Dart packages have two varieties of versioning schemes. We use
semver with a variation that the numbers are shifted to the right
if the package is not stable enough to reach a 1.0.0
release. Note that the
Dart team culture was previously to keep packages at 0.x
for a long time,
whereas now we prefer to publish 1.0.0
as soon as feasible to avoid confusion
over these patterns.
- For packages that are not yet stable:
0.major.minor+patch
. - For most packages:
major.minor.patch
.
Never include a +1
on a version if the first number is not 0
. For breaking
changes, bump the major version. For new features, including all non-breaking
API changes like introducing a new class, bump the minor version. For bug
fixes, documentation changes, or any other change which doesn't impact calling
code, bump the patch version.
Any time the code in the repo does not match exactly the code of a version
published on pub
, both the pubspec.yaml
and CHANGELOG.md
should include a
-wip
version. We bump the version on the first change after a publish, even
if we don't plan on publishing the code in that state or if the change doesn't
impact consumers.
When opening a PR, check if the pubspec
and CHANGELOG
already have a -wip
version listed. If there is already a version, check what variety of version
bump it is from the previous published version and compare to the type for the
change you are making. If necessary you can "upgrade" the version change, for
instance if the current -wip
version is a patch number higher than what is
published, and you are adding a feature, rewrite the version into a minor
version change from what is published.
If the version is not currently a -wip
, perform the appropriate version bump
and add a -wip
. Add a section in the CHANGELOG
to match. Include the -wip
suffix in both places to avoid potential confusion about what versions are
published. If the change has no external impact (for instance a safe refactoring
or lint cleanup) it is OK to leave the CHANGELOG
section empty. If the change
has external impact add an entry to the section for that version.
Whenever you have the occasion to make a breaking version change, check if there are other minor breaking changes that should come with it.
- Search the issue tracker for potential breaking changes. Hopefully these are
tagged with
next-breaking-release
. - Search the code for
@deprecated
members and remove them.
In most packages a breaking version bump should be rare. Before making a breaking change you should weight many factors against the value provided by the change. Some of these factors include:
All of these packages will have to release a new version, and possibly make changes if they were actually broken by the changes.
Be very careful about changes like this - as they can have much larger cascading effects.
Example: Changing a sync API to be async. This could cause other packages to change their APIs to also be async, so they also have to do a breaking change.
Anything which could break internal projects is potentially much more difficult to roll out. The entire internal codebase typically has to be migrated at once at the same time as the roll.
This can cause cascading issues in rolling all the packages that depend on your package internally, so it is important to take this consideration seriously.
For this reason it is recommended that any potentially breaking change be ran through an internal presubmit before publishing externally.
This can also be mitigated by rolling out the breaking change in an incremental way, where both the new behavior can be opted into and the old is the default for some period of time. This allows internal code to be migrated one project at a time and then the old behavior can be removed once everything is migrated.
- Always sync the package into google3 before publishing so that it can be validated against a larger corpus of code. If the package rolls with SDK it should be updated in DEPS and published only after the next successful SDK roll.
- Open a PR which removes the trailing
-wip
from the version in both the pubspec and the changelog. It is OK to do this in the same PR as another change if that change should be published immediately. pub publish
. Don't ignore any warnings or errors reported by pub unless you are completely confident they are safe to ignore.- Add a git tag with the version that was published (e.g.,
git tag v1.2.3
). Check other tags in the repo to decide whether to include a leadingv
in the tag (git tag
orgit tag -l
). New repositories should prefer to include thev
. In a mono repo, start the tag with the name of the package. For examplebuild_runner-v1.7.0
. Be sure togit push --tags
after adding a tag. Note that you can tag a specific commit - instead of just tagging head - withgit tag v1.2.3 <commit-hash>
.
- Pull requests should usually have a 1:1 correspondence with final commits. Use "Squash and merge" after a review.
- Once a pull request is sent for review, commits on that branch should be
considered shared history. Never force push to a branch with an open pull
request. Prefer to merge into that branch to resolve conflicts, prefer not
to rebase.
- Comments are tied to commits, so force pushing also destroys comment history in GitHub prs.
- Pushing a new commit with code review updates makes it easy to review changes since your last review, by looking at the new commits only.
- Add comments from the "Files changed" view, even as the PR author, so that they can sent in a batch rather than replying to each one by one on the "Conversation" view.
- The gh cli tool makes it easy to checkout a PR in cases where a change may be easier to understand in an IDE.