Cycle between `Scope` and `ComponentState`

[Diggsey]

Problem There appears to be a cycle between the Scope and ComponentState types, since each contains an Rc<_> to the other.

I couldn't see anywhere that explicitly breaks the loop, so I'm guessing this causes a memory leak.

[Diggsey]

Having worked with the codebase a bit more, I believe this is incorrect: the cycle is broken when a component is unmounted.

That said, it would probably be a good idea to have some tests for leaks (I have a crate mockalloc which may help with this) and/or to document the design itself, which AFAICT is:

• Scopes are created for each layer internally.
• Each scope contains a strong reference to its parent.
• Each scope contains a strong reference to the component state, which is explicitly broken when the component is unmounted.
• Component states contain a strong reference to their scope.

[lukechu10]

Maybe running some tests in Miri could help?

[mc1098]

That said, it would probably be a good idea to have some tests for leaks (I have a crate mockalloc which may help with this)

Maybe running some tests in Miri could help?

Both sound like great suggestions to me but I think they are owed their own respective issues :)

I've removed the bug label in favour of documentation because I think we should add some comments about this design.