Requires env-logger

[eloff]

I get the error:

error[E0433]: failed to resolve: could not find `env_logger` in the list of imported crates
  --> src/tests/backend_auth_test.rs:10:1
   |
10 | #[test(tokio::test)]
   | ^^^^^^^^^^^^^^^^^^^^ could not find `env_logger` in the list of imported crates

Which goes away when I add env_logger to my dev-dependencies. I think it probably should be a dependency, not a dev-dependency for this crate.

[d-e-s-o]

Unfortunately, that's not how it works, see: db6e700.

[d-e-s-o]

I assume the above answers the question.

[azdle]

From the looks of it, proc-macro crates' dependencies are treated a bit
differently by Cargo than those of library crates. In particular, they
are not "inherited" by (and, hence, available in) the consuming crate.
As such, the dependency to env_logger that we had specified all along
was effectively unused: while we emitted code that requires the
dependency to be available, it still had to be specified explicitly in
the crate making use of test-env-log.
This change fixes the confusion by converting the dependency to a
dev-only one. We require still require it for testing because our tests
execute the emitted code.

Hey, @d-e-s-o. Apologies if you're already aware of this. If you wanted to make this work, you just need to structure this as two different crates. If you moved the current macro crate into a test-log-macros and then just pub use each macro from a new standard test-log crate. Then you could have env_log (and/or tracing-subcriber) as a dependency of the top-level test-log and have it actually get propagated up.

[d-e-s-o]

Ah interesting. Yeah, I didn't think of that -- may be worthwhile to pursue.

[d-e-s-o]

The one downside I see is that at least when using resolver v2 (the default), users would no longer be able to enable additional features of tracing/tracing-subscriber. Colored output, for example, should be configurable this way. On the bright side, we could facade it properly then behind a colored feature that applies to both log and tracing.

[azdle]

Maybe there's some nuance in the v2 resolver that I'm not aware of, but based on rereading the section of the book it sounds like dev-dependencies' features are still unified if the dev-dependencies are used, so a parallel dev-dependency with added features should still work.

Or, maybe I'm misinterpreting what you meant would be a problem?

But, reguarless a test-log feature would still be an even nicer option if you like that.

[d-e-s-o]

Ah indeed; just tried it out, it would still work. Glad we discussed that ;)

[Stargateur]

It would be nice to reopen this issue