Why doesn't tokio::process::ChildStdin implements IntoRawFd

[NobodyXu]

Since tokio::process::ChildStdin already implements AsRawFd, it seems to be only natural that it also implements IntoRawFd like std::process::ChildStdin.

What is the reason for it to not implement IntoRawFd?

[Darksonn]

Well, the conversion is fallible, which the trait doesn't support.

[NobodyXu]

Is it because the conversion into RawFd requires the O_NONBLOCKING flags to be removed?

[Darksonn]

Yes, but also because its registration with epoll needs to be removed.

[NobodyXu]

OK, I got it.

Is it possible to have some alternative API that can return the fd along with its ownership?

I personally is OK with the fd being O_NONBLOCKING.

[Darksonn]

Sure, we could have an into_std function.

[NobodyXu]

Sure, we could have an into_std function.

I don’t think we can support “into_std” since “ChildStdin” doesn’t support FromRawFd.

[Darksonn]

Ok, I guess that's why we don't have it already. I'm sure we could find some other alternative.

[NobodyXu]

Maybe something like fn to_owned_fd(self) -> Result<io_lifetimes::OwnedFd, io::Error> where io_lifetimes::OwnedFd is implementation of the unstable std::os::unix::io::OwnedFd?

[Darksonn]

We would probably use the same return type as the IntoRawFd trait. We are not going to add a dependency on io_lifetimes for this, nor are we going to define a similar type in Tokio just for the stdio types.

[NobodyXu]

I agree that introduce a new dependency is too much and using “RawFd” is fair enough.

I think having a new unix-only trait for this is the way to go, since “ChildStdin”, “ChildStdout” and “ChildStderr” will all have this method implemented.