Should BoxFuture/BoxStream require Sync?

[Nemo157]

I was just looking at testing integrating some code with Hyper's master branch. The first problem I came across was that the existing code uses BoxStream to represent its body type, which cannot be converted into hyper::Body because it additionally wants Sync on the trait object:

pub fn wrap_stream<S>(stream: S) -> Body
where
    S: TryStream + Send + Sync + 'static,
    S::Error: Into<Box<dyn StdError + Send + Sync>>,
    Chunk: From<S::Ok>, 

I don't see why Sync would be necessary, since I see no usecases where the type is accessed concurrently from different threads, but it seems to be pretty common to require both markers for trait objects (e.g. there is Into<Box<dyn Error + Send + Sync>> but not Into<Box<dyn Error + Send>> which I have run into, and seen others run into, multiple times).

I can't think of any Send + !Sync type right now, and it seems like a niche usecase if they do exist.

cc @seanmonstar

[12101111]

hyper require Sync on this method since hyperium/hyper@4441372
Because rust-lang/rust#57017
Maybe it’s a bug in rustc

[Nemo157]

More specifically the bug that Hyper is running into appears to be rust-lang/rust#59245 (via rust-lang/rust#57017 causing the match patterns to create unnecessary references, but it would still be possible to hit #59245 without #57017).

It seems to me that it's definitely a bug in the generator transform, if the reference is internal then there's no need for the referenced value to be Sync.

Either way, is there a downside to adding + Sync here? Skimming the docs I notice that Cell, RefCell and mpsc::{Receiver, Sender} appear to be the only types that are Send + !Sync.

[cramertj]

We now have LocalBoxFuture and LocalBoxStream which I believe resolve this issue. Let me know if there is more wanted here!