Refactoring 101
planet images by Justin Nichol on opengameart.org CC-BY 3.0
Goal of this Tutorial
In this tutorial, you will refactor a space travel text adventure.
Starting with a working but messy program, you will improve the structure of the code. Throughout the tutorial, you will apply standard techniques that make your code more readable and easier to maintain. This tutorial is suitable for junior Python developers.
1. What is Refactoring?
When you are working on your first real-world Python project, the codebase is typically much larger than any textbook or course example. Over time, software entropy kicks in: functions grow longer and longer, the same code gets copy-pasted to multiple places and slowly mutates there, and code that seemed a brilliant idea a few weeks back is now incomprehensible. Refactoring aims to prevent your code from becoming a mess.
Refactoring is improving the structure of code without changing its functionality.
In practice, this means thing like:
- remove redundant code segments
- split long functions into shorter ones
- extract data structures
- encapsulate behavior into classes
In this tutorial, you can try all of these. Let's go!
1.1 Recipe: Generic Refactoring
The basic workflow in refactoring is:
1. run the tests
2. edit the code
3. run the tests
2. Getting Started
Clone or download the space travel game from github.com/krother/refactoring_tutorial:
git clone [email protected]:krother/refactoring_tutorial.git
The game is a text-based command-line app that should run in any Python editor/environment. Make sure it runs:
python space_game.py
Play the game for a few minutes to get a feeling what it is about.
3. Run the Tests
A fundamental rule in refactoring is: do not start without automated tests. The space game already has tests in test_space_game.py
. We will use the pytest library. Please make sure it is installed:
pip install pytest
You can run the tests from the refactoring_tutorial/
folder:
pytest test_space_game.py
You should see a message like:
============================= test session starts ==============================
platform linux -- Python 3.8.10, pytest-6.1.2, py-1.9.0, pluggy-0.13.1
rootdir: /home/kristian/projects/refactoring_tutorial
plugins: flake8-1.0.7, Faker-8.9.1, asyncio-0.15.1, cov-2.10.1, dash-1.18.1, anyio-3.5.0
collected 12 items
test_space_game.py ............ [100%]
============================== 12 passed in 0.04s ==============================
To see the game output, do:
pytest -s test_space_game.py::test_travel
4. Identify problematic Code
Now take a look at the main file space_game.py
. Look for problematic sections that you would want to refactor. Note that the code has been linted (with black). We are not looking for missing spaces or other style issues.
Look for the following:
- long Python modules
- long functions that do not fit on a screen page
- duplicate sections
- code sections that are similar
- code with many indentation levels
- names of functions that are not descriptive
- mixture of languages (e.g. HTML / SQL inside Python code)
- code that mixes different domains together (e.g. user interface + business logic)
- code that could be expressed more simply
- code that you find hard to read
Mark everything you find with a #TODO
comment.
5. Extract a Module
Let's do our first refactoring. The first half of the code consists of a huge dictionary TEXT
. Let's move that variable into a new Python file in the same folder.
- create an empty Python file
text_en.py
- cut and paste the entire dictionary
TEXT
- add an import
from text_en import TEXT
- run the tests again
The tests should still pass.
This refactoring creates a separation of domains. Now it is a lot easier to e.g. add a second language.
6. Extract Functions
The most fundamental refactoring technique is to split a long function into shorter ones.
We will make our toplevel function travel()
easy to read. For that, we chop it into smaller pieces. By creating smaller functions, we either clean up the mess right away or at least create a smaller mess that is contained locally.
We will use the following recipe:
6.1 Recipe: Extract a function
This recipe has a few more steps:
1. Find a piece of code you want to move into a function
2. Give the function a name and create a `def` line
3. Move the code into the new function
4. Make a parameter out of every variable not created inside the function
5. Add a return statement at the end with every variable used later
6. Add a function call where you took the code
7. Run the tests
Let's do this on a few examples:
6.2 Exercise: extract display_inventory
The paragraph labeled display inventory on top of travel()
makes a good refactoring candidate. Create a new function using the signature:
def display_inventory(credits, engines, copilot)
This function does not need a return statement.
Do not forget to run the tests afterwards.
6.3 Exercise: extract select_planet
Extract a function select_planet()
from the last code paragraph from the travel()
function.
This function needs a single parameter and a single return value. Find out what signature the function should have.
Work through the recipe for extracting a function.
7. Extract and Modify
Sometimes, you need to modify a function to move it elsewhere.
7.1 Exercise: extract visit_planets
To get a short and clean travel()
function, it would be good to move the huge block with nested if
statements out of the way. Let's extract a function visit_planets()
. Start with the recipe for extracting a function.
Use the signature:
def visit_planet(planet, engines, copilot, credits, game_end):
...
and the function call:
destinations, engines, copilot, credits, game_end = \
visit_planet(planet, engines, copilot, credits, game_end)
When you refactor the code, the tests should fail!
7.2 The function does not work
When you follow the recipe for extracting functions, the tests break. Something does not quite fit. The code block contains an extra return
statement (in the black hole section).
We need to modify two things to keep the code working:
- Replace the
return
statement bygame_end = True
- Move the line printing end credits into the conditional branch where your copilot saves you
Then run the tests. They should pass now.
7.3 How many functions should you extract?
In an ideal world, each function does exactly one thing. What does that mean?
In his Clean Code Lectures, Uncle Bob (Robert C. Martin) states:
Q: When is a function doing exactly one thing?
A: When you cannot make two functions out of it.
Although this is generally a good idea, you do not have to decompose everything right away. Often there are other, more important refactorings to take care of.
8. Extract Data Structures
After extracting a module and functions, the travel()
function became a lot shorter already. But there are still many things to improve. Let's focus on the data structures:
8.1 Exercise: Extract boolean flags
The function signature of visit_planet()
is not very pretty. It contains a long list of boolean arguments. This was less obvious before. Our refactoring has exposed a problem with the data structures (or lack thereof). Let's take a closer look:
The game progress is controlled by the booleans: copilot
, credits
, engine
and game_end
. These booleans are passed around several times. This is a sign that they could be placed in one data structure.
What Python data structure can we use to store the presence or absence of multiple items?
Let's use a Python set
that we call flags
. We need to modify a lot of code.
First, instead of setting multiple booleans to False
in travel()
, define an empty set.
flags = set()
Create a preset list of values on top of the module (avoids having quotes everywhere):
credits, engine, copilot, game_end = range(4)
To check a flag, we would use its name as a string. So the while
condition in travel()
would become:
while not ('crystal_found' in flags or 'dead' in flags):
Now, we need to change the function display_inventory()
as well:
- replace the boolean arguments by a single argument
flags
- modify the function call accordingly
- modify the function body to use the
in
operator when checking state, e.g.if credits in flags:
We need to do the same with visit_planet()
- replace the boolean arguments by a single argument
flags
- modify the function call accordingly
- remove the booleans from the return values (the set is mutable).
visit_planet()
only returnsplanet
anddestinations
. - remove the booleans from the assigned return in
travel()
as well - modify the function body to use the
in
operator when checking state, e.g.if credits in flags:
- modify the function body of
visit_planet()
. Whenever one of the booleans is modified, add to the set, e.g.flags.add(game_end)
Finally, run the tests again. The tests should pass.
Note that looking up things in the set uses string comparison. This is not very performant, of course, but in a text adventure I frankly don't care. If performance becomes important, one could replace the strings by integers or Enums. Also, if you believe performance is important, how about writing a performance test for it first?
8.2 Extract puzzle functions
The visit_planet()
function is still very long. Now is a good moment to decompose it further. Create a function for the hyperdrive shopping scene on Centauri.
The code left in visit_planet()
should look like this:
if planet == "centauri":
print(TEXT["CENTAURI_DESCRIPTION"])
destinations = ["earth", "orion"]
buy_hyperdrive(flags)
Do the same for the other puzzles:
def star_quiz(flags):
def hire_copilot(flags):
def black_hole(flags):
Now visit_planet()
should approximately fit on your screen.
8.3 Exercise: Extract a dictionary
The destinations can be placed in a data structure as well. With each planet in visit_planet()
there is always a list of destinations returned.
Let's use the following dictionary instead:
STARMAP = {
'earth': ['centauri', 'sirius'],
'centauri': ['earth', 'orion'],
'sirius': ...,
'orion': ...,
'black_hole': ['sirius'],
}
- place the dictionary on top of the Python file
- fill in the two missing positions
- remove the individual definitions of
destinations
- instead, at the end of the
visit_planet()
function, look up the destinations withreturn STARMAP[planet]
- run the tests
The tests should pass.
9. Extract a Class
By now, the visit_planet()
function has not changed much. We managed to save a couple of lines by extracting the STARMAP
dictionary. But there is still has a huge nested if
block. Let's see what we can do.
9.1 Are more dictionaries a good idea?
Should we maybe extract the descriptions of each planet into another dictionary? We would get:
PLANET_DESCRIPTIONS = {
'earth': TEXT['EARTH_DESCRIPTION],
'sirius': TEXT['SIRIUS_DESCRIPTION],
...
}
You could do this, and it would further simplify visit_planet()
. But seeing multiple dictionaries with the same keys is a clear hint that there is a deeper structure in our code. We will extract a class.
9.2 Exercise: The Planet class
We find a couple of things that the planets have in common:
- every planet has a name
- every planet has a description
- every planet has connections to other planets
These are attributes of the new class.
Let's define a new class with the following signature:
class Planet:
def __init__(self, name, description, connections):
self.name = name
self.description = description
self.connections = connections
Run the tests to make sure you didn't mess up anything (even though we do not use the class yet).
9.3 Exercise: Add a method
We will convert the function visit_planet()
into a method of the new Planet
class.
Move the entire code from visit_planet()
into a new method with the signature:
def visit(self, flags):
As the first thing, have the planet print its own description:
print(self.description)
That removes a few lines from the function and makes the code easier to read.
The tests won't pass at this point. You may want to run them to make sure you are editing the right file.
9.4 Exercise: Create instances
Let's create a dictionary of planets. We will do so on the module level, replacing STARMAP
:
PLANETS = {
'earth': Planet('earth', TEXT['EARTH_DESCRIPTION', ['centauri', 'sirius']]),
...
}
We use the Planet
instances in the travel()
function. The code should be
planet = PLANETS['earth']
...
while ...:
planet.visit(flags)
display_destinations(planet)
planet = select_planet(planet.connections)
Note that you need to modify these methods slightly.
At this point, the tests should pass.
9.5 Exercise: Breaking down the visit function
Finally, we have restructured our code to a point where we can decompose the huge block of if
statements.
Some planets have a puzzle. Add a puzzle attribute to Planet.__init__()
Next, we pass these functions as callbacks in the puzzle
argument when creating Planet
objects. One entry in the PLANETS
dict would look like:
'sirius`: Planet('sirius', TEXT['SIRIUS_DESCRIPTION'], star_quiz)
Now in the visit()
method, all you need to do is call the callback:
if puzzle:
puzzle(flags)
And the multiple if
statements should evaporate.
10. Other Refactoring Strategies
10.1 Names matter
"Planet" is not an accurate name from an astronomic point of view. On the other hand, I would refuse to call anything "System" on a computer, because it may mean anything.
From a game design point of view, "Room" or "Location" could be better. These are good questions to discuss with the domain experts and colleagues on your team. Finding common vocabulary is one good side effect successful refactoring may have.
10.2 Programming paradigms
When refactoring Python code, you often have multiple options. It helps if you have a programming paradigm in mind that you are working towards, such as:
- functional programming with stateless functions that can be recombined
- strictly object-oriented programming
- hybrid architecture with core classes and toplevel functions
- look for specific Design Patterns that describe well what your code is doing
- practice TDD and write additional tests when extracting larger units of code
In my experience, refactoring is much about executing a few standard techniques consistently.
You find a great list of refactoring techniques on refactoring.guru by Alexander Shvets.
10.3 Embrace future change
In refactoring, you always want to separate things that are likely to change from things that don't. What might change in a text adventure?
- connections between planets
- puzzles on the planets
- new planets
- almost any text
- a graphical or web interface (replacing the
print()
statements would justify a complete rewrite in this case)
With well-refactored code, any of the above should require changing a single location in the code.
In the end, our rectorings should make it easy to add more planets, puzzles or write a completely new adventure.
Give it a try and have fun programming!
11. Closing Remarks
Refactoring is like washing. It is most effective if repeated regularly.
Of course, one could wait for two weeks, so that taking a shower is really worth it. But in practice this is not such a good idea, at least not if you are working with other people.
It is the same with refactoring.
License
(c) 2022 Dr. Kristian Rother [email protected]
This tutorial is subject to the MIT License. Have fun sharing!
See LICENSE for details.