rmanoka

I added the above test case here: https://github.com/rmanoka/geo/tree/fix/issue-885 . However, something complicated is happening with the code-gen. The test case panics or not depending on if certain trace macros are...

It looks like the issue may be with incorrect usage of `UnsafeCell`; for safety, I've moved to using `RefCell` instead. The issue with the first test-case was indeed a fp...

Just for clarity: we initially added this algo. to support c.hull for non-float types. Since then, we figured how to tweak qhull to do the same. Currently, the graham scan...

I'm also for merging this as is; it helps a reasonable use-case mentioned above. @nickguletskii could you provide us more details on the compact representation that required custom serialization?

Stepping back, it looks like the std data-structs like btreemap are indeed being [serialized](https://github.com/serde-rs/serde/blob/8a4dfa72319667e4f2eab303af96dd6daf0a32d7/serde/src/ser/impls.rs#L223) [as a sequence](https://github.com/serde-rs/serde/blob/master/serde/src/de/impls.rs#L961), and not the tree itself. Ideally we should switch to the packed serialization...

I think this is a standard way to add functionality to a trait: eg. [StreamExt](https://docs.rs/futures/latest/futures/prelude/stream/trait.StreamExt.html), FutureExt. We also have a `where` clause so it's not implemented for all types, only...

Why is it completing this function for non Point types though? That seems to be a bug iiuc.

Yes, we currently have a monotone decomposition algorithm; if that could be used black-box, we can build on top of it.

release 0.7.1

Thanks for the PR @b4l ; looks good except the non-robust impl for `Triangle`; let's use the slightly slower robust impl. on that too. +1 for removing the `Vec` allocations...