You have various options. In my opinion, the best option is to apply the GOOS principle of listening to your tests. When the test becomes difficult to write, it's time to reconsider the design of the System Under Test (SUT). AutoFixture tends to amplify this effect.
Refactor to Value Objects
If you have a requirement that the Email
and FullName
properties should have particularly constrained values, it might indicate that instead of Primitive Obsession, the target API would benefit from defining explicit Email
and FullName
Value Objects. The canonical AutoFixture example is about phone numbers.
Use data annotations
You can also use data annotations to give AutoFixture hints about the constraints of the values. Not all data annotation attributes are supported, but you can use both MaxLength and RegularExpression.
It might look something like this:
public partial class User
{
public int ID { get; set; }
[RegularExpression("regex for emails is much harder than you think")]
public string Email { get; set; }
[MaxLenght(20)]
public string FullName { get; set; }
}
Personally, I don't like this approach, because I prefer proper encapsulation instead.
Use Customize
Instead of using the Build<T>
method, use the Customize<T>
method:
var fixture = new Fixture();
fixture.Customize<User>(c => c
.With(x => x.Email, string.Format("{0}@fobar.com", fixture.Create<string>())
.With(x => x.FullName, fixture.Create<string>().Substring(0,20)));
var newAnon = fixture.Create<User>();
Write a convention-driven Specimen Builder
Finally, you can write a convention-driven customization:
public class EmailSpecimenBuilder : ISpecimenBuilder
{
public object Create(object request,
ISpecimenContext context)
{
var pi = request as PropertyInfo;
if (pi == null)
{
return new NoSpecimen(request);
}
if (pi.PropertyType != typeof(string)
|| pi.Name != "Email")
{
return new NoSpecimen(request);
}
return string.Format("{0}@fobar.com", context.Resolve(typeof(string)));
}
}
This approach I really like, because I can put arbitrarily complex logic here, so instead of having to create a lot of one-off customizations, I can have a small set of conventions driving an entire test suite. This also tends to make the target code more consistent.