I was thinking about arena-based tree representations in Rust and how one thing you lose is that you can't do statically checked split_mut-style partitioning (i.e. a mut ref to a node can be split into mut refs to its child nodes) since the reference granularity is now at the whole arena level. But upon further reflection, that's not quite true. If the index ranges for sibling subtrees are disjoint, you can split_at_mut the slice itself to get subarenas (a slice + index offset).

When you deref a handle index within a subarena, you now have to do both a lower and upper bound check (previously only the upper bound check was needed):

if index >= subarena.offset && index - subarena.offset < subarena.slice.len() {
Some(&subarena.slice[index - subarena.offset])
} else {
None
}

Note that this works just fine with child-to-parent and sibling links, but if they are outside your subarena you will fail the bounds check and get None.

@pervognsen I mean, by the usual identity this is just wrapping_sub(index, subarena.offset) < subarena.slice.len() anyway

@rygorous Sure, I had a deleted reply about that but since the point wasn't really about optimization I left it out. Main point is just that you need to know the subarena offset.