Test more, cry less!
This script is meant to launch a folder of your tests which will behave as one big test suite. It's done this way because of necessity having a fresh python
interpreter for each Kivy application test to run without mistakes (otherwise mess from previous App().run()
interferes).
Each unittest file in a folder consisting of tests must start with test_
prefix and end with .py
Run from console:
python -m kivyunittest --folder "FOLDER" --pythonpath "FOLDER"
Without --folder
flag the file assumes it's placed into a folder full of tests presumably as __init__.py
. It makes a list of files, filters everything not starting with test_
and ending with .py
and runs each test.
Flag --pythonpath
appends a folder to the sys.path
automatically, therefore it's not necessary to include it in each test manually.
If there is an error of whatever kind that unittest recognizes as failure, KivyUnitTest will save the name of the test and its log. When the testing ends all error logs are put together into console divided by pretty headers with test's name.
When the Kivy application starts, it creates a loop and until the loop is there, nothing will execute after App().run()
line. That's why we need to probe the loop.
This can be achieved by a simple time.sleep()
as you've surely noticed sooner when trying to pause the app for a while. That's exactly what a custom unittest for Kivy does - pauses the main loop as much as possible as a scheduled interval and executes the testing run_test
function.
Example:
import unittest
import os
import sys
import time
import os.path as op
from functools import partial
from kivy.clock import Clock
First we need to set up importing of the application set main_path
to be the folder of main.py
e.g. when you have tests in <app dir>/tests/test_example.py
. Or choose the folder with the --pythonpath
flag.
main_path = op.dirname(op.dirname(op.abspath(__file__)))
sys.path.append(main_path)
Import your main class that inherits from App (class My(App):
) or even additional stuff that's not connected with App class or its children.
from main import My
class Test(unittest.TestCase):
# sleep function that catches ``dt`` from Clock
def pause(*args):
time.sleep(0.000001)
# main test function
def run_test(self, app, *args):
Clock.schedule_interval(self.pause, 0.000001)
# Do something
# Comment out if you are editing the test, it'll leave the
# Window opened.
app.stop()
Create an instance of your application, put it as a parameter into partial (so that you could access it later), schedule main function with Clock and launch the application (working Window will appear).
# same named function as the filename(!)
def test_example(self):
app = My()
p = partial(self.run_test, app)
Clock.schedule_once(p, 0.000001)
app.run()
- if __name__ == '__main__':
unittest.main()
This kind of tests is used directly for testing in the Kivy core and might not be easy and/or suitable enough for your needs, however it brings a way extended control over the testing environment. In this test you can render each frame manually, move Clock
each tick on your own and dispatch raw input from (mocked) providers through MotionEvent
.
There is a class with all necessary stuff prepared in the background, so that it launches a Kivy window, but waits for you to move it further e.g. if you decide to render()
a widget.
from kivy.tests.common import GraphicUnitTest
from kivy.input.motionevent import MotionEvent
from kivy.graphics import Color, Point
from kivy.uix.widget import Widget
from kivy.base import EventLoop
from math import sqrt
After you import MotionEvent
, you can create own class that inherits from it and use it later as a mocked input. We will use sx
and sy
which are just positions on X and Y axis in 0 - 1 range (percents, if you will). This class will dispatch a touch
.
class UTMotionEvent(MotionEvent):
def depack(self, args):
self.is_touch = True
self.sx = args['sx']
self.sy = args['sy']
self.profile = ['pos']
super(UTMotionEvent, self).depack(args)
If we know how to assemble a class to create a touch input, we might draw something with it as well. Kivy includes a nice demo, Touchtracer, for showcasing multitouch. We fetch calculate_points
from that example. It basically returns a new set of points we'll input to a drawing function.
# taken from Kivy's Touchtracer
def calculate_points(x1, y1, x2, y2):
dx = x2 - x1
dy = y2 - y1
dist = sqrt(dx * dx + dy * dy)
o = []
m = dist
for i in range(1, int(m)):
mi = i / m
o.extend([
x1 + dx * mi,
y1 + dy * mi
])
return o
For drawing we'll use a very similar thing to the one used in the Touchtracer. Let's draw a Point
on on_touch_down
event. Then, if we move the touch append new points along the line between an old and a new point and draw them.
# core taken from Kivy's Touchtracer
class WidgetCanvasDraw(Widget):
def on_touch_down(self, touch):
win = self.get_parent_window()
ud = touch.ud
with self.canvas:
Color(1, 0, 0, 1)
ud['lines'] = Point(points=(
touch.x, touch.y
))
touch.grab(self)
return True
def on_touch_move(self, touch):
if touch.grab_current is not self:
return
ud = touch.ud
points = ud['lines'].points
oldx, oldy = points[-2], points[-1]
points = calculate_points(oldx, oldy, touch.x, touch.y)
if not points:
return
add_point = ud['lines'].add_point
for idx in range(0, len(points), 2):
add_point(
points[idx],
points[idx + 1]
)
def on_touch_up(self, touch):
if touch.grab_current is not self:
return
touch.ungrab(self)
We have input, drawing behavior, let's set up a test. You might want to get used to this "template" if you intend to use the GraphicsUnitTest
class. It's not that scary though. Set a class attribute framecount
to zero, prepare some debugging behavior (setUp
prepares a new Window, tearDown
purges it). After overriding them with empty functions, such actions won't happen.
class WidgetDrawTestCase(GraphicUnitTest):
framecount = 0
# debug test with / stop destroying window
# def tearDown(self, *_): pass
# def setUp(self, *_): pass
We make sure the Window is available to us with EventLoop
, prepare all out widgets and then call EventLoop.idle()
which makes a lot of internals ready for an application to show like you are used to it. More or less.
def test_touch_draw(self):
# get Window instance for creating visible
# widget tree and for calculating coordinates
EventLoop.ensure_window()
win = EventLoop.window
# add widget for testing
child = WidgetCanvasDraw()
win.add_widget(child)
# get widgets ready
EventLoop.idle()
You can happily start testing now.
The little bit problematic part comes now, because you have to be sure where you want your touch to go and do it in 0 - 1 range, so that the test works even after Window resizing. Absolute values are not the way you want to go. Always try to generalise the movement and find a way how to simplify them into a small list.
# default "cursor" position in the middle
pos = [win.width / 2.0, win.height / 2.0]
# default pos, new pos
points = [
[pos[0] - 5, pos[1], pos[0] + 5, pos[1]],
[pos[0], pos[1] - 5, pos[0], pos[1] + 5]
]
# general behavior for touch+move+release
for i, point in enumerate(points):
x, y, nx, ny = point
# create custom MotionEvent (touch) instance
touch = UTMotionEvent("unittest", 1, {
"sx": x / float(win.width),
"sy": y / float(win.height),
})
The points and touch are ready. Let's dispatch the input in the test. For that we use EventLoop
again and its method post_dispatch_input(event_type, motion_event)
.
- touch down with
begin
event type - touch move with
update
event type - touch up with
end
event type
# dispatch the MotionEvent in EventLoop as
# touch/press/click, see Profiles for more info:
# https://kivy.org/docs/api-kivy.input.motionevent.html#profiles
EventLoop.post_dispatch_input("begin", touch)
# the touch is dispatched and has ud['lines']
# available from on_touch_down
self.assertIn('lines', touch.ud)
self.assertTrue(isinstance(touch.ud['lines'], Point))
# move touch from current to the new position
touch.move({
"sx": nx / float(win.width),
"sy": ny / float(win.height)
})
# update the MotionEvent in EventLoop
EventLoop.post_dispatch_input("update", touch)
# release the MotionEvent in EventLoop
EventLoop.post_dispatch_input("end", touch)
# still available, but released
self.assertIn('lines', touch.ud)
self.assertTrue(isinstance(touch.ud['lines'], Point))
expected_points = [[
x + 0, y, x + 1, y,
x + 2, y, x + 3, y,
x + 4, y, x + 5, y,
x + 6, y, x + 7, y,
x + 8, y, x + 9, y
], [
x, y + 0, x, y + 1,
x, y + 2, x, y + 3,
x, y + 4, x, y + 5,
x, y + 6, x, y + 7,
x, y + 8, x, y + 9
]]
# check if the instruction points == expected ones
self.assertEqual(
touch.ud['lines'].points,
expected_points[i]
)
The less obvious part comes now, because we need to trigger the rendering of our graphics in the application. Fortunately that's easy to do with simple GraphicUnitTest.render()
. You most likely want to put there the root widget like when building an application with App.build()
method.
# render the graphics
self.render(child)
It's quite useful to add unittest.main()
at the end of your test, because if you only try to write a single test then you most likely don't want to run the whole suite. Especially if the suite is large.
if __name__ == '__main__':
import unittest
unittest.main()
Handle class communication through App class via App.get_running_app()
in your application, put every needed widget inside App class like this:
class MyButton(Button):
def __init__(self, **kwargs):
super(<class name>, self).__init__(**kwargs)
self.text = 'Hello Test'
app = App.get_running_app()
app.my_button = self
and then access your widgets in test's run_test()
function via app
parameter like this:
self.assertEqual('Hello Test', app.my_button.text)
Use app.root
to get instance of a class you pass in the build()
function in the App class.
Dispatch events through widgets e.g. <widget>.dispatch('on_release')
to execute function bound to on_release
.
Use Kivy's Inspector module_ as help to navigate down the path of App class and use ids
in kv language
, it'll make targeting a specific widget easier.
Try even Kivy's Recorder module_ to record steps and play them later instead of dispatching events manually. However, this way is heavy time-consuming as it plays the steps exactly as long as they were recorded.
Example:
from kivy.input.recorder import Recorder
# place this inside ``run_test()``
rec = Recorder(filename='myrecorder.kvi')
rec.bind(on_stop=<function>)
rec.play = True
This will play all steps and then executes a function bound to on_stop
. May be useful for testing touch gestures, swipes, dragging and other rather annoying to write manually stuff.
There's also possibility to change time the steps were recorded in in .kvi
file (that long number), which will speed things up.
Also, there's a very interesting Python package made by Mathieu Virbel that allows you to go down the widget tree rabit hole in a more sane way than using this:
my_widget.children[0].children[1].children[2]...
which gets tedious and annoying the more you use it when you navigate the tree from the application's root widget itself through complex layouts. This is where Telenium might save you a lot of minutes instead of typing the same thing over and over.
The MIT License (MIT)