Shorter type names in recursive types

[pkhry]

Description:

This is a workaround for trait bounds generated for recursive types in scale-codec.(see paritytech/parity-scale-codec#603, fixes paritytech/subxt#1603).
To generate short Paths in recursive types we will be temporarily storing parent_path inside the settings struct that's passed around when generating TokenStreams.
So instead of passing just the alloc_crate_path to the TypePath::to_syn_type we will pass all of the settings to make use of parent_path. So when we encounter a Path with a param and it will be equal to the parent_path we will only leave the last segment inside of it.

Examples:

  #[allow(unused)]
    #[derive(TypeInfo)]
    #[scale_info(skip_type_params(A))]
    struct Example<A, B> {
        id: A,
        rest: B,
        inner: Vec<Example<A, B>>,
    }
    ...
    // Registered:  
    let for_codegen = Testgen::new()
        .with::<Example<String, u32>>()
        .with::<Example<u8, u32>>()
        .with::<Example<u16, u32>>()
        .try_gen_tests_mod(Default::default(), true)
        .unwrap();
    ... 
    /// Generated 
        let expected = quote! {
        pub mod tests {
            use super::types;
            pub struct Example1<_1> {
                pub id: ::std::string::String,
                pub rest: _1,
                // Note the short path for `Example<_>`
                pub inner: ::std::vec::Vec<Example1<_1> >,
            }
            pub struct Example2<_1> {
                pub id: ::core::primitive::u8,
                pub rest: _1,
                pub inner: ::std::vec::Vec<Example2<_1> >,
            }
            pub struct Example3<_1> {
                pub id: ::core::primitive::u16,
                pub rest: _1,
                pub inner: ::std::vec::Vec<Example3<_1> >,
            }
        }
    };

[jsdw]

Do we need this or can we add the #[codec(dumb_trait_bound)] attr by default to the codegen?

[pkhry]

Do we need this or can we add the #[codec(dumb_trait_bound)] attr by default to the codegen?

We can, but i'm not sure if we will break something wrt trait bounds, because we will add this attribute to every enum/struct that there is

EDIT:
tested with subxt-codegen with just indiscriminately adding dumb_trait_bound to every struct/enum and it compiles. (tested only for polkadot-full metadata)

[jsdw]

tested with subxt-codegen with just indiscriminately adding dumb_trait_bound to every struct/enum and it compiles. (tested only for polkadot-full metadata)

That's probably a good sign then; I'd be happy if we just went with that approach for now :)

[pkhry]

closing in favor of paritytech/subxt#1630