Add hyper client middleware implementation

[06chaynes]

See docs

[06chaynes]

"Shouldn't take long" haha you fool lol LMAO

If it happens that someone is actually waiting on/wanting this I'm still working on it in the background, in fact my attempts have informed a few changes in the core library... but every attempt has been met with issues that made the implementation straight up goofy.

Still making attempts though!

[robo9k]

The docs you've linked to seem to be the tower::Service implementation of hyper, so this is more a tower-http middleware than "proprietary" hyper-only client middleware, right?

I'm interested in building a custom HTTP client using tower-http, so I guess you can sign me up as wanting something like your crate for caching :)
Is there some info on the diffuculties you've hit, e.g. upstream tracking issues or discussion?

[06chaynes]

I had made some attempts locally but hadn't pushed anything upstream yet. I'll try to revisit it soon and collect more details on the issues I ran into. Think there were some question on error handling in a couple places (left using unwrap to press on at the time), as well as a lifetime issue in the Service implementation.

I think you're right about it being more of a tower::Service middleware rather than specific to hyper. I was just planning on using hyper for the client, but I'm open to any suggestions!

[06chaynes]

I was planning on checking this out over the weekend but it turns out I have an addiction to Elden Ring, so my apologies on the delay

[06chaynes]

@robo9k just pushed what changes I had committed on that attempt to a new branch 'hyper'. I apparently didn't commit my attempt of the Service implementation unfortunately, so only a mostly complete implementation of the http-cache Middleware trait is there.

[06chaynes]

Making another attempt at this on the refactor branch, http-cache-tower Have a few issues to work out still though, like cloning the request body, getting rid of the unwraps, and how to best change things to run async functions inside Future::poll().

[SebRollen]

Hey @06chaynes,

Let me know if you're looking for help to push this along! One idea that might help is to bump the version of http to 1.0.0, as both Request<B> and Response<B> now impl Clone if B: Clone. That also gets rid of most of the unwraps on your branch since you can pass req to the future directly rather than splitting it into parts and reconstituting.

I don't have much experience with http-cache itself, but I feel your pain around cloning the request body. Was working on an implementation of a tower service for an idempotency key cache a few months back and never made it past the request body cloning step.

In terms of how to run the async functions inside the poll, I wonder if adding a state enum that you match on inside poll could work? I found the implementation in tower-etag-cache really helpful when working on my idempotency cache: https://github.com/billythedummy/tower-etag-cache/blob/master/tower-etag-cache/src/future.rs

[06chaynes]

@SebRollen any and all help is greatly welcomed. I haven't taken time to go over the http 1.0.0 changes yet but the addition of Clone to the request and response is absolutely fantastic to hear! I was planning on tackling that once Reqwest was updated to use the 1.0.0 release as well but I suppose it could happen before then.

I will for sure check over that implementation as I did run into issues trying to figure out how to get async functions working in a tower middleware, thanks again!

Another change I was trying to tackle soon was removing the HttpResponse struct and replacing it with a trait of the same name. This way I no longer need to convert back and forth between the response types. I made a pass at this change but ran into some issues that I'll need to think through, seems doable though.
Edit: Well thinking more on it I guess I couldn't make the last change I wanted as I wouldn't be able to implement the trait on the reqwest and surf response types, oh well was fun to try. I do think I can now remove the async-trait crate though which is something.

[SebRollen]

@06chaynes great, happy to hear that! I put up a WIP PR to be able to discuss specific implementation questions. Let me know if you'd rather I branch it off of main, since the refactor branch is now quite far behind