An Exercise in Refactoring - Specification Pattern

« Previous article:   Next article: »
HumanNameParser - Parsing people's names. Blog Home Throwing the Book at ‘em — Knowing Your Storage

This post is the eleventh installment of the 2017 C# Advent Calendar operated by Matthew Groves. Thanks for letting me participate!

This article is based on a video on DimeCasts.Net on the Specification Design Pattern. DimeCasts was a great source of short videos on .NET and other development topics. Unfortunately, they are no longer accepting new videos (and haven’t been for a while). But the videos are still available and many of them, on general topics, like design patterns, are still timely. So, if you got 10 minutes to spare, you might want to check out the Specification Pattern video cited above, before continuing.

For the impatient, I’ll give a quick recap. Using the example of a online retailer trying to determine of an order could be rush-shipped, the author (Tim Barcz) proposes the criteria of a) must be shipped to the US, b) the order total is over $100, c) all items in the order are in stock, and d) none of the items contain hazardous material.

Without the specification pattern, coding that would look something like this:

With the Specification pattern, that can be reduced to just this:

        var spec = new RushOrderSpecification();
        return spec.IsSatisfiedBy(this);

With RushOrderSpecification looking like this:

The video goes on describing the Specification<T> framework, and the DomesticOrderSpecification, HighValueOrderSpecification, HazardousMaterialSpecification, and IsInStockSpecification classes which do the actual work.

Which is the problem. While the code in the Order class in reduced to two lines, you have to write five classes to accomplish that, and those five classes hold a lot of redundant code.

So, let’s see if we can improve this.

(I posted the original code, and all my revisions to my github, with a check-in after each step, so you can compare the changes, by viewing the history)

Each of those classes takes the form of this: Note that there is just one really relevant line (the return statement), and the rest is all boilerplate. And if you were to write a specification class where that line is more than one line, you are probably doing too much in it, and it should be broken up into multiple specifications.

Our first step will be to replace those classes with objects, which the relevant test injected as a lambda expression, so that :

readonly HighValueOrderSpecification highValueSpec =
  new HighValueOrderSpecification();


readonly Specification<Order> highValueSpec =
  Specification<Order>.With(candidate => candidate.OrderTotal > 100);

To handle that, we create one new class to the framework:

And add a static method to the Specifiction<> base class to simplify creating new specifications:

public static SpecificationBuilder<TEntity> With(Func<TEntity, bool> conditional)
    return new SpecificationBuilder<TEntity>(conditional);

This allows us to forgot the SpecificationBuilder class even exists, and we can just work with Specification<>.

OK, that’s all well & good… but, IsSatisfiedBy is in passive voice. In coding, as in speech, one should prefer active voice. So, instead of

        return spec.IsSatisfiedBy(this);

It’s better to say

        return this.Satisfies(spec);

The code for Satisfies() would be the same for every code, so it would be simpliest to move it into a base class, but the derived classes would otherwise be unrelated, so requiring them to use a common base class would be a poor design choice (For one thing, it would prevent them from being derived from any other class, limiting the design)

An alternative would be it put the code into a extension method. One small problem here. Since the this object won’t have a common base class (see above), we’d have to attach the extension method on object which would have the Satisfies() method appearing in the Intellisense for every object, even those which it’s not appropriate. To limit that, we’ll create an ISpecificationTarget<> interface for those classes. Forcing such a class to implement an interface (especially one like this, which will have no members, and so, no implementation), is a far less limiting design decision than a base class.

So we add:

public interface ISpecificationTarget<TEntity>
    // that's right, nothing here.

Add a quick addition to the Order class:

public class Order : ISpecificationTarget<Order>
    // Unchanged

And finally, our extension method:

So, now, we can write IsRush() as:

public bool IsRush()
    var spec = new RushOrderSpecification();
    return this.Satisifies(spec);

OK, homestretch now. We did all this based on the example of one needed specification (“Is this a rush order?””) made up of four criteria. But in a real application we’d have many of each, with some of the specifications being used over and over. We have the top level specification made of member variables for each sub-specification. Once we moved the specifications from classes to objects, that means if two different top-level specifications used the same criteria – say “Does this order earn customer loyalty point?” also needed HighValueOrderSpecification. – we would need to define it separately in our LoyalPointsSpecification, and we’d need to define it exactly the same way as we did in RushOrderSpecification. We really need to define one common definition.

So, let’s make each component specification public properties of a static class, like this:

And now we can make Order.IsRush() just:

public bool IsRush()
    return this.Satisifies(OrderSpec.RushOrder);

PLus, if we then throw in a static using statement,

using static SpecificationPattern.Specifications.OrderSpec;

we can reduce that to

public bool IsRush()
    return this.Satisifies(RushOrder);

Which gives us the peak in clarity and readability.