Time Nick Message 00:03 hecks this is what 800 dist looks like https://a.uguu.se/uBOwidHVHUGt_8002.png 00:03 hecks it's arguably cheaper to render than a health bar :^) 00:04 hecks Some interesting stats: 00:05 hecks a map of valid CIDs to 8 bit colors would be 256k in size 00:05 hecks a rough heightmap of an entire world would only be 16 MB 00:06 hecks a full mapblock is 32 kB if I recall correctly; the first LOD would bring that to 8 kB without lowering the resolution, and a really far one is under 1 kB 00:07 sofar mv hecks heck-yeah 00:07 hecks the definition and position of a deco imposter can be packed into 16 bits 00:08 hecks an XYZ proxy for manmade objects at 16x16x16 is about 768 bytes 00:08 hecks if you allowed, say, a minute of warmup time on server start to pre-generate some useful data, you could do some *crazy* things LOD wise 00:09 hecks and also get yourself some information about the world that is currently impossible to sample without exploring the world, causing a chicken and egg problem 00:11 hecks imagine if dungeons generated in actually reasonable places, and if your lua mapgen could do the same with towns and other interesting things 00:17 paramat 'heightmap': does that suggest a method that assumes a conventional terrain? i feel that LOD should be generalised for any possible world-structure 00:18 paramat for example mgfractal can create insane world-spanning megastructures 00:22 hecks anything involving height and ocean level is just a heuristic for conventional mapgens, so yes 00:22 hecks but you could have an alternative that just keeps generating things at lower and lower resolutions 00:23 hecks and I suppose you'd use that as a fallback for things like floatlands, or caverns 00:23 hecks nevertheless, most of the time you *are* playing a world with regular terrain 00:24 hecks so it makes sense to support that 00:25 hecks it could be the mapgen's responsibility to say "use terrain trickery for mapblocks between -64 and 512, generate the rest normally" 00:25 hecks on a completely unrelated note, I'd love to have the ability to switch mapgens depending on height 00:26 hecks that way I could borrow v7's floatlands for carpathian *and* generate a megastructure starting at -6000 00:27 paramat yes that kind of generalised alternative is essential 00:28 paramat we have discussed stacked mapgen a lot, to use the vertical space 00:39 hecks then the best option is to allow the specification of such "spaces" for LOD purposes 00:47 paramat personally, i would only implement a completely general method, even if slower for normal terrain. simpler and 'purer'. but that's just me 01:13 hecks https://a.uguu.se/CP6iprSlLLgT_chart.png 01:14 hecks the problem is that huge mapblock-sized blocks may look significantly worse for terrain, that's one 01:14 hecks and two, approximating land and water with them would suck 01:15 hecks and in addition, fake sloped terrain with "proxies" lets you see structures from a lot farther than if it all were just huge blocks 01:19 Cornelia This LOD.. would be like perfect for my crazy weather idea which would place actual nodes for clouds.. :p 01:25 hecks if there was a clouds API that would let me specify the noise, that would be good enough for me 01:29 Ruslan1 Hello anyone 01:30 hecks hello world 01:31 Cornelia Oh hecks 01:32 hecks what is it 01:35 hecks playing with obscenely high draw distances made me realize just how bad the culling is 01:37 paramat hmmm, if the generalised method would look that bad then it's not good enough. the generalised approach is the higher priority. especially since LOD becomes extremely useful for viewing non-standard megastructures 01:39 paramat that's what i like about c55's farmap method, it's generalised. also it treats all nodes the same, terrain, builds, decos 01:39 hecks the generalized method does not look bad if the distance is right 01:39 hecks the problem is that it's less efficient 01:40 hecks heightmap and proxy can do the same thing with a fraction of the data 01:42 hecks the matter of the fact is that you *can* make lots of assumptions about certain mapblocks, which probably constitute the majority of all blocks every generated by MT 01:43 hecks so why not use said assumptions to take a shortcut? 01:44 hecks it's possible to seamlessly merge the two approaches as well 01:45 hecks another thing; in mapgens like carpathian, large scale structures can not be as nicely represented by uniform blocks 01:47 paramat ok 01:49 paramat "Keep the common base (the Minetest Engine) as universal as possible. I like to compare everything that goes into the engine for how well it suits a gravitation-less space game without ground surface. If everything is either disableable or suitable for such, then it gives much more freedom for people to experiment and create interesting things, even in ground-based worlds." 01:49 hecks I guess 16x16 might be a good enough resolution for super far LODs even at realistic scales, but slopes are still more efficient to render at that point 01:49 hecks and slopes can be used "earlier" if you know that I mean 01:49 paramat from the MT roadmap http://c55.me/blog/?p=1491 the section about universality 01:50 hecks counterpoint: MT can barely handle non-Y rotations 01:50 hecks and it would be difficult to do anything "spacey" in it due to the nature of nodes themselves; they constrain you to the 6 directions they face 01:51 paramat yeah i know :) 01:51 hecks all in all, if domains were implemented, you could just turn them off 01:52 hecks and last time I checked, most mapgens do fit the model 01:54 paramat mgfractal doesn't if you use custom parameters, which is the whole idea of it 01:57 paramat here's the 'roadmap continued', for your interest https://forum.minetest.net/viewtopic.php?f=7&t=9177 17:08 Krock Will merge #7753 in an hour unless there are objections. 17:08 ShadowBot https://github.com/minetest/minetest/issues/7753 -- Fix temporary path crash in TestAuthDatabase by SmallJoker 18:11 Krock mergign