Blocky terrain issue with fork of ReTerraForged

[Hippopipo]

Describe the bug
All terrain becomes 4x4 cubes when C2ME is run with this fork of ReTerraForged.

To Reproduce
Steps to reproduce the behavior:

1. Create a world with ReTerraForged (Instructions in the repo readme)
2. All terrain is made of 4x4 cubes

Expected behavior

Terrain won't be made of 4x4 cubes but instead 1x1 cubes?

Screenshots

(From https://github.com/ETcodehome/ReTerraForged/issues/91. Not same instance as the logs)

Runtime info (please complete the following information):

• OS: Windows 11
• Minecraft version: 1.21.1
• Mod version: 0.4.0-alpha.0.112+1.21.1
• Mod branch: (fill this if you are not using the default ver branches)

Crash reports / logs

https://mclo.gs/xaM9RoR

Other mods

Cubes Without Borders [3.0.0+mc1.21]
ReTerraForged [0.0.6004R1]
Sodium [0.8.12-beta.2+mc1.21.1]

Checklist

• I am using the official version of the mod.
• I tried the latest development version but the issue persists.
• I searched for similar open issues and could not find an existing bug report on this.

Additional context
Originally reported in the fork's issues: ETcodehome/ReTerraForged#91

[ETcodehome]

For clarity, this was still under active investigation on my end. I consider compatibility with C2me to be very high priority, but i'm not confident that it's not something RTF is doing in a non standard way.

[ishland]

Try reproduce disabling useDensityFunctionCompiler.

[carbonatek]

Yeah it works with DFC off - interestingly C2ME 0.3.x works fine even with DFC on.

[ETcodehome]

Probably also worth noting that the issue doesn't occur when using the fabric equivalent release
c2me-fabric-mc1.21.1-0.4.0-alpha.0.16
not sure how closely aligned the two codebases are but thought this might be a valuable data point diagnostically

[ishland]

Probably also worth noting that the issue doesn't occur when using the fabric equivalent release c2me-fabric-mc1.21.1-0.4.0-alpha.0.16 not sure how closely aligned the two codebases are but thought this might be a valuable data point diagnostically

There's basically no difference between these two codebases outside of adapting api usage.

[InitAuther97]

I just want to point out that the terrain seems to be in 4x4 cubes, which seems to be the characteristics of a flat cache interpolator DF (value is sampled on pos>>2 creating blocky result)? I guess flat cache should usually be used in sampling biome or something?

[ETcodehome]

Is there any advice on what we can do / change to avoid these interpolation issues?

RTF makes extensive use of custom density functions and most the userbase wont think to first check and see if theres an open issue against C2me to find the workaround of changing the config toggle.

Is there any programmatic ways we can flag that usedensityfunctions should be disabled that c2me respects?

[ishland]

Is there any advice on what we can do / change to avoid these interpolation issues?

I cant find any immediate problems in you code base, but github web is horrible to use.
The only thing that might be broken is this, but I can't find any implementations of Noise.Visitor: https://github.com/ETcodehome/ReTerraForged/blob/de37bfbbd25ae0519349f5445ce07b8ac282487f/common/src/main/java/raccoonman/reterraforged/world/worldgen/densityfunction/NoiseFunction.java#L41
My guess is that something is being weird in your own pipeline when combined with dfc.

Is there any programmatic ways we can flag that usedensityfunctions should be disabled that c2me respects?

There's none.

[ETcodehome]

Is there any advice on what we can do / change to avoid these interpolation issues?

I cant find any immediate problems in you code base, but github web is horrible to use. The only thing that might be broken is this, but I can't find any implementations of Noise.Visitor: https://github.com/ETcodehome/ReTerraForged/blob/de37bfbbd25ae0519349f5445ce07b8ac282487f/common/src/main/java/raccoonman/reterraforged/world/worldgen/densityfunction/NoiseFunction.java#L41 My guess is that something is being weird in your own pipeline when combined with dfc.

Is there any programmatic ways we can flag that usedensityfunctions should be disabled that c2me respects?

There's none.

Appreciate you having an eyeball at this issue - I tried a number of workarounds in your suggested area but regrettably didn't find any success.

I eventually doubled back to the inconsistency that fabric works and neoforge didnt while using common shared code.

I wondered if there was a mapping difference between Fabric and Neo on
https://github.com/RelativityMC/C2ME-fabric/blob/de29c92a35d02b419be8485d3ef586c81acfd109/c2me-opts-dfc/src/main/java/com/ishland/c2me/opts/dfc/mixin/MixinChunkNoiseSamplerDensityInterpolator.java#L44

This failing to resolve would cause it to use the fallback path and not interpolate to my understanding. I'm a little out of my depth chasing this much further, but pulling apart the deobf byte code sources as near as I can tell both Neo and Fabric map equivalently here and on the main functions which killed this potential investigation avenue.

Open to any further thoughts/ suggestions anyone has for a further investigative path.