Hidden methods still appear in trait implementation docs

[chris-morgan]

Take, for example, TotalEq:

pub trait TotalEq: Eq {
    // FIXME #13101: this method is used solely by #[deriving] to
    // assert that every component of a type implements #[deriving]
    // itself, the current deriving infrastructure means doing this
    // assertion without using a method on this trait is nearly
    // impossible.
    //
    // This should never be implemented by hand.
    #[doc(hidden)]
    #[inline(always)]
    fn assert_receiver_is_total_eq(&self) {}
}

As is proper, the TotalEq docs make no mention of this hidden method assert_receiver_is_total_eq.

But now refer to an example like the TotalEq impl for Option! Lo! assert_receiver_is_total_eq is shown there.

It shouldn't be.

[mahkoh]

This is broken again.

[mitsuhiko]

Still happens, example down here: http://mitsuhiko.github.io/redis-rs/redis/enum.ErrorKind.html

[SimonSapin]

It looks like this is happening again in #51147 (comment).

[kzys]

@SimonSapin Do you remember what was the issue on #51147? Seems SliceIndex's Rustdoc no longer shows any hidden methods/traits.

https://doc.rust-lang.org/nightly/std/slice/trait.SliceIndex.html

[SimonSapin]

In #51147 we wanted to hide SliceIndex’s methods, but couldn’t because the impl docs had broken links (and the linkchecker wouldn’t let us land that PR).

[kzys]

I see. Hiding get(), get_mut(), get_unchecked(), ... and all of them?

[SimonSapin]

Yes, these methods are permanently-unstable implementation details of methods like [T]::get. The trait itself is only stable so that code that calls [T]::get can itself be generic in the same way. See discussion at #35729 (comment).

[kzys]

Probably these need to strip methods that are inherited from hidden methods.

https://github.com/rust-lang/rust/blob/master/src/librustdoc/passes/strip_hidden.rs
https://github.com/rust-lang/rust/blob/master/src/librustdoc/passes/mod.rs#L191

[jrvidal]

@GuillaumeGomez I was looking into this as a way of better understanding rustdoc internals. Do you plan to tackle it?

[GuillaumeGomez]

Go ahead! I have a lot of other things to do. :)

[jrvidal]

@GuillaumeGomez suggests that making #[doc(hidden)] automatically hide the corresponding items in trait implementations is undesirable, since it is not reversible and the implementer could easily hide the items manually.

If that's the case, should this issue be closed?

[GuillaumeGomez]

From my point of view, it should. Let's close it now and if someone wants to debate, I'll let the comments open.