EcoBuilder13

Can the ticket owner or another authorized user append the following entry onto the table on this ticket? This is the information from MC-241815 that was marked as duplicate of this issue. If it's believed this entry should be kept it's own issue instead, MC-241815 should be reopened.

I just made a copy of the effected world file and ran it on a fresh 21w39a vanilla server and the error in the log about column_placer feature is no longer present. Though there was still a major regression in load times even after first launch. The average start time is about ~17 seconds now; however, after using the --forceUpgrade option when starting the server it has returned to an average of ~7 seconds. That is still worse than 1.17-pre2 though a major improvement from the first few runs after updating to 21w39a. Can someone else that has this issue check if the error in the log is gone? If so I will update this ticket to state the log error has been fixed but not the performance regression.

I believe they prevented the biome from generating in the experiential snapshots and made a feature decorator responsible for adding the plateau part above a certain height. That appears to no longer be the case in 21w37a, as the biome now generates and the plateaus don't generate the same.

This could be the result of the new Data Converter system included in Paper in builds #232 and later. (This was a feature of Tuinity that got added in the merge) Data Converters are a replacement for Data Fixer Uppers (DFU) which is the system which updates older chunk data to newer formats. The issue described here supposedly is caused by DFU upgrading chunks with faulty data which when read causes the errors to appear in the log. Since Data Converter is an entire replacement of DFUs there is a chance that the bug is not present with it's world conversion meaning chunks have the proper data. (No errors in log) If my theory is correct the issue should be fixed in Paper and all of it's downstreams if they have been updated to build #232 or later. Though there will be no fix available for Vanilla and modded servers as this is not a copy and paste solution even though that's not necessary possible anyways due to potential licensing issues. For those who are interested in learning about Data Converters from a technical standpoint you can read about it on the GitHub repo: https://github.com/PaperMC/DataConverter. Do note that the repo contains a Fabric mod that implements Data Converter but it should never be used. While you could compile it your self it if they got it working outside of a dev environment it should not be used unless you fully understand what it is doing or are testing a contribution to the project. I nor the PaperMC team are responsible for any damage caused to your save file if you use this modification.

TLDR

For users of PaperMC or a downstream (Airplane, Purpur, etc) you should upgrade to a build that contains the changes in Paper build #232. This should fix the issue but no promises. For users of the Vanilla server or a modded one (Forge, Fabric, Quilt) a fix is not available and you will have to wait for Mojang to fix the issue unless if the community creates another solution.

It also might be worth trying to generate a new world with the same seed on 1.16.5 and then upgrade it to 1.17.x to see if the issue can be recreated at the same spots. If so people with affected world files just need to provide the seed and coordinates to a location where the issue occurs rather than uploading the world file it's self for people with cases like yours. The only thing we have to be careful of is that the issue may behave differently at particular spots depending on if they are old chunks, newly generated chunks, or up to date chunks.

The only negative effect should be errors in the log and maybe some performance degeneration. Since it appears to be a problem with values used with world gen nothing should happen to the chunks. If you are worried about the chunks having broken values stored in them that should be fixed when the dfu is working properly again since it will replace the problem with the proper value.

I would recommend not to explore new chunks as they will most likely have broken generation. The error can occur in any chunk if the game tries to get that key. It's supposed to have a value but the dfus have appeared to have done something bad to them. So it's the same problem, but the error is just showing up in different places.

Most likely a duplicate of MC-227202, the first log is the same error with a slightly different stack trace. The other logs are probably caused by the same or a similar data fixer upper (dfu). As the column placer decorator throws errors when the server starts in my case but I haven't seen the other areas because I have not updated anything other than spawn chunks. Though it could be two separate dfus causing the two issues. The reason this is believed to be a problem with the dfus is that they are responsible for updating chunk data and they don't properly update some of the keys in world generation which is why the error is thrown.

When you "optimize" your world it runs the dfu on all of the chunks that you have generated to make sure the format is up to date so they load in faster. It is very possible that the same dfu that causes the column placer issue causes other problems which appears to happen in forest biomes as it looks like the tree decorator is what is having problems. If you are wondering what a decorator is it is something the game uses to place features in the world. The tree decorator is what decides where trees go and creates and places the trees and the structure decorators do math to the world seed to get coordinates for where it tries to place structures then generates them when the checks pass.

No, I have left the world to generate with what the default generation for the snapshots. I also skipped updating to the snapshots with the noise caves and world height changes so I haven’t touched anything with the world nbt or used the 1.18 terrain on this server.

Yes it still occurs in pre4 and pre5, but the issue is no longer present if I downgrade the server to pre2 or below.

@tryashtar, the additional elements have the faces inside of the stairs removed so the game only has to render the faces on the outside of the stair. So only model stitching time should be affected but that should be minimal. The thing that is wrong with the vanilla model is that the bottom element is the size of a half slab. That means below the elements which make up the upper part of the stair you can see the texture on that element. Even though the player can only see it if in spectator mode the game still draws that part of the texture while it can not be seen. The fix I made makes use of smaller elements with removed faces so that only the part exposed of the bottom is rendered.

This issue is still present in 21w15a, it affects different sounds than what were originally affected. As of my testing in 21w15a the only remaining duplicated sounds are:

• minecraft:entity.goat.eat is duplicated by minecraft:entity.goat.screaming.eat

• minecraft:entity.goat.long_jump is duplicated by minecraft:entity.goat.screaming.long_jump

• minecraft:entity.goat.ram_impact is duplicated by minecraft:entity.goat.screaming.ram_impact

I have made a fix for this bug. I have added additional elements to the block models that make it so the unnecessary faces do not render. The attached fix is a resource pack that works in 1.16.5 and the 1.17 snapshots. Just ignore the warning for 1.16.5 and below. I also give Mojang Studios full permissions to use the block models in the resource pack directly with or without modification.