Docs prose say Sleep doesn't implement Unpin, but sidebar does

[jrose-signal]

Since v1.1.0, Sleep has been documented as follows:

This type does not implement the Unpin trait, which means that if you use it with select! or by calling poll, you have to pin it first. If you use it with .await, this does not apply.

But the autogenerated sidebar says it does implement Unpin, since v1.0.1. I did find #3267 and #3278 and even #3278 (comment):

We should have a test that asserts that Sleep is !Unpin (same for other types that contain pinned Sleeps).

And yet Sleep is Unpin. I tested locally incorrectly and thought Sleep was actually Unpin, but it's just the autogenerated docs that are confusing.

[sfackler]

The definition of the impl comes from pin project, and is conditional:

impl<'__pin> Unpin for Sleep
where
    __Origin<'__pin>: Unpin,

It ends up meaning that Sleep is not Unpin.

[taiki-e]

This is a rustdoc bug: rust-lang/rust#80481

We may want to mention in the documentation that rustdoc cannot display this correctly.

[Darksonn]

It's a bit annoying, but we could fix this by not using pin-project-lite for Sleep. Or maybe we could get a feature in pin-project-lite that supports this, where we can say "this type should never be Unpin" and have the macro output something that makes the docs say that its never Unpin?

[taiki-e]

I think it is ok to add pin-project's !Unpin equivalent to pin-project-lite and hide the Unpin implementation from the document.

As I said previously, I do not favor making our code vulnerable to human error to avoid rustdoc issue.

[Darksonn]

Yeah, I know. I also agree that it would be vastly preferable to make the macro support it directly.