Poll<T, E> = Result<Async<T>, E> vs Async<Result<T,E>>

[Matthias247]

Hi,
why was the type of poll to be chosen to be Result<Async<T>,E> instead of Async<Result<T,E>>.
Imho the latter signature matches better to the typical definition of a future, which is it's either

• not resolved
• or resolved with either an error or a value

It also matches better the fact that the thing is a boxed (in the sense of 'monadic', not) version of the actual Result of the call.

[Stebalien]

In one word: try!. This way, one can try!(something.poll()) (or, in when it stabilizes, something.poll()?) and then handle the non-error cases. Also, semantically, returning an error doesn't mean that the value has resolved. It could have, e.g., timed out.

Note, this could* possibly be changed in the future depending on what the final form of the Carrier trait looks like. That is, it may eventually become possible to apply the ? operator to Async<Result<T, E>>.

*I'm not a dev; I'm just speculating.

[alexcrichton]

Thanks for the report @Matthias247! This actually was indeed what we used to do, but after the discussion on #108 we decided to settle on what we've got today. @Stebalien hit the nail on the head with the primary rationale being to support the use of try!.

[alexcrichton]

Closing due to previous discussion on this topic.

[agausmann]

I do think this design decision should be reconsidered once the try_trait feature gets stabilized. Async<Result<T, E>> is much easier to read - either a deferred computation has terminated or it hasn't. If it has, then we check if it succeeded.

[Nemo157]

With futures 0.3 removing the error type from the base Future type this has basically happened, TryFuture<Item = T, Error = E> is equivalent to a Future<Output = Result<T, E>> so the return type of Future::poll is Poll<Result<T, E>> (in 0.3 Poll is now what was called Async, so this is similar to 0.1's Async<Result<T, E>>).

There are also some Try implementations included for Poll now.

[agausmann]

That's also a great step forward! I'm glad to see that, and a bit ashamed I didn't see that. Looking forward to the 0.3 rewrite then.