feat: add response::Error

[npmccallum]

This type makes for efficient use of the ? operator when in a function
with multiple return error types that all implement IntoResponse.

Signed-off-by: Nathaniel McCallum nathaniel@profian.com

[jplatte]

I like the idea, but why are there so many separate From impls instead of a single generic one (impl<T: IntoResponse> From<T> for Error)?

[npmccallum]

@jplatte It conflicts with the blanket impl From in the standard library.

[jplatte]

Ah, that's unfortunate. I wonder whether there's a way around it 🤔

[npmccallum]

Perhaps we could have Error not implement IntoResponse and instead impl IntoResponse for Result<T, Error>?

[davidpdrsn]

I think this is an interesting idea as well!

Can you give some examples of how you imagine people would use it? Both for error types that you own (and can implement IntoResponse for) and types you don't (like anyhow).

Specifically it needs to beat .map_err(IntoResponse::into_response)? to be justified imo.

[npmccallum]

@davidpdrsn We have, for example, different traits for different kinds of backends doing different functions. Each of them have their own error types, all of which impl IntoResponse. However, because there isn't one mega error type (it isn't appropriate for our usage) it means the ? operator doesn't work and we have to do .map_err() all over the place. This causes the flow of our (complex) logic to get hidden between the error-handling overhead.

Therefore, we need one common Error type which can work with the ? operator.

[davidpdrsn]

So do you then implement From<Error> for each of your error types and return Result<_, Error> from handlers?

[npmccallum]

@davidpdrsn No. The goal is that each distinct Error type implements IntoResponse. Then everything should just magically combine into this new Error type.

[npmccallum]

I have added a separate PR for a simpler approach. I'll leave this here so we can compare approaches.

[davidpdrsn]

I think we should go with #921 instead as its simpler and more flexible. I'll close this for now.