#[test] Attribute Macro for embassy-executor #4201
Conversation
| fn #fn_name() { | ||
| let mut executor = ::embassy_executor::Executor::new(); | ||
| let executor = unsafe { ::core::mem::transmute::<_, &'static mut ::embassy_executor::Executor>(&mut executor) }; | ||
| executor.run(|spawner| { |
There was a problem hiding this comment.
this will run forever, the test won't complete even if the task returns.
There was a problem hiding this comment.
Wow, you're so fast! I didn't even have time to make the formatter happy.
You're right, that's actually one of my concerns.
Currently, run doesn't return even when there are no more pending tasks in the queue.
I think this could be fixed by changing how the executor behaves — maybe by implementing a custom fn run_and_return, which would, in addition to polling and waiting, also check if the queue is empty.
From what I can see, Executor and its inner components (raw::Executor, SyncExecutor, RunQueue) doesn't seem to have mechanism to check whether the queue is empty. But it it looks feasible.
That said, I'll need to implement all of that for all arch by myself.
What do you think?
| #[test] | ||
| fn #fn_name() { | ||
| let mut executor = ::embassy_executor::Executor::new(); | ||
| let executor = unsafe { ::core::mem::transmute::<_, &'static mut ::embassy_executor::Executor>(&mut executor) }; |
There was a problem hiding this comment.
this is unsound, the executor really needs &'static mut self. Even if all tasks stop running, they keep pointers to the executors and anything can keep a pointer to a task through a waker, so you can't ensure the executor won't be use-after-free'd.
(it's currently sound due to my other comment (run never returrns) but it won't be when that's fixed)
There was a problem hiding this comment.
A little unsound indeed.
What would be the best approach then? Should I make it static inside the function body using a StaticCell?
That would require the end user to either import static_cell themselves, or for static_cell to be re-exported by Embassy.
|
What is the use case for this? On std, you can test async logic that uses embassy-sync, embassy-time etc just fine using On no-std have you seen https://crates.io/crates/embedded-test ? I'm not sure that we should have this in embassy-executor. It doesn't work on no-std, but embassy-executor is designed for no-std so people will inevitably try to use it on no-std and fail and then try to extend it until it does, at which point we'd have reinvented embedded-test. Doing it properly is hard, I'd rather people use embedded-test. |
Resolves #4192