Contributing to Frequenz Repository Configuration¤
Build¤
You can use build to simply build the source and binary distribution:
Local development¤
You can use editable installs to develop the project locally (it will install all the dependencies too):
Or you can install all development dependencies (mypy, pylint, pytest,
etc.) in one go too:
If you don't want to install all the dependencies, you can also use nox to
run the tests and other checks creating its own virtual environments:
You can also use nox -R to reuse the current testing environment to speed up
test at the expense of a higher chance to end up with a dirty test environment.
Running tests / checks individually¤
For a better development test cycle you can install the runtime and test
dependencies and run pytest manually.
python -m pip install .[dev-pytest] # included in .[dev] too
# And for example
pytest tests/test_*.py
Or you can use nox:
The same applies to pylint or mypy for example:
Golden Tests¤
To test the generated files using the Cookiecutter templates, the golden testing technique is used to ensure that changes in the templates don't occur unexpectedly.
If a golden test fails, a diff of the contents will be provided in the test
results.
Failures in the golden tests could indicate two things:
-
The generated files don't match the golden files because an unintended change was introduced. For example, there may be a bug that needs to be fixed so that the generated files match the golden files again.
-
The generated files don't match the golden files because an intended change was introduced. In this case, the golden files need to be updated.
In the latter case, manually updating files is complicated and error-prone, please consult Update the golden test fixtures below for the recommended way to update the golden files.
Modifying Cookiecutter Templates¤
When you need to make changes to the cookiecutter templates (located in
cookiecutter/{{cookiecutter.github_repo_name}}/), there's a specific workflow
to follow. The templates support different project types (actor, api, app, lib,
model) using Jinja2 conditional sections.
1. Make your template changes¤
Edit the files in the cookiecutter/ directory. Keep in mind that some files
have conditional sections for different project types.
2. Update the golden test fixtures¤
After modifying templates, you need to regenerate the golden files so the tests pass:
If you renamed or removed files, it's safest to wipe the golden directory first:
rm -rf tests_golden/
UPDATE_GOLDEN=1 pytest tests/integration/test_cookiecutter_generation.py::test_golden
Please ensure that all introduced changes are intended before updating the golden files.
3. Write a migration script¤
Existing projects using these templates need a way to migrate to the new
version. Update cookiecutter/migrate.py to handle this migration. The script
is reset to a blank template (from .github/cookiecutter-migrate.template.py)
after each release.
A few guidelines for migration scripts:
- Make them idempotent (safe to run multiple times)
- Print clear messages about what's being done
- Use the helper functions:
replace_file_contents_atomically()for simple replacements,apply_patch()for more complex changes
For changes that can't be automated (like GitHub repository settings that
require admin permissions, or changes that would make the script overly
complex), use the manual_step() function to clearly communicate to users
what they need to do manually:
manual_step(
"Update branch protection rules in GitHub settings:\n"
" Settings > Branches > Enable 'Require status checks'"
)
4. Validate with this repository¤
Run the migration script on this repository itself to make sure it works:
Note that some changes (e.g., API-specific features like protolint) won't be
testable on this repository since it's a lib type project.
5. Update the release notes¤
Add an entry to RELEASE_NOTES.md describing what changed in the templates.
6. Commit separately¤
Create two separate commits to keep the history clean:
- First commit: the template changes, migration script, golden tests, and release notes
- Second commit: the changes to this repository from running the migration
This separation makes it easier to review and understand what changed.
Building the documentation¤
To build the documentation, first install the dependencies (if you didn't
install all dev dependencies):
Then you can build the documentation (it will be written in the site/
directory):
Or you can just serve the documentation without building it using:
Your site will be updated live when you change your files (provided that
you used pip install -e ., beware of a common pitfall of using pip install
without -e, in that case the API reference won't change unless you do a new
pip install).
To build multi-version documentation, we use mike. If you want to see how the multi-version sites looks like locally, you can use:
mike works in mysterious ways. Some basic information:
mike deploywill do amike buildand write the results to your localgh-pagesbranch.my-versionis an arbitrary name for the local version you want to preview.mike set-defaultis needed so when you serve the documentation, it goes to your newly produced documentation by default.mike servewill serve the contents of your localgh-pagesbranch. Be aware that, unlikemkdocs serve, changes to the sources won't be shown live, as themike deploystep is needed to refresh them.
Be careful not to use --push with mike deploy, otherwise it will push your
local gh-pages branch to the origin remote.
That said, if you want to test the actual website in your fork, you can
always use mike deploy --push --remote your-fork-remote, and then access the
GitHub pages produced for your fork.
Extra release steps for this repository¤
Releasing below describes the procedure that is shared by every project generated from these templates, and it is kept as close to the generated file as possible so it can be updated automatically. This repository is the template, so it needs a few extra steps around that procedure. They are collected here, and this is where new repository-specific steps should be added.
Before creating the tag¤
- Bump the
frequenz-repo-configversion used by the cookiecutter template to the version you are about to release. Generated projects pin an exact version, so forgetting this ships a template that installs the previous release. This is what happened in v0.18.0, which was released with the template still pinning 0.17.0.
The pin appears four times in
cookiecutter/{{cookiecutter.github_repo_name}}/pyproject.toml
(in [build-system] requires, dev-mkdocs, dev-noxfile and
dev-pytest), and the golden fixtures have to be regenerated afterwards:
-
Consider bumping the other versions pinned by the template (
cookiecutter/{{cookiecutter.github_repo_name}}/pyproject.tomland the actions used incookiecutter/{{cookiecutter.github_repo_name}}/.github/). Dependabot doesn't manage the template, so these only move when someone updates them by hand. Add the corresponding migration steps tocookiecutter/migrate.pyif existing projects need them. -
While cleaning up
RELEASE_NOTES.md, leave the literal<tag>in the migration script URL as it is. It is a deliberate placeholder and it is not substituted at release time.
After the release is published¤
For a patch release cut from a maintenance branch (vX.Y.x), skip the
cookiecutter/migrate.py half of the reset. The migration script on that
branch still describes the migration from the previous minor release, so
resetting it would drop steps that projects upgrading to that branch still
need.
Maintenance branches (vX.Y.x) are created on demand, when the first backport
is needed, and need no ruleset or Dependabot changes: the existing rules
already match that name pattern.
Releasing¤
These are the steps to create a new release:
- Get the latest head you want to create a release from.
If you use milestones, check that the milestone for this release has no open issues left, and move whatever won't make it to a later milestone.
- Update the
RELEASE_NOTES.mdfile if it is not complete, up to date, and remove template comments (<!-- ... -->) and empty sections. Submit a pull request if an update is needed, wait until it is merged, and update the latest head you want to create a release from to get the new merged pull request.
This can't be left for later: the notes published with the release are read
from the RELEASE_NOTES.md committed at the tagged commit, not from the tag
message.
- Create a new signed tag using the release notes and
a semver compatible version number with a
vprefix, for example:
The signature is not optional: the Protect released tags ruleset rejects unsigned tags. It also forbids updating or deleting a released tag, so make sure you are tagging the commit you want.
-
Push the new tag.
-
A GitHub action will test the tag and if all goes well it will create a GitHub Release, and upload a new package to PyPI automatically.
-
Once this is done, reset the
RELEASE_NOTES.mdwith the template:
cp .github/RELEASE_NOTES.template.md RELEASE_NOTES.md
cp .github/cookiecutter-migrate.template.py cookiecutter/migrate.py
Commit the new release notes and migration script, and create a PR (this step should be automated eventually too).
-
If you use milestones, close the milestone for this release, and create the next one if it doesn't exist yet.
-
Celebrate!