"+ style" trait bounds

[cscherrer]

Hi, thank you for your work on this package! I think this can really help with increasing test coverage and decreasing redundancy.

One thing that I found a little tricky to work around is when you want tests associated with some trait bound, instead of just a single trait.

Say I have Trait1 and Trait2, and I want some test for Trait1 + Trait2. As I understand, this currently requires adding some Trait_1_and_2: Trait1 + Trait2 and then making sure implementations are marked explicitly. After this, the package applies as usual.

This works, but it seems like a lot of boilerplate. It would be cleaner if we could somehow apply #[tested_trait] to Trait1 + Trait2, and then similarly for #[test_impl].

I'm new to Rust macros, so I don't have a sense of how involved this would be, or what syntax would be appropriate. Or even better would be if you've already thought of this and might have an approach you've found to work well. I'd appreciate any advice on this.

Thanks again :smile:

[max-heller]

Thanks for the feature request!

This seems rather tricky given the current design of the crate and I don't see an obvious way to implement it. The approach you describe--defining a new trait that has both traits as supertraits--should work, although I agree it's clunky.

A couple thoughts:

• If the traits are expected to be always implemented together, you could add a where Self: Trait2 bound to test functions for Trait1 (or vice versa), similar to how the example tests use where Self: Default. This is more restrictive for implementers, since it means they have to implement both traits to instantiate the tests. Perhaps there should be a way to instantiate only the tests with trait bounds compatible with the implementing type, although then there should be an obvious way to tell which tests are being run and which are being ignored.
• If these tests were to be defined separately from either trait definition (i.e., not contained as associated tests in either definition), there'd have to be both a new way to define test groups and to refer to test groups. trait_tests implements the definition half of this approach

[cscherrer]

Thanks @max-heller for the helpful response. Of your two bullets, I think we have cases of each. I haven't seen trait_tests before, I'll check it out.