Is there a way to use ImplGenerics from split_for_impl() above to avoid this? Not sure whether it will work with our modified lifetimes or not.

You are correct that ImplGenerics takes care of removing the defaults and it takes care of const parameter defaults too, so it's strictly better.

However there's another problem: we replace all lifetimes with 'static and this means that given this input:

struct Goat<'bar, T: SomeTrait = SomeDefault>

we get this in the expansion:

impl<'static, T: SomeTrait>

(which is invalid syntax)

So ImplGenerics is the right tool for the job, but using it will require cloning&munging elsewhere. Not sure what is least worse tbh. :/

I have an idea: remove the replacing of lifetimes and add something like this to make_where_clause:

for lifetime in generics.lifetimes() {
        where_clause
            .predicates
            .push(parse_quote!(#lifetime : 'static));
    }

That way we should be able to use ImplGenerics and remove completely the two mutable bits.

You are completely right. And as an extra bonus we now don't need to clone the input anymore. ❤️

With your suggestion we get an expansion like this:

    struct Assoc<'bar, T: Types> {
        a: T::A,
        b: &'bar u64,
    }
    
    const _: () = {
        impl<'bar, T: Types> ::scale_info::TypeInfo for Assoc<'bar, T>
        where
            'bar: 'static,
            T::A: ::scale_info::TypeInfo + 'static,
            T: Types + ::scale_info::TypeInfo + 'static,
			…
			…

That 'bar: 'static looks a bit funny but it's valid syntax so all good.

?

Ouch, my bad.

But while we're at it: we should probably remove this anyway. The derive crate doesn't need to be no-std I think (only its output must be no-std). And as we're using proc-macro-crate already it's perhaps good to acknowledge the fact that scale-info-derive isn't actually no-std?

Yeah I'm not sure why that's there in the first place, but if we remove it should not be hidden in this PR

See #69 (comment) – I removed it in that PR.