Racing With Select

join! waits for everything. Sometimes you want the opposite: whichever of several things happens first, and none of the rest.

tokio::select! polls every branch and completes as soon as one of them does.

tokio::select! {
    value = fetch_from_cache() => use_it(value),
    value = fetch_from_disk()  => use_it(value),
}

Each arm is pattern = future => expression. Every arm must produce the same type, exactly like a match.

Every loser is cancelled

When one branch wins, the others are dropped mid-flight. Whatever they had done so far is thrown away.

That makes select! in a loop a sharp tool. A future that is safe to drop half-finished and poll again from scratch is called cancel safe. Receiver::recv is: a message is either taken or still in the channel, never lost in between. Reading half a line from a socket into a local buffer is not.

If several branches are ready at once, select! picks between them at random, so never write code that depends on branch order.

Reusing a future across iterations

A future created inside a loop body starts over on every pass. For a deadline that is exactly wrong: the timer resets after each message and never fires. Create it once, pin it, then poll the same one with &mut.

let deadline = sleep(limit);
tokio::pin!(deadline);
 
loop {
    tokio::select! {
        v = rx.recv() => { /* ... */ }
        _ = &mut deadline => break,
    }
}

tokio::pin! pins the future in place on the stack, which is what lets you take &mut of it.

Your Task

Implement first_of

 
pub async fn first_of<A, B>(a: A, b: B) -> String
where
    A: Future<Output = String>,
    B: Future<Output = String>,

Return whichever future produces a value first, and drop the other.

Implement collect_until_deadline

 
pub async fn collect_until_deadline(
    rx: &mut mpsc::Receiver<i32>,
    limit: Duration,
) -> Vec<i32>

Collect messages in arrival order and stop at whichever comes first: the channel closing, or limit elapsing since the call started.

A producer sending steadily forever must not keep the collection running past limit.

Notes
  • The deadline is measured from the start of the call, not from the last message.
  • rx.recv() yields None when every sender is gone. That ends the loop too.
  • Both branches of a select! arm must return the same type, so do the work in a block.

Hints
Click here to reveal hints
  • first_of is one select! with two arms, each binding a value and returning it.
  • In collect_until_deadline, build the sleep and tokio::pin! it before the loop.
  • Match on the Option the receive arm produces: push on Some, break on None.
  • The deadline arm ignores its output and just breaks.
  • If a test hangs, the deadline is being created inside the loop instead of outside it.