Time Nick Message 01:20 MTDiscord Hey everyone, a new blog post is out: https://blog.minetest.net/2022/07/07/June/ 02:08 erle > the next step is to only show settings if they are relevant 02:08 erle > For example, shader settings should only be visible when shaders are enabled, and Android settings should only be on Android. 02:09 erle rubenwardy, i suggest to hide stuff that can never be enabled – but gray out / disable stuff that can be enabled and place it close to the thing that enables it (i.e. shaders) 02:10 rubenwardy That's complex as it requires implementing disabled versions of most components 02:11 rubenwardy in the shaders section, when shaders aren't enable you'll see empty sections like tone mapping etc 02:11 rubenwardy also, I'm still weeks away from doing this, there's not point giving me feedback on this now 02:23 erle oh i just wanted to point out there is a difference between “can not ever work” and context-sensitive options 02:23 rubenwardy (by component I mean settings component, it's possible to do this with formspecs with a lot of tweaking / trial and error) 02:24 erle i have had a lot of issues with context-sensitive options at work and the problem is that many people's ideas about how the UX is easiest to understand do not stand the test of reality 02:24 rubenwardy yeah, it's certainly desirable 02:25 rubenwardy Rather than making disabled versions of all components, I could make a single disabled component that just says the setting title and what it requires 02:25 erle for example, a coworker wanted next to a bunch of checkboxes one that if checked, checks all checkboxes 02:25 erle on the same level 02:26 erle and had a hard time understanding that this makes the interface insanely more complex to understand and implement if a checkbox can check another checkbox. 02:27 erle rubenwardy that sounds good 02:28 rubenwardy That's a common interface pattern, to have bulk setting checkboxes 02:28 erle yeah my coworker said as well 02:28 rubenwardy It's usually seen in tables, with the bulk set checkbox in the header 02:28 erle try implementing it 02:28 rubenwardy I have 02:28 erle it only works in very specific interfaces 02:29 rubenwardy It's easy to implement when your UI is based on data, rather than basing your data on UI 02:29 erle anyway, my current shortcut to explain it to people is “can you implement the enabble/disable or show/hide logic in pure CSS?” 02:29 erle because if you can do that, you have no loops 02:30 erle anyway, most users that think they do want it don't want a checkbox that checks other checkboxes. 02:30 erle they may want a button that checks checkboxes or something similar 02:30 erle rubenwardy, do you have the table example public online? 02:31 rubenwardy No, it's proprietary 02:31 rubenwardy Done at work for a client 02:31 erle btw, in my case the solution was to make the interface better so that it did not have a lot of checkboxes to check to get to a useful result 02:32 erle and instead of a checkbox that checked checkboxes i made a link that did it 02:32 erle so that a) you could go to the previous state easily b) there is no question about what happens if you uncheck one of the checkboxes that got checked or so ;) 02:32 erle because inevitably someone will got the idea that *unchecking* the “check all” checkbox should uncheck all boxes or so 02:33 rubenwardy The checkboxes in the table were for row selection, checking the header box selected or deselected all rows 02:33 rubenwardy It was an indeterminate checkbox as well 02:33 erle that sounds sensible 02:33 rubenwardy If some but not all rows were selected, it showed a bar 02:33 erle i once suggested to dragan espenschied years ago that it should be possible to make a calculus for UI 02:34 erle to determine the power of UI elements 02:34 erle the spinning beachball of death is the halting problem 02:36 erle rubenwardy can i influence the colors you are using for the mod thing btw? 02:36 erle you can desaturate them or use gimps colorblindness filter to see why they are hard to read for some (e.g. me) 02:36 erle i can figure out better colors, but i'll only do it if you want that 02:41 erle rubenwardy thank you for mentioning the license laundering on https://blog.rubenwardy.com/2022/06/22/gpt-3-minetest-modding/ 02:41 erle it was a funny read 02:41 erle the rise of the full-stack-overflow programmer 02:42 erle i wonder how many security issues will be reurgitated that way 13:59 Quandale Good morning guys 14:06 Quandale GPT 3's Minetest mod AI is really something, ain't it? 14:39 erle Quandale just like rubenwardy wrote, the thing is a cargo cult machine 14:39 erle it is certainly amusing 14:39 rubenwardy about as skilled as a beginner dev, they also cargo cult 14:39 Quandale What does cargo cult mean 14:39 erle well, not if you force them not to 14:41 erle Quandale as the story goes, on some island troops had an airfield during a war. the indigenous population was inspired by this and created a new religion in which you make an airfield out of wood or hay or so and do rituals so “john frum” (john from america?) delivers supplies 14:41 erle in terms of programming it means something like … putting together pieces, hoping to get an effect, but not really understanding any of it 14:41 Quandale Ah okay 14:42 erle in other words, the AI understands the minetest API as much as google translate understands language, i.e. not very much 14:43 erle you can easily see that if you compare it to programs that are specifically written to understand languages, e.g. compilers 14:44 erle IMO btw, the existence of sufficiently smart compilers means the chinese room argument is a riddle for non-programmers ;) 14:45 erle rubenwardy i have never seen beginner devs that were as dumb as the AI is for long. so … can it learn beyond what you showed in some way? 14:45 erle with feedback or so? 14:45 rubenwardy It can fix bugs if you get it error messages 14:46 rubenwardy s/get/give 14:46 rubenwardy https://rwdy.uk/MekgP.png 14:46 erle i wonder if it makes common errors, given it is trained on buggy code (all code is buggy until proven otherwise) 14:47 erle or if it makes its own errors as an artifact of the training process or the structure of itself 14:47 MTDiscord It can't "learn" (P stands for Pre-trained), but can improve on it's own output if asked. 14:47 erle i mean i have even written a semgrep rule to find dupe bugs in mods ;) 14:51 Quandale Why is papyrus so rare in MTG? Can't find a single block of it across hundreds of blocks 14:57 erle Quandale others might be able to reproduce it with seed and coordinates 14:59 Quandale 13721338859501107996 14:59 Quandale Every seed I've been getting as of late is snowy and barely any papyrus 15:05 Quandale Sorry abt my absence, some weird stuffs been happening to my hexchat but I fixed it and I'm back 15:06 MTDiscord An AI that programs as well as a 6-year-old is actually pretty impressive considering that it takes WAY less than 6 years to train it. 15:07 erle given the disaster that microsoft tay was, i think maybe you need to raise AIs 15:08 Quandale Considering that the AI was made for text rather than surviving 15:08 erle like, curate the input 15:08 sfan5 "Every seed I've been getting as of late is snowy" that's normal 15:09 erle well, microsoft tay became a simulation of a neonazi teen with daddy issues in about 16 hours, before microsoft pulled the plug on their “let's train our AI on what is on twitter” experiment 15:10 erle i mean you would probably not let a newborn unsupervised on twitter 15:10 erle so a fresh AI should not be fed that either 15:10 MTDiscord Sounds like they tried to make an AI that imitated humanity, and sadly enough, succeeded. 15:10 erle that's the problem, yes 15:10 erle i mean, arguably the problem is where they put it 15:10 Quandale Train the AI on something more civil than Twitter, like Modern Warfare 2 lobbies 15:11 erle on fediverse it would have been a simulation of a queer anarchist teen with daddy issues 15:11 definitelya Newborn: literally doesn't understand what people are. 15:11 erle Quandale that made me smile 15:12 erle i want an AI that was trained on kilbith but only writes comments, so the comments are sufficiently mean 15:12 erle “comment this code in the style of $PERSON” 15:12 MTDiscord It's a novel life-form whose daddy is a civilization made principally out of meat; it'd be hard NOT to have issues. 15:12 erle uh great now you reminded me i am made out of meat 15:12 erle how do you do fellow humans 15:12 definitelya Fascinating 15:14 Quandale What if you could turn the skeleton block that appears when you die into bonemeal 15:15 ROllerozxa how do you do fellow humans made out of meat 15:15 definitelya Quandale: That's illegal! Also useful. 15:30 MTDiscord We're not "made [solely] out of meat". Bones aren't meat. 15:36 celeron55 as it concerns electronics, we're bags of salty water 16:40 MTDiscord Technically bones contain meat if you're willing to stretch the definition to include marrow 16:56 Noisytoot Does Minetest 5.5.1 work with OpenGL 2.1 for anyone else? For me, it fails with https://u.piss.ar/tmp/StiF32R17e.txt. It works with LLVMpipe (LIBGL_ALWAYS_SOFTWARE=true). 16:58 sfan5 https://github.com/minetest/minetest/issues/12467 16:58 sfan5 if you have enough details so I can reproduce this then I can fix it 17:03 sfan5 also consider trying with LIBGL_DEBUG=1 MESA_DEBUG=context and see if that says more 17:05 Noisytoot It does not, except complaining that ~/.drirc does not exist 17:06 sfan5 ok 17:06 sfan5 are you using flatpak as in the issue or? 17:14 Noisytoot I am not using flatpak, I am using the package from GNU Guix 17:24 sfan5 wayland/x11? which gpu? 17:26 Noisytoot X11 and the integrated GPU on the Intel Core 2 Duo P8600 17:33 Krock if llvmpipe works, you might as well blame the Intel drivers 17:33 Noisytoot LLVMpipe uses a different OpenGL version by default. I can also reproduce this if I set LIBGL_ALWAYS_SOFTWARE=true MESA_GL_VERSION_OVERRIDE=2.1, which should make it use software rendering with OpenGL 2.1. 17:33 Krock unless you're missing an opengl library for that 17:34 Krock yes, I can reproduce this 17:35 Krock 3.0 is the first working OpenGL version 17:40 sfan5 if you can the output with --trace might provide a hint 17:40 sfan5 but might also not, in any case this needs a deeper look 17:41 Krock 2022-07-07 19:41:06: VERBOSE[Main]: Irrlicht: GLX >= 1.3 17:41 Krock the only relevant line (and last one) before the error 17:41 sfan5 ¯\_(ツ)_/¯ 17:42 Krock source/Irrlicht/CGLXManager.cpp 17:49 Krock I don't even have an idea how or where CGLXManager even is initialized... would be best to open an issue for that on GitHub, Noisytoot 18:10 Noisytoot Done. 19:57 Noisytoot I can reproduce this with the latest master, but only in release mode. In debug mode, it sometimes works, but sometimes fails due to another error ( https://u.piss.ar/tmp/CPLVfa1oqS.txt ). 19:59 sfan5 ..xcb? 19:59 sfan5 Minetest does not use libxcb 21:07 erle Noisytoot i recently encountered a problem like this on reform2 21:08 Noisytoot reform2? 21:09 erle MNT reform2, an imx8-based laptop that needs no binary blobs and has schematics. a friend of mine makes these: https://mntre.com/media/reform_md/2020-05-08-the-much-more-personal-computer.html 21:09 erle the thing is, i doubt it is an intel driver bug 21:10 erle there are a bunch of devs who will always first suspect it is an intel driver bug though ;) 21:10 erle the intel integrated GPU drivers do not seem to have a good reputation 21:12 erle Noisytoot 00:02.1 Display controller: Intel Corporation Mobile 945GM/GMS/GME, 943/940GML Express Integrated Graphics Controller (rev 03) 21:12 erle should i be able to reproduce it using that? 21:12 erle if so, give me a revision 21:12 erle and i can compile 21:14 MTDiscord Looking at the bare specs - and ignoring feats like OSS - the Reform 2 seems pretty overpriced to me 21:14 Noisytoot 5.5.1 21:30 erle luatic i guess you have to build your own then 21:31 erle some girl from australia is doing it apparently 21:39 Noisytoot 5.5.1 build in debug mode is also unaffected by this bug 21:39 Noisytoot s/build/built/ 21:45 MTDiscord erle: heh, nah, decent laptops exist; I don't need to know the circuitry in great detail 21:46 Quandale Do you know where the textures for MTG are? I look in the textures folder and I only see the texture pack I have equipped 21:54 erle Noisytoot please tell me exactly how to reproduce 21:55 erle luatic the question is if you are willing to put up with asshole behaviour by vendors. like … “genuine batteries” 21:55 erle the thinkpad T450 has a list of “allowed“ screens, so if you upgrade it, the backlight adjustment is disabled 21:55 erle stuff like that 22:01 Noisytoot erle: Compile minetest 5.5.1 in release mode and then run it (with MESA_GL_VERSION_OVERRIDE=2.1 if your GPU supports OpenGL >2.1) 22:12 erle Noisytoot which GPU do you have? 22:12 Noisytoot Intel Core 2 Duo P8600 integrated GPU 22:13 Noisytoot I can reproduce it using LLVMpipe (LIBGL_ALWAYS_SOFTWARE=true) too. 22:39 erle Noisytoot what does lspci say about your GPU? 22:43 MTDiscord Quandale: games/minetest_game/mods/.../textures/ 22:44 Quandale games is empty 23:01 Noisytoot erle: This is the entire output of lspci: https://u.piss.ar/tmp/5wmRCDMB7m.txt 23:01 MTDiscord Oh, didn't you install via flatpak or something, Quandale? Can't help you there 23:02 MTDiscord Just need to find the builtin game dir