BufRead for ring buffers

[darkwater]

I'm writing a crate that parses data directly out of an embassy_stm32 BufferedUartRx. This goes wrong when the internal ring buffer wraps around, causing only the "front" slice to be returned by BufRead::fill_buf() until it is consumed, meaning I can't avoid allocating a second buffer for parsing. I wonder if we can do better than that.

Initially I thought that BufRead shouldn't be implemented for ring buffers, and this implementation would have to move around its contents to keep them contiguous, but that's probably not a good idea. Besides, std implements BufRead for VecDeque.

One idea would be an additional RingBufRead trait, since ring buffers are common enough, especially on embedded:

pub trait RingBufRead: BufRead {
  // mimicking VecDeque::as_slices()
  async fn fill_buf_slices(&mut self) -> Result<(&[u8], &[u8]), Self::Error>;
}

Or, this method could be added to BufRead with a default implementation that calls fill_buf and returns (fill_buf_result, &[]).

Thoughts?

[darkwater]

I implemented a proof of concept here (embassy) and here (embedded-hal). I'm also working on a utility for nom to deal with split buffers.