PowerShell Casting Oddities

6 minute read

PowerShell’s casting can result in some weird results.

Before we begin, check out this C# class:

using System;
public class User
{
    // The user's id (e.g.: StackOverflow id, employee number, or whatever)
    public uint Id { get; set; }
    // The user's name (for display). First and last, nickname, whatever.
    public string Name { get; set; }
    // The user's email (for auth and communications)
    public string Email { get; set; }

    // Get the user from the datastore by Id
    public User(uint id)
    {
        Id = id;
    }

    // Create a new user from their email
    public User(string email)
    {
        Email = email;
    }

    // Create empty users
    // Parameterless default constructors are common in WPF and PowerShell
    public User() { }

    // Prefer user Name but show Email otherwise
    public override string ToString()
    {
        return Name ?? Email;
    }
}

It’s a pretty simple class with three constructors: the default (parameterless) one, and one each taking the email and id. PowerShell will cast using the constructors of a type, so in this case, you can cast either a string or a number –or a hashtable– to User.

We can add this type to our PowerShell session using Add-Type:

Add-Type -path User.cs

The most interesting of PowerShell’s special casting powers is the hashtable cast. Any class that has a default (parameterless) constructor can be created by casting a hashtable of (some of) it’s settable properties:

[User]@{
    Id = 1
    Name = "Jaykul"
}
Id Name   Email
-- ----   -----
 1 Jaykul

One of the weirdest examples is this id constructor.

In the case of this particular User class, when we cast a number, like [User]2 it’s the same as calling [User]::new(2), but the results of either are a little surprising:

Id Name Email
-- ---- -----
 0      2

The number was used for the Email value, instead of the Id that we expected…

The reason for this is actually straightforward, but it’s due to a combination of reasons that aren’t intuitive.

The bottom line is that the numeric constructor we might have thought we were calling accepts a uint (an unsigned int), rather than a signed int, and literal numbers in PowerShell are always [int] (or [double] if they have a decimal place) unless you add a suffix.

A little background.

In programming, “casting” is when you convert an object from one type to another. We talk about “implicit casting” when you set a variable of one type with a value of another, or pass a value to a method (or function) that accepts a different type of value, and the language converts it from one to the other for you without you asking for that. We talk about “explicit casting” when you specify the type you want to convert to:

In general, programming languages (including PowerShell) prefer not to cast implicitly if it’s possible that you’ll lose information. So you can implicitly cast from a smaller to a bigger type (e.g. 32bit integer to 64bit), or from 16bit unsigned to 32bit signed, but not to a smaller type (32bit to 16bit integer), or signed to unsigned.

In this case, our numbers are integers, so PowerShell won’t implicitly cast them to an unsigned integer, and won’t call the method that accepts an unsigned int unless you explicitly make the number an unsigned int.

As usual with PowerShell, there are several ways to do that.

You can specify the type of numbers with a suffix (see about numeric literals) and then cast it or call the constructor:

[User]2u
[User]::new(2u)

Or you can cast it to uint and then either cast it again to User or call the constructor. Either way, if you explicitly specify the type, PowerShell assumes you understand what you’re doing, and lets you.

[User][uint]2
[User]::new([uint]2)

And in any of those cases, you’ll get a user with just the Id set.

Id Name Email
-- ---- --------
 2

Of course, the other part of the puzzle is why you get an unexpected result, instead of an error like:

“InvalidArgument: Cannot convert the “2” value of type “System.Int32” to type “User”.

The reason is yet another PowerShell casting oddity: PowerShell can (and will) cast anything to a string implicitly, by calling it’s ToString() method. That’s really convenient in a shell, where you regularly want to display things as strings, but sometimes… well, something like this happens.

In other words, PowerShell will implicitly cast an integer to string, but will not cast implicitly it to uint – so you get an unexpected constructor and result.

A few notes on designing types for PowerShell

Use PowerShell’s natural types

While it’s fine to use whatever type you want for a property (remember that the hashtable constructor works fine), it’s best to avoid unsigned integers and other types that won’t cast automatically in constructors or methods. That’s also true for types like generic lists. PowerShell has native syntax for arrays, but not for generic collections, and it will unroll generic collections and turn them into arrays if you output them anyway, so it’s usually better in types we write for PowerShell to use arrays at the interface.

One option is to write the class using the “right” types as properties, including unsigned integers or generic collections, but to add a constructor that takes an integer or an array, etc.

For example, in this case, we could add this constructor:

    public User(Int64 id)
    {
        if (id < 0) {
            throw new ArgumentOutOfRangeException("id","id must be a positive integer");
        }
        if (id > UInt32.MaxValue) {
            throw new ArgumentOutOfRangeException("id","id must be less than or equal to " + uint.MaxValue);
        }
        Id = (uint)id;
    }

While it may be a little complicated to write the converting constructor (and you have to think of things like needing to up-size to int64 in order to accept the maximum value of an unsigned int), this will make it “just work” for users without needing to specify the type suffix.

Write Type Converters

What if you’re dealing with a class like that User one up above, that someone else wrote – whether in a 3rd party API or in the .NET framework. The bottom line is that if you are getting integer input from users and you want it to just work, but you can’t modify the class because it came from someone else, you can write a PSTypeConverter.

You can even write it in pure PowerShell:

class UserConverter : System.Management.Automation.PSTypeConverter {
    [bool] CanConvertFrom([PSObject]$psSourceValue, [Type]$destinationType) {
        # Is it an integer number we're trying to convert to a user?
        # NOTE: we're claiming we can convert any integer, but really we can't convert anything bigger than uint
        return $psSourceValue -eq ($psSourceValue -as [int64]) -and $destinationType -eq [User]
    }
    [object] ConvertFrom([PSObject]$psSourceValue, [Type]$destinationType, [IFormatProvider]$formatProvider, [bool]$ignoreCase) {
        return [User]::new([uint]$psSourceValue)
    }

    # The rest of the methods are implemented by calling one of those first two methods:
    [bool] CanConvertFrom([object]$sourceValue, [Type]$destinationType) {
        return $this.CanConvertTo($sourceValue, $destinationType)
    }
    [bool] CanConvertTo([object]$sourceValue, [Type]$destinationType) {
        return $this.CanConvertTo(([PSObject]$sourceValue), $destinationType)
    }
    [object] ConvertFrom([object]$sourceValue, [Type]$destinationType, [IFormatProvider]$formatProvider, [bool]$ignoreCase) {
        return $this.ConvertFrom(([PSObject]$sourceValue), $destinationType, $formatProvider, $ignoreCase)
    }
    [object] ConvertTo([object]$sourceValue, [Type]$destinationType, [IFormatProvider]$formatProvider, [bool]$ignoreCase) {
        return $this.ConvertFrom(([PSObject]$sourceValue), $destinationType, $formatProvider, $ignoreCase)
    }
}

Update-TypeData -TypeName 'User' -TypeConverter 'UserConverter' -Force

Ironically, if you use the type [uint] instead of [int64] in the CanConvertFrom method, the test will be more truthful (it won’t claim it can convert numbers that are actually too big or too small to be converted), but the result will be unexpected: if you try to convert, say, the number 4294967296 the TypeConverter will admit it can’t do the job, and PowerShell will convert it using string constructor, just as it would have without the TypeConverter.

In other words, to get the behavior we want, we write the TypeConverter to lie, and claim it can convert any integer value, which will result in exceptions being thrown when those numbers can’t actually be converted to unsigned integers. Which, of course, is the behavior that we want (and matches the behavior of the integer constructor).

Comments

Joel Bennett

Can you add some thoughts and an example of a better design for classes meant to be used with PowerShell like this?

By the way, you repeated the explanation of “Explicit casting” multiple times with only slight changes in wording. It might be better to remove one of them, or provide a simple and then more detailed explanation.

P.S. This is a test comment.

Leave a comment

Your email address will not be published. Required fields are marked *

Loading...