Display impl for Error eats all the useful information

[apoelstra]

The Display impl for Error passes through to std::error::Error::description(), which is pretty-much useless. For example, all I/O errors become simply "os error" through this lens.

[seanmonstar]

Fair point. A nice fix here would be to use the description for hyper errors, but delegate to fmt of errors from other crates.

Example:

impl fmt::Display for Error {
    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
        match *self {
            Io(ref e) | Uri(ref e) | Ssl(ref e) | Http2(ref e) | Utf8(ref e) => fmt::Display::fmt(e, f),
            e => f.write_str(e.description())
        }
    }
}

[apoelstra]

Yeah, I'd be happy with that.

[sfackler]

I tend to go with something like

impl fmt::Display for Error {
    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
        try!(f.write_str(self.description())
        match *self {
            AllThingsWithInnerErrors(ref e) => write!(f, ": {}", e),
            ....
        }
    }
}

[seanmonstar]

With rust-lang/rust#30312 being merged, at least this should state something more than os error in Rust 1.7, which is an improvement...

[xaviershay]

I'm getting conflicting messages from this thread. If the problem is that os error isn't useful, that's coming from fmt of a non-hyper IO error. So handing off error formatting of non-hyper errors (such as IO) to fmt wouldn't actually address the initial example. Right?

[apoelstra]

@xaviershay the problem is that a Display impl is calling description on the std error instead of Display::fmt, so no useful information is propagated, even though it's accessible from the std error.

[seanmonstar]

@xaviershay it's because the description of io::Error before Rust 1.7 simply says "os error". Instead of printing the description, we could print the full fmt::Debug message of io::Error (or, in 1.7, the description will match the Debug output anyways).

[josephpd3]

Hey all, I am a little confused on this issue. @seanmonstar, did your PR rust-lang/rust#30312 resolve the matter or am I not quite understanding? Apologies, I'm still getting used to Rust and would like to help if I can.

[seanmonstar]

@josephpd3 I'm sure that PR improved things. We could make it a little better in hyper by doing as I suggest in the 2nd comment.