Test Driven Development with pytest

In this session, we will learn the basics of Test Driven Development. Our tool for accomplishing this will be pytest, a powerful and pluggable test runner.

What is TDD

Untested code is broken by design

—Surely Somebody

As developers, we all “know” that we need to write tests. But developing the discipline to do so is a burden. Especially when you are brand new. You have so many other things to deal with, so many things to learn. Surely there are more important things you need to do, right?

NO

There is no more important discipline you can develop than writing tests.

Defining The Problem

So why do so many of us not write tests? I can’t speak for anyone else. But for me the biggest hurdle is that test writing is so different and so unrelated to the work I’m doing in developing software.

I develop interactively, building an understanding of the problem I’m facing by working in the interpreter until I feel I’ve got a good understanding of the problem. Only then do I move to the editor and begin building actual code. By that time, I’m so accustomed to working in the terminal, that interactively “testing” my work by poking at it in an interpreter seems intuitive.

Once I’ve finished a solution, and played with it enough interactively to believe it works, writing tests for it becomes another thing I have to do. Now that I’ve done my primary job, I have to move over to this other file, write a bunch more code, and run it repeatedly. The work feels so extraneous to what I consider my real job that I am sorely tempted to skip it.

(I’m embarassed to say I often give in to the temptation too)

The problem is that the act of testing is too far removed from development. It happens at a different time than my development work. And it happens in a different conceptual space than my development work.

Proposing a Solution

The solution should be pretty clear. If the problem is that development and testing are too far apart, then move them closer together. This is the aim of Test Driven Development. To close both the temporal and conceptual gaps between the worlds of development and testing.

Closing the Conceptual Gap

The conceptual gap is that space between what is perceived development work and what seems to be testing work.

Especially in interpreted languages like Python, it’s easy to get started solving a problem by playing with the code. We open an interpreter and write some lines. We try different approaches, explore possible solutions. We reject failed paths and zero in on a correct answer. It’s a process of coming to understand the shape of the problem we’ve been asked to solve.

What we don’t notice is that this process is not actually one of development but of testing. Thinking of this as development is actually backwards.

What if we were to put the label testing on this type of work instead? Then we could start by writing tests that help us to determine the shape of our problem. Then, implementing code to make the tests pass proves that we have achieved a correct solution.

Let’s imagine that later we discover we have a poor understanding of the problem. Maybe the problem is a little different than we at first believed. Maybe there are edge cases we didn’t anticipate.

At that point, we can return to our tests. update them to better describe our new understanding. Or we can add new tests to cover the unexpected circumstances we are now aware of.

These new tests will fail. So we can return to our implementation code. We update it so that our new tests pass, an ensure the other tests keep passing. And when we are done, we know we have a working system.

This is the cycle of Test Driven Development. Some developers call it “Red-Green-Refactor”

We start by writing a test that shows what what want to have happen (Red)

Then we implement code that makes that test pass (Green)

And then as our understanding of the problem evolves, we update (Refactor)

It’s a nice, neat cycle.

Closing the Temporal Gap

The temporal gap is the space in time between when we write implementation code and when we write tests.

Traditionally, we do our development and then, later, we write tests. When we do this, the large temporal gap makes it hard to actually do the work of writing the tests. It’s entirely too tempting to say our real work is done.

What if we were to move the act of development and the act of testing much closer in time? What if we focused on writing small tests, then small functions, then refactoring? Could we bring the two so close in time that the two separate acts become one?

Test Driven Development seeks to close this gap. In “true” TDD, you are supposed to write no more test than is needed to have it fail. Then you write no more code than you need to make the test pass. Then you repeat.

Again, this is the “Red-Green-Refactor” cycle. If we successfully implement this approach, then the temporal gap between development and testing disappears. The impediments to testing evaporate. And we can write well tested code all the time.

Exploring TDD

Enough theory, let’s set about exploring TDD by building a little project using it. At the same time, we’ll learn a bit about the pytest testing framework.

Project Setup

Begin by creating a project directory, then move into that directory.

Banks:~ cewing$ mkdir tdd-play
Banks:~ cewing$ cd tdd-play
Banks:tdd-play cewing$

Next, let’s create a virtualenv in this directory and activate it.

Banks:tdd-play cewing$ python3 -m venv ./
Banks:tdd-play cewing$ source bin/activate
[tdd-play]
Banks:tdd-play cewing$

Development of pip and setuptools has been very rapid lately. And the new developments in each are usually worth having on board. So I generally follow the creation of a new virtualenv with a quick update of pip and setuptools:

[tdd-play]
Banks:tdd-play cewing$ pip install -U pip setuptools
...
Successfully installed pip setuptools
Cleaning up...
[tdd-play]
Banks:tdd-play cewing$

Now, let’s install our tools for this project. We’ll start with pytest:

[tdd-play]
Banks:tdd-play cewing$ pip install pytest
...
Successfully installed py-1.4.31 pytest-2.8.7
[tdd-play]
Banks:tdd-play cewing$

The pytest package provides a customized test runner with lots of bells and whistles. It also makes it very easy to start writing tests. When you invoke the command it begins looking for Python files with names that start with test_. Once it finds them, it reads them and looks for functions with names that start with test_. After gathering them together, it runs them all, one at a time. At the end, it reports the results to you.

Let’s make a directory to hold our code for this project. We’ll call it src because that’s the conventional name for such directories.

[tdd-play]
Banks:tdd-play cewing$ mkdir src

Then, create a new file in this directory. Call it test_ack.py:

[tdd-play]
Banks:tdd-play cewing$ touch src/test_ack.py

Our First Test

We can get started with just that. Open the src directory in your text editor. In the test_ack.py file, add the following code:

# -*- coding: utf-8 -*-

def test_foo():
    assert 1 == 0

We’ve now written a test. Let’s run it:

[tdd-play]
Banks:tdd-play cewing$ py.test
======================= test session starts ========================
platform darwin -- Python 3.5.1, pytest-2.8.7, py-1.4.31, pluggy-0.3.1
rootdir: /Users/cewing/projects/training/codefellows/tests/tdd-play, inifile:
collected 1 items

src/test_ack.py F

============================= FAILURES =============================
_____________________________ test_foo _____________________________

    def test_foo():
>       assert 1 == 0
E       assert 1 == 0

src/test_ack.py:5: AssertionError
===================== 1 failed in 0.22 seconds =====================

The pytest test runner found our test_ack.py file. It found the test_foo function defined in it. And it ran the test.

The test raised an AssertionError because one is definitely not equal to two. Then pytest captured that error and reported it to us as a failing test.

Next, let’s make our test pass. Change the code in test_foo so that the assertion is true:

# -*- coding: utf-8 -*-

def test_foo():
    assert 1 == 1

And then re-run your tests:

[tdd-play]
Banks:tdd-play cewing$ py.test
======================= test session starts ========================
platform darwin -- Python 3.5.1, pytest-2.8.7, py-1.4.31, pluggy-0.3.1
rootdir: /Users/cewing/projects/training/codefellows/tests/tdd-play, inifile:
collected 1 items

src/test_ack.py .

===================== 1 passed in 0.17 seconds =====================

Great!

We’ve made our first “Red-Green-Refactor” cycle. But it’s kind of a pain to have to go back and re-run our tests after changing that file. Can we get pytest to do that for us? Yes we can!

Test On Save

The pytest system is pluggable. We can install plugins to provide different kinds of functionality. That includes letting us automatically re-run our tests every time a test in our project is updated. To get that particular functionality, let’s install pytest-xdist:

[tdd-play]
Banks:tdd-play cewing$ pip install pytest-xdist
...
Successfully installed apipkg-1.4 execnet-1.4.1 pytest-xdist-1.14
[tdd-play]
Banks:tdd-play cewing$

To use this new feature, we invoke the py.test command with the -f option:

[tdd-play]
Banks:tdd-play cewing$ py.test -f
============================= test session starts ==============================
platform darwin -- Python 3.5.1, pytest-2.8.7, py-1.4.31, pluggy-0.3.1
rootdir: /Users/cewing/projects/training/codefellows/tests/tdd-play, inifile:
plugins: xdist-1.14
collected 1 items
collected 1 items

src/test_ack.py .

=========================== 1 passed in 0.19 seconds ===========================
####################### waiting for changes ########################
### Watching:   /Users/cewing/projects/training/codefellows/tests/tdd-play

If you don’t see color output for the line reporting our test passed, try this:

$ py.test -f --color=yes

Now that that’s in place, let’s begin working on our project.

The Ackermann Function

The Ackermann Function is a recursive mathematical function. Its primary characteristic is that for even very small inputs it produces very large outputs. It also recurses a very high number of times in computing its value. For that reason, it is sometimes used to demonstrate the effectiveness of compiler’s optimizations for recursion.

The function takes two inputs, m and n, and produces a single numeric value. The wikipedia page for the function gives a nice table of output values, given values for the two inputs. Here’s a sampling:

m\n 0 1 2 3 4
0 1 2 3 4 5
1 2 3 4 5 6
2 3 5 7 9 11
3 5 13 29 61 125

So this is the problem domain we have to solve. Without thinking too deeply about it, we can see we need to write a function that will take two numbers and return one. We can also see that, for example, if both of the inputs are 0, then the output is expected to be 1. That’s testable, so lets test it.

Back in your text editor, add a new test to the test_ack.py file. Let’s call it test_ackermann_0_0. We’ll start with the smallest amount of code that will fail:

def test_ackermann_0_0():
    from ackermann import ackermann

Notice that our tests started running again as soon as we saved that file. Here’s what the error says:

# MODIFIED /Users/cewing/projects/training/codefellows/tests/tdd-play/src/test_ack.py
============================= test session starts ==============================
platform darwin -- Python 3.5.1, pytest-2.8.7, py-1.4.31, pluggy-0.3.1
rootdir: /Users/cewing/projects/training/codefellows/tests/tdd-play, inifile:
plugins: xdist-1.14
collected 2 items
collected 2 items

src/test_ack.py .F

=================================== FAILURES ===================================
______________________________ test_ackermann_0_0 ______________________________

    def test_ackermann_0_0():
>       from ackermann import ackermann
E       ImportError: No module named 'ackermann'

src/test_ack.py:9: ImportError
====================== 1 failed, 1 passed in 0.21 seconds ======================
########################## LOOPONFAILING ###########################
src/test_ack.py::test_ackermann_0_0
####################### waiting for changes ########################
### Watching:   /Users/cewing/projects/training/codefellows/tests/tdd-play

Okay, so we have an ImportError. We don’t have an ackermann.py module anywhere. And it certainly doesn’t contain an ackermann function. Let’s fix that, minimally.

Create a new file in the same src folder, called ackermann.py. Then add the following code:

# -*- coding: utf-8 -*-


def ackermann(m, n):
    pass

And again, our tests will run themselves. This time, they’ll run the failing test once. And then, because that passes, they will re-run all the tests to make sure we didn’t break anything else.

# MODIFIED /Users/cewing/projects/training/codefellows/tests/tdd-play/src/ackermann.py
============================= test session starts ==============================
platform darwin -- Python 3.5.1, pytest-2.8.7, py-1.4.31, pluggy-0.3.1
rootdir: /Users/cewing/projects/training/codefellows/tests/tdd-play, inifile:
plugins: xdist-1.14
collected 3 items
collected 3 items

src/test_ack.py .

=========================== 1 passed in 0.01 seconds ===========================
============================= test session starts ==============================
platform darwin -- Python 3.5.1, pytest-2.8.7, py-1.4.31, pluggy-0.3.1
rootdir: /Users/cewing/projects/training/codefellows/tests/tdd-play, inifile:
plugins: xdist-1.14
collected 2 items
collected 2 items

src/test_ack.py ..

=========================== 2 passed in 0.20 seconds ===========================
####################### waiting for changes ########################
### Watching:   /Users/cewing/projects/training/codefellows/tests/tdd-play

Let’s return to our test now and make it test something useful. Let’s assert that when we call that function with 0 and 0 that we get 1 back:

# in test_ack.py
def test_ackermann_0_0():
    from ackermann import ackermann
    assert ackermann(0, 0) == 1

And when our tests run again, there will be one failure.

Now let’s return to our implementation code in ackermann.py. We can write just enough to make that test pass:

# in ackermann.py
def ackermann(m, n):
    return 1

And again, the tests are green!

Let’s try a different number combination. For the inputs 0 and 1 the function should return 2. Let’s add a new test that claims that is true:

# in test_ack.py
def test_ackermann_0_1():
    from ackermann import ackermann
    assert ackermann(0, 1) == 2

That turns our tests red again. Two are passing, but one is not. Let’s add some code that fixes that. Maybe at this point, we should look at the definition of the function. Wikipedia says that if m is equal to 0, then the return value of the function is n + 1. We understand that. Let’s make our function do that.

# in ackermann.py
def ackermann(m, n):
    if m == 0:
        return n + 1

And our tests are back to being green. Let’s add another test and see if that holds true for the next values from our table. For m = 0 and n = 2 the function should return 3.

# in test_ack.py
def test_ackermann_0_2():
    from ackermann import ackermann
    assert ackermann(0, 2) == 3

Oooooh! Our tests are still green! Does that mean we are done? Well, there are still quite a few tests left to write, even to finish out the table we have above. We can’t really be sure we are finished until all of them pass. But who wants to write 16 more tests?

Let’s use the power of our testing framework to help us out of this jam.

In pytest, you can parametrize tests. This allows you to specify a single test, and a list of inputs you will provide. The framework will run the test once for each input in your list. We can use this to write just one test that will test all the values in our handy chart above.

Let’s begin by refactoring our test so we only need one. Back in test_ack.py let’s erase our test_ackerman_m_n tests. We’ll replace them with a single function where m, n and the expected result are paramters of the test:

# in test_ack.py
def test_ackermann(m, n, result):
    from ackermann import ackermann
    assert ackermann(m, n) == result

As expected, our tests are failing again. Let’s fix that. First, at the top of the file, import the pytest package. The tools we want are there.

# at the top of test_ack.py, just below the coding statement
import pytest

Then, on the line just above our test function, add the following code:

@pytest.mark.parametrize('m, n, result', [(0,0,1), (0,1,2), (0,2,3)])
def test_ackermann(m, n, result):
    # our test code here.

And if you’ve done your job correctly, then you should see four tests run and all pass. For each of the input sets in our new line of code, a single test is run.

The pytest.mark.parametrize call takes as its first argument a string that names the parameters for the test, separated by commas. This list should exactly match the parameters you listed in the test below. The second argument to the call is a list of tuples. Each tuple will supply the arguments for one call of the test function. So there must be the same number of items in each as in the parameter list of the test.

We can now add the rest of our chart to the file in the same place. But that will make for a really long line of code. It won’t be very readable. Let’s make a module constant instead. Then we can use that constant in the call to pytest.mark.parametrize.

# in test_ack.py
ACK_TABLE = [
    (0, 0, 1),
    (0, 1, 2),
    (0, 2, 3),
    (0, 3, 4),
    (0, 4, 5),
    (1, 0, 2),
    (1, 1, 3),
    ...
]

Fill out the entire table. There should be twenty tuples. Then, update the test code:

# in test_ack.py
@pytest.mark.parametrize('m, n, result', ACK_TABLE)
def test_ackermann(m, n, result):
    # our test code here.

When you save your file, you should see 21 tests run. Six of them will pass, but the rest will fail. Now we can finish implementing our ackermann function.

Back in ackermann.py, implement the rest of the function. Wikipedia gives us this:

if m > 0 and n == 0 --> ackerman(m - 1, 1)
if m > 0 and n > 0 --> ackerman(m -1, ackerman(m, n - 1))

Armed with that information, can you finish implementing the ackermann function and make the tests pass?

Wrap Up

We’ve learned a lot here. We discussed the reasons that testing has traditionally been so hard to get done. We learned about one approach to solving the problem: Test Driven Development. We also learned about a testing framework in Python that has tools to help us to do TDD. And finally, we implemented a little project using TDD principles.

Now, you go and use these tools for your work tonight and going forward.