为了达到最大并行度，必须编写类似以下的内容

let _1_2 = join!(_1, _2);
let (_3, _4) = select! {
    _3 = _1_2 => {
        (_3, _4.await)
    }
    _4 = _4 => {
        let (_1, _2) = _1_2.await;
        (sum(_1, _2).await, _4)
    }
}
let _7 = sum(_3, _4).await;

代码非常晦涩，如果任务和依赖关系更多，手动调度会很快变得令人疲倦，如果可能的话。

使用 async_dag，可以编写

use async_dag::Graph;

async fn sum(lhs: i32, rhs: i32) -> i32 { lhs + rhs }

async fn run() {
    let mut graph = Graph::new();
    // The closures are not run yet.
    let _1 = graph.add_task(|| async { 1 } );
    let _2 = graph.add_task(|| async { 2 } );
    let _4 = graph.add_task(|| async { 4 } );

    // Sets `_1` as `_3`'s first parameter.
    let _3 = graph.add_child_task(_1, sum, 0).unwrap();
    // Sets `_2` as `_3`'s second parameter.
    graph.update_dependency(_2, _3, 1).unwrap();

    // Sets `_3` as `_7`'s first parameter.
    let _7 = graph.add_child_task(_3, sum, 0).unwrap();
    // Sets `_4` as `_7`'s second parameter.
    graph.update_dependency(_4, _7, 1).unwrap();

    // Runs all the tasks with maximum possible parallelism.
    graph.run().await;

    assert_eq!(graph.get_value::<i32>(_7).unwrap(), 7);
}

use futures::executor::block_on;
block_on(run());

快速失败的图

TryGraph 可以用于用户想要有故障快速策略的可失败任务。

当任何任务完成带有 Err 时，它将终止运行futures。