Real devs don’t use if. This slogan was my first contact with 10Pines, much
earlier than the start of my apprenticeship residence. The first time I heard it
just made a lot of noise in my head. Why would somebody want to eliminate the
What wrong could this little poor word do to us?
My goal with this post is to demonstrate why a real developer wouldn’t want to use
if. For this I will explain some of the benefits obtained by the usage of an
alternative: Let the object take the decisions.
Just as it sounds. The objects are completely capable of handling the task, they just need to be allowed. The key is to take advantage of their polymorphic nature to create better collaborations.
Just an example
Let’s see an exercise solved in Ruby where we can replace all the
by collaborations. We’ll go step by step improving our model and seeing the
benefits of this approach.
This is our assignment:
Make a model of the numbers in OOP. Considering the existence of Integer
numbers and fraction numbers. The numbers must respond properly the messages
Our task seems to be easy, we all know how to add and subtract. However we’ll focus on the model instead of the arithmetic part.
Let’s do a short analysis of the problem. Our goal is to make a model that contemplates different types of numbers(Integers and Fractions) and allows to operate between them. Our model should accept operations like this:
- 1 + 1 = 2
- ½ – ⅓ = ⅙
- 1 – ½ = ½
- ⅔ + ⅓ = 1
It is important to note that the receiver of the message can be any type of number. So does the external collaborator. In consequence we can’t always use the same algorithm, we have to adapt it to the situation. For example: Adding two integers is a straightforward operation, nevertheless adding an integer to a fraction is not so simple. We have to do something extra to solve the operation (for example multiplying the integer by the denominator of the fraction). In the case where the two operands are fractions probably we look for a common denominator. In conclusion, there are many correct algorithms for every situation, but we need one per case.
Having all this in mind we can start build our model. It sounds pretty natural to have a class diagram like this:
The class Number is abstract. Fraction and Integer must know how to add and subtract any number. However the mechanism of the operation can’t be the same for any type of number because the things we mentioned before. A First solution to this problem could be the following:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47
The Number class is abstract. All his methods just fail an tell us that is responsability of the subclass to implement it. The concrete operations follow always the same patron: Ask the collaborator for his class and then do the calc.
It works, but let’s compare it with a solution without if.
The purpose of the if in this case is to decide what to do based on the class of the collaborator. However this isn’t the only way to do it. The object doesn’t know the external collaborators’ classes by himself, but he knows his own class, and therefore he can send the right message to the collaborator with enought context information to do the right thing.
With this in mind 1 – ⅓ = ⅔ could be solved like this:
- We ask 1 to subtract certain number.
- The object 1 doesn’t know what kind of number is the parameter, so he says to him “Hey, I’m an integer, please subtract yourself from me, having that in mind.”.
- The ⅓ know how to handle the specific situation of substrating from an integer, so the operation is performed and the expected result is returned.
Let’s see how it works:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52
Our Number class also has to be adapted to this new approach. Define the methos in the abstract class is very important, because it help to understand the model quickly and expand the behavior easy.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Maybe at first sight it looks difficult to reads, however let’s follow a simple collaboration step by step to see that we obtain exactly what we are looking for.
- The message
#+is sent to the object 3 with the external collaborator ⅕ .
- The 3 send the message
#add_from_an_integerto the ⅕ with himself as external collaborator.
- The ⅕ know specifically what to do because they tell him to add an Integer.
- The correct result is returned.
Just taking advantage of the “go to definition” feature of our favorite IDE we can follow this code without problem. Even debugging is more pleasant because it’s linear: You just need to follow the value of the variables, there is no weird jumps in the code, you don’t have to recreate conditions in your head to find the next line of code to be executed. Note that the objects are handling the decisions for us.
This point is the key. The objects are capable of handling this type of situations (and others much more complex!) by themselves, they don’t need the constant presence of a programmer manifested in the form of an if.Taking advantage of polymorphism we can create objects that care themselves about the situations.
Of course the essential difficulty of the problem doesn’t disappear. We need to
do the things right to get a correct result. For example in the message
have to consider that the minus operation is not commutative. For this reason
consider the external collaborator as a minuend and not as a subtrahend. To
avoid problems the name of the message says it explicity.
Suppose we want to extend our model adding the class ‘IrrationalNumber’. In the
if approach this is just tedious. We would have to travel around all the
methods in every class and add a ‘elsif’ statement with the new behavior
(beside writing the new class). Moreover nothing ensures us that we didn´t miss
if statement in another place.
Now we are gonna see how in the polymorphic version we can add all the necessary code to make our new functionality work in an easy way without touching anything of the existing code and being guided by the model itself.
Just to start we can create the class IrrationalNumber that inherits from
Number. At the beginning this class can define only the messages
1 2 3 4 5 6 7 8 9 10 11
Now is clear that any number should respond the messages
#subtract_from_an_irrational, so we need to define this methos in the number class.
1 2 3 4 5 6 7 8 9 10 11 12 13
At this point if we run any test with our new class the result will be the same:
“Should be implemented in the subclass”. And here it is: The model is not just telling us what
message should be answered, also it is telling us who has to answer it. “An integer
should implement the message
#add_from_an_irrational”. The model itself
guides us over the expansion. It takes only writing some tests to know exactly how to
expand our model.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
In consequence now all our classes need to know how to
The expansion was easy, we didn’t need to touch any of the previous code and the model guided us over the expansion. What else can we ask?
It’s true that at first, all this stuff of delegating the work from an object to another “just to extract an if” can sound a little bit daunting. However… look all we have accomplished. The final model is robust, simple to understand and pretty easy to extend. We did a lot more than just “extracting an if”. In our polymorphic model the objects are responsible of handling the problems and guiding us over the extension of it.
if is not “bad” by itself. In some situations it is just inevitable and
correct. Using the same example but with the
#/ message it’s very difficult
to avoid the
if that verifies that the divisor is not zero. To avoid that
if we would have to make a special class for zero and a specific dispatch,
that would make the solution too complicated. The
if in this case is
I like to see the if as a “bad smell”. Every time it appears I think if I’m losing the chance of using polymorphism and improve my model.
Finally, I’d like to mention something important. In this post I put all my
effort to show the benefits of avoiding the
if. However the
solution I use come out from nowhere, I have never explained the mechanism I used to
came out with the solution. So, there is basically two open questions:
- Does a deductive method or an algorithm exist to remove the
- Can we always replace the
ifwith polymorphism? Is there a limit?
If you are interested in the answer I strongly recomend to see this webinar, where Hernán Wilkinson explain this topic deeply.