In core::ptr, weird thing about impl<T: ?Sized>

[strega-nil]

So the implementation itself is for T: ?Sized, but the actual functions themselves are for T: Sized.

i.e.

#[lang = "const_ptr"]
impl<T: ?Sized> *const T {
    #[inline]
    pub fn is_null(self) -> bool where T: Sized {
        self == 0 as *const T
    }
}

This makes absolutely no sense. First: why would you impl<T: ?Sized> and then not allow you to call any functions on T: !Sized. Second, why can you only call is_null on T: Sized? Makes absolutely no sense.

[alexcrichton]

This is actually currently intentional! The T: ?Sized means that the impl block is maximally generic by default (we only have one). Each method then adds bounds as necessary, and in this case the is_null method currently requires that T is sized.

At the time this function was written, it was an ICE if T was not sized (hence *const T was a fat pointer). It is intended that one day that ICE and/or restriction will be lifted so this method will not have a T: Sized bound.

[strega-nil]

Ah okay cool. We should probably make a comment about that?

[alexcrichton]

I'd be down with that!