Make `'static` part of `ArrayLength` trait bounds?

[cbeck88]

In some recent code, for example here, https://docs.rs/generic-array/0.14.3/src/generic_array/sequence.rs.html#305, we have + 'static bounds on types that represent ArrayLength typenums.

unsafe impl<'a, T, N, K> Split<T, K> for &'a GenericArray<T, N>
where
    N: ArrayLength<T>,
    K: ArrayLength<T> + 'static,
    N: Sub<K>,
    Diff<N, K>: ArrayLength<T>,
{
    type First = &'a GenericArray<T, K>;
    type Second = &'a GenericArray<T, Diff<N, K>>;

    fn split(self) -> (Self::First, Self::Second) {
        unsafe {
            let ptr_to_first: *const T = self.as_ptr();
            let head = &*(ptr_to_first as *const _);
            let tail = &*(ptr_to_first.add(K::USIZE) as *const _);
            (head, tail)
        }
    }
}

Sometimes in my own code that uses generic array, I have to put 'static on the number type bounds, and it's sometimes hard to understand why.

It seems to me that we could change ArrayLength to require 'static

pub unsafe trait ArrayLength<T>: Unsigned + 'static {
    /// Associated type representing the array type for the number
    type ArrayType;
}

Since this trait is only implemented for typenum types and those are all ZWT's anyway, the 'static bound is always true and this breaks nothing in practice. It only helps the compiler to see that some things that are generic parameters are 'static without the user having to spell it out.

Then you could remove + 'static in several places in generic array itself, e.g.

    K: ArrayLength<T> + 'static,

in code like impl ... Split because the + 'static would be implied.

What do you think?

[novacrazy]

This seems like a good idea for a version 1.0 change, or at least a 0.15, just in case.

As far as I know, there is no motivation to try implementing typenum::Unsigned on anything that holds a reference, it's all just ZSTs, so this should be a harmless change.