Honest Return Types

Part 2 of Designing clear method signatures

Posted by Devon Burriss on March 14, 2017
Clean Code OOP Functional

In Part 1 we looked at ways of making your code more descriptive by using custom types instead of simple types like string. In this article we will look at what your return type can tell you about a method.

Updated: 19 March 2017

Honest Return Types

For most of this post let us build on the example of a Person repository. We are not going to dive into implementation but instead focus on the descriptiveness of the return type. Our starting point is this:

public interface IQueryPerson
{
    Person Get(Email email);
}

The return type should be honest about what can happen when you call a method. Does this repository method return null if no record is found? Does it throw and exception? Does it return a special case subtype? Wouldn't it be nice if your return type could tell you this instead of you having to dig into the implementation to find out.

My 2 criteria are:

  1. A return type should be really descriptive of what the possible outcomes are
  2. The interface for interacting with a type should make it difficult for developers to do the wrong thing

Result: A first try

One solution is a Result<T> or some such flavour. It might look something like this:

public class Result<T>
{
    public T Value { get; set; }
    public bool IsSuccess { get; set; }
    public IEnumerable<string> Errors { get; set; }
    public Result()
    {
        Errors = new List<string>();
    }

    public Result(T value)
    {
        if(value == null)
        {
            IsSuccess = false;
        }
        else
        {
            IsSuccess = true;
            Value = value;
        }
    }
}

This could be written in slightly different ways, with error codes instead of string for Errors, or even Exception. Let's discuss the pros and cons of this.

Pros

  • It does acknowledge that something could go wrong
  • Can return some error and state information without throwing an exception (read unexplicit goto statement)

Cons

  • It is not descriptive about what represents a failure
  • Value can be accessed without checking for success
  • The type doesn't convey whether null could still be a valid value

So it is something but doesn't really fulfill either of my criteria very well. We are going to have to take a quick sidebar and talk about representing null. Result<T> doesn't tell us whether we should expect T to be null and whether that is valid.

Functional side-bar

In functional terms an elevated type is like a wrapper. It is a higher level of abstraction that allows us to work with the type in a predictable way. IEnummerable<T>, Option<T>, Exception<T>, Either<L. R>, Validation<T> are all examples of elevated types.

Option: null is None

"It depends" is something you hear a lot in development, and wouldn't it be great if a type conveyed this? Option or Maybe are types often found in more functional languages that highlight the fact that a value could not be present. It allows you to say that there is Some value, or the value is None. This is probably easier to demonstrate...

I am using LanguageExt to get some more functional types. This one is mature and fully featured but pick whatever works for you.

public Option<Person> Get(Email email)
{
    Person person = QueryByEmail(email);//person could be null if no matching email found in the datasource
    return person;
}

//usage example
var person1 = personRepository.Get(email);

//print out last name if person was found otherwise print "Nobody"
person1.Match(
    Some: p => Console.WriteLine(p.LastName),
    None: () => Console.WriteLine("Nobody")
);

//return fullname or Nobody if no one was found
var person1Name = person1.Match(
    Some: p => $"{p.FirstNames} {p.LastName}",
    None: () => "Nobody"
);

The implementation uses implicit conversion to return None if the value is null otherwise the Person is elevated with Some.
I explicitly elevate the result to demonstrate what is happening. Let's also add some error-handling as this will show a problem.

using static LanguageExt.Prelude;
public Option<Person> Get(Email email)
{
    try
    {
        Person person = QueryByEmail(email);
        if(person == null)
            return None;
        return Some(person);
    }
    catch (Exception)
    {
        return None;
    }
}

So this is looking a little better.

Pros

  • Return type is explicit about possibility of no value being returned
  • The API of the type encourages handling of branch between happy and unhappy path

Cons

  • We cannot differentiate between no value and an exception

Exception: return don't throw

The following Exceptional<T> and Validation<T> types are defined in HonestTypes. Check the project page for installation instructions.

So our type needs to be a bit more explicit about what can happen. Let's introduce an Exceptional<T> type. This is similar to Option<Person> but instead of Some and None it has Exception and Success.
For those of you familiar with functional programming it is basically Either<Exception, T> with left set to Exception.

public Exceptional<Option<Person>> Get(Email email)
{
    try
    {
        Person person = QueryByEmail(email);
        Option<Person> result = person;
        return result;
    }
    catch (DbException ex)//only catch expected exceptions
    {
        return ex;
    }
}

//usage
var person1 = personRepository.Get(email);

person1.Match(
    Exception: ex => Console.WriteLine($"Exception: {ex.Message}"),
    Success: opt => opt.Match(
        None: () => Console.WriteLine("Person: Nobody"),
        Some: p => Console.WriteLine($"Person: {p.FirstNames} {p.LastName}")
    )
);

One important point in the repository implementation is you need to assign it to Option<Person> before returning it which implicitly converts to Exceptional<Option<Person>>. You can't go directly from Person to Exceptional<Option<Person>> unfortunately.

The difference in this implementation is in the exception handling. See how we just return the exception? The exception has an implicit conversion to the elevated type of Exceptional<T>.

Pros

  • Return type is very explicit about both errors and no value
  • API of return type encourages good handling of code paths

Cons

  • With the nested generics the type declaration is quite verbose

Conclusion

So with a bit of borrowing from functional programming and some added verbosity to our method signature we managed to move from an admittedly simple signature to a slightly more verbose one that is brutally honest about the possible outcomes.

Person Get(Email email);
Result<Person> Get(Email email);
Option<Person> Get(Email email);
Exceptional<Option<Person>> Get(Email email);

I hope you found something useful in this and if you did I cannot recommend enough the brilliant Functional Programming in C# from Manning. I must warn that some of the chapters in this book are heavy going. Not because they are badly written but because as a C# and Java developer the concepts are so foreign that they take a while to sink in. Like most things worthwhile it takes effort and determination but you will be a better developer for it.

In my following post I will discuss error handling and how logic/validation errors can be represented as return types following the same criteria as in this post.

Recommended Reading

  1. Elevated world
  2. Railway oriented programming


blog comments powered by Disqus