Clone the git repository:

$ git clone
$ cd pglift

Then, create a Python3 virtualenv and install the project:

$ python3 -m venv .venv --upgrade-deps
$ . .venv/bin/activate
(.venv) $ pip install -r requirements/dev.txt

Though not required, tox can be used to run all checks (lint, tests, etc.) needed in development environment.

Linting, formatting, type-checking

The project uses flake8 for linting, black and isort for formatting and mypy for type-checking.

All these checks can be run with tox -e lint or individually.

When creating new files in the source repository, take care to include the license and copyright information. This can be handled with the help of the reuse tool, e.g.:

$ reuse annotate \
    --copyright="Dalibo" --year=2021 --license="GPL-3.0-or-later" \

For data files (even in tests), the --force-dot-license might be also passed to the latter command.

Running tests

The test suite can be run either directly:

(.venv) $ pytest

or through tox:

$ tox [-e tests-doctest|tests-unit|tests-func|tests-expect]

The test suite is quite extensive and can take long to run. It is split into functional tests and more unit ones (including doctests), the former require a real PostgreSQL instance (which will be set up automatically) while the latter do not. Each test suite gets a dedicated tox environment: tests-doctest, tests-unit, tests-func and tests-expect.

When working on a simple fix or change that would be covered by non-functional tests, one can run the following part of the test suite quickly:

(.venv) $ pytest lib/src tests/unit

or through tox:

$ tox -e tests-doctest -e tests-unit

Some unit tests use local files (in test/data) to compare actual results with their expectation. Often, when there is a mismatch that is intentional (e.g. if interface models changed), it’s handy to write back expected files: for this, pass --write-changes option to pytest invocation.

By default, functional tests will not use systemd as a service manager / scheduler. In order to run tests with systemd, pass the --systemd option to pytest command.

Still in functional tests, the PostgreSQL environment would be guessed by inspecting the system to look for PostgreSQL binaries for the most recent version available. If multiple versions of PostgreSQL are available, a specific version can be selected by passing --pg-version=<version> option to the pytest command. Likewise, the --pg-auth option can be used to run tests with specified authentication method.

“Expect” tests (in t directory) should be run with Prysk, not pytest.

If your system uses a specific locale and your tests are failing because of assertion issues with translated messages, you can run the tests with LANG=C.


Tests automatically run in the regular CI don’t use the systemd because of technical limitations. However developers (with appropriate permissions) can launch CI jobs using the sourcehut builds service. Job manifests available in the .builds directory will be run any time commits are pushed to the repository.

If you run this test suite (based on Prysk and port-for) with a non-root user, you will probably have a permission issue for writing to /etc/port-for.conf. As a workaround you can create that file and change the ownership manually to the user running those tests.

$ sudo touch /etc/port-for.conf
$ sudo chown ${UID}:${GID} /etc/port-for.conf

Pre-commit hooks

Some checks (linting, typing, syntax checking, …) can be done for you before git commits.

You just need to install the pre-commit hooks:

(.venv) $ pre-commit install

Working on documentation

To build the documentation in HTML format, run:

(.venv) $ make -C docs html

and open docs/_build/html/index.html to browse the result.

Alternatively, keep the following command running:

(.venv) $ make -C docs serve

to get the documentation rebuilt and along with a live-reloaded Web browser.

Contributing changes

  • Make sure that lint, typing checks pass as well as at least unit tests.

  • If needed, create a news fragment using towncrier create <id>.<type>.rst [--edit] where <id> is a short description of changes and <type> describes the type of changes, within: feature, bugfix, removal, doc or misc.

  • When committing changes with git, write one commit per logical change and try to follow pre-existing style and write a meaningful commit message (see for a quick guide).

Release workflow


Prior to releasing, first, the dependencies for building pglift’s binary with PyOxidizer need to be pinned and compiled. This is done by:

  • running tox -e pin, which, if pyoxidizer/requirements.txt changed, commits the result,

  • in which case, the next step is to create a merge request in which the buildbin job would run (along with, possibly, tests-binary ones), and,

  • then proceeding with next steps after merge.


Assuming we’re releasing version 1.2.3, the following steps should be followed:

  • Build the changelog

    $ towncrier build --version=1.2.3
    $ git commit -m "Prepare version 1.2.3"
  • Create an annotated git tag following the v<MAJOR>.<MINOR>.<PATCH> pattern.

    $ git tag v1.2.3 -a [-s] -m 'pglift v1.2.3' --edit

    then edit the tag message to include a changelog since latest release (as built in the previous step).

  • Push the tag to the main (upstream) repository:

    $ git push --follow-tags
  • Soon after, the CI will build and upload the Python package to PyPI.

Ansible collection

If a release of the Ansible collection is needed, in the ansible/ directory:

  • Bump the version number in galaxy.yml

  • Generate the changelog, using antsibull-changelog, by running antsibull-changelog release

  • Commit the result

  • Make an annotated tag with the ansible/ prefix, e.g.

    $ git tag ansible/v0.10.0 -a [-s] -m 'pglift-ansible v0.10.0'
  • Push to the upstream repository, the CI should then handle publication to Ansible Galaxy.