Waiting for Signals in PySide and PyQt

Posted on July 3, 2014

Recently, I was writing an integration test for a data visualization GUI. The GUI calls out to an external process to run a simulation, waits for the simulator to produce data, and reads the data back in to produce pretty graphs, export data to other formats, etc. I use a PySide signal to inform the GUI that the simulator has produced data. Unfortunately, I couldn’t find a clean way to have my integration test wait for that signal. Luckily, I have found the solution, and I’ll explain it here.

Signals and Slots

PySide and PyQt are Python bindings to the Qt GUI and application framework. One killer feature of Qt is the signal & slot system, which is a way for widgets and objects to communicate events to one another.

An object in Qt can send a signal to other subscribed objects. Signals are used to inform other objects that an event has occurred. For example, buttons fire the QPushButton.pressed signal when they are clicked. When you want to handle a button click, you connect the signal to a function. You can create your own signals, and connect them to arbitrary python functions. In my GUI, I have something like the following:

class GUI:

    def __init__(self):
        self.simulator = Simulator()
        self.data_reader = DataReader()
        self.simulator.finished.connect(self.data_reader.read_data)

        self.startButton = QtGui.QPushButton("Start")
        self.startButton.pressed.connect(self.simulator.run_simulation)

class Simulator(QtCore.QObject):

    """Controls simulation and emits a ``finished`` signal when done."""

    finished = QtCore.Signal()

    def run_simulation(self):
        """Run the simulator in an external process."""
        self.process = QtCore.QProcess()
        self.process.finished.connect(self.finished)
        self.process.start(simulator_exe)
    ...

class DataReader:

    """Reads data files from simulator."""

    def read_data(self):
        load_files()
    ...

In this code snippet, we create the Simulator and DataReader classes, and connect them together using a signal. The Simulator class is a QObject, so it can send signals. It runs a simulation in a separate process, and then emits the finished signal when it’s done. We also happen to use a button that starts the simulation when clicked. We connect the simulator’s finished signal to the DataReader so it knows when to begin reading data.

The Test

So, how do I test the above code? I want to run the simulator and make sure it has the correct output. I initially wrote a test like this:

def test_simulator_output():
    simulator = Simulator()
    simulator.run_simulation()

    assert_correct_output(simulator)

What’s wrong here? The assertion line is executed too early; the simulator probably hasn’t even started in the nanoseconds between when I told it to start and when I check the data! How can we solve this issue? I tried a time.sleep() loop that polls if the simulation is complete. This blocks the main GUI thread, and is not a good solution. I also tried similar things using the threading module, but to no avail. It turns out, we can use a new Qt event loop to stop execution.

Solution

I figured out that you can call QEventLoop.exec_() to create a nested event loop. That is, our QApplication instance event loop is already running, but calling QEventLoop.exec_() will stop execution using a new event loop until QEventLoop.quit() is called. Here is our modified example test:

def test_simulator_output():
    simulator = Simulator()

    loop = QtCore.QEventLoop()
    simulator.finished.connect(loop.quit)
    simulator.run_simulation()
    loop.exec_()  # Execution stops here until finished called

    assert_correct_output(simulator)

We create an event loop that will help us wait for our simulation to complete. When run_simulation() is called, we wait at the loop.exec_() line until finished is emitted. When the simulator finishes, the event loop will exit, and we will advance beyond the loop.exec_() line.

General Solution

There is one problem with this approach. What if simulator.finished is never emitted, like when an error occurs? We would never advance beyond loop.exec_(). If we had this test running on a continuous integration server, we would lock up a job forever until we realized the test never finished!

A solution is to use the QTimer.singleShot() function with a timeout. That means for every time we want to create an event loop, we have to set up the loop, hook up our signals, and create a timer with a suitable timeout in case the signals never fire. Here is a context manager that can handle this for us:

@contextmanager
def wait_signal(signal, timeout=10000):
    """Block loop until signal emitted, or timeout (ms) elapses."""
    loop = QtCore.QEventLoop()
    signal.connect(loop.quit)

    yield

    if timeout is not None:
        QtCore.QTimer.singleShot(timeout, loop.quit)
    loop.exec_()

We can use it like this:

def test_simulator_output():
    simulator = Simulator()

    with wait_signal(simulator.finished, timeout=10000):
        simulator.run_simulation()

    assert_correct_output(simulator)

Isn’t that great? I think so.

pytest-qt

I run my tests using the pytest library. There is a plugin for pytest called pytest-qt, which comes with a fixture called qtbot. A qtbot is used to handle all of the boilerplate that comes along with testing PySide/PyQt code, like:

Once I figured out how to stop test execution to wait for a signal, I created a pull request to add this functionality to pytest-qt. I created a more general solution that allows multiple signals, or none at all (just wait for a timeout). If you use PySide/PyQt, but you think testing your GUI code is a pain, check out pytest-qt! Take a look at the example in the docs to see how to use pytest-qt to block tests for signals.

Thanks for reading!

comments powered by Disqus