Apparently, in PowerShell (ver. 3) not all $null's are the same:
>function emptyArray() { #() }
>$l_t = #() ; $l_t.Count
0
>$l_t1 = #(); $l_t1 -eq $null; $l_t1.count; $l_t1.gettype()
0
IsPublic IsSerial Name BaseType
-------- -------- ---- --------
True True Object[] System.Array
>$l_t += $l_t1; $l_t.Count
0
>$l_t += emptyArray; $l_t.Count
0
>$l_t2 = emptyArray; $l_t2 -eq $null; $l_t2.Count; $l_t2.gettype()
True
0
You cannot call a method on a null-valued expression.
At line:1 char:38
+ $l_t2 = emptyArray; $l_t2 -eq $null; $l_t2.Count; $l_t2.gettype()
+ ~~~~~~~~~~~~~~~
+ CategoryInfo : InvalidOperation: (:) [], RuntimeException
+ FullyQualifiedErrorId : InvokeMethodOnNull
>$l_t += $l_t2; $l_t.Count
0
>$l_t3 = $null; $l_t3 -eq $null;$l_t3.gettype()
True
You cannot call a method on a null-valued expression.
At line:1 char:32
+ $l_t3 = $null; $l_t3 -eq $null;$l_t3.gettype()
+ ~~~~~~~~~~~~~~~
+ CategoryInfo : InvalidOperation: (:) [], RuntimeException
+ FullyQualifiedErrorId : InvokeMethodOnNull
>$l_t += $l_t3; $l_t.count
1
>function addToArray($l_a, $l_b) { $l_a += $l_b; $l_a.count }
>$l_t = #(); $l_t.Count
0
>addToArray $l_t $l_t1
0
>addToArray $l_t $l_t2
1
So how and why is $l_t2 different from $l_t3? In particular, is $l_t2 really $null or not? Note that $l_t2 is NOT an empty array ($l_t1 is, and $l_t1 -eq $null returns nothing, as expected), but neither is it truly $null, like $l_t3. In particular, $l_t2.count returns 0 rather than an error, and furthermore, adding $l_t2 to $l_t behaves like adding an empty array, not like adding $null. And why does $l_t2 suddenly seem to become "more $null" when it gets passed in the the function addToArray as a parameter???????
Can anyone explain this behaviour, or point me to documentation that would explain it?
Edit:
The answer by PetSerAl below is correct. I have also found this stackOverflow post on the same issue.
Powershell version info:
>$PSVersionTable
Name Value
---- -----
WSManStackVersion 3.0
PSCompatibleVersions {1.0, 2.0, 3.0}
SerializationVersion 1.1.0.1
BuildVersion 6.2.9200.16481
PSVersion 3.0
CLRVersion 4.0.30319.1026
PSRemotingProtocolVersion 2.2
In particular, is $l_t2 really $null or not?
$l_t2 is not $null, but a [System.Management.Automation.Internal.AutomationNull]::Value. It is a special instance of PSObject. It is returned when a pipeline returns zero objects. That is how you can check it:
$a=&{} #shortest, I know, pipeline, that returns zero objects
$b=[System.Management.Automation.Internal.AutomationNull]::Value
$ReferenceEquals=[Object].GetMethod('ReferenceEquals')
$ReferenceEquals.Invoke($null,($a,$null)) #returns False
$ReferenceEquals.Invoke($null,($a,$b)) #returns True
I call ReferenceEquals thru Reflection to prevent conversion from AutomationNull to $null by PowerShell.
$l_t1 -eq $null returns nothing
For me it returns an empty array, as I expect from it.
$l_t2.count returns 0
It is a new feature of PowerShell v3:
You can now use Count or Length on any object, even if it didn’t have the property. If the object didn’t have a Count or Length property, it will will return 1 (or 0 for $null). Objects that have Count or Length properties will continue to work as they always have.
PS> $a = 42
PS> $a.Count
1
And why does $l_t2 suddenly seem to become "more $null" when it gets passed in the the function addToArray as a parameter???????
It seems that PowerShell converts AutomationNull to $null in some cases, like calling .NET methods. In PowerShell v2, even when saving AutomationNull to a variable it gets converted to $null.
To complement PetSerAl's great answer with a pragmatic summary:
Commands that happen to produce no output do not return $null, but the [System.Management.Automation.Internal.AutomationNull]::Value singleton,
which can be thought of as an "array-valued $null" or, to coin a term, null enumeration. It is sometimes also called "AutomationNull", for its type name.
Note that, due to PowerShell's automatic enumeration of collections, even a command that explicitly outputs an empty collection object such as #() has no output (unless enumeration is explicitly prevented, such as with Write-Output -NoEnumerate).
In short, this special value behaves like $null in scalar contexts, and like an empty array in enumeration contexts, notably in the pipeline, as the examples below demonstrate.
Given that $null and the null enumeration situationally behave differently, distinguishing between the two via reflection may be necessary, which is currently far from trivial; GitHub issue #13465 proposes implementing a test that would allow you to use $someValue -is [AutomationNull].
As of PowerShell 7.3.0, the following, obscure test is required:
$null -eq $someValue -and $someValue -is [psobject]
Caveats:
Passing [System.Management.Automation.Internal.AutomationNull]::Value as a cmdlet / function parameter value invariably converts it to $null.
See GitHub issue #9150.
In PSv3+, even an actual (scalar) $null is not enumerated in a foreach loop; it is enumerated in a pipeline, however - see bottom.
In PSv2-, saving a null enumeration in a variable quietly converted it to $null and $null was enumerated in a foreach loop as well (not just in a pipeline) - see bottom.
# A true $null value:
$trueNull = $null
# An operation with no output returns
# the [System.Management.Automation.Internal.AutomationNull]::Value singleton,
# which is treated like $null in a scalar expression context,
# but behaves like an empty array in a pipeline or array expression context.
$automationNull = & {} # calling (&) an empty script block ({}) produces no output
# In a *scalar expression*, [System.Management.Automation.Internal.AutomationNull]::Value
# is implicitly converted to $null, which is why all of the following commands
# return $true.
$null -eq $automationNull
$trueNull -eq $automationNull
$null -eq [System.Management.Automation.Internal.AutomationNull]::Value
& { param($param) $null -eq $param } $automationNull
# By contrast, in a *pipeline*, $null and
# [System.Management.Automation.Internal.AutomationNull]::Value
# are NOT the same:
# Actual $null *is* sent as data through the pipeline:
# The (implied) -Process block executes once.
$trueNull | % { 'input received' } # -> 'input received'
# [System.Management.Automation.Internal.AutomationNull]::Value is *not* sent
# as data through the pipeline, it behaves like an empty array:
# The (implied) -Process block does *not* execute (but -Begin and -End blocks would).
$automationNull | % { 'input received' } # -> NO output; effectively like: #() | % { 'input received' }
# Similarly, in an *array expression* context
# [System.Management.Automation.Internal.AutomationNull]::Value also behaves
# like an empty array:
(#() + $automationNull).Count # -> 0 - contrast with (#() + $trueNull).Count, which returns 1.
# CAVEAT: Passing [System.Management.Automation.Internal.AutomationNull]::Value to
# *any parameter* converts it to actual $null, whether that parameter is an
# array parameter or not.
# Passing [System.Management.Automation.Internal.AutomationNull]::Value is equivalent
# to passing true $null or omitting the parameter (by contrast,
# passing #() would result in an actual, empty array instance).
& { param([object[]] $param)
[Object].GetMethod('ReferenceEquals').Invoke($null, #($null, $param))
} $automationNull # -> $true; would be the same with $trueNull or no argument at all.
The [System.Management.Automation.Internal.AutomationNull]::Value documentation states:
Any operation that returns no actual value should return AutomationNull.Value.
Any component that evaluates a Windows PowerShell expression should be prepared to deal with receiving and discarding this result. When received in an evaluation where a value is required, it should be replaced with null.
PSv2 vs. PSv3+, and general inconsistencies:
PSv2 offered no distinction between [System.Management.Automation.Internal.AutomationNull]::Value and $null for values stored in variables:
Using a no-output command directly in a foreach statement / pipeline did work as expected - nothing was sent through the pipeline / the foreach loop wasn't entered:
Get-ChildItem nosuchfiles* | ForEach-Object { 'hi' }
foreach ($f in (Get-ChildItem nosuchfiles*)) { 'hi' }
By contrast, if a no-output commands was saved in a variable or an explicit $null was used, the behavior was different:
# Store the output from a no-output command in a variable.
$result = Get-ChildItem nosuchfiles* # PSv2-: quiet conversion to $null happens here
# Enumerate the variable.
$result | ForEach-Object { 'hi1' }
foreach ($f in $result) { 'hi2' }
# Enumerate a $null literal.
$null | ForEach-Object { 'hi3' }
foreach ($f in $null) { 'hi4' }
PSv2: all of the above commands output a string starting with hi, because $null is sent through the pipeline / being enumerated by foreach:
Unlike in PSv3+, [System.Management.Automation.Internal.AutomationNull]::Value is converted to $null on assigning to a variable, and $null is always enumerated in PSv2.
PSv3+: The behavior changed in PSv3, both for better and worse:
Better: Nothing is sent through the pipeline for the commands that enumerate $result: The foreach loop is not entered, because the [System.Management.Automation.Internal.AutomationNull]::Value is preserved when assigning to a variable, unlike in PSv2.
Possibly Worse: foreach no longer enumerates $null (whether specified as a literal or stored in a variable), so that foreach ($f in $null) { 'hi4' } perhaps surprisingly produces no output.
On the plus side, the new behavior no longer enumerates uninitialized variables, which evaluate to $null (unless prevented altogether with Set-StrictMode).
Generally, however, not enumerating $null would have been more justified in PSv2, given its inability to store the null-collection value in a variable.
In summary, the PSv3+ behavior:
takes away the ability to distinguish between $null and [System.Management.Automation.Internal.AutomationNull]::Value in the context of a foreach statement
thereby introduces an inconsistency with pipeline behavior, where this distinction is respected.
For the sake of backward compatibility, the current behavior cannot be changed. This comment on GitHub proposes a way to resolve these inconsistencies for a (hypothetical) potential future PowerShell version that needn't be backward-compatible.
When you return a collection from a PowerShell function, by default PowerShell determines the data type of the return value as follows:
If the collection has more than one element, the return result is an array. Note that the data type of the return result is System.Array even if the object being returned is a collection of a different type.
If the collection has a single element, the return result is the value of that element, rather than a collection of one element, and the data type of the return result is the data type of that element.
If the collection is empty, the return result is $null
$l_t = #() assigns an empty array to $l_t.
$l_t2 = emptyArray assigns $null to $l_t2, because the function emptyArray returns an empty collection, and therefore the return result is $null.
$l_t2 and $l_t3 are both null, and they behave the same way. Since you've pre-declared $l_t as an empty array, when you add either $l_t2 or $l_t3 to it, either with the += operator or the addToArray function, an element whose value is **$null* is added to the array.
If you want to force the function to preserve the data type of the collection object you're returning, use the comma operator:
PS> function emptyArray {,#()}
PS> $l_t2 = emptyArray
PS> $l_t2.GetType()
IsPublic IsSerial Name BaseType
-------- -------- ---- --------
True True Object[] System.Array
PS> $l_t2.Count
0
Note: The empty parentheses after emtpyArray in the function declaration is superfluous. You only need parentheses after the function name if you're using them to declare parameters.
An interesting point to be aware of is that the comma operator doesn't necessarily make the return value an array.
Recall that as I mentioned in the first bullet point, by default the data type of the return result of a collection with more than one element is System.Array regardless of the actual data type of the collection. For example:
PS> $list = New-Object -TypeName System.Collections.Generic.List[int]
PS> $list.Add(1)
PS> $list.Add(2)
PS> $list.Count
2
PS> $list.GetType()
IsPublic IsSerial Name BaseType
-------- -------- ---- --------
True True List`1 System.Object
Note that the data type of this collection is List`1, not System.Array.
However, if you return it from a function, within the function the data type of $list is List`1, but it's returned as a System.Array containing the same elements.
PS> function Get-List {$list = New-Object -TypeName System.Collections.Generic.List[int]; $list.Add(1); $list.Add(2); return $list}
PS> $l = Get-List
PS> $l.Count
2
PS> $l.GetType()
IsPublic IsSerial Name BaseType
-------- -------- ---- --------
True True Object[] System.Array
If you want the return result to be a collection of the same data type as the one within the function that you're returning, the comma operator will accomplish that:
PS> function Get-List {$list = New-Object -TypeName System.Collections.Generic.List[int]; $list.Add(1); $list.Add(2); return ,$list}
PS> $l = Get-List
PS> $l.Count
2
PS> $l.GetType()
IsPublic IsSerial Name BaseType
-------- -------- ---- --------
True True List`1 System.Object
This isn't limited to array-like collection objects. As far as I've seen, any time PowerShell changes the data type of the object you're returning, and you want the return value to preserve the object's original data type, you can do that by preceding the object being returned with a comma. I first encountered this issue when writing a function that queried a database and returned a DataTable object. The return result was an array of hashtables instead of a DataTable. Changing return $my_datatable_object to return ,$my_datatable_object made the function return an actual DataTable object.
I'm trying to dynamically parse & build-up a data structure of some incoming JSON files I'm to be supplied with (that'll be in non-standard structure) via Powershell to then process the data in those files & hand them over to the next step.
As part of that, I'm trying to build up the data structure of the JSON file into essentially a list of of data-paths for me to parse through & grab the data out of, so that I can cope with arrays, nested JSON objects and so on. So far so good.
Where I fall into some sort of Powershell peculiarity is in handling 2+ levels of depth via a variable. Let me give you a nice code-block to demonstrate the problem...
# Generate a Quick JSON file with different data types & levels
[object]$QuickJson = #'
{
"Name" : "I am a JSON",
"Version" : "1.2.3.4",
"SomeBool" : true,
"NULLValue" : null,
"ArrayOfVersions" : [1.0,2.0,3.0],
"MyInteger" : 69,
"NestedJSON" : {
"Version" : 5.0,
"IsReady" : false
},
"DoubleNestedJSON" : {
"FirstLevel" : 1,
"DataValue" : "I am at first nested JSON level!",
"Second_JSON_Level" : {
"SecondLevel" : 2,
"SecondDataValue" : "I am on the 2nd nested level"
}
}
}
'#
# Import our JSON file into Powershell
[object]$MyPSJson = ConvertFrom-Json -InputObject $QuickJson
# Two quick string variables to access our JSON data paths
[string]$ShortJsonPath = "Name"
[string]$NestedJsonPath = "NestedJson.Version"
# Long string to access a double-nested JSON object
[string]$LongNestedJsonPath = "DoubleNestedJSON.Second_JSON_Level.SecondDataValue"
# Both of these work fine
Write-Host ("JSON Name (Direct) ==> " + $MyPSJson.Name)
Write-Host ("JSON Name (via Variable) ==> " + $MyPSJson.$ShortJsonPath)
# The following way to access a single nested Json Path works fine
Write-Host ("Nested JSON Version (via direct path) ==> " + $MyPSJson.NestedJson.Version)
# And THIS returns an empty line / is where I fall afoul of something in Powershell
Write-Host ("Nested JSON Version (via variable) ==> " + $MyPSJson.$NestedJsonPath)
# Other things I tried -- all returning an empty line / failing in effect
Write-Host ("Alternate Nested JSON Version ==> " + $($MyPSJson.$NestedJsonPath))
Write-Host ("Alternate Nested JSON Version ==> " + $MyPSJson.$($NestedJsonPath))
Write-Host ("Alternate Nested JSON Version ==> " + $($MyPSJson).$($NestedJsonPath))
# Similarly, while THIS works...
$MyPSJson | select-object -Property NestedJSON
# This will fail / return me nothing
$MyPSJson | select-object -Property NestedJSON.Version
... in doing a bunch of research around this, I came across a suggestion to transform this into a Hashtable -- but that has the same problem, sadly. So with the above code-snippet, the following will transform the JSON object into a hashtable.
# Same problem with a hash-table if constructed from the JSON file...
[hashtable]$MyHash = #{}
# Populate $MyHash with the data from our quickie JSON file...
$QuickJson | get-member -MemberType NoteProperty | Where-Object{ -not [string]::IsNullOrEmpty($QuickJson."$($_.name)")} | ForEach-Object {$MyHash.add($_.name, $QuickJson."$($_.name)")}
# ... and even then -- $MyHash."$($NestedJsonPath)" -- fails, while a single level deep string works fine in the variable! :(
So it's pretty clear that I'm running into "something" of a Powershell internal logic problem, but I can't get Powershell to be overly helpful in WHY that is. Adding a '-debug' or similar in an attempt to increase verbosity hasn't helped shed light on this.
I suspect it's something akin to the items raised in this article here ( https://blogs.technet.microsoft.com/heyscriptingguy/2011/10/16/dealing-with-powershell-hash-table-quirks/ ) but just specific with variables.
I've not had any luck in finding anything obvious in the Powershell language specification (3.0 still being the latest from here as far as I can tell -- https://www.microsoft.com/en-usdownload/details.aspx?id=36389 ) either. It may be in there, I may just miss it.
Any advice in how to get Powershell to play nice with this would be greatly appreciated. I'm not sure how / why Powershell is fine with a simple string but seems to have issues with a 'something.somethingelse' type string here.
Thank you.
Further notes & addenda to the original:
It seems there are several issues to attack. One is "dealing with a single nested level". The "quick fix" for that seems to be using "Invoke-Expression" to resolve the statement, so for instance (IMPORTANT - take note of the back-tick with the first variable!):
iex "`$MyPSJson.$NestedJsonPath"
That use of Invoke-Expression also works with multi-nested situations:
iex "`$MyPSJson.$LongNestedJsonPath"
An alternative approach that was mentioned is the use of multiple select statements ... but I've not been able to get that to work with multi-nested objects (Powershell seems to not resolve those properly for some reason).
So for instance in this scenario:
($MyComp | select $_.DoubleNestedJSON | select FirstLevel)
Powershell returns
FirstLevel
----------
... instead of the actual data value. So - for now, it seems that selects won't work with multi-level nested objects due to Powershell apparently not resolving them?
When you write something like
$MyPSJson.Name
this will attempt to retrieve the member named Name from the object $MyPSJson. If there is no such member, you'll get $null.
Now, when you do that with variables for the member name:
$MyPSJson.$ShortJsonPath
this works pretty much identical in that the member with the name stored in $ShortJsonPath is looked up and its value retrieved. No surprises here.
When you try that with a member that doesn't exist on the object, such as
$MyPSJson.$NestedJsonPath
# equivalent to
# $MyPSJson.'NestedJSON.Version'
you'll get $null, as detailed before. The . operator will only ever access a member of the exact object that is the result of its left-hand-side expression. It will never go through a member hierarchy in the way you seem to expect it to do. To be frank, I'm not aware of a language that works that way.
The reason it works with Invoke-Expression is, that you effectively converting the $NestedJsonPath string into part of an expression resulting in:
$MyPSJson.NestedJSON.Version
which Invoke-Expression then evaluates.
You can, of course, define your own function that works that way (and I'd much prefer that instead of using Invoke-Expression, a cmdlet that should rarely, if ever, used (heck, it's eval for PowerShell – few languages with eval advocate its use)):
function Get-DeepProperty([object] $InputObject, [string] $Property) {
$path = $Property -split '\.'
$obj = $InputObject
$path | %{ $obj = $obj.$_ }
$obj
}
PS> Get-DeepProperty $MyPSJson NestedJson.Version
5,0
You could even make it a filter, so you can use it more naturally on the pipeline:
filter Get-DeepProperty([string] $Property) {
$path = $Property -split '\.'
$obj = $_
$path | %{ $obj = $obj.$_ }
$obj
}
PS> $MyPSJson | Get-DeepProperty nestedjson.version
5,0
Why this doesn't work
When you provide the properties that you'd like within a string, like this
[string]$NestedJsonPath = "NestedJson.Version"
Powershell looks for a property called NestedJSon.Version. It's not actually traversing the properties, but looking for a string literal which contains a period. In fact, if I add a property like that to your JSON like so.
[object]$QuickJson = #'
{
"Name" : "I am a JSON",
"Version" : "1.2.3.4",
"SomeBool" : true,
"NULLValue" : null,
"ArrayOfVersions" : [1.0,2.0,3.0],
"MyInteger" : 69,
"NestedJSON.Version" : 69,
"NestedJSON" : {
"Version" : 5.0,
"IsReady" : false
}
}
I now get a value back, like so:
>$MyPSJson.$NestedJsonPath
69
The best way to get your values back is to use two separate variables, like this.
$NestedJson = "NestedJson"
$property = "Version"
>$MyPSJson.$NestedJson.$property
5.0
Or, alternatively, you could use select statements, as seen in the original answer below.
$MyPSJson | select $_.NestedJSON | select Version
Version
-------
1.2.3.4
If you use multiple Select-Object statements, they'll discard the other properties and allow you to more easily drill down to the value you'd like.
I followed Joey's filter example. However, I found it did not support accessing arrays.
Sharing the code that I got to work for this. Hopefully it will help others as well. Awesome thread!
filter Get-DeepProperty([string] $Property) {
$path = $Property -split '\.'
$obj = $_
foreach($node in $path){
if($node -match '.*\[\d*\]'){
$keyPieces = $node -split ('\[')
$arrayKey = $keyPieces[0]
$arrayIndex = $keyPieces[1] -replace ('\]','')
$obj = $obj.$arrayKey[$arrayIndex]
} else {
$obj = $obj.$node
}
}
$obj
}
Example usage:
$path = "nested.nestedtwo.steps[2]"
$payload | Get-DeepProperty $path
I had the same problem, so I wrote a function that does the trick.
It enables accessing any level of the json by variable path (string):
function getNestedJsonValue() {
param(
[Parameter(Mandatory = $true, ValueFromPipeline)] [PSCustomObject] $inputObj,
[Parameter(Mandatory = $true)] [string] $valuePath
)
if (($valuePath -eq $null) -or ($valuePath.length -eq 0) -or ($inputObj -eq $null)) {
return $inputObj
}
[System.Array] $nodes = "$valuePath" -split '\.'
foreach ($node in $nodes) {
if (($node -ne $null) -and ($node.length -gt 0) -and ($inputObj -ne $null)) {
$inputObj = $inputObj.$node
} else {
return $inputObj
}
}
return $inputObj
}
Usage: getNestedJsonValue -valuePath $nestedValuePath -inputObj $someJson
Pipe usage: $someJson | getNestedJsonValue -valuePath $nestedValuePath
An example nestedValuePath would be $nestedValuePath="some.nested.path"
Credit to wOxxOm for getting things on the right track.
Invoke-Expression does seem to work perfectly for this situation (if somewhat expensive, but that's fine in my personal example & situation), and it can cope with multiple levels of nesting.
So as examples for the above code snippet, the following will resolve just fine (Key point - pay attention to the initial back-tick. That caught me off guard):
Write-Host ("Single level JSON test ==> " + (iex "`$MyPSJson.$NestedJsonPath"))
Write-Host ("Double level JSON test ==> " + (iex "`$MyPSJson.$LongNestedJsonPath"))
That'll return our desired results:
Single level JSON test ==> 5.0
Double level JSON test ==> I am on the 2nd nested level
FoxDeploy's answer of using multi-level selects doesn't seem to work with 2+ levels of nesting, unfortunately for some bizarre reason.
Using:
($MyPSJson | select $_.DoubleNestedJSON | select FirstLevel)
We get the following back from Powershell:
FirstLevel
----------
... it seems that Powershell doesn't resolve nested objects in its entirety? We get a similar results if we intentionally use something that doesn't exist:
($MyPSJson | select $_.DoubleNestedJSON | select Doesnotexist)
... also simply returns:
Doesnotexist
------------
So - for now - it seems as if "Invoke-Expression" works most reliably (and most easily, as it's just a case of handing it a variable with the path'ed string).
I still can't explain the WHY of any of this so far (since I've used 'dotwalk'-ing with multiple variables through arrays quite happily), but at least there's a solution for now ... and that is Invoke-Expression !
The best (/least bad?) explanations for Invoke-Expression I've found so far are here (Microsoft's own description of the cmdlet doesn't really make a great job of hinting that it'd help in situations such as this):
http://ss64.com/ps/invoke-expression.html
https://www.adminarsenal.com/powershell/invoke-expression/
I'm trying to dynamically parse & build-up a data structure of some incoming JSON files I'm to be supplied with (that'll be in non-standard structure) via Powershell to then process the data in those files & hand them over to the next step.
As part of that, I'm trying to build up the data structure of the JSON file into essentially a list of of data-paths for me to parse through & grab the data out of, so that I can cope with arrays, nested JSON objects and so on. So far so good.
Where I fall into some sort of Powershell peculiarity is in handling 2+ levels of depth via a variable. Let me give you a nice code-block to demonstrate the problem...
# Generate a Quick JSON file with different data types & levels
[object]$QuickJson = #'
{
"Name" : "I am a JSON",
"Version" : "1.2.3.4",
"SomeBool" : true,
"NULLValue" : null,
"ArrayOfVersions" : [1.0,2.0,3.0],
"MyInteger" : 69,
"NestedJSON" : {
"Version" : 5.0,
"IsReady" : false
},
"DoubleNestedJSON" : {
"FirstLevel" : 1,
"DataValue" : "I am at first nested JSON level!",
"Second_JSON_Level" : {
"SecondLevel" : 2,
"SecondDataValue" : "I am on the 2nd nested level"
}
}
}
'#
# Import our JSON file into Powershell
[object]$MyPSJson = ConvertFrom-Json -InputObject $QuickJson
# Two quick string variables to access our JSON data paths
[string]$ShortJsonPath = "Name"
[string]$NestedJsonPath = "NestedJson.Version"
# Long string to access a double-nested JSON object
[string]$LongNestedJsonPath = "DoubleNestedJSON.Second_JSON_Level.SecondDataValue"
# Both of these work fine
Write-Host ("JSON Name (Direct) ==> " + $MyPSJson.Name)
Write-Host ("JSON Name (via Variable) ==> " + $MyPSJson.$ShortJsonPath)
# The following way to access a single nested Json Path works fine
Write-Host ("Nested JSON Version (via direct path) ==> " + $MyPSJson.NestedJson.Version)
# And THIS returns an empty line / is where I fall afoul of something in Powershell
Write-Host ("Nested JSON Version (via variable) ==> " + $MyPSJson.$NestedJsonPath)
# Other things I tried -- all returning an empty line / failing in effect
Write-Host ("Alternate Nested JSON Version ==> " + $($MyPSJson.$NestedJsonPath))
Write-Host ("Alternate Nested JSON Version ==> " + $MyPSJson.$($NestedJsonPath))
Write-Host ("Alternate Nested JSON Version ==> " + $($MyPSJson).$($NestedJsonPath))
# Similarly, while THIS works...
$MyPSJson | select-object -Property NestedJSON
# This will fail / return me nothing
$MyPSJson | select-object -Property NestedJSON.Version
... in doing a bunch of research around this, I came across a suggestion to transform this into a Hashtable -- but that has the same problem, sadly. So with the above code-snippet, the following will transform the JSON object into a hashtable.
# Same problem with a hash-table if constructed from the JSON file...
[hashtable]$MyHash = #{}
# Populate $MyHash with the data from our quickie JSON file...
$QuickJson | get-member -MemberType NoteProperty | Where-Object{ -not [string]::IsNullOrEmpty($QuickJson."$($_.name)")} | ForEach-Object {$MyHash.add($_.name, $QuickJson."$($_.name)")}
# ... and even then -- $MyHash."$($NestedJsonPath)" -- fails, while a single level deep string works fine in the variable! :(
So it's pretty clear that I'm running into "something" of a Powershell internal logic problem, but I can't get Powershell to be overly helpful in WHY that is. Adding a '-debug' or similar in an attempt to increase verbosity hasn't helped shed light on this.
I suspect it's something akin to the items raised in this article here ( https://blogs.technet.microsoft.com/heyscriptingguy/2011/10/16/dealing-with-powershell-hash-table-quirks/ ) but just specific with variables.
I've not had any luck in finding anything obvious in the Powershell language specification (3.0 still being the latest from here as far as I can tell -- https://www.microsoft.com/en-usdownload/details.aspx?id=36389 ) either. It may be in there, I may just miss it.
Any advice in how to get Powershell to play nice with this would be greatly appreciated. I'm not sure how / why Powershell is fine with a simple string but seems to have issues with a 'something.somethingelse' type string here.
Thank you.
Further notes & addenda to the original:
It seems there are several issues to attack. One is "dealing with a single nested level". The "quick fix" for that seems to be using "Invoke-Expression" to resolve the statement, so for instance (IMPORTANT - take note of the back-tick with the first variable!):
iex "`$MyPSJson.$NestedJsonPath"
That use of Invoke-Expression also works with multi-nested situations:
iex "`$MyPSJson.$LongNestedJsonPath"
An alternative approach that was mentioned is the use of multiple select statements ... but I've not been able to get that to work with multi-nested objects (Powershell seems to not resolve those properly for some reason).
So for instance in this scenario:
($MyComp | select $_.DoubleNestedJSON | select FirstLevel)
Powershell returns
FirstLevel
----------
... instead of the actual data value. So - for now, it seems that selects won't work with multi-level nested objects due to Powershell apparently not resolving them?
When you write something like
$MyPSJson.Name
this will attempt to retrieve the member named Name from the object $MyPSJson. If there is no such member, you'll get $null.
Now, when you do that with variables for the member name:
$MyPSJson.$ShortJsonPath
this works pretty much identical in that the member with the name stored in $ShortJsonPath is looked up and its value retrieved. No surprises here.
When you try that with a member that doesn't exist on the object, such as
$MyPSJson.$NestedJsonPath
# equivalent to
# $MyPSJson.'NestedJSON.Version'
you'll get $null, as detailed before. The . operator will only ever access a member of the exact object that is the result of its left-hand-side expression. It will never go through a member hierarchy in the way you seem to expect it to do. To be frank, I'm not aware of a language that works that way.
The reason it works with Invoke-Expression is, that you effectively converting the $NestedJsonPath string into part of an expression resulting in:
$MyPSJson.NestedJSON.Version
which Invoke-Expression then evaluates.
You can, of course, define your own function that works that way (and I'd much prefer that instead of using Invoke-Expression, a cmdlet that should rarely, if ever, used (heck, it's eval for PowerShell – few languages with eval advocate its use)):
function Get-DeepProperty([object] $InputObject, [string] $Property) {
$path = $Property -split '\.'
$obj = $InputObject
$path | %{ $obj = $obj.$_ }
$obj
}
PS> Get-DeepProperty $MyPSJson NestedJson.Version
5,0
You could even make it a filter, so you can use it more naturally on the pipeline:
filter Get-DeepProperty([string] $Property) {
$path = $Property -split '\.'
$obj = $_
$path | %{ $obj = $obj.$_ }
$obj
}
PS> $MyPSJson | Get-DeepProperty nestedjson.version
5,0
Why this doesn't work
When you provide the properties that you'd like within a string, like this
[string]$NestedJsonPath = "NestedJson.Version"
Powershell looks for a property called NestedJSon.Version. It's not actually traversing the properties, but looking for a string literal which contains a period. In fact, if I add a property like that to your JSON like so.
[object]$QuickJson = #'
{
"Name" : "I am a JSON",
"Version" : "1.2.3.4",
"SomeBool" : true,
"NULLValue" : null,
"ArrayOfVersions" : [1.0,2.0,3.0],
"MyInteger" : 69,
"NestedJSON.Version" : 69,
"NestedJSON" : {
"Version" : 5.0,
"IsReady" : false
}
}
I now get a value back, like so:
>$MyPSJson.$NestedJsonPath
69
The best way to get your values back is to use two separate variables, like this.
$NestedJson = "NestedJson"
$property = "Version"
>$MyPSJson.$NestedJson.$property
5.0
Or, alternatively, you could use select statements, as seen in the original answer below.
$MyPSJson | select $_.NestedJSON | select Version
Version
-------
1.2.3.4
If you use multiple Select-Object statements, they'll discard the other properties and allow you to more easily drill down to the value you'd like.
I followed Joey's filter example. However, I found it did not support accessing arrays.
Sharing the code that I got to work for this. Hopefully it will help others as well. Awesome thread!
filter Get-DeepProperty([string] $Property) {
$path = $Property -split '\.'
$obj = $_
foreach($node in $path){
if($node -match '.*\[\d*\]'){
$keyPieces = $node -split ('\[')
$arrayKey = $keyPieces[0]
$arrayIndex = $keyPieces[1] -replace ('\]','')
$obj = $obj.$arrayKey[$arrayIndex]
} else {
$obj = $obj.$node
}
}
$obj
}
Example usage:
$path = "nested.nestedtwo.steps[2]"
$payload | Get-DeepProperty $path
I had the same problem, so I wrote a function that does the trick.
It enables accessing any level of the json by variable path (string):
function getNestedJsonValue() {
param(
[Parameter(Mandatory = $true, ValueFromPipeline)] [PSCustomObject] $inputObj,
[Parameter(Mandatory = $true)] [string] $valuePath
)
if (($valuePath -eq $null) -or ($valuePath.length -eq 0) -or ($inputObj -eq $null)) {
return $inputObj
}
[System.Array] $nodes = "$valuePath" -split '\.'
foreach ($node in $nodes) {
if (($node -ne $null) -and ($node.length -gt 0) -and ($inputObj -ne $null)) {
$inputObj = $inputObj.$node
} else {
return $inputObj
}
}
return $inputObj
}
Usage: getNestedJsonValue -valuePath $nestedValuePath -inputObj $someJson
Pipe usage: $someJson | getNestedJsonValue -valuePath $nestedValuePath
An example nestedValuePath would be $nestedValuePath="some.nested.path"
Credit to wOxxOm for getting things on the right track.
Invoke-Expression does seem to work perfectly for this situation (if somewhat expensive, but that's fine in my personal example & situation), and it can cope with multiple levels of nesting.
So as examples for the above code snippet, the following will resolve just fine (Key point - pay attention to the initial back-tick. That caught me off guard):
Write-Host ("Single level JSON test ==> " + (iex "`$MyPSJson.$NestedJsonPath"))
Write-Host ("Double level JSON test ==> " + (iex "`$MyPSJson.$LongNestedJsonPath"))
That'll return our desired results:
Single level JSON test ==> 5.0
Double level JSON test ==> I am on the 2nd nested level
FoxDeploy's answer of using multi-level selects doesn't seem to work with 2+ levels of nesting, unfortunately for some bizarre reason.
Using:
($MyPSJson | select $_.DoubleNestedJSON | select FirstLevel)
We get the following back from Powershell:
FirstLevel
----------
... it seems that Powershell doesn't resolve nested objects in its entirety? We get a similar results if we intentionally use something that doesn't exist:
($MyPSJson | select $_.DoubleNestedJSON | select Doesnotexist)
... also simply returns:
Doesnotexist
------------
So - for now - it seems as if "Invoke-Expression" works most reliably (and most easily, as it's just a case of handing it a variable with the path'ed string).
I still can't explain the WHY of any of this so far (since I've used 'dotwalk'-ing with multiple variables through arrays quite happily), but at least there's a solution for now ... and that is Invoke-Expression !
The best (/least bad?) explanations for Invoke-Expression I've found so far are here (Microsoft's own description of the cmdlet doesn't really make a great job of hinting that it'd help in situations such as this):
http://ss64.com/ps/invoke-expression.html
https://www.adminarsenal.com/powershell/invoke-expression/
In Powershell (3.0), I get different results when piping an object to Export-CSV than I do if I use the -IncludeObject parameter with the same object.
Example:
$a = Get-Process | select -first 5
$a | Export-CSV -Path '.\one.csv'
Export-CSV -InputObject $a -Path '.\two.csv'
Why are the contents of one.csv and two.csv different?
(And in case it's just me ...)
one.csv =
#TYPE System.Diagnostics.Process
"__NounName","Name","Handles","VM","WS","PM","NPM","Path","Company","CPU","FileVersion","ProductVersion","Description","Product","BasePriority","ExitCode","HasExited","ExitTime","Handle","HandleCount","Id","MachineName","MainWindowHandle","MainWindowTitle","MainModule","MaxWorkingSet","MinWorkingSet","Modules","NonpagedSystemMemorySize","NonpagedSystemMemorySize64","PagedMemorySize","PagedMemorySize64","PagedSystemMemorySize","PagedSystemMemorySize64","PeakPagedMemorySize","PeakPagedMemorySize64","PeakWorkingSet","PeakWorkingSet64","PeakVirtualMemorySize","PeakVirtualMemorySize64","PriorityBoostEnabled","PriorityClass","PrivateMemorySize","PrivateMemorySize64","PrivilegedProcessorTime","ProcessName","ProcessorAffinity","Responding","SessionId","StartInfo","StartTime","SynchronizingObject","Threads","TotalProcessorTime","UserProcessorTime","VirtualMemorySize","VirtualMemorySize64","EnableRaisingEvents","StandardInput","StandardOutput","StandardError","WorkingSet","WorkingSet64","Site","Container"
"Process","AATray","390","156721152","29769728","10678272","27600","C:\Program Files\IBM\ISAM ESSO\AA\AATray.exe","IBM Corporation","42.4166719","8.2.1.1143","8.2.1.1143","AccessAgent Tray Icon: Component of ISAM ESSO AccessAgent","ISAM ESSO AccessAgent","8",,"False",,"4844","390","7784",".","0","","System.Diagnostics.ProcessModule (AATray.exe)","1413120","204800","System.Diagnostics.ProcessModuleCollection","27600","27600","10678272","10678272","257536","257536","63672320","63672320","29806592","29806592","194101248","194101248","True","Normal","10678272","10678272","00:00:32.8070103","AATray","15","True","1","System.Diagnostics.ProcessStartInfo","8/2/2016 7:20:30 AM",,"System.Diagnostics.ProcessThreadCollection","00:00:42.4166719","00:00:09.6096616","156721152","156721152","False",,,,"29769728","29769728",,
"Process","ac.activclient.gui.scagent","547","155099136","22593536","8478720","33184","C:\Program Files\ActivIdentity\ActivClient\ac.activclient.gui.scagent.exe","HID Global Identity Assurance","2.2932147","7,0,5,17","7,0","ActivClient Agent","${release.product.name}","8",,"False",,"3648","547","7872",".","0","","System.Diagnostics.ProcessModule (ac.activclient.gui.scagent.exe)","1413120","204800","System.Diagnostics.ProcessModuleCollection","33184","33184","8478720","8478720","274408","274408","62431232","62431232","22659072","22659072","168542208","168542208","True","Normal","8478720","8478720","00:00:01.8876121","ac.activclient.gui.scagent","15","True","1","System.Diagnostics.ProcessStartInfo","8/2/2016 7:20:30 AM",,"System.Diagnostics.ProcessThreadCollection","00:00:02.2932147","00:00:00.4056026","155099136","155099136","False",,,,"22593536","22593536",,
"Process","accagt","166","508174336","17592320","25407488","26116",,,,,,,,"8",,,,,"166","2480",".","0","",,,,,"26116","26116","25407488","25407488","159440","159440","25657344","25657344","17694720","17694720","509747200","509747200",,,"25407488","25407488",,"accagt",,"True","0","System.Diagnostics.ProcessStartInfo",,,"System.Diagnostics.ProcessThreadCollection",,,"508174336","508174336","False",,,,"17592320","17592320",,
"Process","acevents","506","140414976","22474752","8048640","30856","C:\Program Files\ActivIdentity\ActivClient\acevents.exe","HID Global Identity Assurance","42.6350733","5,0,4,4","5,0","ActivIdentity Event Service","ActivClient Services","8",,"False",,"3872","506","8256",".","0","","System.Diagnostics.ProcessModule (acevents.exe)","1413120","204800","System.Diagnostics.ProcessModuleCollection","30856","30856","8048640","8048640","249632","249632","61378560","61378560","22528000","22528000","157003776","157003776","True","Normal","8048640","8048640","00:00:24.6013577","acevents","15","True","1","System.Diagnostics.ProcessStartInfo","8/2/2016 7:20:34 AM",,"System.Diagnostics.ProcessThreadCollection","00:00:42.6350733","00:00:18.0337156","140414976","140414976","False",,,,"22474752","22474752",,
"Process","acnamagent","395","98971648","14561280","6586368","28296",,,,,,,,"8",,,,,"395","2012",".","0","",,,,,"28296","28296","6586368","6586368","125784","125784","6647808","6647808","14594048","14594048","101597184","101597184",,,"6586368","6586368",,"acnamagent",,"True","0","System.Diagnostics.ProcessStartInfo",,,"System.Diagnostics.ProcessThreadCollection",,,"98971648","98971648","False",,,,"14561280","14561280",,
two.csv =
#TYPE System.Object[]
"Count","Length","LongLength","Rank","SyncRoot","IsReadOnly","IsFixedSize","IsSynchronized"
"5","5","5","1","System.Object[]","False","True","False"
For context, I'm trying to splat my parameters to Export-CSV, but I run into this when I pass -InputObject, and I can't pipe the input and then splat the rest of the parameters.
Thanks.
This is the expected behavior.
When you pipe in through the pipeline, arrays, collections, enumerable stuff, etc. gets processed item by item. This is usually what you want.
When you use -InputObject, it accepts the array as a single object.
The best way to see this is to use Get-Member:
$a = Get-Process | select -first 5
$a | Get-Member
Get-Member -InputObject $a
In the first invocation, you'll see the data type and members of each element. In the second invocation you'll see the type and members of the collection object.
Depending on the cmdlet, you may not notice difference at all because it's handling both cases (see the pipeline function at the end of my answer).
But in the case of Export-Csv, or ConvertTo-Json, or other serialization type cmdlets, you want this difference; otherwise it's very difficult to serialize the array explicitly when you want to.
Another way to demonstrate it:
$sb = {
$_
Write-Verbose $_.Count -Verbose
}
$a | ForEach-Object -Process $sb
ForEach-Object -Process $sb -InputObject $a
When writing your own pipeline functions, a common way to work around the different ways of receiving the object is to use the Process {} block along with foreach:
function Test-Pipeline {
[CmdletBinding()]
param(
[Parameter(ValueFromPipeline)]
$MyVal
)
Process {
Write-Verbose $MyVal.Count -Verbose
foreach($v in $MyVal) {
$v
}
}
}
$a | Test-Pipeline
# Process block gets called once for each element
Test-Pipeline -MyVal $a
# Process block gets called once total, with the variable being an array
This works well because foreach doesn't fail if you give it a single non-array object, it just executes once.