Add Minecraft 26.1 support - #484
Open
u9g wants to merge 9 commits into
Open
Conversation
Member
|
CI fails. Please release the deps then reun CI here |
u9g
added a commit
to u9g/bot-harness
that referenced
this pull request
Sep 6, 2026
…ds (#4) Pins prismarine-viewer to the v1.33.0-host26.1 release of u9g/prismarine-viewer: the host-object series (PrismarineJS/prismarine-viewer#498-#503) merged with PrismarineJS/prismarine-viewer#484 (26.1 support), with the world bounds file fetched through the host. Recording no longer needs -v 1.21.4.
Three changes needed for 26.x:
- modelsBuilder: unwrap object texture refs. 26.x wraps some model
texture references in objects ({ sprite, force_translucent, ... })
instead of plain strings, which broke the #ref resolution loop.
- atlas: pack tiles at native resolution. 26.1 ships 32x32 block
textures (per-wood shelves); each atlas entry already carries its
own u/v/su/sv extents so consumers are resolution-agnostic.
Animated textures still contribute only their first frame. For
16x16-only versions the generated atlas is equivalent.
- version: add 26.1 to supportedVersions.
Regenerating public/ assets for 26.1 also needs a minecraft-assets
release containing data/26.1 (PrismarineJS/minecraft-assets#50) and
the extractor fix (PrismarineJS/minecraft-jar-extractor#68).
Sections below y=0 were never meshed: worldrenderer's dirty-section loops hardcoded the pre-1.18 0..256 range, and the worker's section-presence guard indexed chunk.sections by y/16 even though ChunkColumn indexes sections from minY (-64 since 1.18), so negative sections resolved to undefined and were skipped. Any camera below y=0 saw only the clear color. Derive minY/worldHeight from the version's chunk implementation, index worker sections by (y - minY) / 16, and gate models.js face culling on world.minY instead of 0.
1.18.0 is the first npm release shipping data/26.1. Exposing it through
mcAssets('26.1') additionally needs PrismarineJS/node-minecraft-assets#54,
so the effective minimum is the release that includes that fix.
1.18.0 ships data/26.1 but has no 26.1 entry in index.js, so
mcAssets('26.1') returns null and viewer/prerender.js throws
during npm install. The entry landed in 1.19.0.
worldrenderer.js reached for prismarine-chunk to read minY/worldHeight, which pulled prismarine-registry and all of minecraft-data into the browser entry bundle. indexConfig only externalizes five minecraft-data files, so webpack tried to inline ~252 MiB of JSON and died at the V8 heap limit before emitting index.js, taking `npm install` down with it via prepare. prerender.js already has prismarine-chunk as a build-time dependency, so it now emits public/worldBounds.json (653 bytes) and the renderer fetches it through the loadJSON it already uses for blocksStates. addColumn/removeColumn still post to the worker immediately and defer only the section-marking loop until the bounds land. The worker's mesh loop is gated on blockStates arriving from the same kind of fetch, so this costs no rendering latency. world.js keeps prismarine-chunk: it runs in the worker and needs Chunk.fromJson, so prismarine-chunk stays the single source of truth for world bounds on both sides.
supportedVersions already lists 26.1, so the suite generates a client 26.1 describe block that nothing was running: the matrix stopped at 1.21.4, which covered the other versions against regressions but left the version this branch adds untested.
The 26.1 server jar is compiled for class file version 69 (Java 25); the pinned JRE 21 only reads up to 65, so it cannot start the server at all. mineflayer runs its whole matrix, 1.8.8 upward, on Java 25.
worldBounds.json is keyed by supportedVersions, but since "Keep the server's version for block data, snap only the assets" (PrismarineJS#492) the version passed to WorldRenderer.setVersion is the server's exact version rather than the snapped one. A 26.1.2 server therefore missed the lookup and silently defaulted to minY 0 / worldHeight 256, undoing the negative-Y rendering this branch adds. Prefer an exact match and fall back to the snapped assets version, which shares the major and so has the same bounds.
.gitignore had no trailing newline, so appending the worldBounds.json entry concatenated it onto the last line as the single dead pattern "test/server_*public/worldBounds.json", ignoring neither path.
u9g
force-pushed
the
26-1-support
branch
2 times, most recently
from
September 12, 2026 15:21
e6a1941 to
dc777aa
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Three changes needed to render 26.x worlds:
modelsBuilder: unwrap object texture refs
26.x wraps some model texture references in objects instead of plain strings:
The
#-reference resolution loop inprepareModelcalls.charAt(0)on these, throwingroot.charAt is not a function. Unwrap to thespritestring. (force_translucentis a vanilla render-type hint — the viewer's single transparent material has no use for it.)atlas: pack tiles at native resolution
26.1 ships 32x32 block textures (per-wood shelves).
makeTextureAtlashard-coded 16x16 tiles and cropped every source to its top-left 16x16, which mangles these. Tiles now keep their native size; each atlas entry already carries its ownu/v/su/svextents, so consumers (modelsBuilder UV math,bu/bvsampling) are resolution-agnostic. Animated textures (taller-than-wide strips) still contribute only their first frame. Versions with only 16x16 textures produce an equivalent atlas.version: add 26.1 to supportedVersions
Verified end-to-end: generated 26.1 assets locally (via the
minecraft-assetsdata from this pipeline) and rendered a live 26.1 world with the standalone viewer — all 1168 blocks resolve, zero missing-texture fallbacks, shelf blocks render at native resolution. Re-verified against the published minecraft-assets 1.18.0 data (with the version-index fix below): atlas + blocksStates generation completes for all 1168 blocks.Depends on:
Unwrap object texture refs in extractModel for 26.x minecraft-jar-extractor#68 (extractor fix for the same object refs)Add Minecraft 26.1 assets minecraft-assets#50 (data/26.1)— shipped in minecraft-assets 1.17.0+data/26.1ships in the 1.18.0 package butmcAssets('26.1')still returns null without it)public/assets are generated at install time byviewer/prerender.js; the devDependency is now pinned tominecraft-assets@^1.18.0(first release shipping the 26.1 data), so once a release containing the version-index fix is out, this PR needs no further changes.