Code reviews are a central part of developing high-quality code for Chromium. All changes must be reviewed.
The general patch, upload, and land process is covered in more detail in the contributing code page. To learn about upcoming code review and OWNERS policy changes, see Mandatory code review and OWNERS.
Ideally the reviewer is someone who is familiar with the area of code you are
touching. Any committer can review code, but an owner must provide a review
for each directory you are touching. If you have doubts, look at the git blame
for the file and the OWNERS
files (see below).
To indicate a positive review, the reviewer provides a "Code-Review +1" in Gerrit, also known as an LGTM ("Looks Good To Me"). A score of "-1" indicates the change should not be submitted as-is.
If you have multiple reviewers, provide a message indicating what you expect from each reviewer. Otherwise people might assume their input is not required or waste time with redundant reviews.
Please also read Respectful Changes and Respectful Code Reviews.
-
Aim to provide some kind of actionable response within 24 hours of receipt (not counting weekends and holidays). This doesn't mean you have to do a complete review, but you should be able to give some initial feedback, request more time, or suggest another reviewer.
-
Use the status field in Gerrit settings to indicate if you're away and when you'll be back.
-
Don't generally discourage people from sending you code reviews. This includes using a blanket "slow" in your status field.
In various directories there are files named OWNERS
that list the email
addresses of people qualified to review changes in that directory. You must
get a positive review from an owner of each directory your change touches.
Owners files are recursive, so each file also applies to its subdirectories.
It's generally best to pick more specific owners. People listed in higher-level
directories may have less experience with the code in question. For example,
the reviewers in the //chrome/browser/component_name/OWNERS
file will likely
be more familiar with code in //chrome/browser/component_name/sub_component
than reviewers in the higher-level //chrome/OWNERS
file.
More detail on the owners file format is provided in the "More information" section below.
Tip: The git cl owners
command can help find owners.
While owners must approve all patches, any committer can contribute to the review. In some directories the owners can be overloaded or there might be people not listed as owners who are more familiar with the low-level code in question. In these cases it's common to request a low-level review from an appropriate person, and then request a high-level owner review once that's complete. As always, be clear what you expect of each reviewer to avoid duplicated work.
Owners do not have to pick other owners for reviews. Since they should already be familiar with the code in question, a thorough review from any appropriate committer is sufficient.
The existing owners of a directory approve additions to the list. It is preferable to have many directories, each with a smaller number of specific owners rather than large directories with many owners. Owners should:
-
Demonstrate excellent judgment, teamwork and ability to uphold Chrome development principles.
-
Be already acting as an owner, providing high-quality reviews and design feedback.
-
Be a Chromium project member with full commit access of at least three months tenure.
-
Have submitted a substantial number of non-trivial changes to the affected directory.
-
Have committed or reviewed substantial work to the affected directory within the last ninety days.
-
Have the bandwidth to contribute to reviews in a timely manner. If the load is unsustainable, work to expand the number of owners. Don't try to discourage people from sending reviews, including writing "slow" or "emeritus" after your name.
The above are guidelines more than they are hard rules, and exceptions are
okay as long as there is a consensus by the existing owners for them.
For example, seldom-updated directories may have exceptions to the
"substantiality" and "recency" requirements. Directories in third_party
should list those most familiar with the library, regardless of how often
the code is updated.
Refer to the source code for all details on the file format.
This example indicates that two people are owners, in addition to any owners
from the parent directory. git cl owners
will list the comment after an
owner address, so this is a good place to include restrictions or special
instructions.
# You can include comments like this.
[email protected]
[email protected] # Only for the frobinator.
A *
indicates that all committers are owners:
*
The text set noparent
will stop owner propagation from parent directories.
This should be rarely used. If you want to use set noparent
except for IPC
related files, please first reach out to [email protected].
You have to use set noparent
together with a reference to a file that lists
the owners for the given use case. Approved use cases are listed in
//build/OWNERS.setnoparent
. Owners listed in those files are expected to
execute special governance functions such as eng review or ipc security review.
Every set of owners should implement their own means of auditing membership. The
minimum expectation is that membership in those files is reevaluated on
project, or affiliation changes.
In this example, only the eng reviewers are owners:
set noparent
file://ENG_REVIEW_OWNERS
The per-file
directive allows owners to be added that apply only to files
matching a pattern. In this example, owners from the parent directory
apply, plus one person for some classes of files, and all committers are
owners for the readme:
per-file [email protected]
per-file foo.*[email protected]
per-file readme.txt=*
Note that per-file
directives cannot directly specify subdirectories, e.g:
per-file foo/[email protected]
is not OK; instead, place a per-file
directive in foo/OWNERS
.
Other OWNERS
files can be included by reference by listing the path to the
file with file://...
. This example indicates that only the people listed in
//ipc/SECURITY_OWNERS
can review the messages files:
per-file *_messages*.h=set noparent
per-file *_messages*.h=file://ipc/SECURITY_OWNERS
"TBR" is our mechanism for post-commit review. It should be used rarely and only in cases where a normal review is unnecessary, as described under "When to TBR", below.
TBR does not mean "no review." A reviewer TBR-ed on a change should still review the change. If there are comments after landing, the author is obligated to address them in a followup patch.
Do not use TBR just because a change is urgent or the reviewer is being slow. Contact the reviewer directly or find somebody else to review your change.
To send a change TBR, annotate the description and send email like normal. Otherwise the reviewer won't know to review the patch.
-
Add the reviewer's email address in the code review tool's reviewer field like normal.
-
Add a line "Tbr: <reviewer's email>" to the bottom of the change list description. e.g.
Tbr: [email protected],[email protected]
-
Type a message so that the owners in the TBR list can understand who is responsible for reviewing what, as part of their post-commit review responsibility. e.g.
TBRing reviewers: reviewer1: Please review changes to foo/ reviewer2: Please review changes to bar/
The most common use of TBR is to revert patches that broke the build. Clean reverts of recent patches may be submitted TBR. However, TBR should not be used if the revert required non-trivial conflict resolution, or if the patch being reverted is older than a few days.
A developer relanding a patch can TBR the OWNERS for changes which are identical to the original (reverted) patch. If the reland patch contains any new changes (such as bug fixes) on top of the original, those changes should go through the normal review process.
When creating a reland patch, you should first upload an up-to-date patchset with the exact content of the original (reverted) patch, and then upload the patchset to be relanded. This is important for the reviewers to understand what the fix for relanding was.
You can use TBR with certain mechanical changes that affect many callers in
different directories. For example, adding a parameter to a common function in
//base
, with callers in //chrome/browser/foo
, //net/bar
, and many other
directories. If the updates to the callers is mechanical, you can:
-
Get a normal owner of the lower-level code you're changing (in this example, the function in
//base
) to do a proper review of those changes. -
Get somebody to review the downstream changes made to the callers as a result of the
//base
change. This is often the same person from the previous step but could be somebody else. -
TBR the owners of the calling code, after the API change is LGTM'ed.
This process ensures that all code is reviewed prior to checkin and that the concept of the change is reviewed by a qualified person, without having to ping many owners with little say in the trivial side-effects they incur.
Note: The above policy is only viable for strictly mechanical changes. For large-scale scripted changes you should:
-
Have an owner of the core change review the script.
-
Use
git cl split
to shard the large change into many small CLs with a clear description of what each reviewer is expected to verify (example).
You can TBR documentation updates. Documentation means markdown files, text documents, and high-level comments in code. At finer levels of detail, comments in source files become more like code and should be reviewed normally (not using TBR). Non-TBR-able stuff includes things like function contracts and most comments inside functions.
-
Use good judgement. If you're changing something very important, tricky, or something you may not be very familiar with, ask for the code review up-front.
-
Don't TBR changes to policy documents like the style guide or this document.
-
Don't mix unrelated documentation updates with code changes.
-
Be sure to actually send out the email for the code review. If you get one, please actually read the changes.