Crates.io | spawns-executor |
lib.rs | spawns-executor |
version | 0.1.1 |
source | src |
created_at | 2024-05-04 11:50:54.081493 |
updated_at | 2024-05-04 16:13:00.794787 |
description | Async executors for thread context task spawner |
homepage | https://github.com/kezhuw/spawns |
repository | https://github.com/kezhuw/spawns |
max_upload_size | |
id | 1229555 |
size | 12,202 |
Thread context task spawner for Rust to ease async runtime agnostic coding.
Currently, Rust does not have a standard async runtime. This exposes us a dilemma to choose one and makes creating runtime agnostic library pretty hard. The most challenging thing we have to face is how to spawn task ?
spawns-core proposes a thread context task spawner for Rust std
and async runtimes. Once delivered, we are able to spawn tasks in runtime agnostic manner. Together with other runtime agnostic io, timer, channel and etc. crates, we are capable to write runtime agnostic code easily.
/// Thin wrapper around task to accommodate possible new members.
#[non_exhaustive]
pub struct Task {
pub id: Id,
pub name: Name,
pub future: Box<dyn Future<Output = ()> + Send + 'static>,
}
/// Trait to spawn task.
pub trait Spawn {
fn spawn(&self, task: Task);
}
/// Scope where tasks are [spawn]ed through given [Spawn].
pub struct SpawnScope<'a> {}
/// Enters a scope where new tasks will be [spawn]ed through given [Spawn].
pub fn enter(spawner: &dyn Spawn) -> SpawnScope<'_>;
Async runtimes have to do two things to accommodate for other runtime agnostic API.
Spawn
to spawn asynchronous task.enter
in all executor threads.impl<T> JoinHandle<T> {
/// Gets id of the associated task.
pub fn id(&self) -> Id {}
/// Cancels associated task with this handle.
///
/// Cancellation is inherently concurrent with task execution. Currently, there is no guarantee
/// about promptness, the task could even run to complete normally after cancellation.
pub fn cancel(&self) { }
/// Attaches to associated task to gain cancel on [Drop] permission.
pub fn attach(self) -> TaskHandle<T> { }
}
impl<T> Future for JoinHandle<T> {
type Output = Result<T, JoinError>;
}
/// Spawns a new task.
///
/// # Panics
/// 1. Panic if no spawner.
/// 2. Panic if [Spawn::spawn] panic.
pub fn spawn<T, F>(f: F) -> JoinHandle<T>
where
F: Future<Output = T> + Send + 'static,
T: Send + 'static;
The API is capable to spawn, join and cancel tasks as what tokio
, smol
and async-std
do.
GlobalAlloc
.try_id()
will return None
. I guess we could provides function to wrap a bit.no_std
? No, it needs thread_local!
currently. We can move this to #[thread_local]
once stabilized.spawn_local
for !Send
future ? No, at least for now. I saw only async-global-executor
is capable to spawn_local
freely. Personally, I think it is Rust's responsibility to not treat futures owning !Send
as !Send
. This way there are little chance for us to create !Send
futures. For futures that capturing !Send
in first place and storing thread local !Send
, they need current thread executor.Package spawns-core
uses linkme
to detect available async runtimes from external packages. Package spawns-compat
provides features tokio
, smol
and async-global-executor
to detect async runtimes for spawns-core
. smol
and async-global-executor
can't coexist as they don't have tokio::runtime::Handle::try_current()
like method to detect thread context aware executor.
Package spawns-executor provides full functional block_on
with both current thread executor and multi-thread executor.
See examples. A minimum runtime agnostic echo server is listed here for demonstration.
use async_net::*;
use futures_lite::io;
pub async fn echo_server(port: u16) {
let listener = TcpListener::bind(("127.0.0.1", port)).await.unwrap();
println!("Listen on port: {}", listener.local_addr().unwrap().port());
let mut echos = vec![];
let mut id_counter = 0;
loop {
let (stream, remote_addr) = listener.accept().await.unwrap();
id_counter += 1;
let id = id_counter;
let handle = spawns::spawn(async move {
eprintln!("{:010}[{}]: serving", id, remote_addr);
let (reader, writer) = io::split(stream);
match io::copy(reader, writer).await {
Ok(_) => eprintln!("{:010}[{}]: closed", id, remote_addr),
Err(err) => eprintln!("{:010}[{}]: {:?}", id, remote_addr, err),
}
})
.attach();
echos.push(handle);
}
}
All you have to do for it to be function is setting up thread context task spawner.