Concurrent Chunk Management Engine - Forge

C2ME-Forge 0.2.0-forge.2 (Minecraft 1.18.2) - build 2026-08-30

Performance
• Client pre-generation, stock settings, C2ME removed vs installed (radius 500, 4,225 chunks): Enigmatica 8 4:10 -> 1:27, Better MC BMC2 3:41 -> 1:19, DawnCraft 5:29 -> 2:07, Vault Hunters 3 3:45 -> 1:32
• Vanilla world, radius 1000: 30.61 -> 76.08 cps with the built-in Starlight engine (+149%)
• More heap still pays - MineColonies at radius 1000 went 3:36 on 8 GB to 3:09 on 12 GB, same seed, no code change

Fixed
• Worldgen could hang outright. A lost wakeup in chunk IO left the storage thread asleep on a queue that had work in it, so chunk reads never completed and every worker sat idle. Enigmatica 8 went from hanging at ~5% with 2,242 read timeouts to finishing in 1m52s with zero - this is also the likely cause of the reported "regions revert to their original state on reload"
• Servers on Forge 40.1.0 could not start (a mixin config failed to initialise during bootstrap)
• Citadel packs generated far slower than they should - Citadel's 3x3 biome scan now runs only when a mod actually registered a feature, removing ~220 million redundant reads over a large pregen. Behaviour is identical
• DeltaFeature's per-block linear scan (~8% of worldgen worker samples) is now a constant-time lookup
• The RAM advisor could report "heap adequate" while the feed was throttled to 71/512, and stayed silent entirely on pre-generations shorter than 90 seconds - exactly the packs people under-allocate

Compatibility
• Thread-unsafe worldgen mods are now detected at load and contained automatically: only the one stage they are unsafe in gets serialized, for that mod only, and everything else keeps generating in parallel. Covers Lost Cities, Blueprint and Large Ore Deposits (features), Skyblock Builder (structure starts) and Ore Tweaker (biomes). Five modpacks that previously could not finish a pre-generation now complete cleanly
• Cross-chunk worldgen writes are named once and counted, instead of one error line per occurrence (Better MC produced over a thousand in a single pregen). Worldgen output is unchanged

Configuration
• Every option in config/c2me.toml is now explained in plain language - what it does, the trade-off, and when you would touch it - with a header saying you should not need to change anything, and which settings can lose data. Existing config files pick up the new descriptions automatically
• New (experimental, off by default): threadedWorldGen.perModFeatureLanes keeps feature generation parallel and serializes only the unsafe mod's features, roughly 2x on decoration-heavy packs

Known issues
• Structure-dense packs are still slower for an unrelated reason: large cross-chunk structures serialize through neighbour dependencies
• Lighting is the remaining ceiling on 1.18.2 - with enough RAM, cores sit about half idle while the feed is at maximum

Drop-in update - recommended for everyone on forge.1. No config changes needed.
Support: https://discord.gg/s7BYs6Njj5

--------------------------------------------------------------------------------

C2ME-Forge 0.2.0-forge.1 (Minecraft 1.18.2) - build 2026-07-16

Initial release for Minecraft 1.18.2 (Forge 40.x).

Engine
• Full port of C2ME to classic Forge 1.18.2: multi-threaded world generation, chunk I/O and lighting, no-tick view distance, worldgen thread-safety fixes
• Uses C2ME's newer scheduling engine, backported from the 1.20.1 line - newer than what upstream ever shipped for 1.18.2
• Auto-tuned worldgen feed: 25.16 -> 38.96 cps with no flags and no config

Lighting
• Ships Starlight's rewritten light engine built in as a C2ME module ((c) Spottedleaf / PaperMC, LGPL-3.0). Vanilla 1.18.2's single-threaded lighting was the measured pre-generation ceiling; with the embedded engine throughput more than doubles
• Auto-disables itself if standalone Starlight is installed, so packs cannot double-apply it