mark tokio::test as test

[PSeitz]

this is to support the excludeTests option to also handle #[tokio::test]

[Veykril]

That's the wrong way to fix this, we need to go up the expansion hierarchy to find the necessary test attribute when categorizing the references

[PSeitz]

I thought that's too slow as mentioned here: #16441 (comment)

Although I would prefer to have a full filtered result which takes a little bit longer

[Veykril]

Right, that was about general traversal which we do, we are missing traversing up the macro expansions still though. Seems that was overlooked in the PR. That won't make much of a difference I guess given most finds shouldn't be in macro calls anyways.

[PSeitz]

I thought it's related to tokio::test, but it's not:
#16768 (comment)

[Young-Flash]

#[tokio::test] is already mark as test, cause it will be expand to #[::core::prelude::v1::test]

[PSeitz]

I thought it's not working because I saw a lot of references in tokio::test, but the problem with detection seems more general.

e.g. a vec![] is enough to break the current filter

struct Point {
    x: f64,
    y: f64,
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn it_works() {
        assert_eq!(2 + 2, 4);
        let vec = vec![Point { x: 1.0, y: 2.0 }];
        assert_eq!(vec.len(), 1);
    }
}

[PSeitz]

@Veykril @Young-Flash
I'm not familiar enough with rust-analyzer.

sema.to_def(&it) what does this do? The it_works function should appear with the func.is_test(sema.db) on the example below, so my theory would be that this look up fails, or the ancestors traversal does not work like expected.

fn is_name_ref_in_test(sema: &Semantics<'_, RootDatabase>, name_ref: &ast::NameRef) -> bool {
    name_ref.syntax().ancestors().any(|node| match ast::Fn::cast(node) {
        Some(it) => sema.to_def(&it).map_or(false, |func| func.is_test(sema.db)),
        None => false,
    })
}

struct Point {
    x: f64,
    y: f64,
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn it_works() {
        assert_eq!(2 + 2, 4);
        let vec = vec![Point { x: 1.0, y: 2.0 }];
        assert_eq!(vec.len(), 1);
    }
}

[Veykril]

The problem is that the name_ref in is_name_ref_in_test is in a macro expansion, so when ascending the tree we will end it at the top of the expansion, but instead we'd need to switch to the caller of the macro and ascend that tree (and so on for nested expansions).

Using

rust-analyzer/crates/hir-expand/src/files.rs

Line 151 in f17a8c7

pub fn ancestors_with_macros_skip_attr_item(

should work