> Feel free to do whatever you want with it, I need to pop out to eat
21:48 sfan5       rubenwardy: gitlab has windows artifacts btw
21:48 sfan5       on that note the win32 is broken but I haven't had time to investigate yet
21:49 rubenwardy  Isn't that only on the master branch?
21:49 sfan5       it is
21:49 rubenwardy  Well, only on official branches. Not PRs