LMMS Architecture - LMMS/lmms GitHub Wiki

Introduction

There should be some documentation about LMMS' core architecture and concepts to help newcomers with some development skills to understand how it works.

This page is an attempt to aggregate the informations found.

Model and ModelView

LMMS uses a design pattern similar to Model–view–controller. In LMMS, most of views also play a role as controllers. Model emits Model::dataChanged signal when its data is changed and Model::propertiesChanged when its property(range, step size, etc.). These signals are used for updating corresponding views(see ModelView::doConnections) and tracking changes in models.

ModelView holds reference to its model. It doesn't own the model except the model is default-constructed. Existing ModelView instance can be reused by setting new model using ModelView::setModel. Views handle model change by overriding modelChanged().

BB tracks

A list of clips by calling Track::getClips for beat/bassline(BB) tracks(type Track::BBTrack) are actually blocks within the song editor, not within the BB editor.

All tracks in BB editor are located within a separate track container BBTrackContainer. You can access the container by Engine::getBBTrackContainer(). Playing a certain BB track (i.e. all clips of the tracks inside at a certain position) is achieved via BBTrack::play(...).

Tracks in BB editor has one clip per each beat/bassline. The index of BB track corresponds to the index of clip for parent BB track in the track.

Audio rendering

LMMS has a class for audio rendering, named AudioEngine (previously called Mixer). AudioEngine::renderNextBuffer contains the main functionality of AudioEngine. Major works that the function does:

  1. Remove all finished PlayHandles, which will be explained
  2. Call Song::processNextBuffer to play songs and add PlayHandles
  3. Queue jobs for AudioEngineWorkerThread and wait for worker threads (PlayHandle, effects in instrument tracks/sample tracks, FX mixing)
  4. Call runChangesInModel to process pending non-automated changes

Audio engine worker threads

AudioEngineWorkerThread (previously called MixerWorkerThread) processes ThreadableJobs queued in AudioEngine::renderNextBuffer. There are three kinds of jobs that AudioEngineWorkerThread runs:

  • AudioPort: Processes intermediate audio signals(Currently used for audio effects for tracks)
  • MixerChannel: Processes audio effects per Mixer channel (previously called FxChannel)
  • PlayHandle: Plays notes/instruments/audio samples

Synchronization with the audio engine

One goal is that the audio engine runs normally without using locks. The audio engine will lock at an appropriate moment when changes to the song are requested, such as removing tracks and changing sample files, avoiding the use of freed data.

When a non-automated change to the song is needed, instead of calling lock() and unlock() on a mutex, the audio engine functions requestChangeInModel() and doneChangeInModel() are used. These functions synchronize GUI threads to do changes when the audio engine deems appropriate. The functions do nothing if they are called from the main thread of the audio engine. If they were called from a worker thread, that would be a design error.

More information is available at GitBook: https://docs.lmms.io/developer-guides/core/