RFC: ResultExt trait

[vtenfys]

This follows discussion in neon-bindings/examples#67

I'm quite new to Rust so hopefully my proposal is clear and I haven't missed any obvious alternatives!

View Rendered RFC

[kjvalencik]

@davidbailey00 The main blocker for this RFC is that in order to support this in a very general fashion, requires Generic Associated Types or GATs, which do not exist in Rust.

We would need to choose between supporting generic Into responses on structs or boxed trait object representations (runtime cost).

This is an interesting idea, but I think it needs a proof of concept that can demonstrate we can provide this in a zero-cost, ergonomic format. I then think we would want to do a holistic update of Neon APIs to expect Into<Throw> or Into<Value> like traits.

[vtenfys]

The main blocker for this RFC is that in order to support this in a very general fashion, requires Generic Associated Types or GATs, which do not exist in Rust.

On that basis, should this be revisited when rust-lang/rust#44265 lands?

[kjvalencik]

@davidbailey00 I would ❤️ to implement this sooner, if possible. I tried a few different hacks to work around the problem, but was unsuccessful. It would be awesome of some contributor could figure out how to solve it.

Maybe there's something we could do with proc macros to make it ergonomic without GATs? Here's my ideal API for JsFunction::new:

trait TryIntoJs {
    type Error: IntoThrow;

    fn try_into_js<'a, C: Context<'a>>(self, cx: &mut C) -> Result<Handle<'a, JsValue>, Self::Error>;
}

trait IntoThrow {
    fn try_into_js<'a, C: Context<'a>>(self, cx: &mut C) -> JsResult<Self>;
}

impl JsFunction {
    pub fn new<'a, C, F, T, E>(f: F) -> JsResult<'a, JsFunction>
    where
        C: Context<'a>,
        F: Fn(FunctionContext) -> Result<T, E>,
        T: TryIntoJs,
        E: IntoThrow,
    {
        todo!()
    }
}

As long as the function returns something that we can always turn into a JsResult, it should be accepted. We can provide noop implementations of TryIntoJs for all of the Js types.

[kjvalencik]

Thanks for the contribution! This idea has been incorporated into #44.