Exactly. This is also necessary for the example code in the keyword generics proposal. You still need to write a .await and treat everything as being potentiall

Exactly. This is also necessary for the example code in the keyword generics proposal. You still need to write a .await and treat everything as being potentially async. Compare this to C++20s coroutine design where you practically have a zero overhead abstraction that might or might not be async without the author of the function having to necessarily know. They just yield control flow back to the caller who might suspend execution or not. I’m not saying that one design is better than the other, just that this seems to be a bandaid on top of existing tradeoffs made in Rust’s async design that is only 2 years old.