Fire clippy::op_ref on PartialOrd but !Ord types

[CAD97]

Some context: this appears to be the only usage of .ord_trait() in the compiler. I've a PR that tries to remove it. It's definitely not correct here, though! Example:

#[derive(PartialOrd, PartialEq, Eq, /*Ord*/)]
struct S(u32);

fn main() {
    let a = S(0);
    let b = S(1);
    dbg!(&a < &b);
}

On the above snippet, clippy::op_ref does not fire unless we also derive Ord due to the use of .ord_trait() here meaning we gate the lint on implementing Ord, rather than just PartialOrd which is the trait actually required to use comparison operators.

changelog: Fix: trigger clippy::op_ref (needlessly taken reference of both operands) when a comparison operator's operands implement PartialOrd but not Ord. Previously, the lint only fired if and only if Ord was implemented.

[CAD97]

CI failure is an extra lint trigger:

error: taken reference of right operand
  --> $DIR/op_ref.rs:20:8
   |
LL |     if b < &a {
   |        ^^^^--
   |            |
   |            help: use the right value directly: `a`

Context:

rust-clippy/tests/ui/op_ref.rs

Lines 17 to 22 in 7a943a9

	let a = "a".to_string();
	let b = "a";
	if b < &a {
	println!("OK");
	}

Relevant arm:

rust-clippy/clippy_lints/src/eq_op.rs

Lines 163 to 180 in 7a943a9

	// foo == &bar
	(_, &ExprKind::AddrOf(BorrowKind::Ref, _, ref r)) => {
	let rty = cx.tables.expr_ty(r);
	let rcpy = is_copy(cx, rty);
	if (requires_ref || rcpy)
	&& implements_trait(cx, cx.tables.expr_ty(left), trait_id, &[rty.into()])
	{
	span_lint_and_then(cx, OP_REF, e.span, "taken reference of right operand", |db| {
	let rsnip = snippet(cx, r.span, "...").to_string();
	db.span_suggestion(
	right.span,
	"use the right value directly",
	rsnip,
	Applicability::MachineApplicable, // snippet
	);
	})
	}
	},

Removing the ref here is wrong due to not having str: PartialEq<String>. I'm not exactly sure why it's firing, though.

[flip1995]

The "FP" is the same as #2597, it is not really a FP tough. In such a case the correct suggestion would be to write *b < *a.

[CAD97]

rust-lang/rust#66941 has been merged and this patch is now required to get clippy building with latest rust again.

Should I just change the test expected output and mark it as a case of #2597? I'm honestly not certain how best to go about fixing that issue (and fixing it in this PR is probably outside of scope).

[flip1995]

I'm on this myself (there was also another PR, that broke Clippy). It stopped linting for this case, which is good (?). But it's weird, that it returned true for &str: Ord<String>, but not for &str: PartialOrd<String>...

[CAD97]

Note that Ord doesn't take parameters. It's just trait Ord {, not trait Ord<Rhs> { like trait PartialOrd<Rhs>. I suspect implements_trait is just ignoring excess generic arguments.