process: Windows tests failing on 1.62

[Noah-Kennedy]

With the new version of rust having landed, windows tests are failing. Specifically, it seems that the pipe_from_one_command_to_another test is failing.

I'm currently investigating what exactly is prompting this to fail on windows. Once the root cause is identified, we can figure out how to proceed from there. It may be an std issue or it may be on us.

CI build failure: https://github.com/tokio-rs/tokio/runs/7155496538?check_suite_focus=true#step:8:228

[pie-flavor]

The error message is generated from this change right here
rust-lang/rust#95469

[Noah-Kennedy]

Thanks!

[ChrisDenton]

Hi, the linked PR fixed a soundness issue reported by Microsoft in rust-lang/rust#81357

The short version is that passing async file handles through stdio is unsound. They must be synchronous. The "fix" was to abort the process when stdio reads or writes do not complete synchronously.

This did however have the effect of breaking code. Including within the stdlib itself. A rather inelegant "pipe relay" (rust-lang/rust#95841) had to be added to turn async pipes synchronous.

Iirc in tokio's case it's mio giving out async pipe handles, which will cause problems when passed to stdin or stdout.

See also: #4670 (aka my failed attempt to use more synchronous pipes in stdlib).

[Noah-Kennedy]

Thanks @ChrisDenton!

I actually read that, and just wrote a change that should fix this failing test, however I now need to write up a bunch of documentation changes.

[Noah-Kennedy]

@ChrisDenton I don't see this in the rust release notes, did I miss something, or was it omitted?

[ChrisDenton]

Hm you're right. The pipe relay thing was added to the notes but the issue it was working around wasn't. That appears to have been an oversight. I'll see to fixing that.

[Noah-Kennedy]

Thanks!