Minecom
Vanilla Minecraft rebuilt from scratch on Minestom. No Mojang server code, no copy-pasted client jar. Mechanics are re-derived from the official data files and verified block-for-block against a real vanilla dedicated server.
Superseded. Work moved to a ground-up Rust implementation and this repository is dormant and incomplete. Left up because the verification approach and the numbers below are worth reading.
What it is
Most vanilla-on-Minestom projects stop at terrain that looks approximately right. This one targets real parity: the same world generation, structure layouts, mob behaviour and redstone timing, driven by the actual Mojang data files (recipes, loot tables, tags, biome parameters, NBT structure templates) extracted from the official server jar.
Where a mechanic was too expensive or out of scope, that is stated in the source and tracked rather than silently faked.
How correctness was verified
- Region diff against real vanilla. A vanilla 26.2 dedicated server
generates a region for a fixed seed, Minecom generates the same region, and every block is diffed. 99.381792% bit-exact across a 1,296-chunk (36x36, 127.4M-block) overworld sample, and 99.419259% in the Nether. Measured deterministically, two identical runs with identical block counts, after a chunk-scheduling race in the measurement path itself was found and fixed. Every mismatch class in the remaining gap is identified and tracked.
--selftest, 250+ deterministic data-engine checks covering world
generation, structures, loot tables, recipes and biome logic, with no live server.
--playtest, 1,000+ headless gameplay checks that boot the real server
wiring, join a fake player and drive combat, mob AI, structures, minecarts, boats, redstone, potions, enchanting, the Nether, the End and raids through the same event pipeline a real client uses.
Nothing landed unless both suites passed, and anything touching world generation needed a fresh region diff. The worldgen number was a ratchet: a change that lowered it did not land.
Performance was measured on identical hardware and seeds under real bot load, side by side with vanilla and Paper. Minecom matched or beat both on tick latency under 15 bots, with spawn-area p99 2.4x better than vanilla. Full numbers in docs/BENCHMARKS.md.
To check the parity claim, scripts/worldgen_region_diff.py regenerates the comparison from scratch against a real vanilla server. See its docstring.
License
AGPL-3.0, see LICENSE. Free to use, run, modify and host. If you run a modified version as a network service you share your modifications with your users. Contributions would have required a CLA.
No Mojang code is in this repository and never has been. Decompiled vanilla source was used strictly as an offline behavioural reference and was uncommittable. All bundled game data (src/main/resources/vanilla/) is mechanically extracted from the official server jar by a re-runnable script.