Cannot hygienically create test functions in macro

[tbu-]

Code:

macro_rules! macro_ {
    ($t:ty) => {
        #[test]
        fn test() { }
    }
}

struct Foo;
struct Bar;
macro_!(Foo);
macro_!(Bar);

Error:

a.rs:4:9: 4:22 error: duplicate definition of value `test` [E0428]
a.rs:4         fn test() { }
               ^~~~~~~~~~~~~
a.rs:1:1: 6:2 note: in expansion of macro_!
a.rs:11:1: 11:14 note: expansion site
a.rs:4:9: 4:22 help: run `rustc --explain E0428` to see a detailed explanation
a.rs:4:9: 4:22 note: first definition of value `test` here
a.rs:4         fn test() { }
               ^~~~~~~~~~~~~
a.rs:1:1: 6:2 note: in expansion of macro_!
a.rs:10:1: 10:14 note: expansion site
error: aborting due to previous error

[tomjakubowski]

The #[test] attribute is irrelevant; fails with the same message without it.

[alexispurslane]

So we can't hygienically create any functions? That seems odd.

[lambda-fairy]

One workaround is to nest the function in a module:

macro_rules! generate_tests {
    ($m:ident, $t:ty) => {
        #[cfg(test)]
        mod $m {
            #[test]
            fn test() { }
        }
    }
}

struct Foo;
generate_tests!(foo, Foo);

[semmaz]

The problem here is not about hygiene of your macro,

macro_!(Foo);
macro_!(Bar);

will expand into two functions with same identifier (name) - test, hence the error message about duplicate definition of value 'test'.

So, as @lfairy pointed out, you can workaround this, I'll only add that module usage is unnecessary.

[tbu-]

@semmaz

I thought this is what macro hygiene is: The identifiers within your macro don't clash with those outside of it?

[jonas-schievink]

@tbu- That's true, but I think only locals are hygienic

[durka]

It says in the book, "This [macro hygiene] holds for let bindings and loop labels, but not for items."

[steveklabnik]

Triage: it's been 18 months, and it's not clear that there's actually a bug here. Closing! If that's wrong, please open a new bug with the exact details today, thank you!

[tbu-]

Yea, probably not a bug. Sorry for not closing it before.