In this article we will employ a functional monadic concept Either to make validation code more expressive and easier to maintain.

Problem

Let's say we get a request from some client code and we need to check if this request is actually valid. If it's not valid, we want to make a detailed description of the problems that we identified. If it is valid, we want to produce a response about the successful acceptance of the request. Let's define the classes:

public class Request { ... }
public class Response { ... }
public class ValidationError { ... }

Now, we need a function which would accept a Request and would return Response or ValidationError. Let's look at some possible solutions.

Throw an exception

Validation error sounds like it could be an exception:

public Response Validate(Request r)
{
   if (!Valid(r))
       throw new ValidationException(new ValidationError(...));

   return new Response(r);
}

This approach is really bad though. You have to declare a special exception class to hold the validation error. But even worse, exception handling is not explicit - you don't see the exception type when you look at method signature. Client processing code is going to be messed up because of exception handling. Never use exceptions for your business logic flow.

Output parameter

We could make an output parameter of ValidationError type:

public Response Validate(Request r, out ValidationError error)
{
    if (Valid(r))
    { 
        error = new ValidationError(...);
        return null;
    }

    error = null;
    return new Response(r);
}

Now the interface is more explicit: client won't be able to completely ignore the fact that an error is possible. But output parameters are not really easy to use in C#, especially in fluent-style client code. Moreover, we are using nulls as a way to represent missing object, which is a smell by itself, because nulls are not explicit. Never use nulls in your business logic.

Return the combined result

We could declare a container class which would keep both Response and ValidationError, and then return it from the method.

public class Both<TData, TError>
{
    public TData Data { get; set; }
    public TErrro Error { get; set; }
}
...
public Both<Response, ValidationError> Validate(Request r)
{
    return Valid(r) 
        ? new Both<Response, ValidationError> { Data = new Response(r) }
        : new Both<Response, ValidationError> { Error = new ValidationError(...) };
}

Looks much nicer, we are getting there. Now it's a pure function with input and output parameters, but we still use null for result state representation. Let's see how we can solve it with Either data structure.

Introducing Either

Instead of returning Both with nullable properties, let's return Either with just one of them. When constructing an object, you can specify either a 'left' or a 'right' argument, but not both.

public class Either<TL, TR>
{
    private readonly TL left;
    private readonly TR right;
    private readonly bool isLeft;

    public Either(TL left)
    {
        this.left = left;
        this.isLeft = true;
    }

    public Either(TR right)
    {
        this.right = right;
        this.isLeft = false;
    }
}

Now, the main difference is in how the client uses it. There are no properties to accept Left and Right parts. Instead we define the following method:

public T Match<T>(Func<TL, T> leftFunc, Func<TR, T> rightFunc)
    => this.isLeft ? leftFunc(this.left) : rightFunc(this.right);

That's the concept of pattern matching implemented in C# world. If a left value is specified, Match will return the result of the left function, otherwise the result of the right function.

Another improvement would be to create explicit operators for easy conversions from left and right types:

public static implicit operator Either<TL, TR>(TL left) => new Either<TL, TR>(left);

public static implicit operator Either<TL, TR>(TR right) => new Either<TL, TR>(right);

Let's have a look at a complete example.

Why it's great

Here is the service code written with Either:

public Either<Response, ValidationError> Validate(Request r)
{
    return Valid(r) 
        ? Data = new Response(r)
        : new ValidationError(...);
}

Clean and nice! Now a simplistic client:

var validated = service.Validate(request);
Console.WriteLine(
    validated.Match(
        result => $"Success: {result}",
        error => $"Error: {error}")
    );

Simple, readable , no conditionals, no null checks, no way to silently ignore the fact that validation may fail.

Show me the code

You can find the definition of Either class in my github repo.

Update. Here is a link to an awesome talk on this topic: Railway oriented programming: Error handling in functional languages by Scott Wlaschin