[0.13.1] Map generation process is very strange

[MrGrim]

After that, I made a test, generating 2 maps using the same seed - 1234567890 - with and without RSO. All settings at default except segmentation (low)

Interesting.. I don't think RSO is suppose to actually modify the terrain. I don't think it did in 0.12. I'll have to double check.

I don't really see much of a difference between the two images to be honest. They're both moderately segmented. The reply by ixnorp early in the thread also shows that turning water size (not segmentation) up causes terrain segmentation to go up. He has several images using the same seed with different water feature sizes that are a good demonstration. The 0.12 save I want to use has very big water features and it's basically switching biomes every 20 tiles or so.

[ytsejam]

After that, I made a test, generating 2 maps using the same seed - 1234567890 - with and without RSO. All settings at default except segmentation (low)

Interesting.. I don't think RSO is suppose to actually modify the terrain. I don't think it did in 0.12. I'll have to double check.

I don't really see much of a difference between the two images to be honest. They're both moderately segmented. The reply by ixnorp early in the thread also shows that turning water size (not segmentation) up causes terrain segmentation to go up. He has several images using the same seed with different water feature sizes that are a good demonstration. The 0.12 save I want to use has very big water features and it's basically switching biomes every 20 tiles or so.

Well, they're both highly segmented considering they were generated with segmentation set to "low", but to me the difference is clear between the 2. The RSO map doesn't look as bad the vanilla one.

I replicated ixnorp's tests in 0.12 to show the difference and posted the results here a few days ago. With the exception of one scenario (water in starting area, segmentation very low), his tests are all with segmentation set to "normal", so we can't say for sure only turning water size is causing higher terrain segmentation. To me all maps ixnorp generated look pretty much the same when it comes to segmentation, they're all very high. The only difference is the amount of water.

Anyway, here's an example of maps generated in 0.13.4 using ixnorp's seed (1946374888) with different water settings, but using "very low" segmentation instead of "normal". All other settings at default:

0.13.4 segmentation very low water very small

0.13.4 segmentation very low water small

0.13.4 segmentation very low water medium

0.13.4 segmentation very low water big

0.13.4 segmentation very low water very big

I didn't generate an "water only in starting area segmentation very low" map because, ixnorp already did that. So, with exception of this "starting area only" scenario where you get the highest segmentation (also a problem in 0.12), it's like the more water you add, segmentation gets higher. Still, all scenarios show very high segmentation. It's even worst considering those were generated with segmentation set to "very low".

[ixnorp]

ytsejam, I noticed that your maps aren't quite generated like mine so I generated my own set off of my seed. Your starting area and lakes don't generate identically and then grow. They ought to if you're using identical settings between the different map settings which is what got me noticing the changes. Here's my map gen process so you can compare your results to mine and see if they're identical or not. Not being identical would be quite concerning since Factorio is deterministic so that should never happen!

Map Generation and Capturing Process

Main Menu -> Play -> New Game
(I actually copy & paste 1946374888 from a text file to make sure I don't make any typos (and it's just faster!))


Once the map has loaded I do these commands while waiting in between the first and second for the map to reveal:

Code: Select all

/c game.surfaces[1].always_day = true; game.speed = 8; game.player.force.chart(game.player.surface,{lefttop = {x = -800, y = -800}, rightbottom = {x = 800, y = 800}})
/c game.take_screenshot{resolution = {x = 5000, y = 5000}, zoom=0.1, path = "MapPreviews-0.13.4/1946374888-Seg01Low-Water05VBig.png"}

The file name changes as per my settings and I use the leading number so that they're viewed in the proper order when flipping through them sequentially.

For uploading to the forums I do a bit of post-processing which consists of reducing the size to 25% of the original capture and converting it to JPEG since the original files are ~50MB
I tried to figure out how to take good screen shots of the map screen but I couldn't come up with a good process for it. The reason I settled on this technique is that I wanted to make sure that everything was aligned perfectly so that I could flip back and forth between the images and see exactly what the differences are.

Seed: 1946374888, Segmentation: Low
Water: Very Small
Water: Small
Water: Medium
Water: Big
Water: Very Big

Notice the small lakes to the North West being static and the receding biome phenomenon.
On the far East of the map you can see how the forest biome shrinks to reveal a set of lakes indicating they're on a lower layer and are overwritten by the upper biome level.
You can see due North of the spawn in the center how the copper is put down before the lakes are iterated upon and then swallowed up by the expanding water. The same can be observed on some Iron and other resources to the South (slightly SouthEast) of spawn.

[ixnorp]

I went and looked at 0.12.35 map generation in order to compare it to the 0.13.x map generation.

These maps are all generated with a fresh install of 0.12.35 using seed 1946374888. Due to the changes in map generation code the seed does not and is not expected to be identical or even similar between 0.12 and 0.13 but I wanted to use the same seed anyhow for consistency.

Water Only In Starting Area generates the very scattered, rough edged biomes that are seen in 0.13 but in all other water amounts a different biome generation is observed.

With Very Low Segmentation and Medium Water we can see there's very low (in fact, essentially no) biome variation.

Segmentation: Normal
Of note here in contrast to the 0.13 map generation, the biomes around the lakes expand as the lakes themselves expand. The biomes are also large and contiguous with relatively smooth edges.
Water: Very Small
Water: Small
Water: Medium
Water: Big
Water: Very Big

Segmentation: Very High
On the Very Big Water map notice how the starting area is still a distinct chunk and all of the resources are found there. Also, even though the segmentation is Very High it doesn't look like Jackson Pollock does Factorio. There are still distinct and contiguous areas where you can walk for more than a dozen paces and not go through 3 different biomes.
Water: Very Small
Water: Small
Water: Medium
Water: Big
Water: Very Big

[ytsejam]

@ixnorp, very interesting stuff. Yeah, I noticed I didn't generate maps the same way you did, your screenshots are much higher quality.

What I did was change the relevant map settings (water and terrain), I leave all other settings at default.

After that I use only this command to reveal a 1024x1024 area

Code: Select all

/c game.forces.player.chart(game.surfaces.nauvis,{{-1024, -1024}, {1024, 1024}})

I noticed it's different than yours, not just in area size (no idea what that "nauvis" means), I just copy/pasted from somewhere else and changed the size settings to 1024. I didn't use commands to stop the day/night cycle and any other because I wanted to take screenshots using the map view, not the game view. My screenshots are regular lower quality steam screenshots.

As for the seed, I also copy/paste yours so there's no typos. Maybe I just forgot to paste your seed in one of my tests. I'll use your settings to make some tests, but yours are more than enough to know something is very wrong with .13 map generation.

EDIT: I screwed up. I was copying your seed in the map exchange string field

Here's your seed in 0.13.4 with very low segmentation and very low water. The only difference is the size I used in the command (1024 instead of 800)

0.13.4 segmentation very low water very low

LARGER VIEW

[Bluelight]

Achievements are now off since we had to put in some oil spots.. There where no oil on the entire map (in relative measures aka. what has been explored)

[cube]

I've made some tweaks to our map generator, a bug or two related to map segmentation and completely replaced how starting areas are generated. Let's see if it will make you happy :-) Moving to fixed for next version.

[Xeteth]

I've made some tweaks to our map generator, a bug or two related to map segmentation and completely replaced how starting areas are generated. Let's see if it will make you happy Moving to fixed for next version.

[Qon]

That's great news cube! Thanks.

[hi_impact]

I've made some tweaks to our map generator, a bug or two related to map segmentation and completely replaced how starting areas are generated. Let's see if it will make you happy Moving to fixed for next version.

[scarhoof]

So excited to see how things look!

[Compboy]

So excited to see how things look!

Just don't look at it too long.
but yes i'm also exited about this new release. as lot of train issues are being looked at.