More stubs, less warnings - Linux Command Wrapping Part 13
Part 12 ended with a to-do list. Stage 3 was going to be the “implement the remaining stubs” stage - turn those Write-Warning "not yet implemented" placeholders into something that actually runs. It also included two new modules from scratch: SmbShare.Linux and PackageManagement.Linux.
Here is what happened, and where it went sideways.
What was on the list
The stubs that looked genuinely implementable going in:
Find-NetRoute,Get-NetNeighbor,Get-NetIPInterface,Test-NetConnectionand a few more - allip-backed, same tool NetTCPIP.Linux already usedGet-PrintConfiguration,Set-Printer,Rename-Printerand a handful more - CUPS vialpoptionsandlpadminNew-Service,Remove-Service,Set-TimeZone- systemd unit file creation andtimedatectlGet-DiskImage,Clear-Disk,Format-Volume,New-Partitionand the rest of the storage write path -losetup,wipefs,mkfs,sfdisk,fsck
Plus two new modules built from nothing.
The storage batch - where gotchas live
The storage cmdlets looked straightforward on paper. Clear-Disk is wipefs -a. Repair-Volume is fsck. Format-Volume is mkfs.* with a switch on filesystem type.
Two PSSA issues caught me before the tests did, and both were a bit embarrassing.
First: $args is an automatic variable in PowerShell. You cannot assign to it. I had written $args = @('mkfs.ext4') to build the mkfs argument list. PSSA flags this as PSAvoidAssignmentToAutomaticVariable. The fix is obvious - rename to $mkfsArgs - but it is the kind of thing you do not think about until the linter tells you. I have been writing PowerShell for years and I still typed $args = without blinking.
Second: PowerShell string interpolation treats ${sizeK}K as a drive expression when sizeK looks like a drive letter to the parser. PSSA raises InvalidVariableReferenceWithDrive. The fix is string concatenation instead: $partNumStr + ': ,' + $sizeSpec. Inelegant, but clean.
Dismount-DiskImage had a subtler problem. The function needs to look up the loop device for a given image path before it can detach it - which means calling losetup -j <imagepath> first. Fine, except that lookup fires before ShouldProcess. So a test calling Dismount-DiskImage -ImagePath '/tmp/test.img' -WhatIf would throw “‘/tmp/test.img’ is not attached as a loop device” before the What if: line ever printed. The fix: check $WhatIfPreference at the top of the ImagePath branch and skip the losetup lookup entirely, going straight to ShouldProcess with the image path as the target.
Get-DiskImage had another one. When there are no loop devices, losetup -l --json returns {"loopdevices":[]} - an empty array, not null. But foreach ($dev in $data) over an empty array produces $null in a $results = foreach... assignment, not @(). The test captured output and tried $result.GetType().IsArray, which fails on null. Fix: wrap the foreach in @(...) to force array semantics regardless of item count. Not the last time I will make that mistake.
PrintManagement - CUPS is not in WSL2
CUPS is entirely absent from the default Ubuntu WSL2 installation. lpstat, lpoptions, lpadmin - none of them exist on the dev machine.
So what do you test when the tool is not there? I settled on: test the error path, and test WhatIf. If lpoptions is not found, Get-PrintConfiguration should emit an error - test that. If lpadmin is not found, Set-Printer should emit an error - test that. For functions with SupportsShouldProcess, the WhatIf path does not touch the tool at all - ShouldProcess is called before any invocation - so those tests run everywhere.
The previous stub tests expected a Write-Warning from each function. All of those had to go.
Rename-Printer deserves its own note. CUPS has no rename operation. The implementation emulates it by copying the device URI to a new queue via lpadmin -p <newname> -v <deviceuri> -E and then deleting the old queue with lpadmin -x <oldname>. This loses the PPD and per-printer option defaults. The synopsis documents this honestly. It is the best available approximation on Linux - the alternative is an error that just says “not supported,” which is less useful.
The two new modules
PackageManagement.Linux wraps three package managers: apt/apt-cache/dpkg-query for Debian/Ubuntu, dnf/rpm for RHEL/Fedora, zypper for openSUSE. Auto-detection picks the right one at runtime. Call Get-Package on Ubuntu and it finds dpkg-query. Same function, different distro, same output shape. That is the point.
One thing that bit me: the built-in PackageManagement module is available on Linux and exports its own Get-Package. When both modules are loaded, there is a name resolution conflict. PackageManagement.Linux\Get-Package works fine when called with the module qualifier, but a test that called Get-Package without qualification got confused in some Pester contexts depending on import order. The fix was mundane: make the test call it in a way that does not depend on resolution order. The underlying conflict is real, though. Worth documenting.
SmbShare.Linux is more constrained. Samba is not installed in WSL2. Neither is nfsstat. Get-SmbConnection wraps smbstatus -b - which means on most development machines, the integration tests will skip. Get-NfsSession tries /proc/net/rpc/nfs first (kernel-provided, no tools needed) and falls back to nfsstat if the proc file is absent. In WSL2, both are absent, so it errors cleanly.
Get-NfsClientConfiguration reads /etc/nfs.conf and /etc/fstab. Those files exist on any Linux system. The /etc/fstab parsing only picks up NFS-type entries, so on a vanilla WSL2 system it returns an empty array - not an error. An empty result is a valid answer. An error means something went wrong. There is a difference, and it matters.
Update.Linux - what is actually implementable vs. what never will be
Update.Linux already had three real implementations from Stage 1. The remaining 19 were stubs. Looking at the stub list honestly, five of them have sensible Linux equivalents:
Get-WURebootStatus- check/var/run/reboot-requiredand/var/run/reboot-required.pkgs. Debian/Ubuntu writes these files after any package operation that needs a reboot.Get-WULastResults- parse/var/log/apt/history.logfor the last operation’s dates, commandline, installed and upgraded packages.Hide-LinuxUpdate-apt-mark hold <package>. Pin a package.Show-LinuxUpdate-apt-mark unhold <package>. Undo a hold.Remove-LinuxUpdate-apt-get remove, with optional--purge.
The other fourteen - Add-WUServiceManager, Get-WUApiVersion, Reset-WUComponents, Invoke-WUJob and so on - are Windows Update Agent concepts. There is no Linux equivalent of the Windows Update Agent. These are not stubs waiting to be implemented; they are stubs that accurately describe the situation. That is a useful distinction. Both look the same from the outside (Write-Warning "not yet implemented") but represent completely different things. The Windows-only ones should probably say so more explicitly - “This cmdlet requires the Windows Update Agent, which has no Linux equivalent” rather than a generic placeholder.
Numbers
| Module | Version | Tests |
|---|---|---|
| Storage.Linux | v0.6.0 | 489 pass |
| PrintManagement.Linux | v0.2.0 | 18 pass, 16 skip (no CUPS) |
| Update.Linux | v0.3.0 | 116 pass |
| PackageManagement.Linux | v0.1.0 | 16 pass, 1 skip |
| SmbShare.Linux | v0.1.0 | 8 pass, 4 skip (no Samba/NFS) |
All at 0 PSSA issues, 0 failures.
One thing that keeps coming up
Tests that run in WSL2 have a persistent friction: some tools are not installed. CUPS, Samba, NFS tools - all absent from the default Ubuntu image. The skip-with-condition pattern works, but it means a significant portion of the test suite is always skipped on the development machine. You push code, see 16 skipped, and have to remember whether that is expected or a sign of something broken.
That is the motivation for Stage 4: a proper multi-distro test matrix. GitHub Actions running the tests inside containers that actually have CUPS, Samba, NFS tools installed. The skips become real test runs. The gap between “passes on my machine” and “actually works” closes.
Next
Stage 4. Infrastructure. Dockerfiles. CI workflows. Container images at ghcr.io/peppekerstens/. One per distro, each with PowerShell and Pester pre-installed and the relevant Linux tools actually present. Then .github/workflows/pester.yml in each module repo, running the full test suite against the matrix.
The goal: every module tested on five distros on every push. Any skip that fires because a tool is missing from the container is a gap to fill, not a condition to accept.