No sizing

[dhardy]

This came out of #287. Review please @burdges?

[burdges]

I don't see anything wrong with removing the Self: Sized bounds, but I've no idea why they ever existed either.

Why did you remove a + ?Sized? It might be + ?Sized belongs in more places, like the subtrait bounds, but not sure. Can we ask Alex or someone?

I find the impl<R: ..., T: DerefMut<Target = R>> for T ... interesting but ambitious. We want impl<'a,R: Rng...> Rng... for &'a mut R ... because we need code to be polymorphic over both the base R: Rng and the &'a mut R. I think your T: DerefMut<Target = R> trick makes sense for exactly the same reasons, thus avoiding all the "deref inheritance is an anti-pattern" arguments : rust-unofficial/patterns@c2cf5e6 Ask around this perhaps though?

[dhardy]

Some of those where Self: Sized bounds were there because originally they were required (in order to support generic methods on a trait which is otherwise object-safe); I'm not sure if the change here is due to a compiler change or due to the generic fns all being moved to an extension trait.

I didn't remove a + ?Sized anti-bound; I added it to an example.

This isn't abusing Deref for inheritance like the anti-pattern you listed; it's implementing a trait for anything Deref-able to a type implementing the trait — @nrc you could maybe even consider this a pattern; are there any catches I'm missing?

[dhardy]

Rebased on #288. I will wait a couple more days for comment then merge.

[pitdicker]

@dhardy Have you seen rust-lang/api-guidelines#158?

[dhardy]

Thanks for asking. The error message issue is certainly a good reason not to use the Deref-based impl.

Edit: no-sizing commit merged directly, other commit abandoned.