nemo157 / embrio-rs Goto Github PK

This might be a conceptual misunderstanding on my part, but I believe you can remove the AtomicBool from Waker and this would remove the static lifetime requirement, as the Waker would then have no state - it could be a nullptr and the vtable static.

The key observation is it is not illegal to call poll again on a future that has returned Pending but not invoked the waker, just ill-advised if you can do better. In this case we can't - all the AtomicBool does is move the location of the busy wait.

On armv6m and armv7m the AtomicBool is ultimately duplicating the functionality already provided by the wfe - assuming it isn't actually treated as a noop, in which case busy waiting is the best you can do anyway.

If my reasoning is wrong, I'd be very interested to know as I'm currently working on a PoC embedded_hal, see here, that somewhat relies on this logic being correct...

It would be a fairly small optimization, but would be helpful for users of the library on platforms other than arm. At worst, it's a NOP on a given platform that would be optimized to the same as it is now.

I see there are two timers implemented in https://github.com/Nemo157/embrio-rs/tree/master/embrio-nrf51/src/timer. But as far as I can see they aren't used anywhere?
How is one supposed to use them?

Not sure if this is something that's fixable or not. If something.await is in the arguments to a macro expression, it can't be expanded since syn represents the arguments to a macro invocation as an un-parsed TokenStream.

For example, write!(output, "{}", something.read().await) would result in a "await used outside of an async block" compile error since our macro wouldn't see the something.read().await as an expression or be able to traverse it.

I think the only way to solve this reliably would be to somehow ensure that we're the last macro to be expanded, which doesn't seem to be a thing, but there's a pre-rfc that sounds like what we'd need.