std: IntoInnerError into_parts, NoStorageSpace

[ijackson]

Hi. I was doing something with std and BufWriterr and I discovered that after getting an IntoInnerError I wasn't able to get an owned copy of the contained io::Error. That seemed wrong, so I set about making an MR to fix it. That is the 2nd commit here.

While doing this I wanted to write a doctest. I wasn't able to think of a reliable portable way of provoking a convenient io error but on Linux at least we have /dev/full so I used that. I think I have marked the doctest with the appropriate attribute. I guess I'll see what the CI says.

But the doctest wants to say something about the actual error, and unaccountably ErrorKind lacks a variant corresponding to ENOSPC. Given that the docs say that existing errors of kind Other may change into specific errors, I thought this was not a breaking change.

I have made both of these things insta-stable. That will definitely need an fcp at least, and also this is my first attempt to (1) do a platform-specific doctest and (2) provide an insta-stable method on something in std, so it will need some review from appropriate people!

Additionally, since I know nothing about Windows I have not added anything to do with ErrorKind::NoStorageSpace there. I don't know if this ought to be a blocker. My git grep also found some stuff to do with vxworks; I don't know if that has ENOSPC.

Thanks for your attention.

[rust-highfive]

r? @m-ou-se

(rust_highfive has picked a reviewer for you, use r? to override)

[jyn514]

I would expect this to be two different PRs, especially since both of these are insta-stable. That said, I'm not on libs, so the team may feel differently.

[jyn514]

1.47 is stable and 1.49 is the current nightly.

Suggested change

	#[stable(feature = "io_error_no_storage_space", since = "1.51.0")]
	#[stable(feature = "io_error_no_storage_space", since = "1.49.0")]

[ijackson]

Right. I'm a bit pessimistic about when this might go in but if it seems right in principle I will change it obviously.

[jyn514]

$ errno ENOSPC
ENOSPC 28 No space left on device

Suggested change

	ErrorKind::NoStorageSpace => "no storage space",
	ErrorKind::NoStorageSpace => "no space left on device",

[ijackson]

Thanks for the suggestion. I'm happy to change the message to be identical to the traditional Unix one if that seems better. But most of the existing messages do not mirror the traditional Unix ones.

[ijackson]

I would expect this to be two different PRs, especially since both of these are insta-stable. That said, I'm not on libs, so the team may feel differently.

If it would be two PRs, one of them would depend on the other. I don't know what the usual way is to handle that situation in rust-lang. I'm happy to go with whatever the usual approach is, obviously.

Thanks for the review, anyway. Let's see what the libs folk think.

[m-ou-se]

If you use a local buffer instead of a File, you can make the example cross platform:

	let mut not_enough_space = [0u8; 10];
	let mut stream = BufWriter::new(not_enough_space.as_mut());
	write!(stream, "this does not fit entirely in the array").unwrap();
        ...

So then you don't need ErrorKind::NoStorageSpace here.

(If you think ErrorKind::NoStorageSpace is useful by itself, please open a separate PR for that.)

[ijackson]

Mara Bos writes ("Re: [rust-lang/rust] std: IntoInnerError into_parts, NoStorageSpace (#78689)"):

If you use a local buffer instead of a File, you can make the example cross platform: let mut not_enough_space = [0u8; 10]; let mut stream = BufWriter::new(not_enough_space.as_mut()); write!(stream, "this does not fit entirely in the array").unwrap(); ... So then you don't need ErrorKind::NoStorageSpace here.

Thanks. I don't know why I didn't think of that. Too much experience of C I guess. I notice that this produces ErrorKind::WriteZero. I think that should be documented, so I will send probably another MR for that.

(If you think ErrorKind::NoStorageSpace is useful by itself, please open a separate PR for that.)

Willdo. Thanks, Ian.

…

-- Ian Jackson <ijackson@chiark.greenend.org.uk> These opinions are my own. Pronouns: they/he. If I emailed you from @fyvzl.net or @evade.org.uk, that is a private address which bypasses my fierce spamfilter.

[crlf0710]

@ijackson Ping from triage! Any updates on this?

[ijackson]

This MR will be confusing If I reuse it for either of the two halves. So I am going to close it and redo this with two separate MRs for the two halves.

Thanks.