lgtm. Tho it could use a test. lmk if you don't have time to add it, I'll give it a look then.

I don't know how to test this. I would either have to make something that's already used log something. I didn't see anything like that when skimming it again. Otherwise, I'd need rustc make the internal write! macros to emit something like they do in the issues, but I assume that cannot be done with the distributed library and compiling stdlib as part of tests here seems a bit too expensive.

Alternatively, I could add a random #[cfg(test)] tracing::error!("Ignore me"); somewhere inside the on_event method, but that didn't seem appropriate either.

So I'd be happy to add tests, but I'm not sure how to go about that here.

Hmm... would it work if we used a custom tracing destination? (i.e. not the stderr default, but a custom writer, that will then just emit a tracing event once)

You're right, that should work, I did not think of that. I'll try it out later today.

we really should get tracing-mock out, which would be a great for this.

I've added the test. I know usually tracing tests use with_default instead of things like set_global_default, but in this case that actually circumvents the bug. tracing internally calls get_default which returns NoSubscriber for nested calls if it is inside a scope with custom subscriber. So this had to go into its own integration test to not interfere with other stuff.

Very nice! Thanks!

Could you also make a release when you get some time? It should help the people in the linked tracing issue.

done