Adam Reichold

So for extensions modules which are shared libraries `ETXTBSY` is irrelevant. Sorry for the noise...

> The direct recursion on the nodes seems simpler as it is more flexible. So something like ```rust fn bottom_up_fold_reduce(tree: &RTree, mut init: I, mut fold: F, mut reduce: R)...

I think what is also nice about this way of doing things is that it has a straight-forward parallel version using Rayon which could come in handy for large data...

> The init parameter should probably take the bounding-box AABB as ref, as it is available anyway, and might help in initializing some structures that help with the reduce/fold. :+1:...

> If so, we might be allowed to consider other libraries to handle numeric types and abstract the details away from users. In my use case, performance is quite important...

> We could provide a bulk_load_unique method which uses e.g. a HashSet to dedupe the provided vec (elements have to be Copy, and it's an O(n) op) and we wouldn't...

> One open question for me: Should we implement insert_unique or rather something like the entry API for HashMaps? I think an entry-style API is definitely preferable but would possibly...

Could you produce a minimal but complete example so we can build and hopefully fix this locally?

Thank you for the worked example. So I think the request here is to make the `PointExt` trait part of the public API of `rstar` which it currently is not?...

> And what do you think about implementing AABB::from_center() by default? If that would suffice to fulfill your use case, then IMHO it would be preferable to exposing all of...