Relax Sized requirements for blanket impls

[fjarri]

Relaxes Sized bound on blanket impls for TryRngCore, TryCryptoRng, UnwrapErr, and UnwrapMut.

This came up when trying to pass a RngCore + ?Sized-bound argument to a method from a third-party library requiring TryRngCore + ?Sized.

I think this is not a breaking change, but not 100% sure.

[fjarri]

Oh wait, I feel dumb, there's already #1592. But I add the UnwrapErr and UnwrapMut part.

[baloo]

I don't believe this a breaking change either, and a 0.9.2 would be appreciated if possible :)
Thanks a lot!

[dhardy]

Other than UnwrapErr this looks fine... though possibly not very useful.

I release now there is something missing from #1589: reborrow support. Basically this method (see the RFC issue):

    /// Reborrow with a new lifetime
    ///
    /// Rust allows references like `&T` or `&mut T` to be "reborrowed" through
    /// coercion: essentially, the pointer is copied under a new, shorter, lifetime.
    /// Until rfcs#1403 lands, reborrows on user types require a method call.
    #[inline(always)]
    pub fn re<'b>(&'b mut self) -> UnwrapMut<'b, R>
    where
        'r: 'b,
    {
        UnwrapMut(self.0)
    }

We may also (or instead) want to publicly export UnwrapErr and UnwrapMut from rand.

@newpavlov thoughts?

[fjarri]

Applied RFCs

though possibly not very useful.

My use case is passing an external RNG through a dyn-safe trait API. Then I might have a dyn RngCore argument that I want to pass to a method that takes dyn TryRngCore.

[newpavlov]

This looks fine to me. Could you add tests which would cover your use case?

[fjarri]

Added a test for RngCore -> TryRngCore. Should I add the same for CryptoRng and UnwrapMut?

[newpavlov]

Yes, I think it's worth to have them to prevent potential future regressions.

[fjarri]

Added the remaining tests