error:Translate Error::Timeout into io::ErrorKind::Timedout

[rawler]

This enables better categorization of errors.

In particular, TcpTransport translates ErrorKind to ureq::Error::Timeout errors on timeout conditions. These should then be re-raised as ErrorKind::Timedout errors behind the io::Error abstraction.

Omitting this error-structure forces downstream error-handlers to resort to string-matching to distinguish timeout-errors from other errors.

[rawler]

Not sure this is necessarily the best way to solve it. But in particular, I've discovered some (but not all) errors disguising themselves as Fetch error: IO(Other, Some("timeout: receive response")) in production. In my application, these are kindof expected occasionally, and deserve slightly less aggressive logging than others.

[algesten]

@rawler thanks for submitting this.

I've been thinking back/forth on whether we can land it. In some respects it can be considered a breaking change.

[rawler]

Good reflection. I briefly touched on that thought too, but my thinking is that anyone relying on the old behavior would (hopefully) not be matching on ErrorKind=Other, but on string-matching the message, which I believe to be unchanged.

[rawler]

Are we reaching any verdict? Is there some other approach that would allow us to classify timeout errors (ideally without string-matching)?

[algesten]

I haven't forgotten, just not made up my mind.