RFC: Limit what types can have destructors

[nikomatsakis]

We need some limits on destructors to prevent bugs like bugs like #3167, #3039, #4170. Also to address the various challenges of destructors and finalizers that have been uncovered in other languages.

My current preference is to say that destructors can only be created for sendable types. Or, put another way, destructors should only be able to access owned content. Perhaps this is too extreme, but it's a nice simple rule.

Anyway, I wanted to open a blanket bug for those other bugs, and perhaps to spark some discussion and thought.

[catamorphism]

I vote for trying the "extreme" rule first (destructors can only access owned memory), and loosening it if it turns out to be too restrictive in practice.

[pcwalton]

Note that I think this kills the icx_popper, and kills using destructors for Go-like defer as well.

[nikomatsakis]

Yes, I think the extended version we talked about was somehow limiting non-sendable types with dtors from being placed into @ boxes, or something like that, with the goal of ensuring that they exist only on the stack, but that sounds like it requires more type system machinery (another kind) to achieve.

[catamorphism]

Can this be closed given d4fee24?

[nikomatsakis]

Yes.