Build failure is due to crate derive checking for the complete feature in the wrong way. This is solved by #131, so let's merge it (or #132 if you insist).

Changes since last push:

• Rebased
• Due to remove 'complete' feature flag, make it unconditional #131, tests are now green :)

Oh I have another question. Should we fix these tests to a specific Rust version in case rustc changes their error format?

Changes since last push:

• rebased on current main
• Changed the arg_missing_metavar test such that the type does not impl Default. So you're absolutely right, thanks @tertsdiepraam! :)

I don't think it's very likely that rustc will change the error messages around these things, since we don't use fancy new features that are likely to change. On the other hand, I don't have enough experience with these things to really know for sure how stable Rust's error messages are. Some points I see are:

• There are new Rust versions constantly. So if we fix it to a particular version, that means we will have to update that piece of information every other 6 weeks or so. That sounds like a lot of pointless work.
• If the output changes, the test will fail, and we're the first to discover that the error messages have changed (and potentially become useless). We'll have ample opportunity to disable or adjust the tests, and/or start implementing the version-fixing. This might even happen before a user runs into one of the (changed, possibly useless) new error messages. I think this is good.

So I suggest we don't do any version fixing right now.