Implement Clone for error enums

[zeenix]

This would allow using crates to embed errors from this crate into their error types that implement Clone.

Unfortunately std::io::Error does not implement Clone so we've to loose some of its state during cloning. The most importat bit (error kind) is preserved though.

[Mingun]

Cloneable errors seems not in a trend in Rust, it is unknown what the long-term consequences of this. I'm not totally against this PR, but I would like to hear what the situations where you need to clone errors?

[zeenix]

Cloneable errors seems not in a trend in Rust, it is unknown what the long-term consequences of this. I'm not totally against this PR, but I would like to hear what the situations where you need to clone errors?

Thanks for the quick response and that's very fair. My own use case: I've a task that owns a socket and reads and parses messages from the socket and then forwards them to all interested parties (streams) through a broadcast channel. Since Result only implements Clone when E: Clone, I can not send a Result the channel. Currently I work around this by using a broadcast channel only for success and a separate channel for error but that means that I only communicate the error to one of the streams.

Please note that in my crate, I'm fixing the std::io::Error issue a bit differently, which won't work for you unless you're ok with breaking API (my error type is non_exhaustive) so I went with the approach in the PR.

[Mingun]

I'm fine with replacing Error::Io(IoError) to Error::Io(Arc<IoError>), which, I think, would be more elegant solution. We also can hide that under a feature flag (say, cloneable-errors), but I think that this is not required (besides, that is anti-pattern, because that create a incompatible API depending on the feature flags, which should be avoided). @dralley, what to you think?

Whatever path we choose, it is important to describe it in the documentation for Error, and add a changelog entry.

I think, that will resolve #360?

[zeenix]

I'm fine with replacing Error::Io(IoError) to Error::Io(Arc<IoError>), which, I think, would be more elegant solution.

I completely agree. I will revise the commits after hearing opinion of @dralley.

Whatever path we choose, it is important to describe it in the documentation for Error,
and add a changelog entry.

For sure. I'll do that as part of the update.

[zeenix]

I think, that will resolve #360?

Not sure. single consumer APIs should not require cloning.

[dralley]

I'm fine with Arc<IoError>. Agreed about not wanting divergent signatures.

[zeenix]

@Mingun Updated the commits. PTALA.

[Mingun]

impl<E: Error> Error for Arc<E> was implemented in Rust 1.52, so need to bump our MSRV. Please add a note about that to the Changelog, before "New Features" section

[zeenix]

Lol, apparently GH wants me to only request review from one person. :) Sorry about adding and removing dance.

[Mingun]

Thanks!