NullableClass

Helper classes and analyzer to avoid usage of null in class types, for C#.

Motivation

We all know about the so-called “billion-dollar mistake”: the existence of a null value for reference types. Among other languages, C# adopted it, and it made sense at the time, but working around all the code and trying to avoid NullReferenceExceptions becomes a real problem.

Fortunately, there's a solution to this problem. Other languages have the concept of a Maybe or Optional type, where the lack of a value is explicitly stated and statically validated by the compiler.

This package provides a wrapper class NullableOf<T> that behaves similar to Nullable<T>, but for classes instead of structs.

Installation

There's two packages to install:

NullableClass: The main classes.
NullableClass.Analyzer: A Roslyn analyzer to forbid the use of a native null.

Usage

You first import the namespace:

using NullableClass;

And then you can use NullableOf in the following ways:

NullableOf<string> withValue = "Hello, World!";
NullableOf<string> empty = null;

string fine = "OK!";
string error = null; // With the analyzer installed, this will be marked as an error and won't compile.

// Same semantics as Nullable<T> for structs
var nullableString = default(NullableOf<string>);
nullableString.HasValue; // Boolean
nullableString.Value; // Throws exception if there's no value
nullableString.GetValueOrDefault(); // Gets default value - see next section

You can also use it on properties and fields (but remember to give them default values):

public class Person
{
    public string FirstName { get; set; } = "";
    public NullableOf<string> MiddleName { get; set; }
    // With the analyzer, this won't compile
    public string LastName { get; set; }
    
    private string userId = "";
    private NullableOf<string> roles;
}

Default values and arrays

Something that always comes up when discussing non-nullable classes is arrays. How do you create arrays, what values do you use?

This library uses the parameterless constructor of a class for its default value. This is used by the Default<T> helper class to get the default value and to create an array full of default values.

var nullableString = default(NullableOf<string>);
nullableString.GetValueOrDefault(); // returns "" -- this is a special case

// Uses implicit constructor. Initializes people with default properties
Person[] arrayOfPeople = Default<Person>.NewArray(10);

// You can create arrays with values just fine
var coolKids = new Person[]
{
    new Person("Sour", "Cream"),
    new Person("Jenny", "Pizza"),
    new Person("Buck", "Dewey"),
};

// This is forbidden by the analyzer
var emptyPeople = new Person[5];

// This is fine
var emptyNullPeople = new NullableOf<Person>[5];

Of course, the NewArray helper method will create a new array and then iterate it and create a new class for each item. Take this into account when creating large arrays so it doesn't affect performance.

LINQ magic

The library includes extensions for both Nullable<T> and NullableOf<T> that allow you to use the LINQ syntax for using their inner values:

Nullable<int> id = 5;
NullableOf<string> category = "books";
var result =
    from _category in category
    from _id in id
    select SearchItemInCategory(_category, _id);

If any of the values is null, the rest of the steps won't be computed, and only the null value is returned from the expression. This works similar to the null conditional operator (?.). This is also how monads work, if you're into that kind of thing.

Enumerable extension methods

You can use LINQ for enumerables as well, but the package includes special methods that instead of returning default (which may or may not be null), explicitly returns either a default non-null value, or a null value.

For example, in this LINQ left join:

var personTable = new[]
{
    new { id = 1, name = "Alice" , locationId = 1 },
    new { id = 2, name = "Bob"   , locationId = 0 },
    new { id = 3, name = "Carol" , locationId = 2 },
};

var locationTable = new[]
{
    new { id = 1, country = "USA" },
    new { id = 2, country = "UK"  }
};

var query =
    from person in personTable
    join location in locationTable on person.locationId equals location.id into location_tmp
    from location in location_tmp.NullIfEmpty()
    select new
    {
        personName = person.name,
        country = (from l in location select l.country)
    };

you can see the use of NullIfEmpty instead of DefaultIfEmpty that guarantees a NullableOf<T> value. There are replacements for FirstOrDefault and others as well.

Known issues

I couldn't name this type Nullable<T> despite how hard I tried. Of course, it's in its own namespace, but you always include System, so including my namespace would only cause a collision between System.Nullable<T> and NullableClass.Nullable<T>. Oh well.
Properties and fields could use a smarter analysis to detect if they're assigned in the constructor. I'll need to study the Roslyn APIs more to be able to implement this.
This will solve the null problem in your own code, but not on any external code. Unfortunately, since this is being written on top of the existing language, you'll still need to be careful interacting with it. If you know a method call can return null, immediately assign its value to a nullable variable.
You can't use null in default parameter values, which means that, except for string's "", all default values will need to be nullable.
If you find something I missed, or have an idea to contribute, please create a new issue.

FAQ

Isn't the C# team already solving this issue in C# 8?

Yes. Well, kinda. They want to keep backwards compatibility, which means they're only issuing warnings when the static analysis can detect incorrect null usage. This, unfortunately, completely misses the array case, and as far as I'm concerned, warnings = mere suggestions, despite what anyone would want. So, I'm creating this as an effort to introduce a true concept of nullable / non-nullable classes.

Aren't there any libraries that already do this?

Yes, it's very easy to find Maybe<T> implementations. However, they focus too much on the functional programming value, and I wanted something that was closer to how Nullable<T> already works. If you like functional programming, maybe (hehe) my implementation won't be the best for you.

License

MIT, of course.

R2D221 / NullableClass