Fix weird rustdoc output when single and glob reexport conflict on a name

[GuillaumeGomez]

Fixes #143107.

The problem was that the second reexport would overwrite the first, leading to having unexpected results. To fix it, I now group items by their original DefId and their name and keep tracks of all imports for this item (should very rarely be more than one though, and even less often more than 2).

cc @lolbinarycat

[rustbot]

r? @notriddle

rustbot has assigned @notriddle.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[GuillaumeGomez]

Ah bad luck. A function I modified was used in a more up-to-date version. ^^'

Anyway, rebased and fixed the CI failure.

[lolbinarycat]

some things could use clarification. i don't entirely understand if things are getting merged or overwritten. the unit test comment is confusing: all glob attrs are ignored, or just those on shadowed items? this should be clarified.

the test could also include coverage for the case where an item is renamed to itself. also what happens if an item is renamed then renamed back to the original name in a chain?

[GuillaumeGomez]

some things could use clarification. i don't entirely understand if things are getting merged or overwritten. the unit test comment is confusing: all glob attrs are ignored, or just those on shadowed items? this should be clarified.

Attributes on glob reexports are always ignored. The comment I added was misleading so I fixed it.

the test could also include coverage for the case where an item is renamed to itself. also what happens if an item is renamed then renamed back to the original name in a chain?

What matters is the duo name+DefId. So if the name is different, then rustdoc considers it's a different item. And if the item as renamed as itself, then nothing happens either. But I'm gonna add a test for this second case just in case.

[GuillaumeGomez]

Applied suggestions, added test and clarified comment.

[lolbinarycat]

What matters is the duo name+DefId.

non-renamed items have name field of None, and there's special logic to make this work with things renamed to themself, so I want there to be a test that makes sure any refactoring of this logic doesn't mess up this edge case.

[lolbinarycat]

Looks like you rebased and responded to review in the the same force push, which was a bit confusing when I clicked on the "Compare" button and saw a bunch of unrelated changes.

Anyways, it looks mostly correct, although if nothing is getting merged in the unit test, maybe it shouldn't have merge in the name? Or maybe it should just also exercise the case where attributes are actually merge.. is there a case where attributes are actually merged?

Also, I think it would be nice to also test what shows on the module docs, since apparently that can get out of sync with the actual item's docs, as shown in the original issue.

[GuillaumeGomez]

Anyways, it looks mostly correct, although if nothing is getting merged in the unit test, maybe it shouldn't have merge in the name? Or maybe it should just also exercise the case where attributes are actually merge.. is there a case where attributes are actually merged?

The two reexports are getting merged. They are merged in case of reexport chain:

mod foo {
    mod bar {
        /// C
        pub struct A;
    }
    /// B
    pub use self::bar::A;
}

/// A
pub use self::foo::A;

The A struct doc will have A, B and C.

Also, I think it would be nice to also test what shows on the module docs, since apparently that can get out of sync with the actual item's docs, as shown in the original issue.

Good idea.

[GuillaumeGomez]

Extended test as suggested.

[lolbinarycat]

Maybe I should clarify: I think the actual contents of the summary/description lines on the module docs need to be tested, since that's what got out of sync in the first hand, not the amount of items.

We don't want to somehow get an inversion of this bug, where the item docs are fine, but the module docs are getting messed up by the glob.

[GuillaumeGomez]

Ah I see. Added a check for short docs on module page.

[GuillaumeGomez]

@bors r=lolbinary rollup

[bors]

📌 Commit 086b13d has been approved by lolbinary

It is now in the queue for this repository.