Why is the render tree implemented with off-heap memory?

[codeinred]

First off: what do you mean by this, and where is the memory coming from? Is it coming from mmap, or are you using memory you haven't allocated yet or reading past the end of an array? Or are you using unallocated stack memory which might get overwritten?

I'm trying to understand the design goals that are achieved or satisfied by this approach, and why it can't be satisfied by an alternative, safer approach

[DaMatrix]

before i begin i should clarify that as of month or so ago, there is no more render tree - the draw commands are now stored in a flat index which is frustum culled on the GPU. i should probably update the README...

the memory is allocated using sun.misc.Unsafe#allocateMemory, which likely delegates to malloc.

the reason i did it off-heap was to reduce indirection - iterating through a sparse tree is already just a pointer-chasing mess, and doing it in "normal" java which wouldn't allow me to flatten generic properties or arrays into the object structure just makes it that much worse. flattening the structure improved tree iterations times by a factor of ~3.

[codeinred]

That makes a lot of sense, thank you. It seems like a sensible use. I had imagined you were doing it in the C++ section of the code, where it'd make less sense and would also be weirder, but what you explained makes complete sense in java