Nick Fitzgerald

Ah good catch! This test is from before we made the increment-to-decrement switch and should also be updated for that.

I don't think we need to do anything specific to detect/inspect the returned pointer. Instead, we just need to make sure we write a byte per page in the resulting...

It is a pretty fundamental assumption in this implementation that chunks of memory come from the global allocator, that we can request new chunks as needed, and that we deallocate...

I think that the standard library `Box` is special cased by the language here. You should be able to create a `&mut dyn Future` from a `&mut MyFuture` that was...

It would look like the first, but additionally the parameter would use the `'a` lifetime: `x: &'a mut T`

There would need to be a version of `Rc` that is specialized to `bumpalo` or (eventually) `Rc` would need to be parameterized by an allocator (and the allocator's lifetime). This...

> Are you suggesting that something like `&'bump RefCell` would work instead? Yep!

If you need to run `MyType`'s destructor, then you'd pass around `&'bump bumpalo::boxed::Box`

A new `"serde"` feature seems reasonable to me. When enabled, it would turn on serialize and deserialize impls for everything in `bumpalo::collections`. I think we would need to implement them...

In addition to the Windows issues you mentioned, this crate is always `no_std`, and therefore doesn't have access to either `OsStr` or `Path` at all. It is probably best to...