Missing short summaries for reexported items in rustdoc

[ollie27]

For example the current nightly docs show:

The summary lines for the reexported modules is missing. They're also missing from the search index and sidebar.

This is a regression from #77519. cc @jyn514 @GuillaumeGomez

[ollie27]

Testing locally, this is fixed by simply removing the newly added DocFragmentKind::Divider but I don't know what that was added for so I don't know if that's the correct fix.

[jyn514]

this is fixed by simply removing the newly added DocFragmentKind::Divider but I don't know what that was added

This was added to distinguish between the attributes on the original item and attributes on the re-export: 8fbfdc5. Previously, collapse_docs would collapse all attributes into one giant attribute:

// crate1
/// Docs in original
pub struct S;

// crate2
/// Docs on re-export
pub use crate1::S;

// how rustdoc sees it: as if _both_ docs were in crate1
/// Docs in original
/// Docs on re-export
pub struct S;

After my change, it distinguishes between the same two in collapse_docs, so they're now separate attributes the way #[doc = "x"] and /// x are not combined:

rust/src/librustdoc/passes/collapse_docs.rs

Line 39 in 8fbfdc5

if matches!(*curr_kind, DocFragmentKind::Include { .. }) || curr_kind != new_kind {

I think the issue is that doc_value is now returning the Divider even if there were no additional docs:

rust/src/librustdoc/clean/types.rs

Lines 653 to 657 in 8fbfdc5

	/// value found.
	pub fn doc_value(&self) -> Option<&str> {
	self.doc_strings.first().map(|s| s.doc.as_str())
	}

There are two possible fixes:

1. Don't add a divider if there were no new docs. I like this one because it's simple.
2. Change doc_value to ignore the first attribute if it's a Divider and look for the first attribute which isn't. This has the drawback there might be other places using attrs, although off the top of my head I don't know where.

I'll try to get a fix in with 1. today.

[jyn514]

And good catch, thanks!

[camelid]

Assigning P-high and removing I-prioritize as discussed in the prioritization working group.

[ollie27]

This was added to distinguish between the attributes on the original item and attributes on the re-export: 8fbfdc5.

Right, I figured that's what it was trying to do but the Divider is being placed at the beginning of the list of DocFragments:

rust/src/librustdoc/clean/types.rs

Line 614 in c71248b

doc_strings.borrow_mut().push(DocFragment::divider());

so it's not actually dividing anything. The result is that collapse_docs is still collapsing the docs as it did before so if a reexport has extra docs, all links are resolved relative to the local crate.

Rather than adding Divider couldn't collapse_docs check for differing parent_modules to decide whether to merge DocFragments? That seems like the simpler solution.

[jyn514]

Oops, you're right on both counts. I guess it didn't show up because of #77200.

couldn't collapse_docs check for differing parent_modules to decide whether to merge DocFragments

That sounds better, yeah, then we could get rid of Divider. Let me try that.