For saddles, carpets, horse armor, and chests for horse-likes
a BlockDispenseEvent handler that always mutated the item without
changing the type would result in a SO error because when it went
to find the replacement dispense behavior (since the item "changed")
it didn't properly handle if the replacement was the same instance
of dispense behavior.
Additionally equippable mob heads, wither skulls, and carved pumpkins
are subject to the same possible error.
Furthermore since 1.21.2, the DISPENSER_REGISTRY map doesn't have a default
return value anymore and some dispense behaviors like equippable and
regular items will not have a defined behavior in that map and might throw
a NPE in that case.
Removes a layer of indirection for EnumDirection.getAdjacent(X|Y|Z)(), which is in the
critical section for much of the server, including the lighting engine.
RegistryAccess is independant from CraftServer and
doesn't require one to be created allowing the
org.bukkit.Registry class to be loaded earlier.
== AT ==
public net.minecraft.server.RegistryLayer STATIC_ACCESS
Normally the JVM can inline virtual getters by having two sets of code, one is the 'optimized' code and the other is the 'deoptimized' code.
If a single type is used 99% of the time, then its worth it to inline, and to revert to 'deoptimized' the 1% of the time we encounter other types.
But if two types are encountered commonly, then the JVM can't inline them both, and the call overhead remains.
This scenario also occurs with BlockPos and MutableBlockPos.
The variables in BlockPos are final, so MutableBlockPos can't modify them.
MutableBlockPos fixes this by adding custom mutable variables, and overriding the getters to access them.
This approach with utility methods that operate on MutableBlockPos and BlockPos.
Specific examples are BlockPosition.up(), and World.isValidLocation().
It makes these simple methods much slower than they need to be.
This should result in an across the board speedup in anything that accesses blocks or does logic with positions.
This is based upon conclusions drawn from inspecting the assenmbly generated bythe JIT compiler on my microbenchmarks.
They had 'callq' (invoke) instead of 'mov' (get from memory) instructions.
