[rustdoc] Check doc(cfg()) even of private/hidden items

[GuillaumeGomez]

Fixes regression found out by @fmease here.

In short: the pass which checks the doc(cfg()) attributes needed to be moved before the private/hidden stripping items passes.

[fmease]

How does this relate to your PR #147905 I just approved? It does the same thing but the tests differ.

[fmease]

Do you want to add these tests to the earlier PR #147905, reapprove it and close this one?

[GuillaumeGomez]

I was able to completely forgot I already had a PR open to fix this bug. Oh well, at least this one is more complete... Closed the other one.

[fmease]

Doubly thanks! @bors r+ rollup

[bors]

📌 Commit e1e851d has been approved by fmease

It is now in the queue for this repository.

[Urgau]

It's weird that the propagate cfg doc pass emits those errors, it would make more sense to me if the check doc cfg pass were to emit those errors.

We probably need to modify the code around cfg_matches to emit those errors.

rust/src/librustdoc/passes/check_doc_cfg.rs

Line 54 in 4d94478

let _ = rustc_attr_parsing::cfg_matches(

[fmease]

@Urgau Yes, this is only a temporary fix. The proper fix should happen once we port over all #[doc(…)] to the new attribute parsing API (cc #147097). Ideally, we would fully validate the #[doc(cfg)] in rustc itself (so it's validated when you run rustc, not just when you run rustdoc) as I've mentioned here #138907 (comment).