Lint for trait methods without bodies

[etaoins]

As discussed in rust-lang/rust#47475 the #[inline] attribute is currently allowed on trait methods without bodies (i.e. without a default implementation). This is misleading as it could be interpreted
as affecting the implementations of the trait method. Add a lint for any use of #[inline] on a trait method without a body.

Fixes rust-lang/rust#47475

[oli-obk]

I think this should some day be caught by the unused_attribute lint from rustc, 'till then, let's lint this special case from clippy

[killercup]

@oli-obk do we have a mechanisms for suggestions that just remove code? We could use that here.

[oli-obk]

Right. Should be as simple as changing the span_lint into a span_lint_and_sugg and using String::new() as the suggestion

[etaoins]

@oli-obk @killercup

This has the downside of only removing the attribute itself, not the line it was on. This is a bit tricky because technically attributes do not need to be on their own line so we can't unconditionally kill the line. Also, rustfmt will not remove the empty line afterwards.

It'd be possible to add something like suggest_item_without_attr that would do the right thing but it wouldn't be trivial.

[killercup]

Hm, what does .next_point() on the attribute's span yield? Maybe we can collect the next points while they are char::is_whitespace? Should cover the typical newline and indent as well as attr and fn on one line with space between them.

[etaoins]

@killercup

Something like this:

let mut remove_span = attr.span;
let fmpos = cx.sess().codemap().lookup_byte_offset(remove_span.next_point().hi());

if let Some(ref src) = fmpos.fm.src {
    let non_whitespace_offset = src[fmpos.pos.to_usize()..].find(|c| {
        c != ' ' && c != '\t' && c != '\n'
    });

    if let Some(non_whitespace_offset) = non_whitespace_offset {
        remove_span = remove_span.with_hi(remove_span.hi() + BytePos(non_whitespace_offset as u32))
    }
}

?

Would that be worth pulling it to a helping function and using it for USELESS_ATTRS and INLINE_ALWAYS as well?

[etaoins]

@oli-obk @killercup

I pushed a commit that suggests removing the attribute. Let me know what you think.

[oli-obk]

The issue is that the normal diagnostics now also hilight the \n. This is an issue in rustc, but should be easy enough to address.

[etaoins]

@oli-obk
Are you referring to how the entire whitespace suggestion span is being rendered instead of just the attribute? Do you think rustc should unconditionally trim whitespace in suggestions? I think that would break suggesting e.g. reindenting.

Would that be a blocker for merging this PR?

[oli-obk]

Do you think rustc should unconditionally trim whitespace in suggestions?

Yes

I think that would break suggesting e.g. reindenting.

No, because it'll only be done in the command line diagnostics, not the json diagnostics

Would that be a blocker for merging this PR?

Nope. We'll see the ui tests breaking when it gets fixed :)

[oli-obk]

You need to rebase/merge the current master and update the ui tests again. Something changed with the error reporting

[oli-obk]

Thanks!