ginkgo refuses to run pre-compiled test binary on Windows

[jgrieger]

If I run ginkgo build \my-package-dir on Windows (21H2), the output is my-package.test binary.

If I run ginko my-package.test, the validation in https://github.com/onsi/ginkgo/blob/7161a9d90a24fd70df1e687b71d4ae186ec85dd7/ginkgo/internal/test_suite.go#L195C33-L195C33 fails, because Go does not provide the executable bit in Windows (see golang/go#41809).

The workaround is to rename the test binary to my-package.exe.

This is related to #529.

[onsi]

hey @jgrieger are you up for a PR for this? i think my preferred approach here would be to only enforce the executable bit check if the OS is not windows, but i don’t have a windows machine so it’s possible something else gets in the way further downstream - if you could try to put something together and validate that it works for you that would be great.

[jgrieger]

hey @jgrieger are you up for a PR for this? i think my preferred approach here would be to only enforce the executable bit check if the OS is not windows, but i don’t have a windows machine so it’s possible something else gets in the way further downstream - if you could try to put something together and validate that it works for you that would be great.

Sure, working on it.

[onsi]

thanks!

[jgrieger]

@onsi I think we have two options here:

1. simply omit the bit check for Windows (runtime.GOOS == "windows")
2. use the Go package golang.org/x/sys/windows for calling GetBinaryTypeW function of Windows' kernel32.dll (I created a POC yesterday, for a test binary built by Ginkgo I received SCS_64BIT_BINARY => "A 64-bit Windows-based application.")

Option 1

Pro

• simple, less error-prone
• fastest

Contra

• Might let pass non-executable files on Windows

Option 2

Pro

• ensures the file in question is recognized as executable by the OS
• workaround for os: executable FileMode bit is not set on Windows golang/go#41809
• required package (golang.org/x/sys) already imported in Ginkgo

Contra

• might take longer in terms of execution time (didn't do any perf checks yet)
• P/invoke into OS with unsafe pointers etc. does never feel right 😂

What do you think?

[onsi]

hey @jgrieger thanks for taking the time to explore options. I’d go the simpler route - option 1 - those non-executable files aren’t actually going to run any tests and succeed and so the user will get an error even if the file in question gets past the guard. Thanks!

[jgrieger]

PR #1301 created.

[onsi]

merged it in, thanks! i’ll cut a release soon.

[jgrieger]

Great, thanks!

[onsi]

2.13.1 is out now. You can close this out if you're happy with where we've ended up. Thanks for the PR!

[jgrieger]

Thank you for your support, much appreciated :-).