Evil use of Maybe monad and extension methods in C#?

edit 2015 This question and its answers are no longer relevant. It was asked before the advent of C# 6, which has the null propagating opertor (?.), which obviates the hacky-workarounds discussed in this question and subsequent answers. As of 2015, in C# you should now use Form.ActiveForm?.ActiveControl?.Name.

I've been thinking about the null propagation problem in .NET, which often leads to ugly, repeated code like this:

Attempt #1 usual code:

string activeControlName = null;
var activeForm = Form.ActiveForm;
if (activeForm != null)
{
    var activeControl = activeForm.ActiveControl;
    if(activeControl != null)
    {
        activeControlname = activeControl.Name;
    }
}

There have been a few discussions on StackOverflow about a Maybe<T> monad, or using some kind of "if not null" extension method:

Attempt #2, extension method:

// Usage:
var activeControlName = Form.ActiveForm
                          .IfNotNull(form => form.ActiveControl)
                          .IfNotNull(control => control.Name);

// Definition:
public static TReturn IfNotNull<TReturn, T>(T instance, Func<T, TReturn> getter)
    where T : class
{
    if (instance != null ) return getter(instance);
    return null;
}

I think this is better, however, there's a bit of syntactic messy-ness with the repeated "IfNotNull" and the lambdas. I'm now considering this design:

Attempt #3, Maybe<T> with extension method

// Usage:
var activeControlName = (from window in Form.ActiveForm.Maybe()
                         from control in window.ActiveControl.Maybe()
                         select control.Name).FirstOrDefault();

// Definition:
public struct Maybe<T> : IEnumerable<T>
      where T : class
{
    private readonly T instance;

    public Maybe(T instance)
    {
        this.instance = instance;
    }

    public T Value
    {
        get { return instance; }
    }

    public IEnumerator<T> GetEnumerator()
    {
        return Enumerable.Repeat(instance, instance == null ? 0 : 1).GetEnumerator();
    }

    System.Collections.IEnumerator System.Collections.IEnumerable.GetEnumerator()
    {
        return this.GetEnumerator();
    }
}

public static class MaybeExtensions
{
    public static Maybe<T> Maybe<T>(this T instance)
        where T : class
    {
        return new Maybe<T>(instance);
    }
}

My question is: is this an evil abuse of extension methods? Is it better than the old usual null checks?

Daniel Earwicker

It's interesting that so many people independently pick the name IfNotNull, for this in C# - it must be the most sensible name possible! :)

Earliest one I've found on SO: Possible pitfalls of using this (extension method based) shorthand

My one (in ignorance of the above): Pipe forwards in C#

Another more recent example: How to check for nulls in a deep lambda expression?

There are a couple of reasons why the IfNotNull extension method may be unpopular.

Some people are adamant that an extension method should throw an exception if its this parameter is null. I disagree if the method name makes it clear.
Extensions that apply too broadly will tend to clutter up the auto-completion menu. This can be avoided by proper use of namespaces so they don't annoy people who don't want them, however.

I've played around with the IEnumerable approach also, just as an experiment to see how many things I could twist to fit the Linq keywords, but I think the end result is less readable than either the IfNotNull chaining or the raw imperative code.

I've ended up with a simple self-contained Maybe class with one static method (not an extension method) and that works very nicely for me. But then, I work with a small team, and my next most senior colleague is interested in functional programming and lambdas and so on, so he isn't put off by it.

Much as I'm a fan of extension methods, I don't think this is really helpful. You've still got the repetition of the expressions (in the monadic version), and it just means that you've got to explain Maybe to everyone. The added learning curve doesn't seem to have enough benefit in this case.

The IfNotNull version at least manages to avoid the repetition, but I think it's still just a bit too longwinded without actually being clearer.

Maybe one day we'll get a null-safe dereferencing operator...

Just as an aside, my favourite semi-evil extension method is:

public static void ThrowIfNull<T>(this T value, string name) where T : class
{
    if (value == null)
    {
        throw new ArgumentNullException(name);
    }
}

That lets you turn this:

void Foo(string x, string y)
{
    if (x == null)
    {
        throw new ArgumentNullException(nameof(x));
    }
    if (y == null)
    {
        throw new ArgumentNullException(nameof(y));
    }
    ...
}

into:

void Foo(string x, string y)
{
    x.ThrowIfNull(nameof(x));
    y.ThrowIfNull(nameof(y));
    ...
}

There's still the nasty repetition of the parameter name, but at least it's tidier. Of course, in .NET 4.0 I'd use Code Contracts, which is what I'm meant to be writing about right now... Stack Overflow is great work avoidance ;)

If you want an extension method to reduce the nested if's like you have, you might try something like this:

public static object GetProperty(this object o, Type t, string p)
{
    if (o != null)
    {
        PropertyInfo pi = t.GetProperty(p);
        if (pi != null)
        {
            return pi.GetValue(o, null);
        }
        return null;
    }
    return null;
}

so in your code you'd just do:

string activeControlName = (Form.ActiveForm as object)
    .GetProperty(typeof(Form),"ActiveControl")
    .GetProperty(typeof(Control),"Name");

I don't know if I'd want to use it to often due to the slowness of reflection, and I don't really think this much better than the alternative, but it should work, regardless of whether you hit a null along the way...

(Note: I might've gotten those types mixed up) :)

In case you're dealing with C# 6.0/VS 2015 and above, they now have a built-in solution for null propagation:

string ans = nullableString?.Length.ToString(); // null if nullableString == null, otherwise the number of characters as a string.

The initial sample works and is the easiest to read at a glance. Is there really a need to improve on that?

The IfNotNull solution is the best (until the C# team gives us a null-safe dereferencing operator, that is).

I'm not too crazy about either solution. What was wrong with ashorter version of the original:

string activeControlName = null;
if (Form.ActiveForm != null)
    if (Form.ActiveForm.ActivControl != null) activeControlname = activeControl.Name;

If not this, then I would look at writing a NotNullChain or FluentNotNull object than can chain a few not null tests in a row. I agree that the IfNotNull extension method acting on a null seems a little weird - even though extension methods are just syntactic sugar.

I think Mark Synowiec's answer might be able to made generic.

IMHO, I think the C# core team should look at the this "issue", although I think there are bigger things to tackle.

Sure, original 2-nested IF is much more readable than other choices. But suggesting you want to solve problem more generally, here is another solution:

try
{
    var activeForm = Form.ActiveForm; assumeIsNotNull(activeForm);
    var activeControl = activeForm.ActiveControl; assumeIsNotNull(activeControl);
    var activeControlname = activeControl.Name;
}
catch (AssumptionChainFailed)
{
}

where

class AssumptionChainFailed : Exception { }
void assumeIsNotNull(object obj)
{
    if (obj == null) throw new AssumptionChainFailed();
}

来源：https://stackoverflow.com/questions/1196031/evil-use-of-maybe-monad-and-extension-methods-in-c

标签

extension-methods

monads