Welcome to Starcraft! We hope this document helps you get started. Before contributing any code, please sign the Canonical contributor licence agreement.
We use a forking, feature-based workflow, so you should start by forking the
repository. Once you've done that, clone the project to your computer using git
clone's --recurse-submodules
parameter. (See more on the git submodules
documentation.)
To set up a development environment, run:
make setup
To test that the environment works, run:
make lint make test
Several targets exist in the Makefile to help with code conventions. They are all
merged into the format
target for automatic formatting and the lint
target
for running all linters. These conventions apply to all files in the repository,
including documentation and metadata files. pre-commit
is used for running many
hooks before committing. Run:
make setup-precommit
to configure it for this repository.
Tests can be run using make test
, which will run all test forms. Specific types
of tests can be run with other testing targets shown in make help
.
Starcraft projects follow the OneFlow git branching model.
The branch name should be brief and less than 200 characters.
Branch names are formatted as <category>/<name>
.
- This naming convention provides a few benefits:
- GitHub workflows can choose which categories they should run on
- GitHub branches rules can be configured per category
- Some IDEs and git tools display categorized branches in a neat and nested format
The main branch containing the latest changes.
Branches created automatically by the Renovate bot.
For any work driven by a ticketing system, the ticket should be part of the branch name. The ticketing system can be built into the repo like GitHub issues or an external ticketing system.
For GitHub issue #100 (a ticket to add a README file), the branch would be
called work/100-add-readme
.
For an external ticketing system like Jira, a ticket labeled
CRAFT-100
would use a branch called work/CRAFT-100-add-readme
.
For any new features, bug fixes, and general work not driven by a ticketing system.
For hotfixes to an existing minor release.
For example, hotfixes for Testcraft 2.1 would go to a branch called
hotfix/2.1
.
As a departure from OneFlow, hotfix branches for *craft applications can be long-lived. This is because *craft applications are built via Launchpad, which uses build recipes that follow specific branches.
After a tagged release of a hotfix branch, the branch should be merged back
to main
.
For commits that merge another branch into the current branch. See the chore section for information on merge commit headers.
For commits that prepare for a release. These commits should update the changelog.
Commit messages are based on the conventional commit style:
<type>[(optional scope)][!]: <description> [optional body] [optional footer]
The commit is divided into three sections: a header, body, and footer.
The header is required and consists of three subsections: a type, optional scope, and description. The header must be 72 characters or less.
Commits that affect the CI/CD pipeline.
Commits that affect the build of an application or library.
This includes dependency updates, which should use the deps
scope
(build(deps):
).
Commits that add a new feature for the user.
Commits that fix a bug or regression.
Commits that improve performance without changing the API or external behavior.
Commits that refactor code.
Using Martin Fowler's definition, refactor means "a change made to the internal structure of software to make it easier to understand and cheaper to modify without changing its observable behavior."
Commits that change the syntax, format, or aesthics of any text the codebase. The meaning of the text should not change.
Examples include:
* automatic changes from tools like black
and ruff format
* changes to documentation that don't affect the meaning
* correcting a typo
Commits that improve, add, or remove tests.
Commits that affect the contents of the documentation.
Changes to how documentation is built should use build(docs)::
.
Changes to how the documentation is built in the CI/CD pipeline should use
the ci(docs):
.
Miscellaneous commits that don't fit into any other type.
Examples include:
- edits to a comment or docstring
- type changes
- accommodating a developer-facing deprecation warning
- many small fixes for an existing PR
- merge commits (
chore(merge): <branch or tag> into <branch>
)- the remote name should not be included (for example, use
main
instead oforigin/main
)
- the remote name should not be included (for example, use
Sometimes, multiple types may be appropriate for a PR.
This may signal that a commit is doing more than one thing and should be broken into multiple smaller commits. For example, a commit should not refactor code and fix a bug. This should be two separate commits.
In other scenarios, multiple types could be appropriate because of the nature
of the commit. This can happen with test
and docs
, which can be used
as types or scopes.
The types above are ordered by descending priority. The first appropriate type should be used.
For example, refactoring a test suite could have the header
test(project): reorganize tests
or
refactor(test): reorganize project tests
. refactor
has a higher
priority than test
, so the latter option is correct.
A scope is an optional part of the commit header. It adds additional context by specifying what part of the codebase will be affected.
It should be a tangible part of the codebase, like a directory, module, or class name.
If a commit affects many areas of the codebase, the scope should be omitted;
many
is not an accepted scope.
If an exclamation point (!
) is inserted after the type/scope, this means
that the commit introduces a breaking change. Including one or more commits
with an exclamation point in a release will trigger a major version increment.
Breaking changes may also be indicated by including the words BREAKING CHANGE
in the commit footer.
The description is written in the imperative mood (present tense, second person). The description should complete the following sentence:
If applied, this commit will <description>.
The description does not begin with capital letter (unless it's a proper noun) and does not end with puncuation mark.
Examples of commit headings:
feat: inherit context from services test: increase unit test stability fix: check foo before running bar feat(daemon): foo the bar correctly in the baz test(daemon): ensure the foo bars correctly in the baz fix(test): mock class Foo ci(snap): upload the snap artefacts to Github chore(deps): update go.mod dependencies
The body is an optional section of the commit to provide more context. It should be succinct (no more than 3-4 sentences) and may reference relevant bugs and issues.
The footer is an optional section of the commit message that can mention the signer and co-authors of the commit.
Example footers:
Signed-off-by: <name> <<email>> Co-authored-by: <name> <<email>>
The changelog is a reference documentation page that gives a human-readable summary of changes to the project that are relevant to users.
Each change should be clear in its purpose, whether it is fixing a bug, adding a feature, or changing an existing behavior.
Internal changes should not be included in the changelog. For example, dev dependency updates, CI updates, and style changes should not be included.
Changes should be written in the imperative mood (present tense, second person) similar to commit headers.
The changelog should link to the project's GitHub releases page, which contains an exhaustive list of all commits added to the release.
Release entries should be sorted by date from newest to oldest.
If an older version gets a hotfix release, subsequent releases should mention when the hotfix is incorporated.
For example, consider a package with a previous release 2.9.0
and a latest
release of 3.0.0
. If the 2.9.0
receives a hotfix release 2.9.1
and
is merged back to main
, then the next 3.x
release should indicate that
it includes the changes from 2.9.1.
Note
3.0.1 includes changes from the 2.9.1 release.