Does the use of filters break the Open-Closed Principle?

Filters are an optional approach to use with specifications. You’ll find samples of them here.

This is a totally valid question. If the intention by using the specifications is to conform to this principle, then by adding the concept of filters, aren’t we doing the opposite? We go back and update the specification by adding additional conditions. As a brief recap, the OCP predicates that we should have constructs that are open to extension and closed to changes. This means, if we need to add a “behavior” to a class, we should be able to do that without changing the class itself. Even more simplified, if you have switch statements and too much conditional logic, it might be a sign that the behavior is too hardcoded, and might be refactored in a better way.

In our case, indeed we have too many conditions within the specification, so this concern is partially true. The catch here is that we have to do with single and atomic business requirement. The user has demanded from us that we change the behavior and add an additional condition by which the customers can be queried. The requirement is that the filters on the customer’s UI page be extended. Whenever you have business requirement changes, undoubtedly you will have code changes in the domain model as well. That particular specification will change only when this exact business requirement changes, and never otherwise; it’s wired up to this functionality only. Even though not the perfect structure, I might say it’s an acceptable solution.

This is quite different from the case when you have to change/add/delete behavior in a “classic” repository, in which case in order to update one business requirement, you’re forced to update a construct that holds a collection of business requirements. That clearly violates SRP and OCP.

Why using ThenInclude in one instance broke the application? What is the proper usage?

In one instance, while describing different usages we updated the specification by adding ThenInclude (as shown below), which in turn resulted in a runtime error.

Query.Include(x => x.Stores).ThenInclude(x => x.Address);

The error here has to do with the fact that the Address property is not a navigation property, but a string property. Obviously, you should not include simple properties. And, this is the same behavior that you will have by using EF directly. Including simple properties won’t result in a compile-time error but a runtime error. It’s up to you to be careful, make proper usage of it, and thoroughly test your queries.

We can constrain this usage and throw an exception, but we don’t want to alter the behavior that much. What if the EF in some further version makes use of this usage? So, the ultimate usage constraints should be the responsibility of the ORM you’re using.

How many Include statements are OK to have?

While creating JOINs in SQL, the real issue is not about how many tables, but the cardinality. If the dependent tables are configured as one-to-one relationships, that’s quite OK. But if you’re including dependent tables, where there are many rows for each principal row, then it can have quite an impact on the performance. On top of that, you should be careful what SQL queries are being produced by EF as well. The EF Core 1&2 uses split queries, while EF Core 3 uses a single query. If you have a lot of 1:n relationships and use a single query, then you might end with a “cartesian explosion” (consider split queries in these cases).

During the stream, I showcased the following specification, and the question was if this is OK?

public class AwbForInvoiceSpec : Specification<Awb>
{
    public AwbForInvoiceSpec(int ID)
    {
        Query.Where(x => x.ID == ID);
        Query.Include(x => x.Packages);
        Query.Include(x => x.AwbCargoServices);
        Query.Include(x => x.AwbPurchase);
        Query.Include(x => x.CargoManifest);
        Query.Include(x => x.Customer).ThenInclude(x => x.Addresses);
    }
}

In the context of that particular application, the Awb has quite significant importance in the overall business workflow, and it might be a bit more complicated than it should be. First of all, the AwbPurchase and CargoManifest represent 1:1 relationships. So, we end up with two 1:n navigations. This is relatively OK if you’re retrieving one Awb record (as in this case). On the other hand, if you’re trying to get a list of records, then you should reconsider if you need the child collections or not. Try to measure the performance, consider the usage of the application, number of users, peak usages, etc, and then you can decide if that meets your criteria or not.

One key benefit of using the specification pattern is that you can easily have different specifiations that include just the related data necessary for a given operation or context.

Anti-pattern usage

In the above example, the actual anti-pattern is not the usage of several Include statements, but including the Customer. That implies that Awb aggregate has direct navigation to the Customer aggregate. If you follow DDD, you should strive to have as independent aggregates as possible. If required, one particular aggregate should hold only the identifier of some other aggregate root and not have a direct reference. The app can then load the other aggregate from its id as necessary.

In our case, it was a deliberate design decision to break this rule (for Customer aggregate), in order to improve the performance in particular cases and to reduce roundtrips to DB. But, it’s not something I would advise you to do. Anyhow, it’s up to you to weigh the pros/cons and make your own elaborate decisions for your applications.

Do I need private constructors for the entities (e.g. for the EF code-first approach)

I got this question related to one particular scenario which happened during the stream. Once we added the DateTime birthdate parameter in the constructor, we were forced to add an additional parameterless constructor so the EF could work properly.

EF requires a parameterless constructor, so it can create the instance of the entity and then populate the properties. So, it might be wise always to add one private parameterless constructor just to be sure EF can instantiate the entity.

EF is smart enough to utilize the constructor and will try to pass the values as ctor arguments. That’s how EF handles the immutability (if your props have only getters). But, the ctor arguments should be named the same as the properties. The first character can be lowercase, and that’s ok, EF will map to it correctly. But, the case of the rest of the characters should be exactly the same. So, in our case, if we have named the argument birthDate instead of birthdate, would have worked with no issues.

public Customer(string name, string email, string address, DateTime birthdate)
{
    Guard.Against.NullOrEmpty(name, nameof(name));
    Guard.Against.NullOrEmpty(email, nameof(email));

    this.BirthDate = birthdate;
    this.Name = name;
    this.Email = email;
    this.Address = address;
}

How can I use the Select operator in Specification?

Use the following syntax to transform elements in a sequence based on a lambda expression:

Query.Select(x => x.Name);

In this case, each element x is being “transformed” into its Name property.

Also make sure that the base class of your Specification using Select is a Specification<T, TReturn>.

See the doc page on Select for a more detailed explanation.

Which versions of EF and EF Core can I use?

Ardalis.Specification by default sets the minimum required version.

Which is: EF Core 6.0.6 and EF 6.4.4.

But you are free to explicitly set higher versions if you like.

ISingleResultSpecification is Obsolete

I’m getting a warning or error telling me that ISingleResultSpecification is obsolete. What am I supposed to do to fix it?

You can just remove the interface usage from any specification that is using it. It was useful with GetBySpec calls in earlier versions of the library but newer versions use the more canonical FirstOrDefault or SingleOrDefault methods on repositories whenever a single result is desired or expected (which eliminates the need for the interface).