TLDR; No, if supported by other tools. Yes, if that is your only target.

Recently I was in a discussion in my current company where a respected colleague told me that aiming for a 100% code coverage is considered bad practice.

For context, my team is developing one of the services in an event driven system. Feeling lucky that I got the chance to develop something from scratch, I set the minimum required code coverage percentage as 100 in our CI pipeline. We never went under 100 since.

My colleague told me that I could find many blog posts around the internet explaining why it is considered bad practice.

Previously, I worked on many projects where there was very little or no code coverage at all. My personal record (excluding personal projects) was 42% and even that project had zero coverage when I joined the company. I had to push and convince the management and the engineers to write tests and increase the coverage.

I know from experience how bad and how hard is to work in an environment where there are no tests at all. Any change has the possibility to break totally unrelated parts of the project and you might never find out until a customer stumbles upon that part of your product. This was so frustrating that when I got the chance to work on a new project from scratch, I set the minimum code coverage requirement to 100.

Going back to our conversation with the colleague; right after our talk I quickly searched for some blog posts explaining why aiming for 100 is considered bad. Now I will go through the process here once again and I’ll do my best to explain how this might actually be a good idea.

So I searched for “100 code coverage bad“. The search was in robot language but it did the job:

I’ve read all the links in the first page and all the answers in the stackoverflow page.

First of all, no post is directly saying 100% code coverage is bad. Most of them say things like:

“It’s bad if done poorly”,

“It’s bad if you start with zero coverage in a big project”,

“Just aiming for 100% coverage is not enough by itself”.

The last post from the screenshot is even defending 100% code coverage and I am sure I would be able to find many other examples for both sides and possibly many other valid views.

I feel like I should point out that I am well aware that blindly aiming for 100% code coverage is bad and it doesn’t guarantee anything. That’s why there are other tools you should integrate to your pipeline and/or development flow to help you understand and later increase the level of quality of your project.

Mutation Tests

Mutation testing is a way to help you determine if your code coverage is good or bad. It measures your tests success to changes in your code so even if you have 100. For example, let’s look at the following code and pretend this is our project:

# main.py
def sum(x, y):
    return x + y

# tests.py
def test_sum():
    assert 0 == sum(0, 0)

In this very simple example we have a 100% code coverage. It is very obvious that the test is bad but still, I cannot say that this won’t ever happen, I can almost guarantee that somewhere in the world some engineer will write a test equivalent (in uselessness) to this one. Maybe s/he was tired that day, no blaming.

To protect ourselves against these kinds of situations we set up mutation tests. The mutation test run will analyze your code and will change things around one by one and run all your tests for each change (it will take a very long time so better to run mutations concurrently in many parallel pipelines). For this example the mutation test run will change to code to “x – y” and see if any of your tests fail. Our test_sum() function won’t fail because zero minus zero is still zero. Now, we have what we call a mutant that managed to stay alive. Our test wasn’t able to kill this mutant. One out of one alive mutants is still alive and this means our test just sucks. We are made aware by the mutation test run that we should either fix our current test or add more tests to cover this specific mutant.

# tests.py
def test_zero_sum():
    assert 0 == sum(0, 0)


def test_sum():
    assert 5 == sum(2, 3)

In the above example we just fixed the issue by adding another test. If we ran the same mutation test again our “test_sum()” would fail and kill the mutant hence giving us 100% mutation test coverage. Having 100% test coverage plus 100% mutation test coverage tells us that our code is better protected against simple changes. I am well aware that even with 100% test coverage and 100% mutation test coverage your complex logic might fail somewhere. This doesn’t mean all this coverage is useless, it just means you need another test to cover your high complexity bug. Aiming for 100 code coverage with a high percentage (above 80% maybe?) mutation test coverage will help you sleep better and be more confident when merging new code.

Code Quality

In my early days as a computer science student the only conventions that I was aware of were camel case vs snake case. If you are using Java, the convention is usually camelCase and if you are using python you use snake_case. You should be consistent in your project by sticking to either one of them so your project doesn’t burn unnecessary brain power of other developers by making them switch from one convention to the other.

Turns out there is much more to this. There are programming styles with tools that check your code style on every commit. Tools that check your import statements (depending on your programming language) and automatically order them alphabetically. Tools to check the overall code quality of your project. There are measurements to calculate your code complexity like cyclomatic complexity and/or cognitive complexity. Tools that parse your code and warn you for duplicate code, long functions and long files and so on…

These tools help you keep things clean and when things are clean every aspect of your development process will be better. You will get more readable code hence less time to understand the code. You will get smaller low complexity functions which are much easier to test.

Engineers will be happier, more motivated and highly confident because everything is easier to do in these kinds of projects.

An Example

Lets say with are working on the following tech stack:

• A Python project.
• Automated checks on every commit with git pre-commit hooks.
• Code quality checks and minimum code coverage check on Gitlab CI/CD.
• Mutations tests with mutmut.

Python Project

For our project we can use the following tools:

• Flake8 for style guide enforcement.
• Black for automated code formatting.
• isort to sort your imports automatically.
• mypy for static type checking.

Set up pre-commit Hooks

The pre-commit project will save you huge amounts of time by automatically checking things you specify at every commit. Here is an example .pre-commit-config.yaml:

repos:
  - repo: https://github.com/pre-commit/pre-commit-hooks
    rev: v3.2.0
    hooks:
      - id: trailing-whitespace
      - id: end-of-file-fixer
      - id: check-yaml
      - id: debug-statements
      - id: check-ast
      - id: check-added-large-files
        exclude: ^static/
  - repo: local
    hooks:
      - id: black
        name: Black
        description: "Black: The uncompromising Python code formatter."
        entry: poetry run black
        exclude: ^.*\b(migrations|schema_registry)\b.*$
        types: [python]
        language: system

      - id: flake8
        name: Flake8
        description: "`flake8` is a command-line utility for enforcing style consistency across Python projects."
        entry: poetry run flake8
        types: [python]
        language: system

      - id: isort
        name: isort (python)
        description: "isort your imports, so you don't have to."
        entry: poetry run isort
        exclude: ^.*\b(migrations|schema_registry)\b.*$
        types: [python]
        language: system

CI Tools

I am not going into much detail here. The point is that you should set up a code quality analyzer in your CI pipeline. For example gitlab offers built in support for code climate. You can set up some configuration files in your pipeline config and you are ready to go.

The code quality tool will automatically generate reports like these:

You just have to set your teams culture so that the team prioritizes code quality issues. This will help you keep your code clean and maintainable thus greatly lowering your development costs.

Mutation Tests

For our python example there is a library called mutmut. For me it did the job out of the box. I just had to add some ignores and other small optimizations for some mutations that didn’t really make sense in our case.

Since every single mutation has to run your entire test suite, mutation tests take a very long time to complete so it’s better if you schedule a pipeline to run concurrently on your code repository of choice. Running pipelines costs some money so to keep things cheaper we run mutation tests once a month but running once a week would be fine if you have the bandwidth to handle them.

Conclusion

Like most subjects in our field there is no definitive answer. It’s another “it depends on other things” case. 100% code coverage is good if you support it with other things such as the ones I described and possibly much more. It is also better if you start your project from scratch and aim for the hundred from the very beginning. If you join a project after years of development and there is no coverage at all then aiming for 100 doesn’t really make sense. It is probably best to try to cover as much as possible and aim for most critical parts of the system in such cases.

I see and hear a lot of “80-85% is enough” statements but I don’t think that’s the case. You see, now you have to decide what stays in the untested 20%. Why wouldn’t you test that? What if there’s a bug in that code? Do you write a test or fix the bug and leave it like that? Since there’s no obligation to satisfy a percentage any engineer can leave or forget to write tests for a critical part of a feature. Code reviews might catch some of these but ultimately the code review process is just another human being trying to find bad stuff in your code.

My conclusion (these days) is: Aiming for 100% code coverage + 100% mutation test coverage + high quality code is the way to go.

I always read about design patterns. I read and read again and again and again… But the feeling that I have to learn more or understand more never goes away. Every time I dive into a pattern I feel like I’m learning something new about it. So this post is about me trying to grasp design patterns better (again). I started writing this post while spending time around this book: Design Patterns: Elements of Reusable Object-Oriented Software. As the book goes on, I will copy and paste some parts of the book here, while adding my opinions or questions; if I have any. I believe this will strengthen my perception of design patterns and at the same time, it may help some others.

Now, I will try to describe what a design pattern is, with my own (english) words, as I understand it now, at this very moment. Then I will go to wikipedia and/or google and copy and paste the description here and see the difference. I’m hoping there won’t be a huge difference :). Here is my definition of a design pattern:

A design pattern is a method to meet a recurring requirement.

So this is what I came up with. Actually it took me a while. First I had to write it in turkish on paper. Then I had to fix the turkish version, then I had to translate it to english. That was the result. So now I’m googling it and here it is:

In software engineering, a design pattern is a general repeatable solution to a commonly occurring problem in software design.

The context of this article is already software engineering and software design so this definition can be shortened to this:

A design pattern is a general repeatable solution to a commonly occurring problem.

I think this sounds better than my definition. I especially like the “commonly occuring problem” part. It is much better than “recurring requirement”. I think I will change my definition after this to: “a method to meet a commonly occuring requirement.” Now the main difference is my “method” is their “general repeatable solution” and my “commonly occuring requirement” is their “commonly occuring problem”. A method vs a solution and requirement vs problem. This can be further discussed but both definitions are pretty close to each other in my opinion.

A much simpler definition can be found in the book:

[A design pattern] is a solution to a problem in a context.

Now that we know what a design pattern is, I will list the design patterns I encountered in the book I mentioned above. This part will be mostly copy/pasting the names and definitions. Here is an overview of 23 design patterns:

• Abstract Factory: Provide an interface for creating families of related or dependent objects without specifying their concrete classes. For a long time I didn’t know the difference between factory pattern and abstract factory pattern.
• Adapter: Convert the interface of a class into another interface clients expect.
• Bridge: Decouple an abstraction from its implementation so that the two can vary independently.
• Builder: Separate the construction of a complex object from its representation so that the same construction process can create different representations. For example, libraries like doctrine use builder pattern to build sql queries.
• Chain of Responsibility: Avoid coupling the sender of a request to its receiver by giving more than one object a chance to handle the request. Chain the receiving objects and pass the request along the chain until an object handles it. Thoughts: for example, this is how django middlewares work. It passes the request to the middleware objects one by one, they all handle it or one of them stops the handling process.
• Command: Encapsulate a request as an object, thereby letting you parameterize clients with different requests, queue or log requests, and support undoable operations. What are undoable operations? I don’t really understand this definition.
• Composite: Compose objects into tree structures to represent part-whole hierarchies. Composite lets clients treat individual objects and compositions of objects uniformly.
• Decorator: Attach additional responsibilities to an object dynamically. Decorators provide a flexible alternative to subclassing for extending functionality. Decorator pattern is among the most popular ones, at least for me.
• Facade: Provide a unified interface to a set of interfaces in a subsystem. Facade defines a higher-level interface that makes the subsystem easier to use.
• Factory Method: Define an interface for creating an object, but let subclasses decide which class to instantiate. Factory Method lets a class defer instantiation to subclasses.
• Flyweight: Use sharing to support large numbers of fine-grained objects efficiently. I had a personal project once where I wanted to code a grid with java swing. I tried to create an object for each square in the grid and ended up with thousands of objects. It killed the process and it was very very slow rendering it. I’m not sure but this pattern may be the solution to that.
• Interpreter: Given a language, define a representation for its grammar along with an interpreter that uses the representation to interpret sentences in the language. I’m curious about this one.
• Iterator: Provide a way to access the elements of an aggregate object sequentially without exposing its underlying representation.
• Mediator: Define an object that encapsulates how a set of objects interact. Mediator promotes loose coupling by keeping objects from referring to each other explicitly, and it lets you vary their interaction independently.
• Memento: Without violating encapsulation, capture and externalize an object’s internal state so that the object can be restored to this state later.
• Observer: Define a one-to-many dependency between objects so that when one object changes state, all its dependents are notified and updated automatically. In django framework, the signals are an example of the observer pattern.
• Prototype: Specify the kinds of objects to create using a prototypical instance, and create new objects by copying this prototype.
• Proxy: Provide a surrogate or placeholder for another object to control access to it.
• Singleton: Ensure a class only has one instance, and provide a global point of access to it.
• State: Allow an object to alter its behavior when its internal state changes. The object will appear to change its class.
• Strategy: Define a family of algorithms, encapsulate each one, and make them interchangeable. Strategy lets the algorithm vary independently from clients that use it. I used this one recently where we had 6 different login scenarios. I divided the strategies to 6 classes with an execute(…) method. I had to create the appropriate strategy object from http post data and call the execute method to log the user in.
• Template Method: Define the skeleton of an algorithm in an operation, deferring some steps to subclasses. Template method lets subclasses redefine certain steps of an algorithm without changing the algorithm’s structure.
• Visitor: Represent an operation to be performed on the elements of an object structure. Visitor lets you define a new operation without changing the classes of the elements on which it operates.

After listing them, there is one last thing to do with this list of design patterns: classifying them. There are three main categories: Behavioral, Structural and Creational.

Behavioral patterns describe how the objects communicate with each other and tell us the responsibilities of the objects. These patterns are: Interpreter, Template Method, Chain of Responsibility, Command, Iterator, Mediator, Memento, Observer, State, Strategy, Visitor.

Structural patterns deal with the composition of classes or objects. These patterns are: Adapter, Bridge, Composite, Decorator, Facade, Flyweight, Proxy.

Creational patterns deal with the process of object creation. These patterns are: Factory Method, Abstract Factory, Builder, Prototype, Singleton.

Even now I feel much better about design patterns. My next quest will be one post for each design pattern. For that, I will probably read the book and support it with some online materials and create a post with the combination of both.

I came across this post on hackernews. Just like the title says, it lists many interesting papers from 1996 to 2018. I said in my first ever post that I am very into refactoring and clean coding these days so from this list one particular paper caught my eye: Understanding Misunderstandings in Source Code written by Dan Gopstein, Jake Iannacone, Yu Yan, Lois DeLong, Yanyan Zhuang, Martin K.-C. Yeh, Justin Cappos.

They talk about something they call “atoms of confusion” that are the smallest sections in source code that have the potential to confuse developers. They have a scientific way of finding these and so help developers and/or companies improve source code and cut development costs coming from these confusions. Here is the paper: https://atomsofconfusion.com/papers/understanding-misunderstandings-fse-2017.pdf