Inner function cannot be tested

[kineticz]

I really would like to be able to this:

fn outer(x: i32) -> i32 {
    fn inner(a: i32, b: i32) -> i32 {
        a + b
    }
    #[test]
    fn test_inner() {
        assert_eq!(inner(3, 5), 8);
    }
    inner(x, 5)
}

#[test]
fn test_outer() {
    assert_eq!(outer(3), 8);
}

[Cobrand]

But why though ? Do you want to be able to test things inner to your function ? How about an assert_test! that would compile / trigger only when in test mode ? Just an idea.

[kineticz]

Because this is convenient for testing nested functions. Currently I'd have to move the fn outside to be able to test it separately.

[kineticz]

@Cobrand Yeah, something like this will be great:

assert!
assert_eq!
debug_assert!
debug_assert_eq!
test_assert!
test_assert_eq!

[Cobrand]

I'm just arguing against my own idea here, but that would require either :

• Require to recompile everything related for "test", which I don't know which is a good idea or not
• Compile this only in debug mode, but only run it when doing test (so it's still in the binary but there is a conditional for whether or not this is in test mode)

My code is horrible so it doesn't have that much tests, is it possible to run tests in release mode ? If that is the case, the second option is a no-no I guess.

[Aatch]

The problem with this is that the test harness needs to access the inner functions, which it can't do because you can't name them.

[Mark-Simulacrum]

I'm marking as a diagnostics issue so that we print a warning in this case at least.

[eddyb]

Why was the #51450 PR accepted, while the RFC is still unmerged?

I disagree with the approach taken (a new lint), we should either (rust-lang/rfcs#2471 (comment)):

• figure out why unused_attribute isn't firing, and let that handle the lint side
• properly support #[test] functions anywhere, as per Add lint warning for inner function marked as #[test] rfcs#2471 (comment)