Note: the CONTRIBUTING.md file recommends updating the CHANGELOG.md but the last entry is very old - over 4 years old; however, the CHANGELOG.md file in the v0.1.x branch is up to date. I plan to make a PR against that branch as well and will update that CHANGELOG.md. I'm happy to update master's as well if that is appropriate.

Thanks. Should I close the other PR then or will maintainers?

No, this PR looks fine, the maintainers are gonna backport everything!

@davidbarsky @hawkw is this good? Want me to rebase on another branch or anything? We'd love to get this fix to unblock some of our scenarios.

Hey, just a heads-up that this is a breaking change in some situations, as global paths don't work with crate re-exports.

In our code base we have a shared library across multiple repositories, and this library re-exports commonly used crates (including tracing) to make version management easier. Previously we could do use lib::tracing::{self, instrument}; to use the instrument macro and have it access tracing, but now each service will have to explicitly specify tracing as a direct dependency.

Makes sense. However, what you were doing before was never meant to be supported, to my understanding.

If you need a new hack, you can add

use lib::tracing::*;
extern crate self as tracing;

to your crate roots. I wouldn't recommend it over the explicit dependency, just pointing out it's possible 😅

I think the problem here is that we can either support users having some module called tracing in the current scope that isn't the real tracing or we can support users only importing tracing via a re-export of the whole crate which brings the real tracing into scope as a module.

However, I don't believe that we can support both cases at the same time. If you know a way, please let me know.

I appreciate that this is a breaking change for you, and I don't want to downplay that, but we're really considering the previous behavior of not using an absolute path as a defect that should be fixed and how the macro should have been implemented all along.

we're really considering the previous behavior of not using an absolute path as a defect that should be fixed and how the macro should have been implemented all along

Yes, I agree, this is an understandable change, and I'm not asking for it to be reverted. Just writing a heads-up in case anyone else encounters this (and perhaps this could also be mentioned in the changelog).

@ilya-zlobintsev Thank you for the heads up and for the understanding. This was missed from the tracing-attributes changelog, but due to another issue (#3306), we'll be releasing a new version soon, so we'll put it in there.