Exit tests crash main process on Windows (testing fails)

[jmschonfeld]

Description

When enabling exit tests on Windows, the tests fail indicating that swift.exe unexpectedly exited:

Error: Process completed with exit code 1.

Reproduction

We're seeing this with any exit test we enable on Windows, to reproduce simply create a test that uses #expect(processExitsWith: .failure) and testing will fail with this message

Expected behavior

I'd expect tests to succeed and not crash (except for the expected crashing code) on Windows like the other platforms

Environment

https://github.com/swiftlang/swift-foundation/actions/runs/16126481664/job/45504492458?pr=1406

Swift version 6.2-dev (LLVM 3c4a54b02dd62b7, Swift a40a7be6945a895)
Target: x86_64-unknown-windows-msvc
Build config: +assertions

→ Testing Library Version: 6.3-dev (25b61ef4667a6d1)

Additional information

Note that the same exit tests in https://github.com/swiftlang/swift-foundation/actions/runs/16126481664?pr=1406 pass on Linux and macOS but hit this issue on Windows. I've also tried to reproduce locally on an arm64 Windows VM and have not reproduced so it may be specific to Intel Windows. I've also tried to reproduce with the swift-testing package dependency and it did not reproduce, so it may be specific to the toolchain swift-testing.

[grynspan]

Our own exit tests all pass in CI (x86-64), and they all pass at desk (ARM) as well.

As we do not have a reproducible case, we need more diagnostic information such as a crash log before we can investigate further.

[compnerd]

Is this related to the recent instability on swift.org CI where we saw flakiness in Foundation tests reporting fatal error: exit code 1 or some such? I thought that @bnbarham narrowed that down to a parallelisation issue (in which case, this might just be uncovering a way to make it more reliably prone to failure).

[grynspan]

× Test windowsDirectoryCreationCrash() recorded an issue at FileManagerTests.swift:1104:6: Caught error: CocoaError(code: FoundationEssentials.CocoaError.Code(rawValue: 513), userInfo: ["NSUserStringVariant": ["Remove"], "NSURL": file:///C:/Users/ContainerAdministrator/AppData/Local/Temp/FilePlayground_894CA4EC-3EBF-4EE3-A010-56716F8A4E8B, "NSUnderlyingError": FoundationEssentials.Win32Error(code: 32), "NSFilePath": "C:/Users/ContainerAdministrator/AppData/Local/Temp/FilePlayground_894CA4EC-3EBF-4EE3-A010-56716F8A4E8B"])

So, for the record, you have a test that's failing, so I would expect SwiftPM to terminate with a non-zero status anyway.

[bnbarham]

I thought that @bnbarham narrowed that down to a parallelisation issue

Don't know about "narrowed down" in so much as "it's now not failing with -j 1" 😓

[grynspan]

https://devblogs.microsoft.com/oldnewthing/20101109-00/?p=12323

welp