Remove type safety from Interface about async-graphql

[sunli829]

@phated I tried this on a new branch.

In addition, thank you very much for the support of async-graphql. I was even more excited when I found out that you are the main maintainer of glup.js.

[sunli829]

The side effect is improved runtime performance.😂

[phated]

I'm not very happy with this idea.

[sunli829]

If the interface field types do not match, there is now a panic at Schema::new and a reason is given.

The following code didn't compile well before, but it does now.

struct MyObj;

#[Object]
impl MyObj {
    async fn a(&self) -> &str {}
}

#[Interface(field(name = "a", type = "String"))]
struct MyInterface(MyObj);

This actually seems reasonable, it simplifies the definition of the interface.

[phated]

Panic is not desirable. I believe there is a solution that involves traits but I'm unable to work on it currently.

[sunli829]

Panic is not desirable. I believe there is a solution that involves traits but I'm unable to work on it currently.

@phated
Thanks for your suggestion.😁 I will make a version as soon as possible and see how it works.

[phated]

@sunli829 is it possible to use a macro to add additional methods to a trait, including a default implementation? That is the solution I am considering.

[sunli829]

@phated I'm actually hoping to solve this problem #29 and #28. I think procedural macros may not be possible.

[phated]

I am hoping to find time to understand macros better, but my other open source responsibilities are requiring my time right now.

Tomorrow I am giving a presentation on async-graphql and I wanted to show type safety on interfaces.

[phated]

🙇 Thank you! Your work on this project is amazing and I very much appreciate it! I am excited to help out more.

[nicolaiunrein]

@sunli829 I think it is a great Idea especially when it panics at application start. I never really liked the implementation of Interface. I think it is somehow reversed. Isn’t there a more natural way where Interface behaves more like a trait? So that Objects and SimpleObjects can "implement" it rather than beeing passed into the interface as struct fields? It just feels wrong and doesn’t allow shareing an interface over crate boundaries.

@phated Is your presentation being recorded? I would love to see it.

[sunli829]

@nicolaiunrein The interface definition is not natural, but it is simple enough to use.😁