1993128 - Subprocess: Pending read of stdout/stderr is discarded on Windows when process exits

Status: RESOLVED FIXED

(Toolkit :: Async Tooling, defect)

Description

[Rob Wu [:robwu]]

When I added test coverage for various scenarios involving subprocess child and stdio pipe communication, I found that stdout is unreadable after process termination, on Windows:

https://searchfox.org/firefox-main/rev/1c6a8b56e6f5f5393ea5ea745ab63ecb67613c93/toolkit/modules/subprocess/test/xpcshell/test_subprocess_child.js#157-163

While reading code, I noticed that there is an implementation flaw in maybeClose: https://searchfox.org/firefox-main/rev/1c6a8b56e6f5f5393ea5ea745ab63ecb67613c93/toolkit/modules/subprocess/subprocess_win.worker.js#177-192

The method is documented as "Closes the pipe if there is a pending read operation with no more buffered data to be read.".
The implementation calls GetOverlappedResult and closes the pipe if it returns FALSE.
But GetOverlappedResult (MSDN) states:

If this parameter is FALSE and the operation is still pending, the function returns FALSE and the GetLastError function returns ERROR_IO_INCOMPLETE.

Since we use async IO, we should expect ERROR_IO_INCOMPLETE and not close the pipe prematurely. When I added the check, the test now behaves like other platforms (macOS/Linux).

Comment 1

[Rob Wu [:robwu]]

Attachment: Bug 1993128 - Don't close InputPipe from maybeClose while IO is pending

Comment 2

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/7f820b9f5375
https://hg.mozilla.org/integration/autoland/rev/3082ff3bf500
Don't close InputPipe from maybeClose while IO is pending r=gstoll

Comment 3

[Atila Butkovits]

https://hg.mozilla.org/mozilla-central/rev/3082ff3bf500