• BB_C
    link
    fedilink
    arrow-up
    4
    ·
    4 months ago

    I skimmed the latter parts of this post since I felt like I read it all before, but I think moro is new to me. I was intrigued to find out how scoped span exactly behaves.

    async fn slp() {
        tokio::time::sleep(std::time::Duration::from_millis(1)).await
    }
    
    async fn _main() {
        let value = 22;
        let result_fut = moro::async_scope!(|scope| {
            dbg!(); // line 8
            let future1 = scope.spawn(async {
                slp().await;
                dbg!(); // line 11
                let future2 = scope.spawn(async {
                    slp().await;
                    dbg!(); // line 14
                    value // access stack values that outlive scope
                });
                slp().await;
                dbg!(); // line 18
    
                let v = future2.await * 2;
                v
            });
    
            slp().await;
            dbg!(); // line 25
            let v = future1.await * 2;
            slp().await;
            dbg!(); // line 28
            v
        });
        slp().await;
        dbg!(); // line 32
        let result = result_fut.await;
        eprintln!("{result}"); // prints 88
    }
    
    fn main() {
        // same output with `new_current_thread()` of course
        let rt = tokio::runtime::Builder::new_multi_thread()
            .enable_all()
            .build()
            .unwrap();
        rt.block_on(_main())
    }
    

    This prints:

    [src/main.rs:32:5]
    [src/main.rs:8:9]
    [src/main.rs:25:9]
    [src/main.rs:11:13]
    [src/main.rs:18:13]
    [src/main.rs:14:17]
    [src/main.rs:28:9]
    88
    

    So scoped spawn doesn’t really spawn tasks as one might mistakenly think!

    • arendjrOP
      link
      fedilink
      arrow-up
      4
      ·
      4 months ago

      I think I would put the emphasis slightly differently: I don’t feel the confusion is around the word “spawn”, but it spawns futures rather than tasks. For tasks you might indeed expect them to be picked up in the background (which is what work-stealing does), but futures only execute when polled.

      • BB_C
        link
        fedilink
        arrow-up
        2
        ·
        4 months ago

        but futures only execute when polled.

        The most interesting part here is the polling only has to take place on the scope itself. That was actually what I wanted to check, but got distracted because all spawns are awaited in the scope in moro’s README example.

        async fn slp() {
            tokio::time::sleep(std::time::Duration::from_millis(1)).await
        }
        
        async fn _main() {
            let result_fut = moro::async_scope!(|scope| {
                dbg!("d1");
                scope.spawn(async { 
                    dbg!("f1a");
                    slp().await;
                    slp().await;
                    slp().await;
                    dbg!("f1b");
                });
                dbg!("d2"); // 11
                scope.spawn(async {
                    dbg!("f2a");
                    slp().await;
                    slp().await;
                    dbg!("f2b");
                });
                dbg!("d3"); // 14
                scope.spawn(async {
                    dbg!("f3a");
                    slp().await;
                    dbg!("f3b");
                });
                dbg!("d4");
                async { dbg!("b1"); } // never executes
            });
            slp().await;
            dbg!("o1");
            let _ = result_fut.await;
        }
        
        fn main() {
            let rt = tokio::runtime::Builder::new_multi_thread()
                .enable_all()
                .build()
                .unwrap();
            rt.block_on(_main())
        }
        
        [src/main.rs:32:5] "o1" = "o1"
        [src/main.rs:7:9] "d1" = "d1"
        [src/main.rs:15:9] "d2" = "d2"
        [src/main.rs:22:9] "d3" = "d3"
        [src/main.rs:28:9] "d4" = "d4"
        [src/main.rs:9:13] "f1a" = "f1a"
        [src/main.rs:17:13] "f2a" = "f2a"
        [src/main.rs:24:13] "f3a" = "f3a"
        [src/main.rs:26:13] "f3b" = "f3b"
        [src/main.rs:20:13] "f2b" = "f2b"
        [src/main.rs:13:13] "f1b" = "f1b"
        

        The non-awaited jobs are run concurrently as the moro docs say. But what if we immediately await f2?

        [src/main.rs:32:5] "o1" = "o1"
        [src/main.rs:7:9] "d1" = "d1"
        [src/main.rs:15:9] "d2" = "d2"
        [src/main.rs:9:13] "f1a" = "f1a"
        [src/main.rs:17:13] "f2a" = "f2a"
        [src/main.rs:20:13] "f2b" = "f2b"
        [src/main.rs:22:9] "d3" = "d3"
        [src/main.rs:28:9] "d4" = "d4"
        [src/main.rs:24:13] "f3a" = "f3a"
        [src/main.rs:13:13] "f1b" = "f1b"
        [src/main.rs:26:13] "f3b" = "f3b"
        

        f1 and f2 are run concurrently, f3 is run after f2 finishes, but doesn’t have to wait for f1 to finish, which is maybe obvious, but… (see below).

        So two things here:

        1. Re-using the spawn terminology here irks me for some reason. I don’t know what would be better though. Would defer_to_scope() be confusing if the job is awaited in the scope?
        2. Even if assumed obvious, a note about execution order when there is a mix of awaited and non-awaited jobs is worth adding to the documentation IMHO.