Implement IntoResponse for Response<()> and response::Parts

[neoeinstein]

Add a simple implementation to use Parts and Response<()> as a source for IntoResponse.

This doesn't fit neatly into the IntoResponseParts paradigm because Parts already contains a status code. The implementation itself is rather trivial.

Motivation

I have a library where I generate some generic template responses using http::response::Builder and wanted to be able to make it one step easier to use those templates (often built with a body of (), and then substituted with the intended body) as responses. This makes it so that the template can be built and specified next to the body to be placed into the response.

Solution

Add impls for Parts and Response<()>. The new Response<()> impl discards the existing () body and uses the provided body to construct the response and then moves the headers and status into the new response.

[jplatte]

I think the Parts one might make sense, but

• it should be done in the macro below, in an equivalent manner to
  impl<R, $($ty,)*> IntoResponse for (StatusCode, $($ty),*, R)
• it should propagate response extensions as well

I think the Response<X> one is too much of a footgun though.

[davidpdrsn]

Hm an issue with this is that we cannot merge two Extensions. So either the Extensions from the Parts or the Response needs to be dropped. I'm a little worried about that. I think in general we should be careful to not remove stuff from either part of the response. That is at least how things work in axum today.

Maybe we should re-open hyperium/http#525?

[neoeinstein]

Yep. The Extensions and Version were both things I looked at, but it's difficult to merge in the extensions as mentioned in that linked issue, as there's no way to iterate and extend at the moment.

I can see Response<X> being a potential footgun. It's a bit difficult because I can't provide an impl for http::response::Response<R> where R: IntoResponse, as that would conflict with the existing one that expects a fully formed http_body::Body. This was my workaround for that, but even just having Parts would make it one step easier.

One potential alternative for that is to impl this for (Response<()>, R) instead of (Response<X>, R). That would make it clear that the body of the left side has no value.

[davidpdrsn]

I think this looks good as is but I don't think we should merge it until hyperium/http#546 has landed.

[davidpdrsn]

I think this is good to go now :)

[jplatte]

What about extra impls for (Parts | Request<()>, $(impl IntoResponseParts)+, impl IntoResponse)?

[jplatte]

Changes look good either way.

[davidpdrsn]

Good idea! I filed #979. I'll look into it.