Hyper's current status

[jolhoeft]

I'm considering revisiting migrating nickel.rs to an async version of hyper. I attempted this a year ago, but put it aside due to work pressure and to let tokio stabilize a bit more. How stable do things look today, in comparison to hyper-0.10.x? In particular, how stable are the interfaces?

If things are not stable yet, would you be open to some PR to 0.10.x to update some packages?

[seanmonstar]

I believe things to be fairly stable. I feel the Client and Server have pretty nice APIs at this point. Things may get small tweaks, but 0.12 is used pretty extensively in large projects, and it feels good.

Some things that are less stable:

• The most unstable is the external dependeny on Future. The trait along with async/await syntax are being merged in libstd. When that is stable and working well, users implementing Future themselves will see some changes. Relevant: async/await #1654
• The Service trait will likely be replaced with that from tower. The concept is similar, with some associated types differing.
• The Payload trait may be replaced with a more general trait, such as tokio_buf::BufStream. Unless you have a custom body type that you implement Payload for (instead of using hyper::Body), this change should be invisible.

(BTW, since this isn't a bug in hyper, I'm going to close. We can still keep discussing though.)

[jolhoeft]

Thank you, I missed this when it first came through. It seems reasonable to start a Nickel migration on a branch, but maintain the current version for a while.