Small, simple CLs are:
- Reviewed more quickly. It's easier for a reviewer to find five minutes several times to review small CLs than to set aside a 30 minute block to review one large CL.
- Reviewed more thoroughly. With large changes, reviewers and authors tend to get frustrated by large volumes of detailed commentary shifting back and forth—sometimes to the point where important points get missed or dropped.
- Less likely to introduce bugs. Since you're making fewer changes, it's easier for you and your reviewer to reason effectively about the impact of the CL and see if a bug has been introduced.
- Less wasted work if they are rejected. If you write a huge CL and then your reviewer says that the overall direction is wrong, you've wasted a lot of work.
- Easier to merge. Working on a large CL takes a long time, so you will have lots of conflicts when you merge, and you will have to merge frequently.
- Easier to design well. It's a lot easier to polish the design and code health of a small change than it is to refine all the details of a large change.
- Less blocking on reviews. Sending self-contained portions of your overall change allows you to continue coding while you wait for your current CL in review.
- Simpler to roll back. A large CL will more likely touch files that get updated between the initial CL submission and a rollback CL, complicating the rollback (the intermediate CLs will probably need to be rolled back too).
Note that reviewers have discretion to reject your change outright for the sole reason of it being too large. Usually they will thank you for your contribution but request that you somehow make it into a series of smaller changes. It can be a lot of work to split up a change after you've already written it, or require lots of time arguing about why the reviewer should accept your large change. It's easier to just write small CLs in the first place.
In general, the right size for a CL is one self-contained change. This means that:
- The CL makes a minimal change that addresses just one thing. This is usually just one part of a feature, rather than a whole feature at once. In general it's better to err on the side of writing CLs that are too small vs. CLs that are too large. Work with your reviewer to find out what an acceptable size is.
- The CL should include related test code.
- Everything the reviewer needs to understand about the CL (except future development) is in the CL, the CL's description, the existing codebase, or a CL they've already reviewed.
- The system will continue to work well for its users and for the developers after the CL is checked in.
- The CL is not so small that its implications are difficult to understand. If you add a new API, you should include a usage of the API in the same CL so that reviewers can better understand how the API will be used. This also prevents checking in unused APIs.
There are no hard and fast rules about how large is "too large." 100 lines is usually a reasonable size for a CL, and 1000 lines is usually too large, but it's up to the judgment of your reviewer. The number of files that a change is spread across also affects its "size." A 200-line change in one file might be okay, but spread across 50 files it would usually be too large.
Keep in mind that although you have been intimately involved with your code from the moment you started to write it, the reviewer often has no context. What seems like an acceptably-sized CL to you might be overwhelming to your reviewer. When in doubt, write CLs that are smaller than you think you need to write. Reviewers rarely complain about getting CLs that are too small.
There are a few situations in which large changes aren't as bad:
- You can usually count deletion of an entire file as being just one line of change, because it doesn't take the reviewer very long to review.
- Sometimes a large CL has been generated by an automatic refactoring tool that you trust completely, and the reviewer's job is just to verify and say that they really do want the change. These CLs can be larger, although some of the caveats from above (such as merging and testing) still apply.
If you write a small CL and then you wait for your reviewer to approve it before you write your next CL, then you're going to waste a lot of time. So you want to find some way to work that won't block you while you're waiting for review. This could involve having multiple projects to work on simultaneously, finding reviewers who agree to be immediately available, doing in-person reviews, pair programming, or splitting your CLs in a way that allows you to continue working immediately.
When starting work that will have multiple CLs with potential dependencies among each other, it's often useful to think about how to split and organize those CLs at a high level before diving into coding.
Besides making things easier for you as an author to manage and organize your CLs, it also makes things easier for your code reviewers, which in turn makes your code reviews more efficient.
Here are some strategies for splitting work into different CLs.
One way to split up a CL without blocking yourself is to write one small CL, send it off for review, and then immediately start writing another CL based on the first CL. Most version control systems allow you to do this somehow.
Another way to split up a CL is by groupings of files that will require different reviewers but are otherwise self-contained changes.
For example: you send off one CL for modifications to a protocol buffer and another CL for changes to the code that uses that proto. You have to submit the proto CL before the code CL, but they can both be reviewed simultaneously. If you do this, you might want to inform both sets of reviewers about the other CL that you wrote, so that they have context for your changes.
Another example: you send one CL for a code change and another for the configuration or experiment that uses that code; this is easier to roll back too, if necessary, as configuration/experiment files are sometimes pushed to production faster than code changes.
Consider creating shared code or stubs that help isolate changes between layers of the tech stack. This not only helps expedite development but also encourages abstraction between layers.
For example: You created a calculator app with client, API, service, and data model layers. A shared proto signature can abstract the service and data model layers from each other. Similarly, an API stub can split the implementation of client code from service code and enable them to move forward independently. Similar ideas can also be applied to more granular function or class level abstractions.
Orthogonal to the layered, horizontal approach, you can instead break down your code into smaller, full-stack, vertical features. Each of these features can be independent parallel implementation tracks. This enables some tracks to move forward while other tracks are awaiting review or feedback.
Back to our calculator example from Splitting Horizontally. You now want to support new operators, like multiplication and division. You could split this up by implementing multiplication and division as separate verticals or sub-features, even though they may have some overlap such as shared button styling or shared validation logic.
To take this a step further, you could combine these approaches and chart out an implementation plan like this, where each cell is its own standalone CL. Starting from the model (at the bottom) and working up to the client:
| Layer | Feature: Multiplication | Feature: Division |
|---|---|---|
| Client | Add button | Add button |
| API | Add endpoint | Add endpoint |
| Service | Implement transformations | Share transformation logic with |
| : : : multiplication : | ||
| Model | Add proto definition | Add proto definition |
It's usually best to do refactorings in a separate CL from feature changes or bug fixes. For example, moving and renaming a class should be in a different CL from fixing a bug in that class. It is much easier for reviewers to understand the changes introduced by each CL when they are separate.
Small cleanups such as fixing a local variable name can be included inside of a feature change or bug fix CL, though. It's up to the judgment of developers and reviewers to decide when a refactoring is so large that it will make the review more difficult if included in your current CL.
CLs should include related test code. Remember that smallness here refers the conceptual idea that the CL should be focused and is not a simplistic function on line count.
Tests are expected for all Google changes.
A CL that adds or changes logic should be accompanied by new or updated tests for the new behavior. Pure refactoring CLs (that aren't intended to change behavior) should also be covered by tests; ideally, these tests already exist, but if they don't, you should add them.
Independent test modifications can go into separate CLs first, similar to the refactorings guidelines. That includes:
- Validating pre-existing, submitted code with new tests.
- Ensures that important logic is covered by tests.
- Increases confidence in subsequent refactorings on affected code. For example, if you want to refactor code that isn't already covered by tests, submitting test CLs before submitting refactoring CLs can validate that the tested behavior is unchanged before and after the refactoring.
- Refactoring the test code (e.g. introduce helper functions).
- Introducing larger test framework code (e.g. an integration test).
If you have several CLs that depend on each other, you need to find a way to make sure the whole system keeps working after each CL is submitted. Otherwise you might break the build for all your fellow developers for a few minutes between your CL submissions (or even longer if something goes wrong unexpectedly with your later CL submissions).
Sometimes you will encounter situations where it seems like your CL has to be large. This is very rarely true. Authors who practice writing small CLs can almost always find a way to decompose functionality into a series of small changes.
Before writing a large CL, consider whether preceding it with a refactoring-only CL could pave the way for a cleaner implementation. Talk to your teammates and see if anybody has thoughts on how to implement the functionality in small CLs instead.
If all of these options fail (which should be extremely rare) then get consent from your reviewers in advance to review a large CL, so they are warned about what is coming. In this situation, expect to be going through the review process for a long time, be vigilant about not introducing bugs, and be extra diligent about writing tests.