On Code Review

Published See discussion on Twitter

I've written a decent amount about code review practices, particularly about Draft Pull Requests and Pair Code Reviews, however I haven't really talked about how I approach reviewing pull requests.

While I don't think I have a particularly interesting process for reviewing pull requests, I do think that documenting it can provide additional insight for myself when I open another pull request. I also would like to see how others approach reviewing code and to see where we might differ in process!

Before diving in, I'll note that essentially all of the code reviews I do these days happen on GitHub, either on GitHub.com or our internal enterprise GitHub instance, so some of the things I refer to may only apply for GitHub and may not map cleanly over to other git hosting services or code review platforms.

Process:

To start, I usually give a read through of the Discussions tab of the PR, this is the default tab that includes the PR description, and a feed of activity on the PR (including other comments, commits, and CI status messages).

At this stage, I may have some understanding of the change before even reading through the PR description, either based on the title of the PR, or if a team member shared the PR with me they may have provided some additional context.

I'll read through the PR description that the author added, noting any interesting assumptions, questions, or callouts that the author added [1].

From there, I ask myself:

Do I have enough information to dive into the rest of the PR?

If the answer to this question is no, I pause reviewing the PR and instead reach out to the author (either on the PR itself or over slack) to have them share a bit more context about the changeset.

While I often review PRs that I don't have complete context over, I generally prefer to have at least a cursory understanding of the context around the changeset as it can help me connect some dots between this PR and the various other work that the team is working on at the moment.

Assuming I have enough context to continue on with the PR, I'll continue down the Discussions tab and review any other PR feedback that has already been added either by the author or another code reviewer.

Catching up on any other active or resolved discussions can be helpful to avoid providing duplicate feedback for the author, and helps to understand other decisions that the author and other code reviewers have made about the changeset [2].

Finally, after reviewing through the Discussions tab, I'm ready to dive into the code changes. Here I'm usually looking for a few things:

  • Are there any assumptions being made within the code changes that haven't been called out in the PR description or the original ticket?
  • Similar to assumptions above, are there possible edge cases that the author might have accounted for?
  • Will I feel confident contributing to this code in ~6 months as-is, essentially, how easily will it be for other contributors to dive in and make a change?

I don't use an explicit checklist for the above bullet points, and frequently I'll have other things in mind when reviewing a change that are specific to the code being changed.

I usually categorize most of my "nitpick"[3] feedback under the third bullet point above, usually trying to frame the feedback as optional and also up for discussion.

Delivery:

Far more important than the process that I follow for code reviews is the delivery of the feedback on the PR.

While providing direct feedback might work sometimes, it usually leads to making the code review process feel hostile and a battle between the author and the code reviewer.

For the first few years at my current job I remember some of the most critical feedback I received during my performance reviews from my peers and managers was around my code review etiquette.

I was super critical, straight to the point, and also frequently delivering my feedback in a way that made it seem like it should be done my way or not be done at all. It took me a while to get around some of this, and I honestly still have a huge opportunity to improve how I deliver feedback on PRs.

Since then, I think I've gotten a bit better at delivering my feedback in a more friendly way, one of the crucial aspects that helped me improve here was reconsidering what code review is for. In the past, I had usually approached code reviews with a very high bar for what we should consider merging into the codebase, frequently applying my own personal preferences and flavors on top of our existing styleguides.

More recently, I've tried to take a more broad view of the changes, devaluaing my own internal preferences in favor of thinking about the code as belonging to everyone (both current contributors and future contributors).

These days, I try my best to both call out opinions as clearly as possible, and also phrase most feedback in the form of a question. I've seen both of these help to clear up the review for the author, and also approaches from a standpoint of the changeset being an opportunity for learning and discussion rather than an opportunity to meet some absurdly high bar for standards.

These frequently look like:

Discussion - non-blocking:

What are your thoughts on changing the approach here to use reduce instead of map and filter?

These signals help the author to more clearly understand if the feedback should be fixed before merging or can be dropped, which usually speeds up the time between review iterations and merging as well.

Wrapping Up

I didn't cover other topics in this post such as running the changed code locally / in a preview environment, or Pair Reviewing so I'll leave those topics to further blog posts.

I'd love to hear how you approach code reviews, both from a process standpoint and from the delivery perspective! Feel free to reach out on Twitter to share your perspective, or write a blog post about it!


Footnotes:

[1] -

I don't usually literally note these in my go-to note taking tool, but usually only take a mental note of these.

[2] -

I don't usually review PRs that another team member is actively reviewing unless I'm interested in seeing the changes or either the author or other reviewers asked for more feedback/insight into the changesets. Part of the reason for avoiding doubling up on a PR is that I have full confidence in all the members of my team (and the other teams I frequently work with) to review any change being made to the codebase.

[3] -

A small, usually insignificant suggested change to make the code more readable