mpl-qtthread Documentation

Theory of operation

When working with Qt and threads you have to be sure to create QObjects on the thread you are going to use them on and you can only draw to the screen from the main thread. Combined, this means that all of the UI creation and work needs to be done on the main thread. However, background threads are able to influence the UI via Qt’s Signal/Slot mechanism which can be safely used for inter-thread communication.

However, when you use the function is matplotlib.pyplot to make new Matplotlib figures it will create a QtWidget subclass instance for the canvas and a MainWindow instance to put the canvas and toolbar in. Thus, if you try to use pyplot from any Python thread but the main thread you can run into significant issue from Qt (never mind that the rest of Matplotlib is not very thread safe) ranging from windows that never render to crashes.

This works by creating a QtCore.QObject on the main thread (manually done via mpl_qtthread.initialize_qt_teleporter). This object has two signals, one for creating the FigureCanvasBase instance and one for creating the FigureManagerBase that we need. As part of creating this (private) object Slots (aka callbacks) are connected to the Signals that will do the actual work of creating the instances.

These Signals are used by the Matplotlib Backend that the package implements. The main thing than is over-ridden from the upstream behavior is that rather than directly instantiating the FigureCanvas and FigureManager instances, it emits enough information to create them to the Signals. Because the callbacks for a Signal are processed on the thread the QObject is affiliated with (and they are by default affiliated with the thread they are created on) the callbacks to create the QWidgets will run on the main thread. Thus, we are able to create and update Matplotlib figures from a background thread while the QApplication runs in the main thread.

Matplotlib still is not actually thread safe so it is important to only update the Figure from one thread. If you are extensively panning or zooming while the background thread is simultaneously updating Figure, it is very likely that you will suffer race conditions. Many of these issues should be render failures due to a temporarily inconsistent Figure state (e.g. updating the x and y values of a Line2D object to be different lengths). It is possible that this package will need to develop a way to add some locking to Matplotilb.


Python Threads

import threading
import time
import mpl_qtthread.backend
import matplotlib
import matplotlib.backends.backend_qt
import matplotlib.pyplot as plt

# set up the teleporter
# tell Matplotlib to use this backend

# import pyplot and make it interactive

# suppress (now) spurious warnings for mpl3.3+

def background():
    # make a figure and plot some data
    fig, ax = plt.subplots()
    (ln,) = ax.plot(range(5))
    # periodically update the figure
    for j in range(5):
        print(f"starting to block {j}")
        ax.set_title(f"cycle {j}")

# start the thread
# start the QApplication main loop