In Progress
Unit 1, Lesson 1
In Progress

Safe Navigation Part 1: Reliable Associations

Video transcript & code

Here's a fairly common scenario when writing a Ruby program. We want to send a message to an object. But we know that in certain situations, the object will be missing, and we'll have nil instead. And so we find ourselves checking to see if the object is nil first, so we don't get a NoMethodError when that's the case.

beer = get_beer # Might return nil :-(
if beer != nil

Of course, our nil-checks are rarely this explicit. Often they look a bit more like this:

beer = get_beer # Might return nil :-(
beer and beer.drink!

And if we ever ne

A more complicated situation arises when we have to navigate through several levels of object, any of which might be nil.


We are pretty sure we have a product. But there might not be a department associated with it, and there might not be a curator attached to the department. So we wind up writing the code more like this:

require "./models"

@product.department &&
  @product.department.curator &&
# => ""

This is long and ugly and a pain to read and maintain.

Now, I'm going to stop right here for a moment and say that this looks ugly for a reason. Every nil-check is a type-check, and well-designed object-oriented programs should almost never need to check an object's type. The whole point of OO is being able to confidently send messages to objects without worrying about how they will handle those messages. And without constantly branching on object types.

Code like this usually points to a deeper problem: structural coupling. Every time we add a line of code that steps through this chain of connections, we tie another part of our program to the deep structure of the objects as they exist right now. If the relationships of objects should ever change, it will break every line where we tied our code to the deep structure.

In the next few episodes we're going to explore some strategies for performing so-called "safe-navigation" across networks of objects like this one. But before we get into that, I think we should talk a little bit about how we can avoid the need for "safe navigation" in the first place.

In various episodes we've examined patterns for eliminating spurious nil values from our code. For instance, in episode #114 we learned about the Null Object pattern, where we provide do-nothing objects whose job is to respond to expose the expected interface even in the absence of a "real" object.

Here are the model classes for this little example. They are extremely basic. We know that in order to use the Product#department attribute confidently, we need to be sure that it will always return something that behaves like a department—even if no department has been explicitly assigned yet.

At the same time, it is not Product's job to know exactly what a department should behave like. So let's punt the question to the Department class, and assign a default value of Department.null.

What is Department.null? That's what we have to decide next. Let's start its definition.

As suggested by the name, we want to return a "null object" of some kind; some object which represents the absence of an assigned department, without being nil.

Now, there are lots of different choices in how to define a null object. And if you're interested in learning about some of those options, check out the naught gem that I wrote.

But today, we're going to use what is perhaps the simplest possible form of null object: we're just going to define a special instance of the class to play the roll of null.

We only need one null instance, so we conditionally assign to a class instance variable called @null. In it we put a new Department instance, with a name that makes it clear that this is the null department.

Then we go into the initializer, and just as we did for Product's department attribute, we give Department objects a default non-nil curator association. Once again, we delegate to the associated class to define a suitable null version.

We go through the whole process over again for Curator. The only difference this time, is that we're not sure what kind of default value to give to the email_address attribute of a missing curator. We decide that since the email_address is used more for its value rather than for its behavior, it makes sense in this case to leave it with the default nil value.

Let's see what these modifications have bought us. We instantiate a new Product, but we don't assign it to a department.

If we ask it about its department, we get the null department. If follow the chain into department curator, we get the null curator. And if we follow the chain all the way to email address, we get nil.

require "./models_null"

product ="Bass-O-Matic")

# => #<Department:0x0055d351bb5cb8 @name="<Missing Department>", @curator=#<C...
# => #<Curator:0x0055d351bb5c18 @name="<Missing Curator>">
# => nil

Along the way, we don't have to check for the presence of any of the intervening associations. We are able to confidently navigate all the way to curator email address, even when parts of the chain are missing.

And if we ever need to, we can still check to see if a particular association is null. This works because we ensured that there would only be one null instance for each class.

require "./models_null"

product ="Bass-O-Matic")

product.department == Department.null
# => true

Again, this is the most rudimentary possible implementation of the null object pattern. But even so, it's enough to ensure that we don't have to worry about special methods or operators to "safely" navigate object associations.

In the next episode, we'll start looking at strategies for cases when we can't modify the objects being navigated to make them reliable. Until then: happy hacking!