This is a mobile version, full one is here.
3 December 2019
Does Code Review Involve Testing?
When you review a pull/merge request from someone, do you check out the branch and run the build? I usually don’t, but some people do. Their obvious reason is: running a build, or even testing the product manually, helps find more important errors. Just looking at the source code may not reveal all visual defects recently introduced to the HTML/CSS, for example. It’s better to check out the branch, start Apache, open the site in Chrome, and see what’s broken. Then, make a screenshot, attach it to the pull request, and return it back to the author. But I disagree with this, and here is why.
The discussion is not new, check this and this on SO. However, it seems that all the answers there miss the key point.
As we all know, there are two conflicting roles in any software project: constructors and destructors, also known as programmers and testers. Programmers add new features and fix bugs. Their result is the amount of functionality created: the more, the better. While testers are breaking the product and reporting bugs, doing anything they can to prove that it’s not yet ready to be shipped to the customer. At some point of time the team (or the management) decides that the fight is over and the product can be delivered.
Thanks to this essential conflict quality is achieved.
Programmers finish their part of the fight when the code passes the merge pipeline: the changes are made at their laptops, unit tests pass locally, static analysis doesn’t complain, the build is clean, and the branch is merged to the trunk. This is where programmers stop and get their bonuses.
Testers finish their part of the fight when they manage to find a new defect in the product deployed to the staging or production environment: the bug is found, reported, and accepted by the project. This is where testers stop and get their bonuses.
This is obvious. If it’s not, you may want to read The Art of Software Testing by Glenford Myers or Code Ahead by Yegor Bugayenko. You may also want to watch this video.
Now, where is the place for a code reviewer in this conflict?
I believe code review is part of the merge pipeline, together with unit tests, static analyzers, linters, coverage control tools, mutation testers, and everything else the project may want to put in there, in order to make the life of programmers more difficult and the quality of the source code higher. The objective of a merge pipeline is to protect the code repository from programmers.
Code reviewers are no different than linters or static analyzers: They are blocking the branches with bad code from getting into the trunk. When they block, they give their reasons. Their job is done (similar to linters) when they manage to find a problem in the branch and explain it. A linter, when it rejects a pull request, says exactly where the problem is, which line, and how it can be fixed. This is what a good code reviewer must do: find a problematic line, explain the problem, and suggest a solution.
This is what code reviewers must be paid for: completed reviews.
What is a completed review? Does “Everything is OK” sound like a completed review? For a linter — yes; for a code reviewer — no. This one is much better: “I found three problems, explained them, and they were either debated or fixed.” This is how the job description of a code reviewer may sound: Find the three most critical problems, explain them, and make sure they are either fixed or correctly argued.
How reviewers find these three problems — that’s up to them. They may do visual inspections of the code or run the build. However, they will still be paid when they manage to find problems and make sure the author of the code understands them and either fixes them or explains why they are not fixable (or not problems). A visual inspection of the code is fast, while checking the branch out and running the build is a much more time-consuming operation. Moreover, a bug found after running the build locally is very difficult to explain in the code review format. The discussion with the author of the branch will take longer, which means more time to complete the code review, which means lower effectiveness of the code reviewer.
My point is that smart code reviewers don’t do that, because it’s not efficient. Pay attention, not because they don’t care, but because they know a better way to contribute to the project. Let’s face it, when the branch in front of us is passing all automated checks like linters and unit tests, and yet still has some bugs which we can reproduce only by executing the code, something is not right with our automated tests. An effective, responsible and greedy code reviewer will not explain to the author of the code what is wrong. Instead, a new bug will be created, to blame the merge pipeline for being too weak. And, of course, this new bug will be rewarded.
Thus, as a code reviewer, you can work with the branch locally, test it and report your findings to the author. But this will be against your personal interests and not beneficial for the project. Instead, you should complain to the project about the low quality of the automated tests and put the review on hold. When the complaint is resolved, the tests get stronger, you get back to the review, which will be rejected by the merge pipeline, not by you.
In this case, everybody wins: the pipeline gets stronger, you earn an extra bonus for the reported bug, and the review is rejected with a very specific reproducible reason.
P.S. The idea of this blog post was suggested by Robert Sösemann.
You review a pull request and suspect that it doesn't work, even though the CI status is green. What do you do?— Yegor Bugayenko (@yegor256) February 23, 2020