check for misplaced headers in missing_safety/error_doc

[llogiq]

This fixes #5593.

changelog: none

[rust-highfive]

r? @phansch

(rust_highfive has picked a reviewer for you, use r? to override)

[ebroto]

Hey @llogiq, I took a look at this issue before but never finished, I thought I could share some ideas if it's OK!

In general, I think the problem comes from a difference in how rustdoc and clippy parse the doc comments. Clippy adapted strip_doc_comment_decoration but rustdoc also uses an ad-hoc algorith for unindenting the comments here:
https://github.com/rust-lang/rust/blob/45127211566c53bac386b66909a830649182ab7a/src/librustdoc/passes/unindent_comments.rs#L49-L105

It would be great if clippy could adapt that code to behave in the same way.

[ebroto]

The problematic example in the issue has an indentation of 4 spaces (here there are only 3). I think this is important because in markdown 4 spaces start a code block, see my other comment.

[ebroto]

We should check that we are not in_code, otherwise this passes the test:

/// This is not a safety section.
/// ```
///    # Safety
/// ```
pub unsafe fn ceci_n_est_pas_un_titre() {
    unimplemented!();
}

If we add the check (and the missing space in the other test to make it 4) the header is still not recognised.

[llogiq]

Yeah, not duplicating this would be nice. However, we don't need to unindent here, just check if there's an unindented header, so our code can do less.

[phansch]

@llogiq I might have misread your last comment - Is that something you still want to fix for this PR or something for a separate PR?

[flip1995]

ping from triage @llogiq. There's still the open question if you want to address this: #5659 (comment) in this PR?

[bors]

☔ The latest upstream changes (presumably #5912) made this pull request unmergeable. Please resolve the merge conflicts.

[flip1995]

ping from triage @llogiq. Closing this issue, since it is stale for quite some time now. Feel free to reopen anytime you want to continue working on this!

[brightly-salty]

I'm willing to revisit this in a new PR if that's okay.

[jonboh]

@rustbot label -S-inactive-closed