Game Crash on World Creation

[DaleHat]

Context

modpack works without Sable but when i enable sable it crashes when trying to create a world

Crash Report

https://mclo.gs/PRRiHSL

[DaleHat]

According to Claude - "the actual defect lives in Sable's VoxelNeighborhoodState cache — it isn't safe against being re-entered mid-lookup, and Conquest's leaf mixin is just what happens to trigger it (vanilla fences/panes/walls already do neighbor-aware collision too, so in principle this could lurk elsewhere, just less likely to line up)"

[DaleHat]

https://mclo.gs/tnlfbIn
here is a updated log, with a handful of mods to confirm the conflict was between Conquest & Sable

---> ChatGPT thinks the issue is on Conquest

"The crash occurs inside a Conquest mixin:

handler$bgp000$conquest$leaves1_getCollisionShape

That code eventually does

world.getBlockState(...).getBlock();

without checking whether

world.getBlockState(...)

returned null.

That is the actual NullPointerException."

But Claude thinks the issue is on Sable's Side

The chain of events

A new chunk generates and LevelChunk.postProcessGeneration applies its pending block placements — this is where trees/leaves get set for the first time.
Placing a leaf block fires Sable's mixin hook on setBlockState → SableCommonEvents.handleBlockChange → RapierPhysicsPipeline.handleBlockChange.
That asks Sable's VoxelNeighborhoodState.isSolid() whether a neighboring voxel is solid, which does Int2BooleanOpenHashMap.computeIfAbsent(...) — a cache lookup.
To answer that, the lookup calls getCollisionShape() on the neighbor — and Conquest Reforged's AbstractBlockStateMixin (Reforged.mixins.json, leaves_getCollisionShape/leaves1_getCollisionShape) intercepts it, because Conquest gives its leaves custom carpet-style collision that depends on another neighboring block.
To check that other neighbor, Conquest's handler calls back through Sable's own LevelAccelerator.getBlockState() → .getChunk(). If that neighbor sits in a chunk that isn't loaded yet, this blocks and pumps the server's task queue — which can drag in another chunk's postProcessGeneration, re-entering the exact same call stack, into the same Int2BooleanOpenHashMap instance, before the first call has finished.

[Taffy278]

我也遇到了同样的问题