The current implementation of the BlockPos offset "helper" methods (BlockPos#up(int n), BlockPos#up(), etc.), can be improved by reducing the number of function calls. While the performance increase may seem insignificant, due to the sheer number of calls related to BlockPos checks any improvement is valuable.

Instead of complicating the offset through the use of EnumFacing and the likes, the methods could instead directly create a new BlockPos with the offset. A full example can be found here:
>https://github.com/cabaletta/baritone/blob/3a68009194ee46930d9609947c93863a82c7d180/src/main/java/baritone/bot/utils/pathing/BetterBlockPos.java#L29

Benchmarking our BetterBlockPos against MC's BlockPos shifting methods, we see a clear performance increase (see the two tables below, and generated from https://github.com/cabaletta/baritone/blob/3a68009194ee46930d9609947c93863a82c7d180/src/test/java/baritone/bot/utils/pathing/BetterBlockPosTest.java#L23). This is even including the Hotspot VM optimizations, which we can see kick in after the first two rows.

Also, to be clear, our implementation creates the exact same number of objects that the Vanilla version does, so any concerns about garbage collection performance shouldn't be considered around this specific fix.

Benchmark for up()

Benchmark for up(int n)