Remastered Mod Suite — Changelog 30.5.1

Minecraft: 1.21.1 (Fabric)

Mods in This Release

• BCLibRemastered 30.5.1
• BetterEndRemastered 30.5.1
• BetterNetherRemastered 30.5.1
• WorldWeaverRemastered 30.5.1

Performance and memory release. No gameplay, world generation or visual changes: the same seed gives the same End and Nether biomes at every position as 30.5.0.

Fixes by Mod

WorldWeaverRemastered

• End biome lookups no longer run one at a time. WoverEndBiomeSource.getNoiseBiome held one lock around the whole lookup, including BetterEnd's island test. Every thread that asked for an End biome (structure placement at world load, End chunk generation, /locate biome) waited for that lock. The lookup now reads an immutable snapshot of the biome maps and runs on all threads in parallel.
  • Measured with 30.5.0 in a large modpack: a datapack structure set with 219 ring positions in the End (711 531 biome lookups per world load) kept all worker threads queued for 29-33 s in every world load, and about 70 % of the End chunk generation worker time was lock wait.
  • The same 711 531 lookups on a test server with only the suite installed: 22-31 s with 30.5.0, about 2-4 s with 30.5.1, same results.
• Fixed a memory leak when several worlds are opened in one game run (singleplayer). WorldWeaver kept the registries and the world folder of the last world after its server had stopped, and the biome sources of the world presets of the next world bound themselves to those old registries. Every world therefore kept the registries of all worlds opened before it in memory (measured in a large modpack: about 47 MB more per world). The stored registries and world folder are now released when the server has stopped, and a world load starts with an empty state, exactly like the first load of a game run.
• A biome picker that is used before it was rebuilt is now rebuilt by one thread only (it used to rely on the biome source lock).
• BiomeDecider documents its threading contract: deciders are called from many worldgen threads at once and must keep working state per thread.
• The Nether biome source is unchanged (it still uses its lock).

BCLibRemastered

• Runtime block models can be shared between block states. A block that builds the same model for every state (new opt-in RuntimeBlockModelProvider.hasStateIndependentModel()) is asked once per model load instead of once per state. Default behaviour for every other block is unchanged.
• Fixed a memory leak: the data exchange connector kept the last server (with its levels) after it had stopped. It is released when the server has stopped.

BetterEndRemastered

• Flower pot models are built once per pot. Each of the 8 flower pots has 4 352 block states and the same multipart model was rebuilt for every one of them at every game start and every resource reload (F3+T). Measured with 30.5.0 in a large modpack: 18 s per start and about 190 MB of heap held by the duplicates. The pots look the same as before.

BetterNetherRemastered

• No changes. Version aligned with the suite.

Update Recommendation

• Keep all four Remastered mods on 30.5.1 together: they are built and tested as one suite. The flower pot change needs both BCLib and BetterEnd 30.5.1.
• Existing worlds need no migration: chunks generated with 30.5.0 and with 30.5.1 fit together without a seam (checked with two seeds: 60 921 End and 66 564 Nether biome samples per seed are identical between the two versions).