Fix generics tokenization in Error derive macro

[zohnannor]

Synopsis

#[derive(Error)] on enums or structs that have both const generics and generic types...

#[derive(Debug, Display, Error)]
enum ErrConstGeneric<const X: usize, E> {
    Inner(E),
}

...panics:

error: proc-macro derive panicked
   --> tests/error/derives_for_generic_enums_with_source.rs:270:30
    |
270 |     #[derive(Debug, Display, Error)]
    |                              ^^^^^
    |
    = help: message: expected one of: `for`, parentheses, `fn`, `unsafe`, `extern`, identifier, `::`, `<`, square brackets, `*`, `&`, `!`, `impl`, `_`, lifetime

Solution

For some reason, iterating over syn::Generics::params that consists of only const generics or only generic types, tokenizes them as identifiers only, but when we mix them, syn decides to output const generics like they were defined, with const keyword, type annotation, etc. So when we are rendering the trait impl, adding bounds for an item in where clause, we get this:

#[automatically_derived]
impl<const X: usize, E> ::std::error::Error for ErrConstGeneric<X, E>
where
    E: ::std::fmt::Debug + ::std::fmt::Display + ::std::error::Error + 'static,
    ErrConstGeneric<const X: usize, E>: ::std::fmt::Debug + ::std::fmt::Display, // obv wrong
{
    fn source(&self) -> Option<&(dyn ::std::error::Error + 'static)> {
        match self {
            ErrConstGeneric::Inner(source) => {
                Some(source as &(dyn ::std::error::Error + 'static))
            }
        }
    }
}

Use syn's built-in Generics::split_for_impl to obtain a type that properly tokenizes all generic parameters.

As for the tests, I've added a "compile fail" tests for Error derive macro on generic enums and structs with combinations of different generics. AFAIU, this happens only in Error macro, tell me if I should add tests for some others.

This should be listed in the changelog under Fixes section, am I right?

Checklist

• Documentation is updated (if required)
• Tests are added/updated (if required)
• CHANGELOG entry is added (if required)

[JelteF]

Thanks for the contribution and nice catch. We use split_for_impl for this exact reason in most other places, but it seems this one was missed. Changelog should indeed be under the Fixes section.

[zohnannor]

Thank you! I'll make corrections and re-request a review when I can.

[tyranron]

@zohnannor thanks!