This page describes a common development setup for working with Mopidy and Mopidy extensions. Of course, there may be other ways that work better for you and the tools you use, but here's one recommended way to do it.
The following steps help you get a good initial setup. They build on each other to some degree, so if you're not very familiar with Python development it might be wise to proceed in the order laid out here.
Install Mopidy the regular way. Mopidy has some non-Python dependencies which may be tricky to install. Thus we recommend to always start with a full regular Mopidy install, as described in :ref:`installation`. That is, if you're running e.g. Debian, start with installing Mopidy from Debian packages.
Make a directory to be used as a workspace for all your Mopidy development:
mkdir ~/mopidy-dev
It will contain all the Git repositories you'll check out when working on Mopidy and extensions.
Make a Python virtualenv for Mopidy development. The virtualenv will wall off Mopidy and its dependencies from the rest of your system. All development and installation of Python dependencies, versions of Mopidy, and extensions are done inside the virtualenv. This way your regular Mopidy install, which you set up in the first step, is unaffected by your hacking and will always be working.
To create a virtualenv in the Mopidy workspace directory, run:
python3 -m venv ~/mopidy-dev/.venv
Now, each time you open a terminal and want to activate the mopidy
virtualenv, run:
. ~/mopidy-dev/.venv/bin/activate
There are lots of ways to set up your shell to automatically activate the virtualenv,
e.g. when changing directory into ~/mopidy-dev/
or a subdirectory.
As this is just convenience and not strictly required,
it is left as an exercise for the reader.
Once inside the virtualenv, it's time to clone the mopidy/mopidy
Git repo
from GitHub:
cd ~/mopidy-dev/ git clone https://github.com/mopidy/mopidy.git
When you've cloned the mopidy
Git repo, cd
into it:
cd ~/mopidy-dev/mopidy/
With a fresh clone of the Git repo, you should start out on the main
branch. This is where all features for the next feature release land. To
confirm that you're on the right branch, run:
git branch
Next up, we'll want to run Mopidy from the Git repo. There's two reasons for this: first of all, it lets you easily change the source code, restart Mopidy, and see the change take effect. Second, it's a convenient way to keep at the bleeding edge, testing the latest developments in Mopidy itself or test some extension against the latest Mopidy changes.
Assuming you're still inside the Git repo, use pip to install Mopidy from the Git repo in an "editable" form:
pip install --upgrade --editable .
When using the --editable
flag, the source code is not copied into the
virtualenv's site-packages
directory, but instead creates a link there
pointing to the Git repo. This way, you can change the source code in the Git
repo and the changes will be visible inside the virtualenv without having to
reinstall Mopidy.
It will also create a mopidy
executable inside the virtualenv that will
always run the latest code from the Git repo:
$ cat ~/mopidy-dev/.venv/bin/mopidy ...
This file is on the path when the virtualenv is active, so you can run it from anywhere, simply by running:
mopidy
Note
It is also possible to run Python apps directly,
e.g. using python3 src/mopidy
directly on the
:file:`~/mopidy-dev/mopidy/src/mopidy/` Python package directory.
However, if you don't run the install command above,
the extensions bundled with Mopidy will not be registered and made available
for use, making Mopidy quite useless.
Third, the install command will register the bundled Mopidy extensions so that Mopidy may find them through :mod:`importlib`. The result of this can be seen in a file named :file:`entry_points.txt` which can be found inside the virtualenv dir, e.g. :file:`~/mopidy-dev/.venv/lib/python3.11/site-packages/Mopidy-4.0.0.dist-info/entry_points.txt`. The :file:`entry_points.txt` file is of special interest as it shows both how the above executable and the bundled extensions are connected to the Mopidy source code:
[console_scripts]
mopidy = mopidy.__main__:main
[mopidy.ext]
file = mopidy.file:Extension
http = mopidy.http:Extension
m3u = mopidy.m3u:Extension
softwaremixer = mopidy.softwaremixer:Extension
stream = mopidy.stream:Extension
Before continuing, you will probably want to install the development tools we use as well. These can be installed into the active virtualenv by running:
cd ~/mopidy-dev/mopidy/ pip install --upgrade --editable ".[dev]"
Note that this is the same command as you used to install Mopidy from the Git
repo, with the addition of the [dev]
suffix after .
. This makes pip
install the "dev" set of extra dependencies. Exactly what the "dev" set
includes are defined in :file:`pyproject.toml`.
To upgrade the development tools in the future, just rerun the exact same command.
As long as the virtualenv is activated, you can start Mopidy from any directory. Simply run:
mopidy
To stop it again, press Ctrl+C.
Every time you change code in Mopidy or an extension and want to see it live, you must restart Mopidy.
If you want to iterate quickly while developing, it may sound a bit tedious to restart Mopidy for every minor change. Then it's useful to have tests to exercise your code...
Mopidy has quite good test coverage, and we would like all new code going into Mopidy to come with tests.
You need to know at least one command; the one that runs all the tests:
tox
This will run exactly the same tests as our CI setup runs for all our branches and pull requests. If this command turns green, you can be quite confident that your pull request will get the green flag from CI as well, which is a requirement for it to be merged.
As this is the ultimate test command, it's also the one taking the most time to run; up to a minute, depending on your system. But, if you have patience, this is all you need to know. Always run this command before pushing your changes to GitHub.
If you take a look at the tox config file, :file:`tox.ini`, you'll see that tox
runs tests in multiple environments, including a ruff
environment that
lints the source code for issues and a docs
environment that tests that the
documentation can be built. You can also limit tox to just test specific
environments using the -e
option, e.g. to run just unit tests on Python 3.11:
tox -e py311
To learn more, see the tox documentation .
Before submitting a pull request, we recommend running:
tox
This will locally run similar tests to what we use in our CI runs and help us to merge high-quality contributions.
Under the hood, tox -e py311
will use pytest
as the test runner. We can also use it directly to run all tests:
pytest
pytest has lots of possibilities, so you'll have to dive into their docs and plugins to get full benefit from it. To get you interested, here are some examples.
We can limit to just tests in a single directory to save time:
pytest tests/http/
With the help of the pytest-xdist
plugin, we can run tests with four Python
processes in parallel, which usually cuts the test time in half or more:
pytest -n 4
Another useful feature from pytest-xdist
, is the possibility to stop on the
first test failure, watch the file system for changes, and then rerun the
tests. This makes for a very quick code-test cycle:
pytest -f # or --looponfail
With the help of the pytest-cov plugin, we can get a report on what parts of
the given module, mopidy
in this example, are covered by the test suite:
pytest --cov=mopidy --cov-report=term-missing
Note
Up to date test coverage statistics can also be viewed online at Codecov.
If we want to speed up the test suite, we can even get a list of the ten slowest tests:
pytest --durations=10
By now, you should be convinced that running pytest directly during development can be very useful.
Mopidy uses GitHub Actions for automatically running the test suite when code is pushed to GitHub. This works both for the main Mopidy repo, but also for any forks. This way, any contributions to Mopidy through GitHub will automatically be tested, and the build status will be visible in the GitHub pull request interface, making it easier to evaluate the quality of pull requests.
For each successful build, the CI setup submits code coverage data to Codecov. If you're out of work, Codecov might help you find areas in the code which could need better test coverage.
We're quite pedantic about :ref:`codestyle` and try hard to keep the Mopidy code base a very clean and nice place to work in.
Luckily, you can get very far by using the ruff linter to check your code for issues before
submitting a pull request. Mopidy's ruff rules are configured in :file:`pyproject.toml`.
You can either run the ruff
tox environment, like our CI setup will do on
your pull request:
tox -e ruff
Or you can run ruff directly:
ruff .
If successful, the command will not print anything at all.
Note
In some rare cases it doesn't make sense to listen to ruff's warnings. In
those cases, ignore the check by appending # noqa: <warning code>
to
the source line that triggers the warning. The # noqa
part will make
ruff skip all checks on the line, while the warning code will help other
developers lookup what you are ignoring.
To write documentation, we use Sphinx. See their site for lots of documentation on how to use Sphinx.
Note
To generate a few graphs which are part of the documentation, you need to install the graphviz package. You can install it from APT with:
sudo apt install graphviz
Other distributions typically use the same package name.
To build the documentation, go into the :file:`docs/` directory:
cd ~/mopidy-dev/mopidy/docs/
Then, to see all available build targets, run:
make
To generate an HTML version of the documentation, run:
make html
The generated HTML will be available at :file:`_build/html/index.html`. To open it in a browser you can run either of the following commands, depending on your OS:
xdg-open _build/html/index.html # Linux open _build/html/index.html # OS X
The documentation at https://docs.mopidy.com/ is hosted by Read the Docs, which automatically updates the documentation
when a change is pushed to the mopidy/mopidy
repo at GitHub.
Much of the above also applies to Mopidy extensions, though they're often a bit simpler. They don't have documentation sites and their test suites are either small and fast, or sadly missing entirely. Most of them use tox to run various linters, and pytest can be used to run their test suites.
As always, the mopidy
virtualenv should be active when working on
extensions:
. ~/mopidy-dev/.venv/bin/activate
Just like with non-development Mopidy installations, you can install extensions using pip:
pip install Mopidy-Scrobbler
Installing an extension from its Git repo works the same way as with Mopidy itself. First, go to the Mopidy workspace:
cd ~/mopidy-dev/
Clone the desired Mopidy extension:
git clone https://github.com/mopidy/mopidy-spotify.git
Change to the newly created extension directory:
cd ~/mopidy-dev/mopidy-spotify/
Then, install the extension in "editable" mode, so that it can be imported from anywhere inside the virtualenv and the extension is registered and discoverable through :mod:`importlib`:
pip install --editable .
Every extension will have a README.rst
file. It may contain information
about extra dependencies required, development process, etc. Extensions usually
have a changelog in their GitHub relases page.
Extensions often have a much quicker life cycle than Mopidy itself, often with daily releases in periods of active development. To find outdated extensions in your virtualenv, you can run:
pip list --outdated
To upgrade an extension installed with pip, simply use pip:
pip install --upgrade Mopidy-Scrobbler
To upgrade an extension installed from a Git repo, it's usually enough to pull the new changes in:
cd ~/mopidy-dev/mopidy-spotify/ git pull
Of course, if you have local modifications, you'll need to stash these away on a branch or similar first.
Depending on the changes to the extension, it may be necessary to update the metadata about the extension package by installing it in "editable" mode again:
pip install --editable .
Before you being, make sure you've read the :ref:`contributing` page and the guidelines there. This section will focus more on the practical workflow.
For the examples, we're making a change to Mopidy. Approximately the same workflow should work for most Mopidy extensions too.
Assuming we already have a local Git clone of the upstream Git repo in
:file:`~/mopidy-dev/mopidy/`, we can run git remote -v
to list the
configured remotes of the repo:
$ git remote -v origin https://github.com/mopidy/mopidy.git (fetch) origin https://github.com/mopidy/mopidy.git (push)
For clarity, we can rename the origin
remote to upstream
:
$ git remote rename origin upstream $ git remote -v upstream https://github.com/mopidy/mopidy.git (fetch) upstream https://github.com/mopidy/mopidy.git (push)
If you haven't already, fork the repository to your own GitHub account.
Then, add the new fork as a remote to your local clone:
git remote add myuser git@github.com:myuser/mopidy.git
The end result is that you have both the upstream repo and your own fork as remotes:
$ git remote -v myuser git@github.com:myuser/mopidy.git (fetch) myuser git@github.com:myuser/mopidy.git (push) upstream https://github.com/mopidy/mopidy.git (fetch) upstream https://github.com/mopidy/mopidy.git (push)
Fetch the latest data from all remotes without affecting your working directory:
git remote update --prune
Now, we are ready to create and checkout a new branch off of the upstream
main
branch for our work:
git checkout -b fix-crash-on-foo upstream/main
Do the work, while remembering to adhere to code style, test the changes, make necessary updates to the documentation, and making small commits with good commit messages. All as described in :ref:`contributing` and elsewhere in the :ref:`devenv` guide.
When everything is done and committed, push the branch to your fork on GitHub:
git push myuser fix-crash-on-foo
Go to the repository on GitHub where you want the change merged, in this case https://github.com/mopidy/mopidy, and create a pull request.
When the pull request is created, our CI setup will run all tests on it. If something fails, you'll usually get a notification from GitHub. You might as well just fix the issues right away, as we won't merge a pull request without all CI builds being green. See :ref:`running-tests` on how to run the same tests locally as our CI setup runs on your pull request.
When you've fixed the issues, you can update the pull request simply by pushing more commits to the same branch in your fork:
git push myuser fix-crash-on-foo
Likewise, when you get review comments from other developers on your pull request, you're expected to create additional commits which addresses the comments. Push them to your branch so that the pull request is updated.