is_instantiable check is broken by nested types

[arielb1]

STR

struct Foo<T>(T, Box<Bar<Bar<Foo<T>>>);
struct Bar<T>(T);

fn main() {
    println!("Hello, world!");
}

Results

Foo is uninhabitable, but the uninhabitability error doesn't trigger. Removing one of the Bar-s does make it trigger. This is rather suprising.

The check was added in the olden days when typeck was much less robust, and can be bypassed very easily by associated types anyway - we may want to remove it.

cc @nikomatsakis

[arielb1]

is_instantiable is also broken when dealing with empty types - it treats them as instantiable, meaning that struct Foo(Box<Foo>, Void) is allowed.

[eefriedman]

is_instantiable errors are already disabled when the static_recursion feature gate is turned on. I agree that it isn't something we really need to check.

[nikomatsakis]

I'm trying to remember what motivated us to add that check in the first place, but I cannot.

[arielb1]

@nikomatsakis

The old type-checker was far less robust then. Can it be removed?

[nikomatsakis]

@arielb1 seems ok to me, but cc @rust-lang/lang

[pnkfelix]

(I'm okay with removing the check, without going through the RFC process.)

[nikomatsakis]

@arielb1 after discussing in lang subteam meeting, general conclusion was "seems fine to experiment with removing the is_instantiable check and see what happens". I agree it was dubious to start and is more dubious now with associated types etc.