Powershell with Git Command Error Handling - automatically abort on non-zero exit code from external program

The $ErrorActionPreference variable does not apply to calling external utilities ([console] applications) such as git, invariably in Windows PowerShell and PowerShell (Core) 7 up to 7.3.x, and by default in 7.4.+

• In PowerShell 7.4+, opt-in integration with PowerShell's error handling is available, via the $PSNativeCommandUseErrorActionPreference preference variable.

  • If you set this variable to $true, any external-program call that reports a nonzero exit code automatically triggers a PowerShell error in response, which then integrates with PowerShell's own error handling.
    That is, with the $ErrorActionPreference preference variable set to 'Stop', any external program reporting a nonzero exit code will then result in a script-terminating (fatal by default) error.

There are only two ways to determine success vs. failure of an external utility:

• By examining the automatic $LASTEXITCODE variable, which PowerShell sets to the exit code reported by the external utility.
  By convention, a value of 0 indicates success, whereas any nonzero value indicates failure. (Do note that some utilities, e.g., robocopy.exe, use certain nonzero exit codes to communicate non-error conditions too.)

• If you're not interested in the specific exit code reported, you can examine the Boolean automatic variable $?, which reflects $True for exit code 0, and $False for any nonzero exit code.

  • [Fixed in PowerShell 7.2+] Caveats: As of PowerShell (Core) 7.1, there are two bugs; both these are tentatively fixed via the PSNotApplyErrorActionToStderr experimental feature (!) in preview versions of 7.2 (all experimental features are turned on by default in newly installed preview versions).

    • $? can falsely reflect $false when $LASTEXITCODE is 0 and a stderr redirection (2> or *>) is used and the command emitted stderr output (which in itself doesn't necessarily indicate failure) - see GitHub issue #10512.

      • It is therefore a good habit to form to query only $LASTEXITCODE, not $? to infer failure vs. succcess.

    • If $ErrorActionPreference is set to 'Stop' and you happen to use 2> or *> with an external program and that program produces actual stderr output, $ErrorActionPreference unexpectedly does apply and causes a script-terminating error (whether or not the external program signals actual failure via a nonzero exit code) - see GitHub issue #4002

• If you want to capture stderr output lines, you have to redirect them to a file as of PowerShell 7.2.x (e.g., 2>errs.txt; see caveat above re PowerShell 7.1 and below); GitHub issue #4332 suggests introducing a syntax that allows in-memory capturing, in a variable. However, there is an indirect way of capturing them in memory, namely by merging stderr into the success output stream with 2>&1, which allows you to filter lines that came from stderr by the .NET type of the objects in the success output stream - see this answer.

Acting on a failure requires explicit action, typically by using the Throw keyword to generate a script-terminating error.

Clearly, checking $LASTEXITCODE / $? after every external-utility call is cumbersome, so here's a wrapper function that facilitates this process:

Note: While this function works, it does not try to compensate for the broken-up-to-at-least-v7.2.x passing of arguments with embedded double quotes to external programs. To get this compensation, you can use the iee function from the Native module, installable from the PowerShell Gallery via Install-Module Native.

function Invoke-Utility {
<#
.SYNOPSIS
Invokes an external utility, ensuring successful execution.

.DESCRIPTION
Invokes an external utility (program) and, if the utility indicates failure by 
way of a nonzero exit code, throws a script-terminating error.

* Pass the command the way you would execute the command directly.
* Do NOT use & as the first argument if the executable name is not a literal.

.EXAMPLE
Invoke-Utility git push

Executes `git push` and throws a script-terminating error if the exit code
is nonzero.
#>
  $exe, $argsForExe = $Args
  # Workaround: Prevents 2> redirections applied to calls to this function
  #             from accidentally triggering a terminating error.
  #             See bug report at https://github.com/PowerShell/PowerShell/issues/4002
  $ErrorActionPreference = 'Continue'
  try { & $exe $argsForExe } catch { Throw } # catch is triggered ONLY if $exe can't be found, never for errors reported by $exe itself
  if ($LASTEXITCODE) { Throw "$exe indicated failure (exit code $LASTEXITCODE; full command: $Args)." }
}

Now you just need to prepend Invoke-Utility  to all your git calls, and if any of them reports a nonzero exit code, the script is aborted.

If that is too verbose, define an alias for your function: Set-Alias iu Invoke-Utility, in which case you only need to prepend iu :

iu git init
iu git remote add origin '%env.gitFolder%'
iu git fetch
# ...

I believe the problem here is that the error thrown by git is not trappable by PS.

Illustration:

try {
    git push
    Write-Host "I run after the git push command"
}
catch {
    Write-Host "Something went wonky"
}

Note the missing Write-Host from the catch block!

This is where we need to look at the exit code of the git commands.

The easiest way (I know) in PowerShell is to check the value of $? (more information on $? here: What is `$?` in Powershell?)

try {
    git push
    if (-not $?) {
        throw "Error with git push!"
    }
    Write-Host "I run after the git push command"
}
catch {
    Write-Host "Something went wonky"
    throw
}

Check our custom error (now caught by the catch block)!