Consider Migrating to Failure crate

[stephanbuys]

I had an opportunity to play with the failure crate a couple of weeks ago and was thoroughly impressed. I believe it would be a worthwhile exercise pre 1.x to migrate to it.

[KodrAus]

I absolutely agree 👍 thanks for bringing this up! Instead of attempting to translate errors from one type to another we can just wrap them with Contexts and reduce a lot of boilerplate and leaking of error variants.

[KodrAus]

This would also be a bit nicer for adding the correlation id to errors coming out of the Sender machinery and getting the Send + Sync bounds consistent. Overall it should be a big win.

[mwilliammyers]

I am fairly new to Rust and the whole ecosystem, but it seems like there is somewhat of a debate about the future of failure especially considering that fixing std::error::Error has recently seen some new momentum (it looks like).

Should we just use the downcasting support that just landed in 1.33 or go forward with Failure?

In the meantime, I opened #340 and will continue to improve errors as I run into them.

[KodrAus]

@mwilliammyers Yeh the state of the world is a little different now. failure is still a nice library for error management in applications (where you can easily #derive(Fail)]), but some of the holes in std::error::Error for libraries have been filled.

Sticking with std::error::Error for now seems ok. And thanks for working on improving the API errors!

[KodrAus]

We now use the new error::Error APIs for properly downcastable errors so I'll go ahead and close this one. We can open new issues for other deficiencies with the errors.