Code review is an essential software development practice that we use at Liefery. It’s so central to how we work that I cannot even imagine my day at work without it. In this blog post I’d like to describe why we do it, how we designed our process, and what to look for when you are the reviewer.
Why do we make sure that every line of code is reviewed by at least one other team member? Besides the standard reason–i.e. code quality–the most important reason is that it helps us build a shared ownership of the application. It’s not an “Adam’s feature” that “has bugs” and “Adam has to fix it now”. When I reviewed and accepted a pull request I’m now equally responsible for the code. This reduces the fear of making changes by replacing “programmer mistakes” with “team mistakes”. And I make mistakes every single day. This approach is a shift from “blame” culture to “we have to fix it” culture.
Our application gets bigger and bigger, and it’s no longer possible that everybody works on every feature. Thanks to code reviews I am still aware of what other people are working on. You can think of it as a way of sharing the domain knowledge across the team.
We also share the technical knowledge during code reviews. The reviewer can suggest an alternative way to solve the problem, a useful method from Ruby standard library that simplifies the code, or ask questions about new technologies that they encounter for the first time.
Process - DIY
It’s important to design a code review process that suits your company, not just blindly copy what other companies are doing. They probably have a different team size, different structure, different constraints and work on different problems. I can give you some perspective by comparing the code review process at Liefery with the one at GitLab where I worked before.
My experience at GitLab
GitLab is a large open source project that ships a new minor version on a
fixed date every month. This means that anything that lands in
to be “fully correct”, because your follow-up pull request may or may not
land in time for the next release. A code review in most cases involves
getting the approval of two team members - a Reviewer and a Maintainer. The
Reviewer is a role at the company which is handed after acquiring some
experience with the codebase to people that are particularly interested in
performing code reviews. The Maintainer role means push access to
and acts as a “gatekeeper of the application” - code changes cannot get
through without the approval of one of the Maintainers.
It’s the responsibility of the person submitting the pull request to find and assign the appropriate Reviewer. The criteria is the experience with the particular area of the code. The application is huge and Reviewers may have a hard time looking at the code they never saw before.
All the feedback coming from the Reviewer and the Maintainer has to be fully addressed - either by changing the code or by convincing the person that the code is correct.
This creates a hierarchical structure where the room for disagreement with people reviewing your pull request is quite limited. It works well for a company where roughly 70 people commit to one codebase (and this number excludes numerous contributors from the open source community).
How we do it at Liefery
We have a very different situation at Liefery. Our team consists of 10 programmers right now. We value the speed of delivering new features, incremental improvements and want to empower the whole team to review code.
When you submit a pull request you don’t assign anybody to review it, instead you just leave it open. Everybody is encouraged to grab and review it, although in many cases it’s one of the people that you work together on the current feature group, because they have the necessary context already in their head. If the pull request introduces a non-trivial change we ask other people to take a look at it. We don’t want the knowledge about certain areas of the application to be clustered.
Addressing feedback on the pull request is not so much “making Adam happy”, but rather “making sure that you’re not the only person that holds this opinion”. You don’t have to adjust your pull request to match exactly the programming habits of the reviewer, it’s enough when another team member agrees with your opinion. This requires a lot of trust in the team as a whole, but saves us from having long discussions on things that don’t really matter. For example, making a method shorter by extracting a private method is often a matter of style and should not block a pull request if other people agree that it’s optional.
As you can see our review process doesn’t depend that much on individuals and we don’t get blocked on waiting for a response from a particular person. This allows us to merge code quickly, without going back and forth for multiple days.
When there are things that are nice to have, but not strictly necessary
for the code to land in
master, we tend to merge the pull request
without them and then create follow-up tickets that will be addressed in
the future. Thanks to this we keep our pull requests smaller, and smaller
pull requests are much easier to review. This again requires a lot of
trust that your colleagues will remember to go back to the follow-up
The Reviewer’s Manual
So you decided to review a pull request, you have the code diff right in front of you, what do you do now? I have an informal list of steps in my mind that I follow every time I perform a code review. I’ll write it down for you now.
- Read the associated issue / ticket first. You need to have a good understanding of what had to be done before you see how it was done. This forces you to think about how you would solve the problem without being biased by the solution in the pull request.
- Go through all the changes and try to understand the proposed solution on the high level. Do you understand how it solves the problem? Do you think that it’s better, worse or equivalent to what you would do? Does it leave the codebase in a better shape than it was before?
- Don’t leave any comments yet during the first pass. Chances are that your questions will be answered by the code that follows. Instead create draft comments for these remarks.
- Go back to the first file and start over. This time read every single line and focus on individual methods and classes.
- Do not leave comments related to the code formatting. The code formatting should be enforced by automatic tools (like Rubocop) and having a discussion about it in every pull request is a waste of time.
- Think about the readability of the code. Is this method / class easy to read? Is it immediately clear what it does? Will it benefit from being split into multiple methods / classes? Does it have a good name?
- Think about edge cases. Is it possible that the argument is
nil? Will the code still work when there are no users with the provided email? What if the sum of the elements is zero?
- Think about your business domain. Does the language used in the code match the terms that your business specialists use? Does the code feel like a natural solution in the context of your business?
- Think about the duplication. Do you already have code that tries to do the same thing? Is it the right moment to introduce an abstraction that reduces the duplication?
- Think about the security. Do we filter the allowed parameters properly? Do we scope the available records to the ones associated with the current user? Do we sanitize the input?
- Think about the performance. Does this action happen in the request life-cycle or in the background job? Does it take a long time? Can it be moved to the background? Does the code produce an efficient SQL query?
- Look at the tests. Are there tests for the newly introduced classes and public methods? Are changes of behaviour documented by corresponding test cases? Does this pull request require higher level integration tests?
- Pick at least one thing that you particularly like in the pull request (for example perfect method name, simple solution or improved test) and point it out. Positive feedback is also important!
- Ask questions instead of pointing out mistakes. It happens to me often that when I confidently say that something is wrong, I missed some important piece of information and it’s in fact me who is wrong. That’s why it is much better to ask questions instead: “why did you decide to implement it like that?” or “don’t we have to consider that case when customer is missing here?”.
- Be nice. Always assume the best intentions of the person submitting the pull request. When describing things to improve, talk about the code (“This code is a bit unclear. What do you think about renaming the method?”) not about the person (“You failed to communicate your intent here. Can you learn how to use better naming conventions?”).
- Ask questions when you have no experience with certain technologies or techniques. When I review a pull request that introduces static typing with Flow, and I’ve never used Flow before, I have two options: do my own research or ask my colleague to point me to the resources that they know. Now the chances are that whatever they suggest will be a much better source of knowledge that the thing I can find after 5 minutes of googling. That’s because they already have this knowledge so they can judge the quality of resources more easily. Asking for help doesn’t make you a “less credible reviewer”.
- When you go back and forth in a text discussion without reaching an agreement, switch to one-on-one, synchronous communication. It’s much more efficient when resolving conflicts. Write down the outcome of the pairing review as a comment in the pull request for other people reviewing it.
This is how we do code reviews. And how about your company? What is your process? What are the things that you pay attention to? I encourage you to think about it :)