Investigate implementing embedded-hal traits for UARTE

[jonathanpallant]

I looked at implementing embedded-hal Serial traits for UARTE, but it wasn't immediately obvious how we might do that. All of my implementations so far have been with a hardware FIFO and so writing byte by byte makes sense. Sending a string of 100 bytes as 100 single-byte DMA operations would clearly be sub-optimal.

Has anyone given this any thought? Does it need to just wait for the interrupt-driven UARTE driver?

[jonathanpallant]

Any thoughts on this one? @jamesmunns? @hannobraun?

[hannobraun]

• I think we should implement the byte-by-byte trait, regardless of how bad that implementation is going to be. It's the standard interface, and a bad implementation is better than not supporting the respective drivers at all.
• embedded-hal also has higher-level blocking traits with an opt-in default implementation. For most other targets, that default implementation is pretty optimal, but in our case, we should implement that trait directly.
• I assume that at some point (once we have stable async/await, async fn in traits, and whatever else), it will make sense to add high-level async traits to embedded-hal. Once that happens, we should implement those too, and hope people start using is in favor of the stinking byte-by-byte implementation :-)

[lulf]

I think this issue can be closed now that #343 was merged?