Hacker News Reader

Comment by rland | Hacker News Reader

It's been a while since I've played Minecraft, so I can't speak to the newer terrain. But I always thought they could significantly improve the world gen.

Perlin noise is cool, but why not try to fully simulate plate tectonics, weather patterns, drainage? It would go a long way if deserts were in the rainshadow of mountain ranges, islands corresponded to real volcanic activity, etc.

I guess there is a limit to what you can do with the chunk-by-chunk approach, but I think a certain type of minecraft player stays basically put within a couple dozen chunks. Having worlds "load" on exploration would be a worthwhile tradeoff.

I get that this is a hard, but it is interesting, and it's a problem that would captivate an intelligent developer. They have like a million dollars, why not?

View on HN

What do you mean having then load on exploration? That's the current state - a chuck isn't generated until you need to load it. This means when you take an old save to a new version any already-loaded chucks you're stuck with but any new exploration causes new world gen on the new version.

Yes, I think (maybe) the reason they use perlin noise is because it’s fast. I’m saying have an actual loading screen when you go to a new area and spend some CPU cycles doing advanced plate tectonics or whatever.

Oh wow. I did not realise this was the case. Thank you!

I think this actually might be a slight negative tbh - since everything is split into chunks and generated on demand, you lose the ability to do larger passes over multiple chucks at gen time. No real way to do tectonics or weather patterns as you don't necessarily know what's in your chunk neighborhood.

You can by generating Metadata for an area larger than a chunk itself.

Also frequently when using things like peeling noise is to use different scale and resolution of the noise itself. While on smallest scale there might be 1 noise value per minecraft block. You can can also use a different scale in that 1 noise value is used per 10x10 minecraft chunk (16x16 blocks?)

Also not sure how familiar you are with noise generation like perrin but things have slopes to them and isn't just completely random per point. So even without seeing generated data. You can infer a little bit about the next value depending on the intensity and resolution.

Lastly, you can just not use noise at all and just place things at given coordinates and make the game incorporate them into the proc gen at chunk time. E.g. you want a city always at 1000,1000 coordinate far away from 0 0. When you get near you can apply smoothing filters in and around the 1000 1000 position thru various math and layering systems so no matter the proc gen at that point. City can be there and it not look completely out of place.

> No real way to do tectonics or weather patterns as you don't necessarily know what's in your chunk neighborhood.

Sure you do. When you want to apply tectonics to a chunk, it's easy to calculate (and cache) the pre-tectonics version of all the neighbors.

The problem isn't effects that need to know about their neighborhood, it's effects that need detailed info from 50 chunks away.

There is a biome system that solves this issue. You can compute biome borders and use them for tectonics or w/e else.

You could do much larger chunks on gen.

It is stunningly difficult to do something like this within procedural generation. Remember a given xyz coordinate has to generate given only the coordinate number itself and a seed. You couldn't do stuff like plate tectonics within an "infinite" world because that would require generating every block which is akin to generating every digit of pi

Hmm, I'm not sure that's true.

If you generate a noise function that dictates where the "plates" are, then the location of the "plates" is already predetermined from the construction of the function. Then the output of that function could be used to alter the mountain/terrain functions.

Thus, no need to pregenerate everything. You pregenerate all of the functions in advance, but not all of the outputs.

You could set up domain boundaries where everything in that domain is relatively isolated from other domains. When nearing an ungenerated domain, do the procedural + evolutionary generation and handle the borders blending. Domains could be tectonic plates. This would cause hitches on exploration, but they would be relatively infrequent and overall few since domains would be so large.

I also find it stunningly difficult, however, when I look at Sebastian Lague's channel [0] (no affiliation, just one of the best channels I follow), I feel like this could be a not so stunningly difficult project for him.

[0] https://www.youtube.com/channel/UCmtyQOKKmrMVaKuRXz02jbQ

The article suggests that "more realistic" is correlated with "boring".

"Real" is an interesting concept. Any such simulation couldn't really be based on Earth geophysics. Who wants to spend months crossing the steppes? Real magma isn't something you find by accident. Real tunnels can collapse. And so on. So these intelligent developers will need to create a new system of geophysics.

I've read that Dwarf Fortress uses a process more like what you describe.

I think a SimClimate could become a hit. Maybe the dynamics could be gamified for kids, but I bet there are lots of adults who'd also want to just "spend months crossing the steppes" following the seasons pretty much.

Hmm are there any Climate Change strategy games?

There are games for nuclear war and pandemics that are fun and educational about the globe-spanning dynamics of those catastrophes. There are also games like Universal Paperclips that provide an intuition for the exponential dynamics of complex systems.

Seems like there should be a way to fuse these concepts around climate change.

Civilization VI tacked on climate mechanics with one of its addons.

https://civilization.fandom.com/wiki/Climate_(Civ6)

> The article suggests that "more realistic" is correlated with "boring".

I wouldn't say so. With the increased mountain height I'd argue the terrain is more realistic than ever, and not boring.

I'm referring to this part of the article:

> In the b1.8 update [September 2011], the terrain generator was subject to a rewrite aimed to make the code simpler, and make the terrain more realistic. ... at the time it caused quite a controversy, a thread on the Minecraft Forums decried the new terrain as "boring".

What does "realistic" mean to you? Are you looking at it like a geologist? If we follow rland's idea to completion, Minecraft players should look for kimberlite pipes to find diamonds, etc.

I'm certain some people would love virtual prospecting like this. But I strongly doubt the game would be as popular.

Perhaps I should've re-thought that part. Realism was widely interpreted to be the goal at the time. But the changelog only mentions code simplification.

I'd say if you compare the first two set of screenshots. It is more realistic since you don't get as crazy land formations. But falls short of creating significant mountain ranges to compensate, making it less realistic in a way.

That "in a way" is what I refer to. rland proposed a more physics based model. So I'll look at things through that lens.

The first image, 2011, makes me think of tower carst formations, like Ao Phang Nga National Park or the Li River. It's indeed a "crazy land formation" in that most of the planet is not like that. But well within what a more physics based model might generate.

The second image, 2021, makes me wonder what's going on. How did that concavity appear in the bottom left? Was it limestone? Sandstone, like NM's Natural Amphitheater or CO's Cliff Palace? How is the top of the concavity supported, given that it appears to carry a lot of earth?

What is the geological connection to the ridge above it? How is the ridge supported? It appears highly eroded. Where did the scree go? (Shouldn't some have filled in the exposed concavity?) Why is the transition from light-colored to green so clearly delineated?

In the real-world, large cave systems are rare, and walkable ones even less common. But they are also part of what make Minecraft fun.

So yes, I think a more realistic (in the physics sense) would be less fun than what Minecraft does. (Excepting, perhaps, a small number of more geophysically inclined players.)

Other people are pointing out that real == boring, but I disagree. Perhaps we wouldn’t want a literal real world, but hyper interactive terrain like you suggest could be used to explore new and interesting gameplay ideas.

You can do something like this by layering. E.g. generate a high level coarse map that sets some features like average elevation, and calculates climate from that and derives biome from that, as an input into the lower level worldgen

I believe this is exactly how minecraft's new "3D" worldgen works. It calculates a high-level terrain map, then a temperature map, and then biomes are placed on top of them that fit the characteristics.

Reminds me of the AutoBiomes paper! Waiting for something to be integrated into EG a terrain creation tool (no!-MC games), should produce pretty interesting results.

https://cgvr.cs.uni-bremen.de/papers/cgi20/AutoBiomes.pdf

Doesn't the ultra-realistic mod terrafirmcraft already do exactly that?

To my knowledge, no. Terrafirmacraft does have strata of different stone types that stretch for large areas, but these aren't tectonic plates. It does generate biomes based on "lattitude" however.

As another comment mentions, dwarf fortress does do all of this though. I don't think this would work in minecraft though, I think minecraft needs a mix of terrain types near each other to be most interesting, and tectonic plates would spread it out.

Reply on news.ycombinator.com