Lifetimes are not hygenic

[RalfJung]

I would expect the following to not compile (Playground):

trait Trait {}

macro_rules! impl_trait_for {
    ($name:path) => { impl<'tcx> Trait for $name {} }
}

impl_trait_for!(Option<i32>);
impl_trait_for!(Option<&'tcx i32>);

The lifetime 'tcx is bound in the macro, and used in a macro argument. This violates hygiene.

I am sure this has already been reported but I cannot find an issue that matches...

[RalfJung]

Type parameters have the same issue. This compiles as well:

trait Trait {}

macro_rules! impl_trait_for {
    ($name:path) => { impl<T> Trait for $name {} }
}

impl_trait_for!(Result<T, T>);

[petrochenkov]

This is by design in macro_rules!, but fixed in macro.

[pnkfelix]

There's some truly ancient history here, see e.g. #7743, #12451 and meeting from 2014-06-10

[eddyb]

I thought we had to pass lifetimes to macros to make things work, are they hygienic sometimes?

[petrochenkov]

Only labels are hygienic (if those can be counted as lifetimes).

[petrochenkov]

(Closing as not an issue.)