wip: update to embedded-io

[lulf]

@rmja Currently awaiting rmja/buffered-io#1

[rmja]

I have released buffered-io v 0.2.1 that now works with embedded-io 0.5.
There is one thing that I am not certain about over there, and that is how to handle the new WriteAllError: https://github.com/rmja/buffered-io/blob/master/src/asynch/write.rs#L62
Any good ideas?

[lulf]

For embedded tls i went with propagating it https://github.com/drogue-iot/embedded-tls/blob/main/src/lib.rs#L138-L139

But for buffered.. maybe its ok with zero?

[rmja]

Year, in your case you already have a custom error type in which you can have a variant where the case can be assigned to. I do not have that in my case, as the error type is transparent to the buffering layer. I can see multiple places where this new error type can cause annoyances. I am really uncertain what the best way to handle this case is.

[rmja]

Right now I simply ignore the error and discard the currently buffered bytes which is clearly not correct.

[lulf]

Hmm, yeah, that is problematic. I guess there is no way around creating an error type specific to buffered-io then? Or maybe it would help if ErrorKind had a variant for WriteZero so that it WriteAllError could provide a kind() ?

[rmja]

I have now released buffered-io v. 0.3 which now fixes the outstanding issue where the WriteAllError::ZeroWrite error variant is ignored. It however adds a constraint on the error type to the below transport: https://github.com/rmja/buffered-io/blob/master/src/asynch/write.rs#L29-L30. This is really annoying, as it kind of cascades up to the consumer. I have tried to align reqwless to this change, and it seems to add a-lot of constraints all over the place. For tests that uses Vec as a transport, this also requires rust-embedded/embedded-hal#491.

I have looked at e.g. https://github.com/MabezDev/embedded-fatfs for comparison. It does something similar to circumvent this, but it is really annoying to work with.

[rmja]

I think that the WriteAllError/WriteFmtError error types will be removed again in the next release of embedded-io, at least that seemed to be the consensus at the last wg meeting. Lets wait for a final decision there and a possible release, before we proceed with this.

[lulf]

Agreed, thanks for following up on that!

[rmja]

buffered-io v 0.4 is now released implementing embedded-io 0.6, hence this should not block any further work here.

[lulf]

Continued in #41