PyScaffold comes with a lot of elaborated features and configuration defaults to make the most common tasks in developing, maintaining and distributing your own Python package as easy as possible.
Configuration, Packaging & Distribution¶
All configuration can be done in
setup.cfg like changing the description,
url, classifiers, installation requirements and so on as defined by setuptools.
That means in most cases it is not necessary to tamper with
The syntax of
setup.cfg is pretty much self-explanatory and well commented,
check out this example or setuptools’ documentation.
In order to build a source, binary or wheel distribution, just run
python setup.py sdist,
python setup.py bdist or
python setup.py bdist_wheel (recommended).
Uploading to PyPI
Of course uploading your package to the official Python package index PyPI for distribution also works out of the box. Just create a distribution as mentioned above and use twine to upload it to PyPI, e.g.:
pip install twine twine upload dist/*
Please also note that PyPI does not allow uploading local versions for practical reasons. Thus, you have to create a git tag before uploading a version of your distribution. Read more about it in the versioning section below.
Optionally, namespace packages can be used, if you are planning to distribute a larger package as a collection of smaller ones. For example, use:
putup my_project --package my_package --namespace com.my_domain
my_package inside the namespace
com.my_domain in java-style.
Package and Files Data
Additional data, e.g. images and text files, that reside within your package and
are tracked by Git will automatically be included
include_package_data = True in
It is not necessary to have a
MANIFEST.in file for this to work. Just make
sure that all files are added to your repository.
To read this data in your code, use:
from pkgutil import get_data data = get_data('my_package', 'path/to/my/data.txt')
Starting from Python 3.7 an even better approach is using importlib.resources:
from importlib.resources import read_text, read_binary data = read_text('my_package', 'path/to/my/data.txt')
from pkg_resources import resource_string data = resource_string(__name__, 'path/to/my/data/relative/to/module.txt')
Yes, actually “there should be one– and preferably only one –obvious way to do it.” ;-)
Versioning and Git Integration¶
Your project is already an initialised Git repository and
the information of tags to infer the version of your project with the help of setuptools_scm.
To use this feature you need to tag with the format
python setup.py --version to retrieve the current PEP440-compliant version.
This version will be used when building a package and is also accessible through
my_project.__version__. If you want to upload to PyPI you have to tag the current commit
before uploading since PyPI does not allow local versions, e.g.
for practical reasons.
Best Practices and Common Errors with Version Numbers
How do I get a clean version like 3.2.4 when I have 3.2.3.post0.dev9+g6817bd7? Just commit all your changes and create a new tag using
git tag v3.2.4. In order to build an old version checkout an old tag, e.g.
git checkout -b v3.2.3 v3.2.3and run
python setup.py bdist_wheel.
Why do I see `unknown` as version? In most cases this happens if your source code is no longer a proper Git repository, maybe because you moved or copied it or Git is not even installed. In general using
python setup.py install(or
develop) to install your package is only recommended for developers of your Python project, which have Git installed and use a proper Git repository anyway. Users of your project should always install it using the distribution you built for them e.g.
pip install my_project-3.2.3-py3-none-any.whl. You build such a distribution by running
python setup.py bdist_wheeland then find it under
Is there a good versioning scheme I should follow? The most common practice is to use Semantic Versioning. Following this practice avoids the so called dependency hell for the users of your package. Also be sure to set attributes like
Is there a best practise for distributing my package? First of all, cloning your repository or just coping your code around is a really bad practice which comes with tons of pitfalls. The clean way is to first build a distribution and then give this distribution to your users. This can be done by just copying the distribution file or uploading it to some artifact store like PyPI for public packages or devpi, Nexus, etc. for private packages. Also check out this article about packaging, versioning and continuous integration.
Using some CI service, why is the version `unknown` or `my_project-0.0.post0.dev50`? Some CI services use shallow git clones, i.e.
--depth N, or don’t download git tags to save bandwidth. To verify that your repo works as expected, run:
git describe --dirty --tags --long --first-parent
which is basically what setuptools_scm does to retrieve the correct version number. If this command fails, tweak how your repo is cloned depending on your CI service and make sure to also download the tags, i.e.
git fetch origin --tags.
Unleash the power of Git by using its pre-commit hooks.
This feature is available through the
After your project’s scaffold was generated, make sure pre-commit is
pip install pre-commit, then just run
It goes unsaid that also a default
.gitignore file is provided that is well
adjusted for Python projects and the most common tools.
Build the documentation with
python setup.py docs and run doctests with
python setup.py doctest after you have Sphinx installed.
Start editing the file
docs/index.rst to extend the documentation.
The documentation also works with Read the Docs.
The Numpy and Google style docstrings are activated by default. Just make sure Sphinx 1.3 or above is installed.
Dependency Management in a Breeze¶
PyScaffold out of the box allows developers to express abstract dependencies
and take advantage of
pip to manage installation. It also can be used
together with a virtual environment to avoid dependency hell during both
development and production stages.
In particular, PyPA’s Pipenv can be integrated in any PyScaffold-generated
project by following standard setuptools conventions. Keeping abstract
setup.cfg and running
pipenv install -e . is basically
what you have to do (details in Dependency Management).
Experimental Feature - Pipenv support is experimental and might change in the future
Unittest & Coverage¶
python setup.py test to run all unittests defined in the subfolder
tests with the help of py.test and pytest-runner. Some sane
default flags for py.test are already defined in the
[pytest] section of
setup.cfg. The py.test plugin pytest-cov is used to automatically
generate a coverage report. It is also possible to provide additional
parameters and flags on the commandline, e.g., type:
python setup.py test --addopts -h
to show the help of py.test.
JUnit and Coverage HTML/XML
For usage with a continuous integration software JUnit and Coverage XML output
can be activated in
setup.cfg. Use the flag
--travis to generate
templates of the Travis configuration files
tests/travis_install.sh which even features the
coverage and stats system Coveralls.
In order to use the virtualenv management and test tool tox
--tox can be specified.
If you are using GitLab you can get a default
.gitlab-ci.yml also running pytest-cov with the flag
Managing test environments with tox
tox to generate test virtual environments for various python
environments defined in the generated
tox.ini. Testing and building
sdists for python 2.7 and python 3.4 is just as simple with tox as:
tox -e py27,py34
Environments for tests with the the static code analyzers pyflakes and pep8 which are bundled in flake8 are included as well. Run it explicitly with:
tox -e flake8
With tox, you can use the
--recreate flag to force tox to create new
environments. By default, PyScaffold’s tox configuration will execute tests for
a variety of python versions. If an environment is not available on the system
the tests are skipped gracefully. You can rely on the tox documentation
for detailed configuration options.
Management of Requirements & Licenses¶
Installation requirements of your project can be defined inside
install_requires = numpy; scipy. To avoid package dependency problems
it is common to not pin installation requirements to any specific version,
although minimum versions, e.g.
sphinx>=1.3, or maximum versions, e.g.
pandas<0.12, are used sometimes.
More specific installation requirements should go into
This file can also be managed with the help of
pip compile from pip-tools
that basically pins packages to the current version, e.g.
The packages defined in
requirements.txt can be easily installed with:
pip install -r requirements.txt
All licenses from choosealicense.com can be easily selected with the help
PyScaffold comes with several extensions:
- If you want a project setup for a Data Science task, just use
--dsprojectafter having installed pyscaffoldext-dsproject.
- Create a Django project with the flag
--djangowhich is equivalent to
django-admin.py startproject my_projectenhanced by PyScaffold’s features.
- Create a template for your own PyScaffold extension with
--custom-extensionafter having installed pyscaffoldext-custom-extension with
- Have a
README.mdbased on MarkDown instead of
--markdownafter having installed pyscaffoldext-markdown with
- Add a
pyproject.tomlfile according to PEP 518 to your template by using
--pyprojectafter having installed pyscaffoldext-pyproject with
- With the help of Cookiecutter it is possible to further customize your project
setup with a template tailored for PyScaffold. Just use the flag
--cookiecutter TEMPLATEto use a cookiecutter template which will be refined by PyScaffold afterwards.
- … and many more like
--gitlabto create the necessary files for GitLab.
There is also documentation about writing extensions. Find more
extensions within the PyScaffold organisation and consider contributing your own.
All extensions can easily be installed with
Deprecation Notice - In the next major release both Cookiecutter and
Django extensions will be extracted into independent packages. After
PyScaffold v4.0, you will need to explicitly install
pyscaffoldext-django in your
system/virtualenv in order to be able to use them.
Keep your project’s scaffold up-to-date by applying
putup --update my_project when a new version of PyScaffold was released.
An update will only overwrite files that are not often altered by users like
setup.py. To update all files use
An existing project that was not setup with PyScaffold can be converted with
putup --force existing_project. The force option is completely safe to use
since the git repository of the existing project is not touched!
Also check out if configuration options in
setup.cfg have changed.
Updates from PyScaffold 2¶
Since the overall structure of a project set up with PyScaffold 2 differs quite
much from a project generated with PyScaffold 3 it is not possible to just use
--update parameter. Still with some manual efforts an update from
a scaffold generated with PyScaffold 2 to PyScaffold 3’s scaffold is quite easy.
Assume the name of our project is
old_project with a package called
old_package and no namespaces then just:
- make sure your worktree is not dirty, i.e. commit all your changes,
putup old_project --force --no-skeleton -p old_packageto generate the new structure inplace and
cdinto your project,
- move with
git mv old_package/* src/old_package/ --forceyour old package over to the new
git statusand add untracked files from the new structure,
git difftoolto check all overwritten files, especially
setup.cfg, and transfer custom configurations from the old structure to the new,
- check if
python setup.py test sdistworks and commit your changes.
With the help of an experimental updating functionality it is also possible to
add additional features to your existing project scaffold. If a scaffold lacking
.travis.yml was created with
putup my_project it can later be added by issuing
putup --update my_project --travis. For this to work, PyScaffold stores all
options that were initially used to put up the scaffold under the
setup.cfg. Be aware that right now PyScaffold provides no way to
remove a feature which was once added.