ThrottleLimit calculation is incorrect on systems with more than one physical processor, breaking the build. (Added bonus: inf2cat travels through time to break the build)

[gsuberland]

The ThrottleLimit argument is currently implemented as follows:

Windows-driver-samples/Build-SampleSet.ps1

Lines 20 to 25 in 7bc5255

	$ThrottleFactor = 5
	$LogicalProcessors = (Get-CIMInstance -Class 'CIM_Processor' -Verbose:$false).NumberOfLogicalProcessors
	if ($ThrottleLimit -eq 0) {
	$ThrottleLimit = $ThrottleFactor * $LogicalProcessors
	}

However, this CIM query actually returns an array of values representing the logical processor cound per physical socket, rather than just a single integer:

PS C:\Users\Graham> $procs = (Get-CIMInstance -Class 'CIM_Processor' -Verbose:$false).NumberOfLogicalProcessors
PS C:\Users\Graham> $procs.GetType()

IsPublic IsSerial Name                                     BaseType
-------- -------- ----                                     --------
True     True     Object[]                                 System.Array

PS C:\Users\Graham> $procs
56
56

As such, attempting to run the build without specifying the -ThrottleLimit argument results in an error when the script attempts to multiply the array by 5:

InvalidOperation: C:\Users\Graham\source\repos\Windows-driver-samples\Build-SampleSet.ps1:24
Line |
  24 |      $ThrottleLimit = $ThrottleFactor * $LogicalProcessors
     |      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
     | Method invocation failed because [System.Object[]] does not contain a method named 'op_Multiply'.

The code should probably be something like this instead:

$LogicalProcessorsList = (Get-CIMInstance -Class 'CIM_Processor' -Verbose:$false).NumberOfLogicalProcessors
$LogicalProcessors = ($LogicalProcessorsList | Measure-Object -Sum).Sum

Except... this doesn't really work. For some inexplicable reason, making this change causes the powershell script itself to work just fine without needing to specify -ThrottleLimit, but suddenly every single target fails to build due to inf2cat.exe exiting with a bad return code. Example error message:

Windows-driver-samples\packages\Microsoft.Windows.WDK.x64.10.0.26100.4204\c\build\10.0.26100.0\WindowsDriver.common.targets(1432,5): error MSB6006: "inf2cat.exe" exited with code -2. [C:\Users\Graham\source\repos\Windows-driver-samples\tools\dv\samples\DV-FailDriver-WDM\driver\defect_toastmon.vcxproj]

Initially I thought that it was something downstream not liking that I got rid of it being an array, so I swapped the code to this:

$LogicalProcessorsList = (Get-CIMInstance -Class 'CIM_Processor' -Verbose:$false).NumberOfLogicalProcessors
$LogicalProcessors = @(($LogicalProcessorsList | Measure-Object -Sum).Sum)

This did not fix it, but instead I noticed that a completely random subset of projects would fail to build each time, seemingly nondeterministically.

I looked into the log files and found the following:

  Building 'defect_toastmon' with toolset 'WindowsKernelModeDriver10.0' and the 'Universal' target platform.
  Stamping x64\Debug\defect_toastmon.inf
  Stamping [Version] section with DriverVer=07/24/2025,0.32.47.39
  defect_toastmon.c
  wmi.c
  Generating Code...
  defect_toastmon.vcxproj -> C:\Users\Graham\source\repos\Windows-driver-samples\tools\dv\samples\DV-FailDriver-WDM\driver\x64\Debug\defect_toastmon.sys
  Done Adding Additional Store
  Successfully signed: C:\Users\Graham\source\repos\Windows-driver-samples\tools\dv\samples\DV-FailDriver-WDM\driver\x64\Debug\defect_toastmon.sys
  
  Driver is 'Universal'.
  ........................
  Signability test failed.
  
  Errors:
  22.9.7: DriverVer set to a date in the future (postdated DriverVer not allowed) in defect_toastmon\defect_toastmon.inf.
  
  Warnings:
  None
C:\Users\Graham\source\repos\Windows-driver-samples\packages\Microsoft.Windows.WDK.x64.10.0.26100.4204\c\build\10.0.26100.0\WindowsDriver.common.targets(1432,5): error MSB6006: "inf2cat.exe" exited with code -2. [C:\Users\Graham\source\repos\Windows-driver-samples\tools\dv\samples\DV-FailDriver-WDM\driver\defect_toastmon.vcxproj]

Somehow these fixes are making inf2cat think the DriverVer is... in the future? What? I'm now deeply confused. I feel like maybe parallelism is somehow breaking the order of operations here and resulting in a weird race condition with inf2cat, but I'm very much questioning my sanity here.

[gsuberland]

Investigating further, this seems to persistently break the build. I had to nuke the whole directory, git clone and restore packages again, and only then would it build successfully again (with the original powershell code, but manually supplying -ThrottleLimit 112).

Applied the powershell fix again, and ran it without the -ThrottleLimit argument... it works fine. Huh.

Ran the build again for good measure, and it's still fine.

Ran the build again... no such luck, it falls over. Same issue - bunch of projects randomly having inf2cat think it's in the future.

My suspicion is that since I have 112 logical processors on this box, and it multiplies that by five, I end up with so many parallel build processes that something randomly trips a race condition somewhere in the toolchain. My guess is that this is exceedingly rare on the first build because there's more work to do, but much more common on rebuilds because the work completes fast enough to trigger whatever the race condition is.