Théodore Prévot

> https://github.com/ibraheemdev/matchit/pull/49 Merged and released with [matchit 0.8.2](https://docs.rs/matchit/latest/matchit/struct.Router.html#method.remove)

Hey, indeed to implement webtransport, we need h3 to support it as you said. For the moment because it is not used a lot I prefer to wait for this....

After digging this issue a bit, the best solution would be to implement something based on `serde` directly. The following are possible ideas that need to be explored: * Have...

There is no problem to break backward compatibility. The library is in its beta state and it is explicitely stated in the readme that API will change over the time....

Sure ! Don't hesitate to add benchmarks and unit tests to check how it behaves compared to the current packet deserialization.

> My next thought was to just get rid of the idea of an intermediate representation, and have it pass a T: Deserialize into those trait methods. The problem there...

> Now looking at ack.rs, I see instances of serde_json::Value there exposed publicly, like in AckInnerStream, which according to docs needs to be public for people who implement Adapter. Indeed...

> Yes, can do that, but then that would make my custom Value enum public, since FromMessageParts/FromMessage/MessageHandler are public. If we think that's ok after all (as I'm starting to...

Yes, it is a good idea, but it implies to also apply `maybe_async` on `Operators` and `Socket` structs to maybe return futures from the adapter.

Sure, the compatibility with previous sync code and the possibility to use both are my main blocking points to move on on this feature. Currently one of the approaches I...