Deep null checking, is there a better way?

We have considered adding a new operation "?." to the language that has the semantics you want. (And it has been added now; see below.) That is, you'd say

cake?.frosting?.berries?.loader

and the compiler would generate all the short-circuiting checks for you.

It didn't make the bar for C# 4. Perhaps for a hypothetical future version of the language.

Update (2014): The ?. operator is now planned for the next Roslyn compiler release. Note that there is still some debate over the exact syntactic and semantic analysis of the operator.

Update (July 2015): Visual Studio 2015 has been released and ships with a C# compiler that supports the null-conditional operators ?. and ?[].

I got inspired by this question to try and find out how this kind of deep null checking can be done with an easier / prettier syntax using expression trees. While I do agree with the answers stating that it might be a bad design if you often need to access instances deep in the hierarchy, I also do think that in some cases, such as data presentation, it can be very useful.

So I created an extension method, that will allow you to write:

var berries = cake.IfNotNull(c => c.Frosting.Berries);

This will return the Berries if no part of the expression is null. If null is encountered, null is returned. There are some caveats though, in the current version it will only work with simple member access, and it only works on .NET Framework 4, because it uses the MemberExpression.Update method, which is new in v4. This is the code for the IfNotNull extension method:

using System;
using System.Collections.Generic;
using System.Linq.Expressions;

namespace dr.IfNotNullOperator.PoC
{
    public static class ObjectExtensions
    {
        public static TResult IfNotNull<TArg,TResult>(this TArg arg, Expression<Func<TArg,TResult>> expression)
        {
            if (expression == null)
                throw new ArgumentNullException("expression");

            if (ReferenceEquals(arg, null))
                return default(TResult);

            var stack = new Stack<MemberExpression>();
            var expr = expression.Body as MemberExpression;
            while(expr != null)
            {
                stack.Push(expr);
                expr = expr.Expression as MemberExpression;
            } 

            if (stack.Count == 0 || !(stack.Peek().Expression is ParameterExpression))
                throw new ApplicationException(String.Format("The expression '{0}' contains unsupported constructs.",
                                                             expression));

            object a = arg;
            while(stack.Count > 0)
            {
                expr = stack.Pop();
                var p = expr.Expression as ParameterExpression;
                if (p == null)
                {
                    p = Expression.Parameter(a.GetType(), "x");
                    expr = expr.Update(p);
                }
                var lambda = Expression.Lambda(expr, p);
                Delegate t = lambda.Compile();                
                a = t.DynamicInvoke(a);
                if (ReferenceEquals(a, null))
                    return default(TResult);
            }

            return (TResult)a;            
        }
    }
}

It works by examining the expression tree representing your expression, and evaluating the parts one after the other; each time checking that the result is not null.

I am sure this could be extended so that other expressions than MemberExpression is supported. Consider this as proof-of-concept code, and please keep in mind that there will be a performance penalty by using it (which will probably not matter in many cases, but don't use it in a tight loop :-) )

I've found this extension to be quite useful for deep nesting scenarios.

public static R Coal<T, R>(this T obj, Func<T, R> f)
    where T : class
{
    return obj != null ? f(obj) : default(R);
}

It's an idea I derrived from the null coalescing operator in C# and T-SQL. The nice thing is that the return type is always the return type of the inner property.

That way you can do this:

var berries = cake.Coal(x => x.frosting).Coal(x => x.berries);

...or a slight variation of the above:

var berries = cake.Coal(x => x.frosting, x => x.berries);

It's not the best syntax I know, but it does work.

Besides violating the Law of Demeter, as Mehrdad Afshari has already pointed out, it seems to me you need "deep null checking" for decision logic.

This is most often the case when you want to replace empty objects with default values. In this case you should consider implementing the Null Object Pattern. It acts as a stand-in for a real object, providing default values and "non-action" methods.

Update: Starting with Visual Studio 2015, the C# compiler (language version 6) now recognizes the ?. operator, which makes "deep null checking" a breeze. See this answer for details.

Apart from re-designing your code, like this deleted answer suggested, another (albeit terrible) option would be to use a try…catch block to see if a NullReferenceException occurs sometime during that deep property lookup.

try
{
    var x = cake.frosting.berries.loader;
    ...
}
catch (NullReferenceException ex)
{
    // either one of cake, frosting, or berries was null
    ...
}

I personally wouldn't do this for the following reasons:

• It doesn't look nice.
• It uses exception handling, which should target exceptional situations and not something that you expect to happen often during the normal course of operation.
• NullReferenceExceptions should probably never be caught explicitly. (See this question.)

So is it possible using some extension method or would it be a language feature, [...]

This would almost certainly have to be a language feature (which is available in C# 6 in the form of the .? and ?[] operators), unless C# already had more sophisticated lazy evaluation, or unless you want to use reflection (which probably also isn't a good idea for reasons of performance and type-safety).

Since there's no way to simply pass cake.frosting.berries.loader to a function (it would be evaluated and throw a null reference exception), you would have to implement a general look-up method in the following way: It takes in an objects and the names of properties to look up:

static object LookupProperty( object startingPoint, params string[] lookupChain )
{
    // 1. if 'startingPoint' is null, return null, or throw an exception.
    // 2. recursively look up one property/field after the other from 'lookupChain',
    //    using reflection.
    // 3. if one lookup is not possible, return null, or throw an exception.
    // 3. return the last property/field's value.
}

...

var x = LookupProperty( cake, "frosting", "berries", "loader" );

(Note: code edited.)

You quickly see several problems with such an approach. First, you don't get any type safety and possible boxing of property values of a simple type. Second, you can either return null if something goes wrong, and you will have to check for this in your calling function, or you throw an exception, and you're back to where you started. Third, it might be slow. Fourth, it looks uglier than what you started with.

[...], or is it just a bad idea?

I'd either stay with:

if (cake != null && cake.frosting != null && ...) ...

or go with the above answer by Mehrdad Afshari.

P.S.: Back when I wrote this answer, I obviously didn't consider expression trees for lambda functions; see e.g. @driis' answer for a solution in this direction. It's also based on a kind of reflection and thus might not perform quite as well as a simpler solution (if (… != null & … != null) …), but it may be judged nicer from a syntax point-of-view.