unnecessary_unwrap not produced in 1.76

[Miha-Rozina]

Summary

The following code doesn't produce a warning for the unnecessary_unwrap lint. It works as intended in Rust 1.73

Lint Name

unnecessary_unwrap

Reproducer

I tried this code:

fn foo(option: Option<String>) {
    if option.is_some() {
        println!("{}", option.unwrap());
    }
}

I expected to see this happen:
To get the unnecessary_unwrap lint's warning.

Instead, this happened:
No warnings emitted.

Version

rustc 1.76.0 (07dca489a 2024-02-04)
binary: rustc
commit-hash: 07dca489ac2d933c78d3c5158e3f43beefeb02ce
commit-date: 2024-02-04
host: aarch64-apple-darwin
release: 1.76.0
LLVM version: 17.0.6

[y21]

It looks like the lint avoids visiting any nested expressions as soon as it finds code from an external macro (since it return;s and doesn't call walk_expr), which means that it won't see that option.unwrap(), since it is somewhere within the expansion of println!:

rust-clippy/clippy_lints/src/unwrap.rs

Lines 302 to 306 in 237fbdd

	fn visit_expr(&mut self, expr: &'tcx Expr<'_>) {
	// Shouldn't lint when `expr` is in macro.
	if in_external_macro(self.cx.tcx.sess, expr.span) {
	return;
	}

... but that logic was added 4 years ago, so "It works as intended in Rust 1.73" is surprising to me.

We might just be able to add another call to walk_expr in that if 🤔

FWIW, I did make changes to this lint, and the PR date lines up with 1.73 where the regression could have started, but this particular false negative doesn't look at all related to my changes. Was this reproducer broken down or slightly modified in a way that hides a different regression?