quinn main update using async traits

[FrankSpitulski]

Quinn has removed the Open* Future structs that h3 was storing for polling. I tried storing it myself, but the lifetimes got very complicated. Then I tried using async traits and removing polling but got stuck again with lifetimes. I'll share my work in case this is the path you want to go down for keeping up to date with quinn, but in any case a breakage is coming with 0.9.0.

[seanmonstar]

Is the short of it that quinn is removing polling methods and only exposing async fns for everything, now?

[FrankSpitulski]

There are still some polling methods left. @Ralith or @djc may be able to answer about future plans.

[FrankSpitulski]

Some discussion is in here https://gitter.im/djc/quinn

[Ralith]

See also previous discussion in #55.

Is the short of it that quinn is removing polling methods and only exposing async fns for everything, now?

That's the current state; it's much more maintainable. We could offer polling methods, but doing so is not a trivial amount of work.

[seanmonstar]

I very much understand that. Encapsulating them in a trait is still annoying, but exposing them is too.

[Ralith]

We're discussing alternative approaches in quinn-rs/quinn#1357, and input would be welcome. Pure polling methods on Connection (for example) are difficult to provide since it's often highly valuable to maintain state specific to the attempt to perform an operation as is possible with returned futures, but there's open questions about to what extent those futures should be 'static, especially with GATs looming on the horizon, though apparently with some delays...

[DemiMarie]

GATs are now stable.

[djc]

We just merged quinn-rs/quinn#1357.

[FrankSpitulski]

I'll close this since it appears quinn is keeping named futures instead of async traits.