fix: new approach for select_buffer

this is definitely bloatier but should work and not deadlock?

Cargo.toml

@@ -25,7 +25,6 @@ serde_json = { version = "1", optional = true }
 tokio-stream = { version = "0.1", optional = true }
 # global
 lazy_static = { version = "1.4", optional = true }
-futures = "0.3"
 
 [build-dependencies]
 tonic-build = "0.9"

src/client.rs

@@ -3,8 +3,8 @@
 //! codemp client manager, containing grpc services
 
 use std::{sync::Arc, collections::BTreeMap};
-use futures::stream::FuturesUnordered;
 
+use tokio::sync::mpsc;
 use tokio_stream::StreamExt;
 use tonic::transport::Channel;
 
@@ -154,19 +154,33 @@ impl Client {
 
 
 	pub async fn select_buffer(&self) -> crate::Result<String> {
-		let mut futures = FuturesUnordered::new();
 		match &self.workspace {
 			None => Err(Error::InvalidState { msg: "join workspace first".into() }),
 			Some(workspace) => {
+				let (tx, mut rx) = mpsc::unbounded_channel();
+				let mut tasks = Vec::new();
 				for (id, buffer) in workspace.buffers.iter() {
-					futures.push(async move {
+					let _tx = tx.clone();
-						buffer.poll().await?;
+					let _id = id.clone();
-						Ok::<&String, Error>(id)
+					let _buffer = buffer.clone();
-					})
+					tasks.push(tokio::spawn(async move {
+						match _buffer.poll().await {
+							Ok(()) => _tx.send(Ok(_id)),
+							Err(_) => _tx.send(Err(Error::Channel { send: true })),
+						}
+					}))
+				}
+				loop {
+					match rx.recv().await {
+						None => return Err(Error::Channel { send: false }),
+						Some(Err(_)) => continue, // TODO log errors
+						Some(Ok(x)) => {
+							for t in tasks {
+								t.abort();
+							}
+							return Ok(x.clone());
+						},
 					}
-				match futures.next().await {
-					None => Err(Error::Deadlocked), // TODO shouldn't really happen???
-					Some(x) => Ok(x?.clone()),
 				}
 			}
 		}