Move `std::io::Error` out of `std`.

[indolering]

Having std::io::Error permeates much of std::io, blocking many crates from being used in no_std environments or re-implementing functionality.

I believe Using std::io::{Read, Write, Cursor} in a nostd environment is the canonical tracking issue.

I'm a Rust newbie 🐣, so my apologies if this ticket is not helpful.

[yaahc]

@indolering are you looking to also avoid having it bring in a dependency on alloc? std::io::Error has a lot of API dependencies on Box.

[indolering]

I am super unqualified to answer that question, but the aforementioned ticket and many of the tickets linking to it address that question. I just know that this is on a lot of wishlists and didn't see it covered in the Zulip chat or other tickets.

[elichai]

I'd love to see this but there are 2 problems I currently see:

1. downcast returns a Box.
2. It means a bunch of the From impls it contain will violate the orphan rule

solving the first one will probably require a backwards incompatible breakage, and solving the second one might not be possible

[sfackler]

It also has a Box internally...

[elichai]

It also has a Box internally...

only in the impl of these two though

[sfackler]

The type io::Error contains a Box inside of it, regardless of what traits it implements: https://github.com/rust-lang/rust/blob/master/library/std/src/io/error.rs#L58-L73

[yaahc]

The type io::Error contains a Box inside of it, regardless of what traits it implements: https://github.com/rust-lang/rust/blob/master/library/std/src/io/error.rs#L58-L73

Wondering if this could maybee be worked around the way panic does with BoxMeUp... Seems sketch tho. I don't think I want to focus on this short term over #3 which completely blocks this issue already.

[indo-dev-0]

Cross referencing:

• Can we move std::error to core? forum thread.
• Consider moving std::error::Error to alloc ticket, which discusses backtrace.
• Move ::std::error to ::core::error PR, which ends with a suggestion to use the Failure crate in alloc-free environments.

[lygstate]

I am trying to move std::io into alloc:io, Error is the most fundenmental peace of code, any prototype design of core::Error are appeared?

[burdges]

I think pub type Error = TinyBox<dyn Error>; should work without alloc using #20 albeit only for a trimmed Error trait without backtrace.

[ColinFinck]

I've raised a few questions if something like core::io::Error can ever be incorporated into Rust at all: rust-lang-nursery/portability-wg#12 (comment)
Please reply and let me know if this is the correct working group after all.

[yaahc]

This is the right place. And to answer your question, I don't know if core::io::Error could ever be a thing. I could see some hacks like the ones we're trying to add to Backtrace to let it have an API that exists in core with the implementation provided by std working for core::io::Error as well, so I don't think it's for sure something we can't do. We already have some prior art of core types (PanicInfo) secretly using a Box as a raw pointer. But we've not seriously considered this yet because we view moving std::error::Error into core and integrating it with panic!s as a higher priority.