Thank you!!!

@tianon @TBBle @corhere FYI; you seem like candidates that are interested in (reviewing) this 😅 🤗

Need more work on windows build side: https://ci-next.docker.com/public/blue/organizations/jenkins/moby/detail/PR-43431/1/pipeline/282#step-283-log-343

[2022-03-26T18:31:09.692Z] At D:\gopath\src\github.com\docker\docker@tmp\durable-4df80df9\powershellWrapper.ps1:3 char:1
[2022-03-26T18:31:09.692Z] + & powershell -NoProfile -NonInteractive -ExecutionPolicy Bypass -Comm ...
[2022-03-26T18:31:09.692Z] + ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[2022-03-26T18:31:09.692Z]     + CategoryInfo          : NotSpecified: (cmd\dockerd\bin_windows_amd64.go:String) [], RemoteException
[2022-03-26T18:31:09.692Z]     + FullyQualifiedErrorId : NativeCommandError
[2022-03-26T18:31:09.692Z]  
[2022-03-26T18:31:09.692Z] cmd\dockerd\bin_windows_amd64.go:4: running "goversioninfo": exec: "goversioninfo": executable file not found in %PATH%

But cross is ok: https://github.com/moby/moby/runs/5704937248?check_suite_focus=true

One failure; looks related (but could be due to how the test is run);

Stacktrace
=== RUN   TestVersion
    version_test.go:19: assertion failed: version.Version is ""
--- FAIL: TestVersion (0.03s)

One failure; looks related (but could be due to how the test is run);

Stacktrace
=== RUN   TestVersion
    version_test.go:19: assertion failed: version.Version is ""
--- FAIL: TestVersion (0.03s)

Yes version string is not propagated the same way in the powershell script. Should be ok now.

@thaJeztah Looks unrelated: https://ci-next.docker.com/public/blue/organizations/jenkins/moby/detail/PR-43431/18/pipeline/288#step-289-log-1466

[2022-03-27T09:44:02.016Z] === Failed
[2022-03-27T09:44:02.016Z] === FAIL: github.com/docker/docker/libnetwork/networkdb TestNetworkDBCRUDTableEntries (8.81s)
[2022-03-27T09:44:02.016Z]     networkdb_test.go:303: Entry existence verification test failed for node1(3ed9271ef286)
[2022-03-27T09:44:02.016Z] 2022/03/27 09:42:27 Closing DB instances...

@crazy-max that's a genius way of figuring out the syntax to instruct go-winres to add the message table (as a custom resource, in go-winres parlance) into the resources! Note that windmc was still needed to produce the bin file: the message table extracted from dockerd.exe 20.10.14 was compiled by windmc. As go-winres does not know how to compile message tables from an editable format, windmc would still be needed in the future to recompile the message table if one ever needed to add new entries to it. That's knowledge which could easily be forgotten if the bin file is committed to the repo directly and windmc removed from the build scripts, so if we go that route we should keep event_messages.mc around and document how to recompile it.

I think instead we could use the Manifest-based provider to publish the exact same event schema available in event_messages.mc.

It doesn't look like that would help. https://docs.microsoft.com/en-us/windows/win32/wes/writing-an-instrumentation-manifest says

After writing your manifest, use the message compiler to validate the manifest and generate the resource and header files that you include in your provider.

The message compiler it is referring to is MC.exe, the Windows SDK tool which GNU binutils windmc is a reimplementation of. There's no avoiding the message compiler, it seems.

That's knowledge which could easily be forgotten if the bin file is committed to the repo directly and windmc removed from the build scripts, so if we go that route we should keep event_messages.mc around and document how to recompile it.

💯 agreed; if instructions are needed on how to create this, we should add docs for that or a script (if possible). Not too long ago, I had to do a lot of digging to figure out how some fixtures were created (what options to pass), which was a bit of fun (see #42389)) Its ok to be hacky / minimal, as long as it's easy to discover "how to do this".

Switch the implementation to use go-winres which supports event logging manifests. Also added instructions in case we want to regenerate it. Note that the event_messages.bin targets amd64 arch atm so cross against windows/arm64 might import a bad message table. I will try to generate one for arm64 if necessary.

PTAL @corhere @thaJeztah

Edit: Message table entries can be generated for any windows platform. Tried to run dockerd on a windows 11 arm64 machine as a service:

$ docker buildx bake --set *.args.DOCKER_CROSSPLATFORMS=windows/arm64 cross

Events are registered:

Issue on Jenkins looks unrelated: https://ci-next.docker.com/public/blue/organizations/jenkins/moby/detail/PR-43431/40/pipeline/279#step-280-log-1466

[2022-04-12T11:29:08.873Z] === Failed
[2022-04-12T11:29:08.873Z] === FAIL: github.com/docker/docker/libnetwork/ipam TestRequestReleaseAddressDuplicate (0.03s)
[2022-04-12T11:29:08.873Z]     allocator_test.go:1531: IP 198.168.1.215/23 was previously allocated

Kicked CI earlier, and all green, except for one flaky test on Windows;

=== RUN   TestNetworkDBCRUDTableEntries
    networkdb_test.go:310: Entry existence verification test failed for node2(e7bb4e7b34e3)
2022/04/20 11:21:02 Closing DB instances...
--- FAIL: TestNetworkDBCRUDTableEntries (8.04s)

Let's get this one in