Due to recent changes in core teams, some have reduced in size to the level where the 2 '+2' vote requirement to approve patches is not sustainable anymore. This patch adds notes to the Code Submission Guide to ease that requirement, when necessary. Change-Id: I4982c74af658f28a9d28203d82d4e53df8cbcad4 Signed-off-by: Ildiko Vancsa <ildiko.vancsa@gmail.com>
10 KiB
Code Submission Guidelines
StarlingX follows the OpenStack developer contribution guidelines. This guide contains StarlingX-specific tasks and guidelines.
Pre-review and pre-submission testing
For the majority of cases, it is expected that the author completes their testing before posting a review.
Update existing automated unit tests and add new ones when applicable.
Make sure the new code compiles and builds successfully.
For each package being modified, update the
TIS_PATCH_VER
variable in the centos/build_srpm.data. IfTIS_PATCH_VER=PKG_GITREVCOUNT
is already present, then changes to increment is for testing only; and is not required in commit as the version is automatically incremented on commit. This ensures that packages are versioned correctly and the latest version will be used. If up-versioning a package, then resetTIS_PATCH_VER
to 0.Check build dependencies of a new or modified package using
build-pkgs --dep-test <pkg>
(see details below).Run tox tests (flake8, py27, etc) successfully. These can all be run manually prior to launching a review.
Verify basic functional testing on a live system to ensure the new code gets executed and functions correctly.
If the code changes are related to config/stx-puppet/ansible, impact on the following should be assessed. If appropriate, additional tests should be executed.
- Bootstrap
- Backup and restore
- Upgrade
If needed, consult with the core reviewers or send questions to the StarlingX discuss mailing list (starlingx-discuss@lists.starlingx.io) regarding required/recommended testing.
The commit message should include a Test Plan section that lists the test case titles of manual tests executed on the update. In addition to the success path test cases, additional test plan subsections are recommended. For example.:
Test Plan: <success path test cases for new or changed behaviors> PASS: Verify new functional behavior 1 PASS: Verify affected functional behavior 2 Failure Path: <failure path, i.e. negative tests> PASS: Verify feature error path 1 is handled properly PASS: Verify feature error path 2 is handled properly PASS: Verify feature error path n is handled properly Regression: <test cases that should still work but should be confirmed> PASS: Verify system install PASS: Verify feature alarm handling PASS: Verify no core dumps and no memory leaks PASS: Verify feature logging Notes on the above "Execution Status Key": FAIL: test failed; author is investigating while review is in progress PASS: test passed : no status means the test is not yet run.
After completing testing, when the code is ready for submission, it's not necessary to add expected results. Just the titles are sufficient. The formatting of the Test Plan section must conform to standard commit message rules. The goal is to inform reviewers about the testing done as part of this new code.
Code reviews
- Use Gerrit for StarlingX code reviews
- Follow the OpenStack Git Commit Good Practice for Git Commit Messages
- Add the core reviewers for the affected sub-project to the review,
as well as any other interested reviewers
- The core reviewers are listed in each sub-project repository. Refer to the list of sub-project repos
- In order for code to get merged, two core reviewers must give the
review +2. A final Workflow +1 from one core reviewer will allow the
code to merge. Typically, the final W+1 is done by the second core
reviewer.
- While the requirement to have at least 2 ‘+2’ votes from core reviewers is set to advocate for higher code quality through a more diverse set of eyes, in certain cases it can also become a bottleneck. When the core team of a sub-project becomes too small to be able to sustain the 2 ‘+2’ rule, they can reduce the requirement to 1, however, a core reviewer should not approve their own changes, when that is possible.
- In the above case it is further advised to seek for a higher number of ‘+1’ votes. This can ensure the changes are still thoroughly reviewed, while can also help with increasing the number of core reviewers with a more engaged team.
- If a core reviewer sets a -2, the code cannot be merged until that reviewer removes their -2.
- Authors should not review their own code and, therefore, should not
+2 or W+ their code submissions
- If an exception is needed (ex: emergency fix for a broken build), the author should send an email to the mailing list to let the community know or contact the core reviewers on IRC.
Code review process
- Any contributor can review commits by any other contributor, optionally adding comments and/or setting approvals of -1/0/+1. For regular contributors, these are considered opinion only, though the core reviewers will generally take them into account.
- Reviewers should set "-1" to indicate there are things which must be fixed before the review should be approved. If a reviewer sets -1, they must leave one or more comments indicating specific issues.
- If line-specific issues are raised by a reviewer, it is expected that the committer leaves line-specific comments indicating that it has been addressed (this could be a simple as replying with "Done"), or if not, why not. General comments applying to the whole review should be replied to in the general comments.
- Reviewers may prefix comments with "nit:". This indicates a fix that would be nice but would not block merging the commit. If a new revision is needed to fix any non-nit issues, it is expected that any nits should be fixed at that time.
- For more details, refer the OpenStack code review process, documented in the OpenStack Developer Guide
Link reviews to story or bug
- For traceability, always link your code change to a story or bug. The story/bug will give reviewers context for the code changes. This will also be used to help determine the relative priority of the code changes.
- Gerrit will update the status of the story/bug automatically once the code is merged.
- Linking to StoryBoard Stories: Specify the story and task ID in the
commit message as follows:
- Story: $story_id
- Task: $task_id
- Example: https://review.openstack.org/#/c/590083/
- Linking to Launchpad Bugs: Specify the Bug ID in the commit message
as follows:
- Closes-Bug: $bug_id -- use 'Closes-Bug' if the commit is intended to fully fix and close the bug being referenced.
- Partial-Bug: $bug_id -- use 'Partial-Bug' if the commit is only a partial fix and more work is needed.
- Related-Bug: $bug_id -- use 'Related-Bug' if the commit is merely related to the referenced bug.
- If a fix requires multiple commits, use "Partial-Bug" with only the final commit using "Closes-Bug"
- Example: https://review.openstack.org/596305
Check build dependencies
When you upversion a package or make significant changes to its build scripts (spec files, make files, auto-config, etc.), you must test the build dependencies of the modified package.
First, complete a full build using build-pkgs
.
Next, use build-pkgs --dep-test <pkg>
to test the
build dependencies.
You may think that if your package passes a full build
(build-pkgs
), that all dependencies have been checked,
however, this is not the case. When doing a full build,
the build environment is not wiped clean between packages. This means
that the environment might (or might not) have a tool or library
required by your package, which is not listed as a
BuildRequires
in its spec file. Your package may build
successfully one time, but might not build the next time, depending on
which packages were scheduled to build in the same environment before
your package.
The --dep-test
option rebuilds one specific package in a
clean environment and provides an effective test of the BuildRequires
for that package.
Early review and feedback
- In specific cases, changes can be posted for early review prior to testing (ex: need early feedback on detailed design/coding approach)
- Such changes should be marked as WIP in the commit message and given a Workflow -1 immediately by the author
- The author should also include a comment in the review explaining the purpose of the review and why the testing is deferred.
- Reviewing code early and often helps catch design and coding errors sooner and shows us following the Four Opens.
Cherry-picking
- All code changes must be pushed to master first and then cherry-picked to the appropriate release branch as needed
- When cherry-picking updates using “git cherry-pick”, include the '-x' option. This automatically adds the “(cherry picked from commit XXXXX)” line to your commit message, which is helpful to code reviewers.
- Exception: Feature branches used during development
Patch rebase
- During patch re-base, there is a chance that patches can be applied by treating the patch line numbers as approximate, rather than a strict requirement, just so long as the before/after context seems to be correct. They require fuzzing during the patch apply, and an .orig file will be created as the consequence of applying patches that are not clean.
- In StarlingX, we will not accept fuzzing patches. All patches are required to be re-based cleanly so that no fuzzing and no .orig files are generated.