As you've discovered, PowerShell refuses to run scripts that contains class definitions that reference then-unavailable (not-yet-loaded) types - the script parsing stage fails.
- As of PSv5.1, even a
using assembly statement at the top of a script does not help in this case, because in your case the type is referenced in the context of a PS class definition - this may get fixed in PowerShell Core, however; the required work, along with other class-related issues, is being tracked in GitHub issue #6652.
The proper solution is to create a script module (*.psm1) whose associated manifest (*.psd1) declares the assembly containing the referenced types a prerequisite, via the RequiredAssemblies key.
See alternative solution at the bottom if using modules is not an option.
Here's a simplified walk-through:
Create test module tm as follows:
Create module folder ./tm and manifest (*.psd1) in it:
# Create module folder (remove a preexisting ./tm folder if this fails).
$null = New-Item -Type Directory -ErrorAction Stop ./tm
# Create manifest file that declares the WinSCP assembly a prerequisite.
# Modify the path to the assembly as needed; you may specify a relative path, but
# note that the path must not contain variable references (e.g., $HOME).
New-ModuleManifest ./tm/tm.psd1 -RootModule tm.psm1 `
-RequiredAssemblies C:pathoWinSCPnet.dll
Create the script module file (*.psm1) in the module folder:
Create file ./tm/tm.psm1 with your class definition; e.g.:
class Foo {
# As a simple example, return the full name of the WinSCP type.
[string] Bar() {
return [WinSCP.Protocol].FullName
}
}
Note: In the real world, modules are usually placed in one of the standard locations defined in $env:PSMODULEPATH, so that the module can be referenced by name only, without needing to specify a (relative) path.
Use the module:
PS> using module ./tm; [Foo]::new().Bar()
WinSCP.Protocol
The using module statement imports the module and - unlike Import-Module -
also makes the class defined in the module available to the current session.
Since importing the module implicitly loaded the WinSCP assembly thanks to the RequiredAssemblies key in the module manifest, instantiating class Foo, which references the assembly's types, succeeded.
If you need to determine the path to the dependent assembly dynamically in order to load it or even to ad-hoc-compile one (in which case use of a RequiredAssemblies manifest entry isn't an option), you should be able to use the approach recommended in Justin Grote's helpful answer - i.e., to use a ScriptsToProcess manifest entry that points to a *.ps1 script that calls Add-Type to dynamically load dependent assemblies before the script module (*.psm1) is loaded - but this doesn't actually work as of PowerShell 7.2.0-preview.9: while the definition of the class in the *.psm1 file relying on the dependent assembly's types succeeds, the caller doesn't see the class until a script with a using module ./tm statement is executed a second time:
# Create module folder (remove a preexisting ./tm folder if this fails).
$null = New-Item -Type Directory -ErrorAction Stop ./tm
# Create a helper script that loads the dependent
# assembly.
# In this simple example, the assembly is created dynamically,
# with a type [demo.FooHelper]
@'
Add-Type @"
namespace demo {
public class FooHelper {
}
}
"@
'@ > ./tm/loadAssemblies.ps1
# Create the root script module.
# Note how the [Foo] class definition references the
# [demo.FooHelper] type created in the loadAssemblies.ps1 script.
@'
class Foo {
# Simply return the full name of the dependent type.
[string] Bar() {
return [demo.FooHelper].FullName
}
}
'@ > ./tm/tm.psm1
# Create the manifest file, designating loadAssemblies.ps1
# as the script to run (in the caller's scope) before the
# root module is parsed.
New-ModuleManifest ./tm/tm.psd1 -RootModule tm.psm1 -ScriptsToProcess loadAssemblies.ps1
- Now, still as of PowerShell 7.2.0-preview.9, trying to use the module's
[Foo] class inexplicably succeeds only after calling using module ./tm twice - which you cannot do in a single script, rendering this approach useless for now:
# As of PowerShell 7.2.0-preview.9:
# !! First attempt FAILS:
PS> using module ./tm; [Foo]::new().Bar()
InvalidOperation: Unable to find type [Foo]
# Second attempt: OK
PS> using module ./tm; [Foo]::new().Bar()
demo.FooHelper
The problem is a known one, as it turns out, and dates back to 2017 - see GitHub issue #2962
If your use case doesn't allow the use of modules:
- In a pinch, you can use
Invoke-Expression, but note that it's generally better to avoid Invoke-Expression in the interest of robustness and so as to avoid security risks[1]
.
# Adjust this path as needed.
Add-Type -LiteralPath C:pathoWinSCPnet.dll
# By placing the class definition in a string that is invoked at *runtime*
# via Invoke-Expression, *after* the WinSCP assembly has been loaded, the
# class definition succeeds.
Invoke-Expression @'
class Foo {
# Simply return the full name of the WinSCP type.
[string] Bar() {
return [WinSCP.Protocol].FullName
}
}
'@
[Foo]::new().Bar()
- Alternatively, use a two-script approach:
- A main script that loads the dependent assemblies,
- which then dot-sources a second script that contains the
class definitions relying on the types from the dependent assemblies.
This approach is demonstrated in Takophiliac's helpful answer.
[1] It's not a concern in this case, but generally, given that Invoke-Expression can invoke any command stored in a string, applying it to strings not fully under your control can result in the execution of malicious commands.
This caveat applies to other language analogously, such as to Bash's built-in eval command.
