Bikeshed: Should (Embedded) Linux Be Able to Implement These Traits? about embedded-hal HOT 2 CLOSED

The short answer is: yes.

The long answer is: but probably not the lowest level traits because they are tailored for MCUs where non-blocking operations are retry-able without having to track state in the program (the state is in the registers). That form of non-blocking operation is covered by the nb crate and the block!, try_nb! macros.

More concretely apart from the low level traits which are already in tree we want to add traits tailored for blocking operation and another set of traits for interoperation with futures. Where possible we want those traits to built on top of the existing low level traits to avoid rewriting register manipulation code in the implementation of the higher level traits.

I was originally envisioning something like this:

// write half the Serial trait we already have
trait Write<Word> {
    type Error;

    fn write(&self, word: Word) -> nb::Result<(), Self::Error>;
}

// Blocking write operations
trait BWrite {
    type Error;

    fn bwrite_all(&self, buffer: &[u8]) -> Result<(), Self::Error>;
    // ..
}

// overridable blanket implementation
default impl<S> BWrite for S where S: Write<u8> {
    type Error = S::Error;

    fn write_all(&self, buffer: &[u8]) -> Result<(), S::Error> {
        for byte in buffer {
            block!(self.write(*byte))?;
        }
        Ok(())
    }
}

However the default impl doesn't work today because of limitations of the feature (it doesn't work when associated types are involved) which throws a wrench into my plans. However, I think I have come up with a reasonable workaround. I'll test it and send a PR to show the approach.

from embedded-hal.

Closing as this now exists in the form of the linux-embedded-hal crate.

from embedded-hal.