Make `graphql_scalar` a proc macro

[LegNeato]

Make graphql_scalar a proc macro to match everything else...currently it is only a macro with some wonky syntax.

[andrisak]

I'd like to give this a try.

[andrisak]

Would this be a suitable approach?

Create a new trait that gets the information needed for the macro:

trait ParseCustomScalarValue<'a, T> {
    const DESCRIPTION: &'a str;
    fn resolve(&self) -> juniper::Value;
    fn from_input_value(value: &juniper::InputValue) -> Option<T>;
    fn from_str<'b>(value: ScalarToken<'b>) -> ParseScalarResult<'b, DefaultScalarValue>;
}

Then implement the trait for the custom scalar type when using the new macro:

struct UserId(String);

juniper::graphql_scalar! {
    impl<'a> ParseCustomScalarValue<'a, UserId> for UserId {
        const DESCRIPTION: &'a str = "my description";

        fn resolve(&self) -> juniper::Value {
            Value::string(&self.0)
        }

        fn from_input_value(value: &juniper::InputValue) -> Option<UserId> {
            value.as_string_value().map(|s| UserId(s.to_owned()))
        }

        fn from_str<'b>(value: ScalarToken<'b>) -> ParseScalarResult<'b, DefaultScalarValue> {
            <String as ParseScalarValue>::from_str(value)
        }
    }
}

[LegNeato]

Close! We'll want to to be a proc macro, not a macro (so annotate with #[juniper::graphql_scalar] not the macro juniper_graphql!).

We have a general function to parse an impl block and handle a lot of this. For example, the description can be pulled from doc comments rather than making it part of the proc macro.

Also, take a look at how graphql_object is implemented.

[andrisak]

I've created a WIP PR just to verify the approach. What do you think?