Daniel Egger

> Also if we return an iterator it seems to me like the .next() function would hide the bus timing? I'm sorry I don't follow. You don't see the bus...

> I was imagining the returned iterator being returned directly and passed into other read functions or something, but, maybe that's not possible? Oh, in theory it is possible but...

> I'd agree that we don't need size specifications if we were to only allow single reads in a transaction. But that would entirely miss the point of transactions: To...

What do you mean by synchronize? Open PRs on both and then we'll merge them, couldn't be easier. 😅

I think the `flush()` is not only a mistake API-wise but also superfluous: You can only `read()` data when you have `send()` data before, it's the job of the HAL...

> If an implementation only provides the Send trait (e.g. because the miso line isn't connected), there's no other way to determine if the write transfer is done, so we...

> Yeah, but if you don't have an miso line and can thus only implement Send, the implementation shouldn't provide a read function, as there's no real data to read...

> HalfDuplex? No, I'm not reading anything, only writing. > If we had a separate Read trait, then it'd be able to represent a half duplex line by implement Write...

> (2) & (3) are both (usually) a subset of (1). "Officially" (2) and (3) do not exist so they're not a subset but a special case. > Which means...

> Yeah, but then you have the issue that you have to verify all the code you depend on for every single update to make sure it doesn't utilize it....