Reviewing Zulip code¶
Code review is a key part of how Zulip does development! If you’ve been contributing to Zulip’s code, we’d love for you to do reviews. This is a guide to how. (With some thoughts for writing code too.)
Principles of code review¶
Anyone can review¶
Anyone can do a code review – you don’t have to have a ton of experience, and you don’t have to have the power to ultimately merge the PR. If you
- read the code, see if you understand what the change is doing and why, and ask questions if you don’t; or
- fetch the code (for Zulip server code, tools/fetch-rebase-pull-request is super handy), play around with it in your dev environment, and say what you think about how the feature works
those are really helpful contributions.
Please do reviews¶
Doing code reviews is an important part of making the project go. It’s also an important skill to develop for participating in open-source projects and working in the industry in general. If you’re contributing to Zulip and have been working in our code for a little while, we would love for some of your time contributing to come in the form of doing code reviews!
For students participating in Google Summer of Code or a similar program, we expect you to spend a chunk of your time each week (after the first couple of weeks as you’re getting going) doing code reviews.
Fast replies are key¶
For the author of a PR, getting feedback quickly is really important for making progress quickly and staying productive. That means that if you get @-mentioned on a PR with a request for you to review it, it helps the author a lot if you reply promptly.
A reply doesn’t even have to be a full review; if a PR is big or if you’re pressed for time, then just getting some kind of reply in quickly – initial thoughts, feedback on the general direction, or just saying you’re busy and when you’ll have time to look harder – is still really valuable for the author and for anyone else who might review the PR.
People in the Zulip project live and work in many timezones, and code reviewers also need focused chunks of time to write code and do other things, so an immediate reply isn’t always possible. But a good benchmark is to try to always reply within one workday, at least with a short initial reply, if you’re working regularly on Zulip. And sooner is better.
Things to look for¶
- The Travis CI build. The tests need to pass. One can investigate
any failures and figure out what to fix by clicking on a red X next
to the commit hash or the Detail links on a pull request. (Example:
in #1219, click the red
X next to
f1f474eto see the build jobs for that commit, at least one of which has failed. Click on the link for Travis continuous integrations details to see the tests Travis ran on that commit, at least one of which failed, and go to one of the failing tests to see the error.)
- Technical design. There are a lot of considerations here: security, migration paths/backwards compatibility, cost of new dependencies, interactions with features, speed of performance, API changes. Security is especially important and worth thinking about carefully with any changes to security-sensitive code like views.
- User interface and visual design. If frontend changes are involved, the reviewer will check out the code, play with the new UI, and verify it for both quality and consistency with the rest of the Zulip UI. We highly encourage posting screenshots to save reviewers time in getting a feel for what the feature looks like – you’ll get a quicker response that way.
- Error handling. The code should always check for invalid user input. User-facing error messages should be clear and when possible be actionable (it should be obvious to the user what they need to do in order to correct the problem).
- Testing. The tests should validate that the feature works
correctly, and specifically test for common error conditions, bad
user input, and potential bugs that are likely for the type of
change being made. Tests that exclude whole classes of potential
bugs are preferred when possible (e.g., the common test suite
test_bugdown.pybetween the Zulip server’s frontend and backend Markdown processors, or the
GetEventsTesttest for buggy race condition handling).
- Translation. Make sure that the strings are marked for translation.
- Clear function, argument, variable, and test names. Every new piece of Zulip code will be read many times by other developers, and future developers will grep for relevant terms when researching a problem, so it’s important that variable names communicate clearly the purpose of each piece of the codebase.
- Duplicated code. Code duplication is a huge source of bugs in large projects and makes the codebase difficult to understand, so we avoid significant code duplication wherever possible. Sometimes avoiding code duplication involves some refactoring of existing code; if so, that should usually be done as its own series of commits (not squashed into other changes or left as a thing to do later). That series of commits can be in the same pull request as the feature that they support, and we recommend ordering the history of commits so that the refactoring comes before the feature. That way, it’s easy to merge the refactoring (and minimize risk of merge conflicts) if there are still user experience issues under discussion for the feature itself.
- Completeness. For refactorings, verify that the changes are
complete. Usually one can check that efficiently using
git grep, and it’s worth it, as we very frequently find issues by doing so.
- Documentation updates. If this changes how something works, does it update the documentation in a corresponding way? If it’s a new feature, is it documented, and documented in the right place?
- Good comments. It’s often worth thinking about whether explanation
in a commit message or pull request discussion should be included in
/docs, or other documentation. But it’s better yet if verbose explanation isn’t needed. We prefer writing code that is readable without explanation over a heavily commented codebase using lots of clever tricks.
- Clear commit messages. See the Zulip version control documentation for details on what we look for.
Some points specific to the Zulip server codebase:
- Testing – Backend. We are trying to maintain ~100% test coverage on the backend, so backend changes should have negative tests for the various error conditions.
- Testing – Frontend. If the feature involves frontend changes, there should be frontend tests. See the test writing documentation for more details.
- mypy annotations. New functions should be annotated using mypy
and existing annotations should be updated. Use of
ignore, and unparameterized containser should be limited to cases where a more precise type cannot be specified.
To make it easier to review pull requests, if you’re working in the
Zulip server codebase, use our git tool
tools/fetch-rebase-pull-request to check out a pull request locally
and rebase it against master.
If a pull request just needs a little fixing to make it mergeable, feel free to do that in a new commit, then push your branch to GitHub and mention the branch in a comment on the pull request. That’ll save the maintainer time and get the PR merged quicker.
We also strongly recommend reviewers to go through the following resources.
- The Gentle Art of Patch Review article by Sarah Sharp
- Zulip & Good Code Review article by Sumana Harihareswara
- Code Review - A consolidation of advice and stuff from the sinternet article by James J. Porter
- Zulip Code of Conduct