Code review gets a bad rap. People worry that code reviews slow teams down, create conflict between team members, or disempower team members by taking away decision making authority.
I believe these concerns are anti-patterns that indicate a team is using code review for criticism as opposed to critique. What's the difference? Critique is focused on specific parameters, where criticism is unbound.
That said, many programmers are unclear how to perform judgement free code reviews. The following are a series of suggestions which may help move code reviews from criticism to critique:
- Start with compassion
- Working code wins
- Discuss clarity, coupling, cohesion, and encapsulation
- Celebrate intentionally
- Follow up
Start With Compassion
I like to begin each review by reflecting on the context of the person submitting the code. Are they experienced team members or novices? Do you have good rapport with them? Are they under schedule or personal or professional pressure?
I follow that up by reminding myself of a variation on the retrospective prime directive; the submitter is doing the best they can given their past experiences and current context. This isn't to say the work is above reproach; but that the submitter is doing the best they can.
Working Code Wins
Code review may be used as a gate to prevent technically concerning code from entering the system. This is a flawed approach to code review as it introduces an unpredictable amount of latency in a workflow and implies distrust. I prefer to allow anyone to merge the code right away, so long as the merger has confidence that the change causes benefits from the user or business perspective. I ask myself the following questions:
- Do I understand the goals of the change?
- Does it meet my understanding of user or business needs?
- What are the changes potential unintended side effects?
- Does the change update the user facing documentation?
- Do tests demonstrate the features functionality? These don't need to go through the UI, especialls when there's reasonable confidence in the code connecting the interface to the feature.
- Is there an image, gif, or video demonstrating the feature in use attached to the change? This makes it much easier for team members to evaluate the change from a users perspective.
If the code can be said to be working; then it may be merged ASAP. Further code review for a piece of work may continue past the change being released into production.
Allowing anyone to carry forward or merge a change at any time encourages low cycle times, high flow, and tight collaboration between team members.
Discuss Clarity, Coupling, Cohesion, and Encapsulation
Code reviews often focus on style over substance. I prefer to use linters or automated style checkers in the build to pre-empt stylistic comments. This frees up team members to focus on how a change impacts maintainability of the sofware. The four aspects I focus on are clarity, coupling, cohesion, and encapsulation.
Clarity - What Does This Code Mean?
Clarity is all about names, documentation, tests, types, etc. Clear code helps future maintainers to understand the intent of the code as well as the implementation. Are commit messages and notes coherent? Are the names of variables and methods reasonable? Are there usage examples? Are types or shapes of data demonstrated or enforced?
Unclear code indicates a change lacks cohesion, has unnecessary coupling, or poorly encapsulates the goal of the change. By paying attention to the clarity, we make it easier to pick up the code and make changes later on.
Coupling - What Does This Code Work With?
Coupling is when changes to other components may require changes in this section of code. How many external collaborators are used to do the work? How deeply does this code reach into it's collaborators? How many of their functions are being used? In cases where coupling with external collaborators is both wide and deep it can be valuable to write an intermediary to wrap the collaborators and expose a smaller surface area to the new code.
By drawing attention to coupling, we may reduce the likelihood that we'll have to change this code when another component changes.
Encapsulation - How Will Others Use This Code?
While coupling is how a component interacts with outside collaborators; encapsulation is how outside consumers interact with a given component. One indicator of a loosely encapsulated component is difficulty in testing. A component that is hard to test is difficult to use. This is likely because the the code unintentionally exposes it's implementation. Sometimes the difficulty in testing is hidden behind test helpers; where a more effective approach may be to improve the interface of the component itself.
Long argument lists, callbacks, option hashes with required keys, and parameter objects are all indicators of loose encapsulation. In many cases these are completely reasonable; after all reducing the surface area of an interface likely diminishes it's utility. In other cases, it's a sign the component may be doing too much.
When we draw attention to codes encapsulation during review, it creates an opportunity to simplify interfaces and decrease the cost of re-using the component.
Cohesion - What Is The Codes Job?
The clearest example of non-cohesive code is a horror story from my youth. There was a SQL query which had hardcoded values it would return in order to change the color of a label. The argument could be made that this encapsulated the behavior of the feature; but in doing so it mixed the responsibility of retrieving data with the responsibility of presenting the data. Cohesive code eschews additional responsibilities wherever it can; resulting in horizontal layers and vertical slices.
By drawing attention to the cohesion of code we create opportunities to improve clarity by grouping related functionality and un-grouping un-related functionality. This un-related functionality may become either a collaborator or a consumer of the component it was extracted from.
When we frame our code review in terms of clarity, coupling, encapsulation, and cohesion we change the focus from the programmer and place it squarely on the artifact. It also moves us out of the world of moral judgement such as what's the "best" way to do something and into a conversation about the tradeoffs that we make as we build a software system together.
Despite our best intentions; no matter how well a code review goes it is still a high pressure situation for the submitter. No matter how I try, I still feel a bit sad when someone critiques my code. To counter the inevitable bummer of critical feedback I try to comment on parts of the code that are of high quality or teach me something. It can also help to use celebratory emoji.
The important part is to always be genuine in your celebration; as it is often impossible to discern between sarcasm or mockery and genuine excitement or positivity. Especially for people who struggle to value their own contributions already.
If there's valuable changes to be made from code review but the work was merged already; it's OK to add the desired changes in your task or project tracking system. I generally prefer that the team has sole discretion on when to complete those tasks; as they'll have a better understanding of the cost of delaying clean up as well as the impact that maintainability improvements will have on short and long term goals.
Overall, I believe a compassionate, critique-oriented code review practice results in high quality code and allows companies to include team members who struggle with pair or mob programming.