Code duplication in unit tests

Let me start with a statement many developers tend to agree:

Code duplication is bad

But is it true for code in unit tests? Should we refactor our unit tests and remove duplications there as well?

From my personal experience, code duplication in unit tests is not only acceptable but desirable.

This is not a dogma or absolute truth. Here I am only sharing what usually works for me personally and why. It may not be the same for everybody. Each of us has unique background, coding style and screen wallpaper.

Anybody home?

First of all, let's ask a question. What is the main thing a unit test does? My answer is - unit test verifies behaviour of some 'code'. Important nuance here is the direction. It is the unit test that verifies its 'code' and not the other way around. The 'code' does not verify its unit test.

Ok, then what verifies the unit test? The answer is - nothing. There is nothing that checks if our unit test is bug-free. Except if we decide to write a unit test for our unit test. Which sounds like a good plot for another Inception movie.

Coffee powered debugger

But it is not quite correct, right? There is one thing that validates a unit test. This 'thing' is called a human being. In order to make sure that our test is valid we just look at it. Our eyes scan the unit test line by line. We run a little debugger in our brain simulating execution of the unit test.

And here is the main problem for me. My brain is pretty bad at debugging. This process of checking the unit test with my eyes is hard. It requires a lot of discipline, concentration and time. There is a big risk that I will miss a bug.

The problem

Ok, so now I have a issue:

  • To verify a unit test I need to look at it.
  • Verifying a test by looking at it is far from reliable.

How do we solve this issue? To be honest, I have no solution. The best I can do is to make unit test as simple as possible in order to make it easier for my eyes to catch the bugs.

I am very poor at debugging in my brain. That's why I try to make unit test code explicit. In most cases I prefer using simple assertions that verify actual and expected values.

Simple assertion example

Let's look at a real life example of an assertion in a unit test. I will use Ruby RSpec syntax in the examples.

Imagine that I need to check a string variable str with expected value 'Hello world'.

I do not compare str with a variable:

# Avoid
str_expected = 'Hello world'
expect(str).to eq str_expected

Instead I prefer to explicitly write 'Hello world' text on the same line with the str variable:

# Good practice
expect(str).to eq 'Hello world'

Example explained

I prefer the last approach because it makes the unit test easier to read and understand. When I check this unit test with my eyes it is obvious what it does. The test ensures that str is 'Hello world'. And it is easy to notice a problem in this test, because it is so simple.

What happens if I use a variable str_expected in the assertion?

When we look at

expect(str).to eq str_expected

we need to jump with our eyes and search for line where str_expected is initialised. It does not sound like much work, but it is additional work. Which requires more of our energy. Which is a limited resource. Especially at 5pm on Monday.

Simple duplication example

Now let's get closer to the main topic - code duplication in unit tests.

I will start with a simple example - duplication of expected value in assertions. Imagine that we now have two assertions that ensure that variables str1 and str2 both equal 'Hello world':

My first instinct is to see that 'Hello world' is mentioned twice, put it to a variable and use this variable in both assertions:

# Avoid
str_expected = 'Hello world'
expect(str1).to eq str_expected
expect(str2).to eq str_expected

This approach has the same problem as I described above. It requires more effort to verify the unit test with one's eyes. It becomes even harder if these two assertions are separated in a file or if they are located in different files.

There is another problem. What if our program logic changed and str2 now equals 'My lovely horse'? Now we need to update our unit test, which becomes even harder to read:

# Avoid (str2 value changed)
str1_expected = 'Hello world'
expect(str1).to eq str1_expected
str2_expected = 'My lovely horse'
expect(str2).to eq str2_expected

To make things simple, I prefer just duplicating 'Hello world' in both assertions.

# Good practice
expect(str1).to eq 'Hello world'
expect(str2).to eq 'Hello world'

And if str2 changes, I just change one line:

# Good practice (str2 value changed)
expect(str1).to eq 'Hello world'
expect(str2).to eq 'My lovely horse'

Win-win! Easy to write. Easy to read.

More serious duplications

Now let's get to the main point. What if we have two unit tests with similar logic:

# Good practice
# Test 1:
person = db.find_person('Peter')
assert(person.name).to eq 'Peter'

# Test 2:
person = db.find_person('Ivan')
assert(person.name).to eq 'Ivan'

It is quite tempting to remove duplication by extracting the common behaviour into a helper method verify_person and then call that method from both tests:

# Avoid
# Test 1:
verify_person('Peter')

# Test 2:
verify_person('Ivan')

# A helper method
def verify_person(name)
  person = db.find_person(name)
  assert(person.name).to eq name
end

I try to void this approach for exactly the same reason I mentioned two times already. This reason is so important so I will repeat it again.

In order to make sure the test is bug-free I need to verify it with my eyes. Inside my brain I go through the unit test code line by line, just like in a IDE debugger. Here is what is happening inside my brain:

  1. Start reading test 1.
  2. Read verify_person('Peter').
  3. I need to know what verify_person does.
  4. I search for verify_person method.
  5. I try to locate verify_person on the same screen, scroll the page and/or search the current file or entire project.
  6. After finding verify_person helper method, I read it.
  7. Return to test 1, and check it as "verified" in my brain.

Now I do the same routine for test 2. Again, it does not seem to be much work and happens very quickly especially after practice. But it is not the point. My point is - it requires more physical effort. Which is a limited resource. Especially at 5pm on Monday.

Now I want to repeat the preferred way of writing these tests:

# Good practice
# Test 1:
person = db.find_person('Peter')
assert(person.name).to eq 'Peter'

# Test 2:
person = db.find_person('Ivan')
assert(person.name).to eq 'Ivan'

Reading these tests is easy. You just go line by line. No need to jump. No need to return back. Stupid simple. And this is exactly what I need at 5pm on Monday.

Conclusion

Unit tests are supposed to reveal bugs and not to add new ones. Right?

Here is the list of best practices that I try to follow:

  • Make test code as simple as possible.
  • Use simple assertions to compare actual and expected values.
  • Put actual and expected value on the same line.

Things to Avoid

  • Avoid using variables for expected values.
  • Avoid extracting common functionality into helper methods
  • In general, avoid creating complex code in your tests.

Forget the rules. Do it your way

Here I shared my own experience in unit testing. I mentioned approaches that help me write better unit tests. I try to follow these principles most of the time. But sometimes there are exceptions. Where it makes sense I do write helper methods, loops and other more complex code in unit tests. It really depends on the situation.

And most importantly - coding should be fun.

Reference







Wooooow!