I’ve been trying to create a simple acceptance test for a pyglet application. A thorough suite of acceptance tests, verifying the correctness of all the shapes drawn to the screen by OpenGL, sounds like far more work than I want to do. But a couple of simple acceptance tests would be valuable, to check out basic things: that the application runs; opens a fullscreen window; responds correctly to some basic inputs and command-line options; and has an acceptable framerate. Especially if I could quickly run this basic smoke test on multiple operating systems.

I tried writing an acceptance test which ran the application-under-test on a new thread. This didn’t work for me *at all* (perhaps because pyglet is not intended to be used with multiple threads), so for the time being I’d given up, and was proceeding without acceptance-level tests.

A couple of days ago I had the idea of a test that didn’t involve threading. Instead, it takes a list of test conditions (as lambdas), and uses pyglet’s own clock and scheduler to request a callback to a test function - try_condition() - on every frame:

    def try_condition(self, dt):
        if self.condition():
            self.time += dt
            if self.time > self.timeout:
                self.fail("timeout on %s" % self.condition)

When try_condition() gets called by pyglet on the first frame, it evaluates the value of self.condition. If True, then that first part of the test has passed, and it gets the next condition off the list. If False, then this function simply returns, to let the application continue running. When try_condition() is called again on the next frame, it resumes where it left off, testing out the same condition again. After it has been trying the same condition for a long enough time, it deems that condition to have failed, and raises an assertion error.

Here is the rest of the class, which sets up the scheduled calls to try_condition().

from unittest import TestCase
from pyglet import app, clock

class AcceptanceTest(TestCase):

    timeout = 1.0

    def set_conditions(self, conditions):
        self.conditions = conditions
        clock.schedule(lambda dt: self.try_condition(dt))

So self.conditions is a list of lambdas that will be provided by the acceptance test. Calling self.next_condition() merely plucks the next condition off the list, into self.condition. If there are no more conditions left, then the test has entirely passed and it requests the application to terminate, by setting the pyglet member window.has_exit to True.

    def next_condition(self):
        if len(self.conditions) > 0:
            self.condition = self.conditions.pop(0)
            self.time = 0.0

    def terminate(self):
        win = self.get_window()
        win.has_exit = True

    def get_window(self):
        windows = self.get_windows()
        if len(windows) == 1:
            return windows[0]
        return None

    def get_windows(self):
        return [w for w in app.windows]

In future, I should probably allow the acceptance test code to specify a different timeout value for each condition.

Anyhow, we can then write an acceptance test by inheriting from this AcceptanceTest class, providing an appropriate list of conditions, and then just calling the application’s main() function. This function won’t return until the application exits, either when one of the conditions times out and raises an assertion error, or else when all conditions have passed and the test framework sets window.has_exit.

from testutils.testcase import run_test
from acceptance_test import AcceptanceTest
from sole_scion import main

class AT001_app_opens_a_fullscreen_window(AcceptanceTest):

    def is_window_visible(self):
        win = self.get_window()
        return win and win.visible

    def is_window_fullscreen(self):
        win = self.get_window()
        return win and win.fullscreen

    def test_window(self):
        conditions = [

if __name__ == "__main__":

The conditions don’t simply have to be expressions to verify the application state. They could stimulate the application by raising mouse or keyboard events, etc, and then simply return True so that the test harness would move right on to the next condition.

Early days with this idea, but it seems to work, and thus far I’m very happy with it.