Development environment

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.

Initial setup

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

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 Installation. That is, if you’re running e.g. Debian, start with installing Mopidy from Debian packages.

Make a development workspace

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 virtualenv

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.

Most of us use the virtualenvwrapper to ease working with virtualenvs, so that’s what we’ll be using for the examples here. First, install and setup virtualenvwrapper as described in their docs.

To create a virtualenv named mopidy which uses Python 2.7, allows access to system-wide packages like GStreamer, and uses the Mopidy workspace directory as the “project path”, run:

mkvirtualenv -a ~/mopidy-dev --python `which python2.7` \
  --system-site-packages mopidy

Now, each time you open a terminal and want to activate the mopidy virtualenv, run:

workon mopidy

This will both activate the mopidy virtualenv, and change the current working directory to ~/mopidy-dev.

Clone the repo from GitHub

Once inside the virtualenv, it’s time to clone the mopidy/mopidy Git repo from GitHub:

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 develop branch. This is where all features for the next feature release land. To confirm that you’re on the right branch, run:

git branch

Install development tools

We use a number of Python development tools. The dev-requirements.txt file has comments describing what we use each dependency for, so we might just as well include the file verbatim here:

# Automate tasks
invoke

# Build documentation
sphinx

# Check code style, errors, etc
flake8
flake8-import-order

# Mock dependencies in tests
mock
responses

# Test runners
pytest
pytest-capturelog
pytest-cov
pytest-xdist
tox

# Check that MANIFEST.in matches Git repo contents before making a release
check-manifest

# To make wheel packages
wheel

# Securely upload packages to PyPI
twine

Install them all into the active virtualenv by running pip:

pip install --upgrade -r dev-requirements.txt

To upgrade the tools in the future, just rerun the exact same command.

Install Mopidy from the Git repo

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 --editable .

This will not copy the source code into the virtualenv’s site-packages directory, but instead create a link there pointing to the Git repo. Using cdsitepackages from virtualenvwrapper, we can quickly show that the installed Mopidy.egg-link file points back to the Git repo:

$ cdsitepackages
$ cat Mopidy.egg-link
/home/user/mopidy-dev/mopidy
.%
$

It will also create a mopidy executable inside the virtualenv that will always run the latest code from the Git repo. Using another virtualenvwrapper command, cdvirtualenv, we can show that too:

$ cdvirtualenv
$ cat bin/mopidy
...

The executable should contain something like this, using pkg_resources to look up Mopidy’s “console script” entry point:

#!/home/user/virtualenvs/mopidy/bin/python2
# EASY-INSTALL-ENTRY-SCRIPT: 'Mopidy==0.19.5','console_scripts','mopidy'
__requires__ = 'Mopidy==0.19.5'
import sys
from pkg_resources import load_entry_point

if __name__ == '__main__':
    sys.exit(
        load_entry_point('Mopidy==0.19.5', 'console_scripts', 'mopidy')()
    )

Note

It still works to run python mopidy directly on the ~/mopidy-dev/mopidy/mopidy/ Python package directory, but if you don’t run the pip install command above, the extensions bundled with Mopidy will not be registered with pkg_resources, making Mopidy quite useless.

Third, the pip install command will register the bundled Mopidy extensions so that Mopidy may find them through pkg_resources. The result of this can be seen in the Git repo, in a new directory called Mopidy.egg-info, which is ignored by Git. The Mopidy.egg-info/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]
http = mopidy.http:Extension
local = mopidy.local:Extension
mpd = mopidy.mpd:Extension
softwaremixer = mopidy.softwaremixer:Extension
stream = mopidy.stream:Extension

Warning

It’s not uncommon to clean up in the Git repo now and then, e.g. by running git clean.

If you do this, then the Mopidy.egg-info directory will be removed, and pkg_resources will no longer know how to locate the “console script” entry point or the bundled Mopidy extensions.

The fix is simply to run the install command again:

pip install --editable .

Finally, we can go back to the workspace, again using a virtualenvwrapper tool:

cdproject

Running Mopidy from Git

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...

Running tests

Mopidy has quite good test coverage, and we would like all new code going into Mopidy to come with tests.

Test it all

You need to know at least one command; the one that runs all the tests:

tox

This will run exactly the same tests as Travis CI 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 Travis 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, tox.ini, you’ll see that tox runs tests in multiple environments, including a flake8 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:

tox -e py27

To learn more, see the tox documentation .

Running unit tests

Under the hood, tox -e py27 will use pytest as the test runner. We can also use it directly to run all tests:

py.test

py.test 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:

py.test 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:

py.test -n 4

Another useful feature from pytest-xdist, is the possiblity 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:

py.test -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:

py.test --cov=mopidy --cov-report=term-missing

Note

Up to date test coverage statistics can also be viewed online at coveralls.io.

If we want to speed up the test suite, we can even get a list of the ten slowest tests:

py.test --durations=10

By now, you should be convinced that running py.test directly during development can be very useful.

Continuous integration

Mopidy uses the free service Travis CI 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 by Travis CI, 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, Travis submits code coverage data to coveralls.io. If you’re out of work, coveralls might help you find areas in the code which could need better test coverage.

Style checking and linting

We’re quite pedantic about Code style 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 flake8 linter to check your code for issues before submitting a pull request. Mopidy passes all of flake8’s checks, with only a very few exceptions configured in setup.cfg. You can either run the flake8 tox environment, like Travis CI will do on your pull request:

tox -e flake8

Or you can run flake8 directly:

flake8

If successful, the command will not print anything at all.

Note

In some rare cases it doesn’t make sense to listen to flake8’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 flake8 skip all checks on the line, while the warning code will help other developers lookup what you are ignoring.

Writing documentation

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 some additional dependencies. You can install them from APT with:

sudo apt-get install python-pygraphviz graphviz

To build the documentation, go into the 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 _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.

Working on extensions

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 and flake8, and py.test can be used to run their test suites.

Installing extensions

As always, the mopidy virtualenv should be active when working on extensions:

workon mopidy

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:

cdproject    # or 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-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 pkg_resources:

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 the readme file.

Upgrading extensions

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 search mopidy

This will list all available Mopidy extensions and compare the installed versions with the latest available ones.

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 .

Contribution workflow

Before you being, make sure you’ve read the 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.

Setting up Git remotes

Assuming we already have a local Git clone of the upstream Git repo in ~/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)

Creating a branch

Fetch the latest data from all remotes without affecting your working directory:

git remote update

Now, we are ready to create and checkout a new branch off of the upstream develop branch for our work:

git checkout -b fix/666-crash-on-foo upstream/develop

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 Contributing and elsewhere in the Development environment guide.

Creating a pull request

When everything is done and committed, push the branch to your fork on GitHub:

git push myuser fix/666-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.

Updating a pull request

When the pull request is created, Travis CI will run all tests on it. If something fails, you’ll get notified by email. You might as well just fix the issues right away, as we won’t merge a pull request without a green Travis build. See Running tests on how to run the same tests locally as Travis CI 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/666-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.

Note

Setup the remote as the default push target for your branch:

git branch --set-upstream-to myuser/fix/666-crash-on-foo

Then you can push more commits without specifying the remote:

git push