`.ResolveTarget() | gci -Recurse` buggy when target contains `[]`

[Hashbrown777]

Prerequisites

• [X] Write a descriptive title.
• [X] Make sure you are able to repro it on the latest released version
• [X] Search the existing issues.
• [X] Refer to the FAQ.
• [X] Refer to Differences between Windows PowerShell 5.1 and PowerShell.

Steps to reproduce

cat test.ps1

#!/bin/pwsh
Param($dir = 'A [B C-2]', $link = 'B')
mkdir $dir
touch "$dir/bob"
ln -s $dir $link

try {
         gi -LiteralPath $dir                            | %{ "$($_.GetType().Name)`t$($_.FullName)" }
         gi -LiteralPath $dir                            | gci -Recurse

         gi -LiteralPath $link                           | %{ "$($_.GetType().Name)`t$($_.FullName)" }
         gi -LiteralPath $link                           | gci -Recurse

        (gi -LiteralPath $link).ResolveLinkTarget($True) | %{ "$($_.GetType().Name)`t$($_.FullName)" }
        (gi -LiteralPath $link).ResolveLinkTarget($True) | gci

        (gi -LiteralPath $link).ResolveLinkTarget($True) | %{ "$($_.GetType().Name)`t$($_.FullName)" }
        (gi -LiteralPath $link).ResolveLinkTarget($True) | gci -Recurse
}
finally {
        rm -rf $dir $link
}

Expected behavior

PS> ./test.ps1 'A' 'B'
DirectoryInfo   /home/hashbrown/A
{bob}
DirectoryInfo   /home/hashbrown/B
{bob}
DirectoryInfo   /home/hashbrown/A
{bob}
DirectoryInfo   /home/hashbrown/A
{bob}

PS> ./test.ps1 'A [B C-2]' 'B'
DirectoryInfo   /home/hashbrown/A [B C-2]
{bob}
DirectoryInfo   /home/hashbrown/B
{bob}
DirectoryInfo   /home/hashbrown/A [B C-2]
{bob}
DirectoryInfo   /home/hashbrown/A [B C-2]
{bob}

PS> ./test.ps1 'A [C]' 'B'
DirectoryInfo   /home/hashbrown/A [C]
{bob}
DirectoryInfo   /home/hashbrown/B
{bob}
DirectoryInfo   /home/hashbrown/A [C]
{bob}
DirectoryInfo   /home/hashbrown/A [C]
{bob}

Actual behavior

PS> ./test.ps1 'A' 'B'
DirectoryInfo   /home/hashbrown/A
{bob}
DirectoryInfo   /home/hashbrown/B
{bob}
DirectoryInfo   /home/hashbrown/A
{bob}
DirectoryInfo   /home/hashbrown/A
{bob}

PS> ./test.ps1 'A [B C-2]' 'B'
DirectoryInfo   /home/hashbrown/A [B C-2]
{bob}
DirectoryInfo   /home/hashbrown/B
{bob}
DirectoryInfo   /home/hashbrown/A [B C-2]
DirectoryInfo   /home/hashbrown/A [B C-2]
The specified wildcard character pattern is not valid: A [B C-2]

PS> ./test.ps1 'A [C]' 'B'
DirectoryInfo   /home/hashbrown/A [C]
{bob}
DirectoryInfo   /home/hashbrown/B
{bob}
DirectoryInfo   /home/hashbrown/A [C]
DirectoryInfo   /home/hashbrown/A [C]
<#FREEZE#>

Error details

No response

Environment data

Name                           Value
----                           -----
PSVersion                      7.4.2
PSEdition                      Core
GitCommitId                    7.4.2
OS                             Fedora Remix for WSL
Platform                       Unix
PSCompatibleVersions           {1.0, 2.0, 3.0, 4.0…}
PSRemotingProtocolVersion      2.3
SerializationVersion           1.1.0.1
WSManStackVersion              3.0

Visuals

No response

[mklement0]

The root cause is that - unfortunately - pipeline input to provider cmdlets such as Get-ChildItem is bound (a) invariably as a string and (b) to the -Path rather than the -LiteralPath parameter.

This means that path strings as input - which includes implicitly stringified objects that do not have a .PSPath or .LiteralPath property - are invariably interpreted as wildcard patterns.

Workarounds:

• Wrap the strings in an object with a .LiteralPath (or .PSPath) property; e.g.:

'hi [123].txt' | Select-Object @{ Name='LiteralPath'; Expression={ $_ } } | Get-ChildItem

• More simply, use a delay-bind script block with the -LiteralPath parameter (of course, you can pass strings also directly to -LiteralPath, as arguments, without using the pipeline):

'hi [123].txt' | Get-ChildItem -LiteralPath { $_ }

Note that piping Get-Item / Get-ChildItem directly to another such call does bind to -LiteralPath, courtesy of the .PSPath property added to each output item by the provider cmdlets, which binds pipeline input by property name to the -PSPath parameter, which is simply an alias of -LiteralPath.

[Hashbrown777]

Why doesn't the output of ResolveLinkTarget also have PSPath populated given it purports to be an identical DirectoryInfo?

Delay-bind script blocks is something I thought up and wanted but didn't know was already supported by the language itself. I thought only cmdlets that explicitly optionally ask for scriptblocks (and thus handle it themselves) accepted it (like Sort-Objects -Property), neat!

[Hashbrown777]

Are you saying that the earlier cmdlets are looking ahead in the pipeline and modifying their output instead of the later cmdlets accepting and handling different types of inputs? Like gci only accepts strings in, and without gi changing its output, the pipeline is being cast? Why doesnt gci just accept FileSystemInfo as well as strings and handle those cases appropriately on input?

[mklement0]

Why doesn't the output of ResolveLinkTarget also have PSPath populated given it purports to be an identical DirectoryInfo?

Because, unfortunately, it is only the DirectoryInfo and FileInfo instances emitted by provider cmdlets that have a .PSPath property (among other provider-related instance ETS properties).

This problem would go away if these provider-related ETS properties were defined at the type level, which would also speed up things; see the following issue and the one it links to; the discussion around which stalled a long time ago:

• 4347

[mklement0]

Why doesn't gci just accept FileSystemInfo

That is an excellent question, but you'll have to ask the original designers; I wish it did.

Are you saying that the earlier cmdlets are looking ahead in the pipeline and modifying their output instead of the later cmdlets accepting and handling different types of inputs?

No, no cmdlet does nor is even capable of looking ahead in the pipeline (at least not without substantial manual effort).

Get-ChildItem's pipeline-binding parameters are as follows:

Name        Aliases      ParameterType   PipelineBinding
----        -------      -------------   ---------------
LiteralPath {PSPath, LP} System.String[] By property name
Path        {}           System.String[] By value, By property name

[Hashbrown777]

4347 needs to be done. This revelation has a lot of implications for basically all of my scripts, I thought the whole point of powershell was that using the pipeline was the safest thing to do; you're piping objects, not just strings. Not only converting these objects to strings, but then using those strings as a glob pattern instead of a path was incredibly shortsighted, then needing to work around it using scriptblocks destroys the elegance that pwsh has :(

How many other cmdlet-usecases have caveats akin to this, it would be impossible to know.. I feel like the auto casting in the language itself was a bad idea, and it needed to be obvious to the programmer. Not necessarily mandating manual casting, but maybe when pwsh was starting out it needed a "autocasting permitted" operator.

gci.Resolve() | gci
#Error: Sorry, I only accept strings!

[](gci.Resolve()) | gci
#empty cast declaration, the programmer has opted into auto-casting

gci.Resolve() [|] gci
#some sort of syntax to allow this via the pipeline not only maintaining visual elegance but performance in terms of not blocking the whole generation of the former gci

..alas

[mklement0]

Re potential challenges around Get-ChildItem, ... directly accepting FileInfo / DirectoryInfo input:

• Given that Get-ChildItem is a provider cmdlet, it must accept an open-ended set of types, not just file-system items:
  • Conceivably, an interface that all provider output types must implement would help, but that precludes direct use of FileInfo / DirectoryInfo; while argument-transformation attributes could help with that, it would necessitate creating a different object for each such input object.
  • A - less desirable - interface-less alternative would be to simply accept [object] and look for a PSPath ETS property to identify provider items.
  • In fact, the simple proof-of-concept in https://github.com/PowerShell/PowerShell/issues/6057#issuecomment-400107991 demonstrates the latter approach, albeit without trying to preserve the input objects, as an intermediate fix for the fundamental asymmetry between pipeline and argument parameter binding discussed there.

Apart from that, I think the existing pipeline-based argument binding is sufficient except that discovery of which parameters are pipeline-binding and, if so, how is called for; see:

• 4135