Idea: "Maybe Trait" Object and Bounds (an alternative form of specialization)

Ah, I see. It hadn't occured to me that whether a trait is implemented for a type or not could depend on a lifetime. Perhaps because I don't think I have ever written such a trait impl myself. Presumably:

impl<'a> Bar for &'a u32 {}
fn func<T: ~Bar>(t: T) -> bool {

would also work, as the trait is implemented for all possible lifetimes.

If so, then I agree with your assessment that this would be good enough for the vast majority of use cases. It would probably require a bit of effort to be put into the error messages to explain the situation, but that's nothing new for rustc :slight_smile:

I think it would need to be impl Bar for for<'a> T<'a> {} (which isn't possible spelled this way at least; the closest I could get to work is impl<T: for<'a> Lt<'a>> Bar for T {} where Lt<'a> is another trait) because Bar could have a method that works on 'a and that feels like an easy "not sound" determination to me at least.