Lint for unnecessary trait bounds

[huonw]

It's easy to accidentally leave an unnecessary (e.g.) Clone bound when refactoring a function, making the function less generic/useful than it otherwise could be.

This would warn for something like:

// warning: unnecessary trait bound. No method from `Clone` is used.
fn foo<T: Clone>(x: T) -> T { 
    x
}

It would presumably have to be very careful about super-traits i.e. using a function from a super-trait, but not one from the listed trait bound, would either not warn at all, or suggest that the constraints could be weakened to that trait (handling multiple supertraits would be even trickier, i.e. if a function from 2 different super-traits was called, is it sensible/useful to suggest weakening to a combination of those two traits?)

Presumably a far-future feature, if at all.

[thestinger]

Sometimes the bounds are there only as a restriction, like most uses of Send and Freeze. I guess this would only apply to non-kinds, but that will get blurrier when we allow inheritance from kinds.

[huonw]

I was going to mention that, but clearly forgot. Presumably it could have some smarts like that, or just require #[allow(unused_bounds)] on those functions (less nice).

[bblum]

Any function that does unsafe code inside could be assumed to need all declared trait bounds. That would cover the cases of ARC and pipes and so forth.

[alexcrichton]

I'm going to close this for now. This seems very similar to unsafe on a function that doesn't actually do anything unsafe. The trait bound may not be used, but it may be a bound for different reasons that the compiler cannot know about.