Shall we extend the abstraction to box like pointers?

[cark]

While we're at it, it might make sense to create a PointerFamily super-trait, and base everything on it.

• The project scope increases
• Is it feasible
• What does it buy us
• What consequences for end user ergonomics.

[cark]

Started the work on this, in a new branch.

That's the kind of ergonomic issues I was fearing

Compared to master branch

But at least we see this is feasible.

[Eh2406]

I see your point about the ergonomics. It can be simplified a little bit better than the error message suggests to <P::Pointer<Self> as RefCounted<Self>>::WeakPointer, but it is not much better.

I also noticed that the StrongMark can be removed without affecting any tests. How was it intended to be used?

[cark]

I'm not sure it still does this in that context, but the intent was to help the type inference and rust-analyzer... There are still cases where it feels like inference should work but it doesn't. GATs are new, we'll see how it evolves on that front.

[cark]

Oh but you're talking about the StrongMark in RefCounted ! This one is a remnant of the master branch that can be safely be removed.

Edit: and it looks like the one in WeakPointer can go too !