rustdoc: Inconsistant decision to include impl with docs but no visible items

[aDotInTheVoid]

Since #90905, rustdoc has impl blocks with docs but no items: But it's logic doesn't make (user facing) sense

pub struct Foo;

/// 1. Empty
impl Foo {}

/// 2. Private
impl Foo {
    fn private() {}
}

/// 3. Public
impl Foo {
    pub fn public() {}
}

/// 4. Public Hidden
impl Foo {
    #[doc(hidden)]
    fn public_hidden() {}
}

/// 5. Private Hidden
impl Foo {
    #[doc(hidden)]
    fn priviate_hidden() {}
}

Running this code across various flag's gets us:

Header		--document-private-items	--document-hidden-items	--document-private-items --document-hidden-items
Empty	✔️	✔️	✔️	✔️
Private	❌	✔️	✔️	✔️
Public	✔️	✔️	✔️	✔️
Public Hidden	✔️	✔️	✔️	✔️
Private Hidden	✔️	✔️	✔️	✔️

It doesn't make sense that adding #[doc(hidden)] to a private function makes it's block visible, or that running --document-hidden-items makes a block with only private functions visible. I'm not sure what the exact behavior should be here

[GuillaumeGomez]

In the code, the comment says:

Impl blocks can be skipped if they are: empty; not a trait impl; and have no
documentation.

There is one special case: if the impl block contains only private items.

The logic makes sense to me: impl blocks with only private items should very likely not be visible to readers. The question here I think should be more "should we do the same with impls containing only hidden items?". What do you think?

[lolbinarycat]

@GuillaumeGomez the "private hidden" case definitely seems like a bug.