process: Windows tests failing on 1.62 #4801
Comments
|
The error message is generated from this change right here |
|
Thanks! |
|
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). |
|
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. |
…implementation This fixes #4801, where, as a result of rust-lang/rust#95469, our implementation of cat used for this test no longer works, as stdio functions on windows now can abort the process if the pipe is set to nonblocking mode. Unfortunately in windows, setting one end of the pipe to be nonblocking makes the whole thing nonblocking, so when, in tokio::process we set the child pipes to nonblocking mode, it causes serious problems for any rust program at the other end. Fixing this issue is for another day, but fixing the tests is for today.
…implementation (#4803) This fixes #4801, where, as a result of rust-lang/rust#95469, our implementation of cat used for this test no longer works, as stdio functions on windows now can abort the process if the pipe is set to nonblocking mode. Unfortunately in windows, setting one end of the pipe to be nonblocking makes the whole thing nonblocking, so when, in tokio::process we set the child pipes to nonblocking mode, it causes serious problems for any rust program at the other end. Fixing this issue is for another day, but fixing the tests is for today.
|
@ChrisDenton I don't see this in the rust release notes, did I miss something, or was it omitted? |
|
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. |
|
Thanks! |
With the new version of rust having landed, windows tests are failing. Specifically, it seems that the
pipe_from_one_command_to_anothertest 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
The text was updated successfully, but these errors were encountered: