Lint against useless &mut (i.e., uplift clippy::unnecessary_mut_passed)

[matklad]

In rust-analyzer, I had the following code:

impl Completions {
    pub(crate) fn add_field(
        &mut self,
        ctx: &CompletionContext,
        field: hir::StructField,
        ty: &Type,
    ) {
        let is_deprecated = is_deprecated(field, ctx.db);
        let mut completion_item = CompletionItem::new(CompletionKind::Reference)
            .kind(CompletionItemKind::Field)
            .build();

        compute_score(&mut completion_item, ctx);

        self.add(completion_item);
    }
}

pub(crate) fn compute_score(
    completion_item: &mut CompletionItem, 
    ctx: &CompletionContext,
) {
   ...
}

I've changed the compute_score function to

pub(crate) fn compute_score(
    completion_item: &CompletionItem, 
    ctx: &CompletionContext,
) -> Option<CompletionScore> {
   ...
}

After that, the code compiled without warnings, was broken, because compute_score(&mut completion_item, ctx); statement no longer did what it was supposed to be doing.

The smell here I think is that we use explicit &mut, which gets silently reborrowed to &. I think we should lint against, roughly, all programs where &mut e expression can be replaced with &e without breaking the compilation.

[fmease]

If this is about code like drop::<&str>(&mut String::new()), then this is covered by Clippy already, namely by clippy::unnecessary_mut_passed which was added 10 years ago.

Are you requesting to uplift the Clippy lint into rustc? I'm not sure if that's super warranted. Related: clippy::needless_borrow (drop::<&str>(&"")).

[matklad]

Yes, I suggest uplifting the lint to rustc