2 releases

0.1.1 Oct 10, 2020
0.1.0 Sep 8, 2020

#524 in Asynchronous

Download history 12618/week @ 2024-08-16 12659/week @ 2024-08-23 13722/week @ 2024-08-30 13335/week @ 2024-09-06 12434/week @ 2024-09-13 15746/week @ 2024-09-20 14768/week @ 2024-09-27 14602/week @ 2024-10-04 13622/week @ 2024-10-11 16132/week @ 2024-10-18 13546/week @ 2024-10-25 16325/week @ 2024-11-01 8520/week @ 2024-11-08 5549/week @ 2024-11-15 5641/week @ 2024-11-22 5268/week @ 2024-11-29

28,081 downloads per month
Used in 55 crates (18 directly)

Apache-2.0 OR MIT

8KB
53 lines

spin_on

This crate contains what aims to be the simplest possible implementation of a valid executor. Instead of nicely parking the thread and waiting for the future to wake it up, it continuously polls the future until the future is ready. This will probably use a lot of CPU, so be careful when you use it.

assert_eq!(12, spin_on::spin_on(async {3 * 4}))

The advantages of this crate are:

  • It is really simple
  • It should work on basically any platform
  • It has no dependency on std or on an allocator
  • It only has one dependency

The Design

This crate intentionally violates one of the guidelines of Future: as of Rust 1.46, the runtime characteristics of core::future::Future says:

The poll function is not called repeatedly in a tight loop -- instead, it should only be called when the future indicates that it is ready to make progress (by calling wake()).

When no Future can make progress, a well-behaved executor should suspend execution and wait until an external event resumes execution. As far as I know, though, there is not a cross-platform way to suspend a thread. With Rust's std, this would be done by using thread::park. But, for example, if you're on an embedded board with an ARM Cortex M processor, you would instead use a WFE or WFI instruction. So, an execution-suspending executor would need to be adapted for each different platform.

What price do we pay for violating this guideline? This executor is a "resource hog," since it continually runs the CPU at 100%. On an embedded system, this could cause increased power usage. In a situation where many programs are running, this could make your application waste CPU resources that could otherwise be put to good use by other applications.

When might this be useful?

  • Running async applications on a platform that doesn't have an executor
  • Testing that an async crate works with no_std
  • Educational purposes?
  • Implementing an application where you don't care about performance

License: Apache-2.0 OR MIT

Dependencies

~6KB