10 years of Mopidy

Ten years ago today, on December 23, 2009, Mopidy was born. While chatting with my friend and then-colleague Johannes Knutsen, we came up with the idea of building an MPD server that could play music from Spotify instead of local files.

This is the story of the first decade of Mopidy.

After a brief discussion of how it could work and what we could build upon, Johannes came up with the name “Mopidy.” The name is, maybe quite obviously, a combination of the consonants from “MPD” combined with the vowels from “Spotify.” At the same time, the name is different enough from both of its origins not to be mixed up with them. Even during the first few hours we had some thoughts about maybe adding file playback and support for other backends in the future. Thus we quickly appreciated that the “Mopidy” name would still work, even if Spotify wasn’t always the sole focus of the project.

Within a couple of hours we had a Git repo with some plans written up. We joined the #mopidy IRC channel on Freenode and we had recruited Thomas Adamcik to the project. Over the next few years, he designed many of our most essential components, including the extension system. Today, ten years later, Thomas is still involved with Mopidy and many of its extensions.

After a couple of days, it worked! We had built a primitive MPD server in Python that at least worked with the Sonata MPD client. On the backend side, we used the reverse-engineered “despotify” library to interface with Spotify as it already had some Python bindings available. For all three of us, coming mostly from web development and Django, I believe we already had a feeling of achievement and expanding horizons. If we could pull this off, we could build anything.

The story of Mopidy is a story of thousands of small iterative improvements that, over time, add up to something far greater than the sum of its parts. It was a hack, but a hack with good test coverage from the very start, making changes and iteration safe and joyful.

In March 2010 we released our first alpha release. Over the next decade this would become the first of 74 releases, not counting the numerous releases of the extensions we later extracted from Mopidy or built from the ground up.

Later in 2010, as we added an alternative Spotify backend using the official libspotify library, we started seeing the first traces of our current system architecture. At that time, you switched between the single active backend by manually changing the config file and restarting Mopidy. Support for multiple active backends wouldn’t come until much later, after we had built the muxing “core layer” in-between the backend and frontend layers. However, multiple frontends were supported from the start. The MPD server was the first frontend, and during the first year, we added a second one with the Last.fm scrobbler.

By the end of 2010 we had made GStreamer a requirement and thrown out the support for direct audio output to OSS and ALSA. In 2011 we built upon the great power of GStreamer to support multiple outputs; this allowed us to play audio locally whilst also streaming it to a Shoutcast/Icecast server at the same time. This support for multiple outputs survived for about a year before we removed it. Instead, we exposed the GStreamer pipeline configuration directly, just like we still do today with the audio/output configuration. Streaming to Icecast is also still possible but is involved enough to require specific documentation for the setup. However, by exposing the GStreamer pipeline directly, we didn’t have to guess what kind of installations people would use Mopidy in and exposed the full power of GStreamer to our end-users.

We added support for Python 2.7, but we definitely didn’t plan to stay with it for eight years. During the winter of 2010/2011, I designed and built the Pykka actor library based on the concurrency patterns we had established in Mopidy. When we started using Pykka in Mopidy in March 2011, it already supported Python 3.

Towards the end of the year, we added support for the Ubuntu Sound Menu and the MPRIS D-Bus specification in our third frontend.

Most of 2012 went by without much happening other than a few maintenance releases. However, in November, we released the almost revolutionary Mopidy 0.9 after finally building out the muxing core layer in-between frontends and backends. Depending on the type of request from a frontend, the core layer would either forward the request to the correct backend or, e.g., in the case of search, fan out the request to all backends and then merge the returned search results before passing the result back to the frontend. We had accomplished one of our original goals from the very first day of development: we had a music server that could play music from both Spotify and local files.

Less than a month later, the wheel turned again and we released Mopidy 0.10 with the HTTP frontend. This exposed the full Core API using JSON-RPC over a WebSocket. With this it was suddenly possible to build clients for Mopidy directly, instead of going through MPD.

In a lazy and ingenious moment, we decided that we had no interest in manually keeping the Core API and the JSON-RPC API in sync for the indefinite future. Thus, we based the JSON-RPC API on introspection of the Core API and included a JSON-RPC endpoint which returned a data structure that described the full API. On top of this, I built the Mopidy.js library and released it together with Mopidy 0.10. Mopidy.js uses the API description data structure to dynamically build a mirror of our Core API in JavaScript, working both on Node.js and in the browser.

Even as the JSON-RPC implementation and the Mopidy.js library became the foundation for several popular Mopidy web clients over the next few years, no bug was ever reported that originated in this library. To this day, I testament this to two things: proper test-driven development and excellent code review by Thomas, making these few weeks in November-December 2012 one of the highlights of my years as an open source maintainer.

Jumping just a few months ahead to Easter 2013, the next revolution was about to happen. In a single long and intense day, Thomas and I hashed out and implemented Mopidy’s extension support. Up to this point, there was just one Mopidy. Today, a search for “mopidy” on PyPI returns 127 results.

The Stream backend was created early in 2013, and later in the year, it learned how to parse several playlist formats to find the streamable URL they contained. This made it easy to build backend extensions for music services that exposed playable URIs, like SoundCloud, Google Music, and thousands of radio stations. The backends only had to find and present the playable streams as a playlist or as a virtual file hierarchy; the heavy-lifting of actually playing the audio could be fully delegated to Mopidy and its Stream backend.

By the end of 2013 we had performed the first round of shrinking Mopidy’s core. The Spotify support, Last.fm scrobbler and MPRIS server were all extracted to new extensions living outside the core project. I believe that pulling extensions out of core has helped reduce the amount one must juggle in one’s head to effectively develop on Mopidy.

In 2013 we eased the on-ramping for new users by automatically creating an initial configuration file on the first run. Mopidy also got support for announcing its servers through Zeroconf so they could be autodetected by mobile apps, like Mopidy-Mobile.

Elsewhere in 2013, Wouter van Wijk built the first iterations of the Pi MusicBox distribution for Raspberry Pi. Pi MusicBox provided a turn-key jukebox setup built on Mopidy. This made Mopidy more approachable for the masses that didn’t know Mopidy, Python, or even Linux; allowing them to create their own hi-fi setups.

The next year, in 2014, Fon launched a Kickstarter campaign to build a “modern cloud jukebox” named Gramofon. It turned out that they based their prototype on Mopidy and Javier Domingo Cansino from their development team started submitting patches and becoming active in Mopidy development.

In the summer of 2014 we had our first real-life development sprint at EuroPython in Berlin. Javier and I were joined by several newcomers that got up and running with Mopidy development and squashed a few bugs.

Mopidy 0.19.5 was released on Mopidy’s fifth anniversary in December 2014. A few months later, we released Mopidy 1.0. The release of Mopidy 1.0 did not mark a breaking change, but rather the decision that we could commit to the current APIs for a while and bring stability to the extension ecosystem.

In the summer of 2015, Thomas joined us as we had our second development sprint at EuroPython in Bilbao, Javier’s hometown.

The year between the sprints of 2014 and 2015, and the 0.19, 1.0, and 1.1 releases, were quite significant when looking back at the project’s history. They were not as notable in features as when we added multi-backend, the HTTP API, and extension support back in 2012/2013. Still, they were significant in that the project garnered lots of interest. Up to twenty different people contributed code to each of these three releases.

More than five years ago, in July 2014, I opened Mopidy’s issue #779 for tracking the port to Python 3. There were three large buckets of work that had to be completed. First, our libspotify Python bindings had no Python 3 support. Second, we needed to upgrade to GStreamer 1. Finally, we had to port Mopidy and all of its extensions.

Over the next year, I rewrote pyspotify from scratch using CFFI. Just a couple of weeks before I shipped the final version, Spotify deprecated the libspotify API. However, since they’ve never provided a replacement API for audio playback, we’re still using pyspotify 2 and libspotify to play music from Spotify more than four years later.

During the autumn of 2015 I ported Mopidy from GStreamer 0.10 to 1.x. GStreamer 0.10 was quickly being deprecated, so this work was necessary solely for Mopidy to continue being packaged in Debian and Ubuntu. It was a nice bonus that the new PyGObject wrapper also supported Python 3. With the release of Mopidy 2.0 in February 2016, the move to GStreamer 1 was complete, and thanks to Thomas, we finally supported gapless playback.

Shortly after the 2.0 release, life caught up with several of the most active contributors. A mix of more kids and more work threw Mopidy into a three-year-long period with lower activity and almost exclusively maintenance releases. The only significant development of Mopidy during this period was the support for persisting playback and tracklist state across restarts, contributed by Jens Lütjen, and released in 2.1 in 2017.

Jumping ahead to 2019. Five years after I wrote the tracking issue for moving Mopidy to Python 3, the world looks quite different. Python 2.7’s announced end-of-life is looming at the end of the year, and Python 3 is the standard for all new projects.

So, finally, in the middle of October, we got started on the third and final step toward Python 3.

Once we had a small part of the test suite running on both Python 2 and 3, it took Nick Steel and myself about three weeks of porting modules one by one until suddenly the full test suite was running. Once Mopidy without extensions ran well on Python 3, we axed all Python 2 support, cleaned up all hacks left over from the porting process, and reformatted the code with Black. All of a sudden Mopidy felt like a modern Python codebase.

The six weeks since then have mostly been spent on extensions.

We’ve built a new Mopidy extension registry. We believe that the new registry will ease the discovery of extensions in general. Short-term, we also hope it will help users navigate the extension ecosystem while it is temporarily split in two between Python 2 and 3.

All extensions in the Mopidy GitHub organization are now running on Python 3, as well as a few popular extensions elsewhere. In total, we have almost 20 extensions compatible with Mopidy 3.0 on the day of the release.

Some extensions have also recently received some extra tender loving care.

The Mopidy-Local extension was pushed out of core early in the Python 3 porting. After becoming an independent extension, Thomas Kemmer’s excellent Mopidy-Local-SQLite and Mopidy-Local-Images were merged into Mopidy-Local. We now have a single comprehensive extension for using Mopidy with pre-indexed local music collections.

Next, Nick did a great job fixing up the Mopidy-Spotify extension. Since Spotify suddently broke the playlist part of libspotify a while ago, Mopidy-Spotify has been without functional playlist support. It now uses the Spotify Web API for everything related to playlists and Nick has continued adding several new features using the Web API. Some are shipped in the release that went out together with Mopidy 3.0, and more are right around the corner. Once complete, Mopidy-Spotify will support everything Mopidy-Spotify-Web provides, once again leaving us with a single comprehensive extension for using Mopidy with Spotify.

Just a few days ago, the Mopidy-MPD frontend was pushed out to an extension too. With this, we’ve come full circle from Mopidy being named after “MPD” and “Spotify.” As of Mopidy 3.0, both Mopidy-MPD and Mopidy-Spotify are independent extensions, and Mopidy core knows nothing of either. I wasn’t entirely sure whether moving the MPD server would bring any benefits. Still, once the move was complete, it was evident that Mopidy-MPD deserves to be a project by itself. The split reduced the amount of code in Mopidy by 25% and cut the test suite run time in half. Hopefully this will also make it easier for newcomers to start contributing to either Mopidy-MPD or Mopidy itself.

Finally, December 22, the day before the Mopidy project’s 10th anniversary, we published Mopidy 3.0.

Together with Mopidy 3.0, we uploaded ten updated extensions to PyPI, with at least six more having compatible pre-releases, and, hopefully, final releases over the next few days. For an up to date overview of what’s ready for Python 3 right now, the extension registry is the place to look.

It’s still early days for Mopidy 3.0. Our Homebrew tap is up to date with the new releases, and parts of Arch Linux are already up to date. Updated Debian packages, both at apt.mopidy.com and in Debian, are still some days away. Nonetheless, Mopidy 3.0 will definitively be a part of Ubuntu 20.04 LTS come spring.

Now, go forth and update your extensions to work with Python 3.7+ and Mopidy 3.0.

In a week, Python 2.7 reaches end-of-life. Mopidy, however, is ready for the next decade.

Thanks to Nick Steel for reviewing this blog post.




Stein Magnus Jodal