unexpected rustdoc result about auto-trait

[bvanjoi]

code:

pub struct Inner<T>(T);

trait MyTrait {
    type MyItem;
}

trait OtherTrait {}

impl<T> Unpin for Inner<T>
where
    T: MyTrait<MyItem = bool>,
{
}

pub struct Foo<K> {
    inner_field: Inner<K>,
}

because Unpin is not implement for Foo explicitly, I expected to see this happen after execute rustdoc :

Instead, this happened:

I am not sure the rendered if right or not, so I created the issue to seek some answer.

related #110740

[fmease]

This is the intended output (as far as I am aware).

because Unpin is not implement for Foo

I'm not sure what you mean by that. Unpin is implemented for Foo<K> for all types K where
the type of field inner_field Inner<K> implements Unpin (since no explicit impl exists).
Whether a struct impls an auto-trait doesn't depend on the generic parameters (as your 1st picture shows)
but on the field types. You can read more about that here.

So technically speaking, the synthetic impl of Unpin for Foo looks like

impl<K> Unpin for Foo<K>
where
    Inner<K>: Unpin

but since one can argue that the constraint Inner<K>: Unpin is not that helpful (“actionable”) to the reader, we simplify the constraint. In this case Inner<K>: Unpin is implied by the explicit impl impl<T> Unpin for Inner<T> as long as the constraint T: MyTrait<MyItem = bool> holds. Thus we propagate that constraint and end up with the synthetic impl as seen in your 2nd screenshot.

[fmease]

impl<K> Unpin for Foo<K> where K: MyTrait<MyItem = bool> is not “more correct” than
impl<K> Unpin for Foo<K> where Inner<K>: Unpin (what the compiler actually sees), it's
simply a different way to display the constraint.

impl<K> Unpin for Foo<K> where K: Unpin on the other hand is not correct in this case
because of the existence of the explicit impl impl<T> Unpin for Inner<T> where …. If you were
to remove it, then that synthetic impl would be correct since

1. for all types K, Foo<K>: Unpin is implied by
2. for all types K, Inner<K>: Unpin is implied by
3. for all types T, T: Unpin (where T corresponds to the type of field 0 of the tuple struct Inner)

[jyn514]

It looks like this isn't a bug, so I'm going to close the issue.