Doc alias checks: ensure only items appearing in search index can use it

[GuillaumeGomez]

Following the discussion in #73721, I added checks to ensure that only items appearing in the search are allowed to have doc alias.

r? @ollie27

[ollie27]

The search index also doesn't include associated types and consts in trait impls so we'll need to error on those too. I'm not actually sure why they're not included though.

[GuillaumeGomez]

I think it's in case of conflicts between two implementations. I'll add them as well, good catch!

[GuillaumeGomez]

@ollie27 Updated!

[GuillaumeGomez]

cc @rust-lang/rustdoc

[ollie27]

Ideally this check would be done inside rustc with the other checks. Unfortunately I think #73566 meant that the checks moved in #74148 no longer run in rustdoc. That should have bee caught by CI but the test was moved rather than copied.

Assuming it's possible to get the check_mod_attrs pass running in rustdoc again then we should do that and move the checks from this PR there.

[GuillaumeGomez]

Good point! I'll move those checks then.

[jyn514]

Assuming it's possible to get the check_mod_attrs pass running in rustdoc again then we should do that and move the checks from this PR there.

It should be possible to add it back as long as it doesn't need typeck. See https://github.com/rust-lang/rust/blob/master/src/librustdoc/core.rs#L460 for an example.

[jyn514]

So the discussion is in one place: @GuillaumeGomez tried this and found that checking for unused attributes is in the same pass as checking for unused code. So it currently requires typeck:

 Documenting std v0.0.0 (/home/imperio/rust/rust/src/libstd)
error[E0599]: no method named `into_handle` found for struct `sys::unix::thread::Thread` in the current scope
--> src/libstd/sys/windows/ext/thread.rs:19:27
|
19 | self.into_inner().into_handle().into_raw() as *mut _
| ^^^^^^^^^^^ method not found in `sys::unix::thread::Thread`
| ::: src/libstd/sys/unix/thread.rs:14:1
|
14 | pub struct Thread {
| ----------------- method `into_handle` not found for this
error: internal compiler error: src/librustc_passes/dead.rs:122:13: no type-dependent def for method

It will have to be split into two passes for this to work: one for unused attrs, and the other for unreachable expressions.

[ollie27]

check_mod_attrs is called here:

rust/src/librustc_interface/passes.rs

Line 847 in 2186722

tcx.ensure().check_mod_attrs(local_def_id);

so I tested adding:

for&module in tcx.hir().krate().modules.keys(){let local_def_id = tcx.hir().local_def_id(module);
tcx.ensure().check_mod_attrs(local_def_id);};

just after

rust/src/librustdoc/core.rs

Lines 460 to 462 in 2186722

	sess.time("missing_docs", || {
	rustc_lint::check_crate(tcx, rustc_lint::builtin::MissingDoc::new);
	});

and it seems to work just fine. I don't think that requires typeck but I could be wrong.

[GuillaumeGomez]

Updated! I added the equivalent tests both in test/rustdoc-ui and in test/ui to be sure that the pass is run in rustdoc. I also moved the checks in rustc directly.

[GuillaumeGomez]

ping @rust-lang/rustdoc

[Manishearth]

Do you want review from others? Seems like jyn and Ollie are reviewing this PR

[GuillaumeGomez]

You can provide one as well if you want. But it was more like a reminder for the actual reviewers. :)

[Manishearth]

Nah, I don't have many opinions on doc(alias) and I trust the others

[bors]

☔ The latest upstream changes (presumably #74932) made this pull request unmergeable. Please resolve the merge conflicts.

[GuillaumeGomez]

Rebased.

[GuillaumeGomez]

ping @ollie27

[GuillaumeGomez]

@ollie27 The impl const items are now allowed to have a #[doc(alias)]. :)

[GuillaumeGomez]

Forgot that we tested in both test/ui and test/rustdoc-ui...

[GuillaumeGomez]

ping @ollie27

[ollie27]

Sorry for the delay.

As I've said associated consts in trait impls also don't appear in the search index so should also produce errors.

[ollie27]

ImplTraitItem should be removed from here as well now it's been removed from doc-alias.rs.

[GuillaumeGomez]

That's why the search for it is returning nothing below. The goal was to check that nothing was returned.

[ollie27]

ImplTraitItem doesn't appear anywhere in the source file anymore so of course it's not going to produce any search results. This isn't a particularly useful test.

[GuillaumeGomez]

But that's the point of the test: it's supposed to not return anything, so we ensure that it doesn't.

[GuillaumeGomez]

I guess we're good to go then! \o/

@bors: r=ollie27

[bors]

📌 Commit fc6fb3f has been approved by ollie27

[bors]

⌛ Testing commit fc6fb3f with merge e7f6ed1...

[bors]

☀️ Test successful - checks-actions, checks-azure
Approved by: ollie27
Pushing e7f6ed1 to master...