hyper-ish 2022 in review
A quick personal note, I wrote much more this year!2 Yay!
Let’s start with some general things about hyper.
In an effort to share the lovely work those humans do, I started writing monthly hyper updates.
And we celebrated 10 thousand stars this year! ⭐
We started with a v1.0 timeline (of which we’re in the final steps). After much discussion with users, we put into writing hyper’s VISION, which defined where we were going. Shortly aftewards came the v1.0 ROADMAP, outlining how to get there.
And then, a summer flurry of coding and hairy diffs.
On the other side, late last year, we published v1.0.0-rc.1. This represented the likely working library that we’d like set down into a stable parking spot for a while. While people check it out and give us feedback, we entered the hyper polish period, making sure everything feels good. That’s where we still are, for a short period more.
hyper in curl
The work to make hyper an HTTP backend in curl from last year continued throughout this year. There’s just a few remaining tests in curl’s large HTTP suite that didn’t work when hyper is enabled. Several wonderful people showed up to dig in and find out exactly why.
To try to empower others to do, I wrote up a help-us-finish guide, explaining step-by-step how anyone could help us finish this all-important work. Later in the year, I streamed a hyper-in-curl debug session, where you could watch me follow the guide, and then bash the keyboard randomly while trying to understand what the issue is. You can still watch the recording, or see these details about what test was solved and the pull requests that came out of it.
I also joined Daniel Stenberg virtually by presenting about hyper in curl at curl-up 2022.
We’ve been working on the h3 crate, providing HTTP/3 that is generic over any QUIC implementation, with the goal of integrating into hyper directly. The repository has had a working server and client which already interoperates with other implementations.
I’ve written about it in the monthly updates, but here’s some highlights I’m excited about:
- We added 3 new collaborators who have been driving the work: @eagr, @g2p, and @Ruben2424.
- @stammw implemented graceful shutdown for the server and the client.
- @Ruben2424 added GREASE support via an default-on option which sends random reserved identifiers, such as frames, settings and streams, to help prevent ossification that would make future extensions harder.
- @eagr made it so we now track compliance with the HTTP/3 specifications, by using special comments that are compared with the spec text, and it even outputs a report updated as part of our continuous integration.
- @Ruben2424 also added
h3specto CI, and fixed the missing pieces it noticed.
- @g2p documented the entire API.
While there’s still specific work that can be done on the
h3 crate itself, it’s time to consider next steps to get it into user hands. To that end, there’s even a pull request for reqwest to use
h3! There’s just some details to work out around publishing unstable versions so reqwest can depend on it. We hope this experimental support will help us iron out any usage annoyances, so we can start landing it in hyper proper.
While tower isn’t tied to hyper, we’ve always meant for people to easily combine the two libraries to make powerful, opinionated HTTP stacks.
In the later half of the year, we started having discussions about making tower easier. Up until now, it has mostly felt like expert mode. But if done right, we shouldn’t be telling users “no, you don’t hold it that way” when they try to implement retries slightly wrong and storm their servers. So, to that end, Lucio put together a big issue outlining how we can make retries better.
Another discussion started about the path to tower 1.0. This brought some interesting questions around how
Service handles backpressure, whether we could make that any better. Certainly, something else we would want to consider is if
Service can make use of async fn in traits.
tower-http repository continued to see additions. Mostly middleware that are specific to HTTP that many people would find beneficial, such as
I’ve also kicked around the idea loosely about cracking open the reqwest crate, and turning its various features into tower middleware. Then, reqwest is just a single opinionated way to build up a client stack. The community would be more empowered to customize the order of layers, adding or removing or swapping, and still have the power that they come to expect from using reqwest.
What are some possiblities in 2023?
Besides launch hyper 1.0, of course.
These are are all things that many people have asked for, and I’d like to see done. But realistically, most will require help from you!
- Improved middleware
- HTTP/3 in hyper
- Tracing and Metrics
- An even-lower level
I’ll likely be focused at the top of that list, but would welcome anyone interested jumping into an issue (or discussing on Discord if you prefer). Really, the biggest success would be empowering others to be the leaders and owners on these things. Do you want to be one of them?
Sorry for the delay, illness struck right at the beginning of the year. ↩︎
I wrote barely anything in 2021, 2020… actually for quite a few years. I used to blog multiple times a month back in 2013, but kind of teetered off the following year. Anyways, I’m really liking it (again), so here’s to more! ↩︎