Fix server compatibility: use modern post-1.18 chunk format by louis-e · Pull Request #855 · louis-e/arnis

louis-e · 2026-03-28T11:46:12Z

Arnis-generated worlds appeared empty on servers (PaperMC, etc.) because chunks used the deprecated pre-1.18 Level wrapper format and were missing critical metadata. Servers don't auto-repair chunks like the singleplayer client does, so they treated the chunks as corrupt/ungenerated.

Changes:

Remove Level compound wrapper, write chunk data at root level
Add DataVersion (3955/MC 1.21.1), Status ("minecraft:full"), yPos (-4)
Emit all 24 sections (Y=-4 to Y=19) with biomes, not just sections with blocks
Compute MOTION_BLOCKING/WORLD_SURFACE/OCEAN_FLOOR heightmaps from block data
Add required empty structures, PostProcessing, block_ticks, fluid_ticks
Set isLightOn=1, InhabitedTime=0, LastUpdate=0

github-actions · 2026-03-28T11:48:12Z

⏱️ Benchmark run finished in 0m 34s
🧠 Peak memory usage: 1115 MB

📈 Compared against baseline: 30s
🧮 Delta: 4s
🔢 Commit: 3a317b4

🟢 Generation time is unchanged.

📅 Last benchmark: 2026-03-31 18:53:41 UTC

You can retrigger the benchmark by commenting retrigger-benchmark.

Copilot

Pull request overview

Updates Java Anvil chunk serialization to the modern (post-1.18) root-level chunk NBT layout to improve dedicated server compatibility (e.g., PaperMC), addressing cases where servers treat generated chunks as ungenerated/corrupt.

Changes:

Switch chunk NBT output from the deprecated Level wrapper to root-level fields (e.g., DataVersion, Status, yPos).
Emit a full vertical section set (24 sections) and add required chunk metadata scaffolding (structures, ticks, post-processing).
Add computed heightmaps to the written chunk NBT.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Arnis-generated worlds appeared empty on servers (PaperMC, etc.) because chunks used the deprecated pre-1.18 Level wrapper format and were missing critical metadata. Servers don't auto-repair chunks like the singleplayer client does, so they treated the chunks as corrupt/ungenerated. Changes: - Remove Level compound wrapper, write chunk data at root level - Add DataVersion (3955/MC 1.21.1), Status ("minecraft:full"), yPos (-4) - Emit all 24 sections (Y=-4 to Y=19) with biomes, not just sections with blocks - Compute MOTION_BLOCKING/WORLD_SURFACE/OCEAN_FLOOR heightmaps from block data - Add required empty structures, PostProcessing, block_ticks, fluid_ticks - Set isLightOn=1, InhabitedTime=0, LastUpdate=0 Fixes #844 Co-Authored-By: Claude Opus 4.6 <[email protected]>

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 3 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

djredflame · 2026-03-28T21:47:38Z

PR #855 validation report

I tested PR #855 (Fix server compatibility: use modern post-1.18 chunk format) on my local stack and it appears to fix the original Java/Paper world-compatibility problem that I was seeing.

Test environment

Arnis: local build including PR Fix server compatibility: use modern post-1.18 chunk format #855
Base upstream version before applying the PR: v2.5.1
Minecraft server: Paper 1.21.11
Java runtime image used by server tooling: itzg/minecraft-server:java25
World flow tested:
- generate Arnis world
- start Paper server
- verify whether prime-world / repair-world are still needed
- inspect resulting chunk NBT
- validate with my local quick-check scripts

Important caveat:
This means the PR works on my setup. It does not prove that it works in every environment, every Paper build, or every Minecraft version. But on my stack, the difference is very clear and reproducible.

What was broken before

Older Arnis-generated Java worlds on my side produced chunk NBT that looked like this:

root keys only:
- isLightOn
- sections
- xPos
- zPos

Typical missing fields in raw old worlds:

DataVersion
Status
Heightmaps
structures
yPos
section biomes

Example old raw worlds checked:

Arnis World 11
Arnis World 12
Arnis World 14

Validation results for old raw worlds:

Arnis World 11: generated quick-check OK, primed-format quick-check FAIL
Arnis World 12: generated quick-check FAIL, primed-format quick-check FAIL
Arnis World 14: generated quick-check OK, primed-format quick-check FAIL

Typical primed-format failure on old worlds:

chunk-root-still-minimal
missing-data-version
missing-heightmaps
missing-status
missing-biomes

So the old issue was real on my side: the raw Java output was too minimal for direct modern Paper use.

What changed with PR #855

After rebuilding Arnis with PR #855, newly generated worlds looked fundamentally different.

Representative new chunk structure now contains fields like:

DataVersion
Status
Heightmaps
structures
PostProcessing
block_ticks
fluid_ticks
yPos
section biomes

This is exactly the category of metadata that was missing before.

Representative example from a new generated world:

Status = minecraft:full
DataVersion = 3955 or later after Paper touches it
Heightmaps present
structures present
biomes present across the emitted sections
full vertical section range emitted instead of the old minimal section set

New-world test results

I generated and tested these new worlds with the patched build:

paris_test
test_pr855
hawaii_small
tokyo_small
london_small
london_small2

Primed-format quick-check results:

paris_test: OK
test_pr855: OK
hawaii_small: OK
tokyo_small: OK
london_small: FAIL
london_small2: OK

So on my side, 5 out of 6 newly generated test worlds passed directly.

Important nuance about the one failure

london_small still failed, but for a different reason:

failure: too-few-non-air-sections

That world still had the new modern chunk format:

DataVersion present
Status present
Heightmaps present
structures present
biomes present

So this does not look like the old server-compatibility issue anymore.
It looks more like a very small / thin BBOX edge case, not the original broken Java chunk format problem.

Runtime behavior on Paper

This was the most important practical result for me:

With the new worlds generated from the patched Arnis build, I no longer saw the old behavior where Paper had to rely on expensive fallback preparation for every new map.

In multiple cases:

no prime-world
no repair-world
server started directly
world loaded correctly in Paper

That is the behavior I originally wanted.

Side-by-side summary

Old raw Arnis output

minimal chunk root
missing DataVersion
missing Status
missing Heightmaps
missing structures
missing section biomes
not reliably direct-loadable on my Paper setup

New output with PR #855

modern post-1.18 chunk structure
DataVersion present
Status present
Heightmaps present
structures present
section biomes present
directly usable on my Paper 1.21.11 setup in most tested cases

Conclusion

On my local stack, PR #855 appears to fix the original Paper/Java compatibility issue.

More specifically:

it fixes the old minimal Java chunk NBT problem
newly generated worlds are now generally direct-loadable in Paper
the old prime/repair fallback is no longer needed for most of my test worlds
the remaining failure I found looks like a small-BBOX/content-density edge case, not the original chunk-format issue

So my conclusion is:

PR #855 solves the main issue for me on Paper 1.21.11, but I would still describe it as “confirmed on my environment”, not “universally confirmed everywhere”.

djredflame · 2026-03-30T10:46:32Z

PR #855: Chunk Validation Re-Review (Large Maps)
I re-tested PR #855 with larger maps, generated without subsurface-fill (so missing-subsurface-fill is expected and not problematic).

The only validation issues reported are:

missing-heightmaps
too-few-non-air-sections
Both are not blocking—worlds load and run fine in Paper.
Just noting these for completeness; everything else works as expected until now, will check it more out

Avoids recomputing has_air, bits_per_block, values_per_long, and mask for the same section on every (x,z) column iteration. Precomputes all section metadata once and sorts by Y descending for the top-down scan. Co-Authored-By: Claude Opus 4.6 <[email protected]>

louis-e · 2026-03-31T18:29:09Z

retrigger-benchmark

louis-e · 2026-03-31T18:32:42Z

retrigger-benchmark

louis-e · 2026-03-31T18:41:29Z

retrigger-benchmark

louis-e · 2026-03-31T18:48:00Z

retrigger-benchmark

louis-e · 2026-03-31T18:51:45Z

retrigger-benchmark

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 4 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

- Set isLightOn=0 (no light arrays emitted, server computes lighting) - Derive PostProcessing length from sections.len() - Rename level_data -> chunk_nbt for clarity - Fix doc comment to list all four heightmap types Co-Authored-By: Claude Opus 4.6 <[email protected]>

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot AI review requested due to automatic review settings March 28, 2026 11:46

Copilot started reviewing on behalf of louis-e March 28, 2026 11:46 View session

Copilot AI reviewed Mar 28, 2026

View reviewed changes

Comment thread src/world_editor/java.rs

Comment thread src/world_editor/java.rs Outdated

Comment thread src/world_editor/java.rs Outdated

Comment thread src/world_editor/java.rs

Comment thread src/world_editor/java.rs

louis-e force-pushed the fix/server-compatible-chunk-format branch from 5c9bad1 to 4c32a1f Compare March 28, 2026 12:04

louis-e requested a review from Copilot March 28, 2026 12:56

Copilot started reviewing on behalf of louis-e March 28, 2026 12:56 View session

Copilot AI reviewed Mar 28, 2026

View reviewed changes

Comment thread src/world_editor/java.rs

Comment thread src/world_editor/java.rs

Comment thread src/world_editor/java.rs Outdated

GMart mentioned this pull request Mar 31, 2026

[BUG] Terrain not showing up in external tools/distant horizons #858

Closed

fix cargo fmt

c32b420

louis-e requested a review from Copilot March 31, 2026 19:12

Copilot started reviewing on behalf of louis-e March 31, 2026 19:13 View session

Copilot AI reviewed Mar 31, 2026

View reviewed changes

Comment thread src/world_editor/java.rs

Comment thread src/world_editor/java.rs Outdated

Comment thread src/world_editor/java.rs Outdated

Comment thread src/world_editor/java.rs Outdated

louis-e requested a review from Copilot April 1, 2026 14:30

Copilot started reviewing on behalf of louis-e April 1, 2026 14:30 View session

Copilot AI reviewed Apr 1, 2026

View reviewed changes

louis-e merged commit 684cc6f into main Apr 1, 2026
5 of 6 checks passed

louis-e deleted the fix/server-compatible-chunk-format branch April 1, 2026 14:35

Conversation

louis-e commented Mar 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Mar 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

djredflame commented Mar 28, 2026

PR #855 validation report

Test environment

What was broken before

What changed with PR #855

New-world test results

Important nuance about the one failure

Runtime behavior on Paper

Side-by-side summary

Old raw Arnis output

New output with PR #855

Conclusion

Uh oh!

djredflame commented Mar 30, 2026

Uh oh!

louis-e commented Mar 31, 2026

Uh oh!

louis-e commented Mar 31, 2026

Uh oh!

louis-e commented Mar 31, 2026

Uh oh!

louis-e commented Mar 31, 2026

Uh oh!

louis-e commented Mar 31, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

louis-e commented Mar 28, 2026 •

edited

Loading

github-actions Bot commented Mar 28, 2026 •

edited

Loading