io::reader should expose OS errors to users

[erickt]

io::reader's read_* functions need to expose a mechanism to report an underlying OS errors. Perhaps the read functions should return a result<..., ~str> or option<...>?

[eholk]

@graydon suggested an error handling system a while back that is very similar to LISP's condition system. If we end up adding that, would it be a better way to handle this? IO operations are common enough and errors rare enough that we may not want to have to surround every read with an alt to handle the error. On the other hand, this might encourage programmers to make more robust software.

[graydon]

Yeah, my intention is to sketch this out in this release cycle now that tls is in.

[brson]

I've got a WIP on the new I/O system that overhauls error handling. It will use conditions in combination with 'nullable' return values (Option, empty vectors, sentinel values) and some clever trait implementations to hopefully provide an interface that isn't obnoxious to use while also providing error information and recoverability when it is desired. I'll have more details on the mailing list soon.

[pnkfelix]

I believe brson was forward-referencing this mailing list message:

https://mail.mozilla.org/pipermail/rust-dev/2013-April/003746.html

which has a link to the core io::mod source that talks about error handling strategy:

https://github.com/brson/rust/blob/io/src/libstd/rt/io/mod.rs#L128

See also #6163. Actually, I am just going to close this as I think issue #6163 subsumes it.