Tracking issue for closures in futures 0.4 combinators

[taiki-e]

This is a tracking issue for consistency in APIs that use closures in futures 0.4.

For individual cases, see below. (Especially #1889, #2470, and #2518 have some discussion on this)

• Should closures of StreamExt::map and TryStream::map_{ok, err} return a Future? #1889
• Why does StreamExt::filter and StreamExt::filter_map require functions that returns futures #1987
• Missing synchronous version of StreamExt::filter_map #2237
• Should TryStream::try_filter accept closures that return Results? #2375
• Review TryFutureExt combinators #2470
• Various inconsistencies around sync/async in TryStreamExt and StreamExt #2518

(I created this issue to avoid duplication of discussion between them.)

In most cases, the APIs where people are unhappy seem to be filter-related. (Note that this does not necessarily mean that "many people are unhappy with the current filter-related API" since those who are not unhappy with the current API will not leave comments.)

[tisonkun]

@taiki-e Today I found a scenario where a TryStreamExt::try_flat_map may help. That is, take an Ok variant, returns a Stream (or TryStream?), and then the result streams will be flatten.

The method signature would be:

    fn try_flat_map<Fut, F, T>(self, f: F) -> TryFlatMap<Self, U, F>
    where
        F: FnMut(Self::Ok) -> U,
        U: TryStream<Ok = Self::Ok, Error = Self::Error>,
        Self: Sized,
    {
        assert_stream::<Result<Self::Ok, Self::Error>, _>(TryFlatMap::new(self, f))
    }

Is it worth to start preparing a patch?

P.S. I'm redirected to here from #2518.

UPDATE - Seems it can be combined with map_ok + stream_iter + try_flatten:

input
    .into_stream()
    .and_then(move |rs| future::ready(partial_sort.apply(&rs))) // ignored, flatten the Error produced by partial_sort
    .map_ok(|rss| stream::iter(rss.into_iter().map(Ok)))
    .try_flatten()

[Plebshot]

I was looking for try_flat_map when I stumbled upon this issue. @tisonkun workaround works for me (thanks), but without this issue I wouldn't have found out a clean way to do so myself.