Test Driven Development in Python

Kevin Dahlhausen
[email protected]

My (pythonic) Background

learned of python in 96 < Vim Editor Fast-Light Toolkit python wrappers PyGallery one of the early online photo gallery generators wxGlade plugin despise HDD Hype-Driven Design

Could there be something behind the hype?

[...] I had been a programmer for nearly thirty years before I was introduced to TDD. I did not think anyone could teach me a low level programming practice that would make a difference. Thirty years is a lot of experience after all. But when I started to use TDD, I was dumbfounded at the effectiveness of the technique. -- Robert Martin

Test Driven Development

design methodology -> artifacts for testing short cycle repeated many times:

write a test watch it fail make it compile make it pass refactor the code refactor the test (and elaborate) rinse and repeat

What does this do for you?

your api is determined by using it written minimal amount of application code

total application + tests is probably more

objects: simpler, stand-alone, minimal dependencies tends to result in extensible architectures instant feedback safety net same feeling as w/scc

Python Makes TDD Easier

typing: dynamic & duck dynamic object changes reflection

flexible testing infrastructures reduced developer overhead

rapid edit-compile-run cycles ease of on-the-fly class definition its python

Tools for test-driven Python

there are many. we will cover:

unittest pymock nose nosy py.test

Grig Gheorghius Python Testing Tools Taxonomy:

http://pycheesecake.org/wiki/PythonTestingToolsTaxonomy tool, category, author, claim to fame

unittest

batteries included since 2.1 PyUnit before that python version of JUnit TestCase

assert_, equality, almost equality, raises exception assertEqual( a, b, msg) setUp / tearDown methods doc strings addTest, run auto-discovery tools vs. maintenance by hand of test-suites

TestSuite

unittest/TDD example

Game needs a Tank Pygame -> Sprite -> image, rect attr start with test:

testTankImage:

create a tank object and assert that tank has image and image is not None

unittest/TDD: first step

TestTank.1.py: import unittest class TestTank(unittest.TestCase): def testTankHasImage(self): tank = Tank() assert tank.image != None, "tank image is None" if __name__=="__main__": unittest.main()

unittest/TDD: first step results

c:/cygwin/bin/sh.exe -c "python TestTank.1.py" E ==================================================== ERROR: testTankHasImage (__main__.TestTank) ---------------------------------------------------Traceback (most recent call last): File "TestTank.1.py", line 6, in testTankHasImage tank = Tank() NameError: global name 'Tank' is not defined ---------------------------------------------------Ran 1 test in 0.000s

unittest/TDD: second step

TestTank.2.py import unittest from Tank import Tank class TestTank(unittest.TestCase): def testTankHasImage(self): tank = Tank() assert tank.image != None, "tank image is None" if __name__=="__main__": unittest.main()

unittest/TDD: second step results

~/: python TestTank.1.py Traceback (most recent call last): File "TestTank.1.py", line 2, in ? from Tank import Tank ImportError: No module named Tank

unittest/TDD: third step

add missing module doing the simplest thing that could possibly work: Tank.py: pass yes, kind of silly a this point, but important for TDD to stick to this model of coding

unittest/TDD: third step results

~: python TestTank.2.py Traceback (most recent call last): File "TestTank.2.py", line 2, in ? from Tank import Tank ImportError: cannot import name Tank

unittest/TDD: fourth step

Tank.py: class Tank: pass Results:
c:/cygwin/bin/sh.exe -c "python TestTank.2.py" E ====================================================== ERROR: testTankHasImage (__main__.TestTank) -----------------------------------------------------Traceback (most recent call last): File "TestTank.2.py", line 8, in testTankHasImage assert tank.image != None, "tank image is None" AttributeError: Tank instance has no attribute 'image' -----------------------------------------------------Ran 1 test in 0.000s

unittest/TDD: fifth step

Tank.py: class Tank: def __init__(self): self.image = pygame.image.load("tank.png") Results:
c:/cygwin/bin/sh.exe -c "python TestTank.2.py" . ---------------------------------------------Ran 1 test in 0.000s OK

unittest/TDD: sixth step

sprites in pygame must also have a rect attribute. so lets add a test for this:
def testTankHasRectangle(self): tank = Tank() assert tank.rect != None, "tank rectangle is None

test fails -> Tank instance has no attribute rect

unittest/TDD: sixth step

doing the simplest thing:

import pygame class Tank: def __init__(self): self.image = pygame.image.load("tank.png") self.rect = None

now the assertion fails:

tank rectangle is None

unittest/TDD: seventh step

so make the test pass:
import pygame class Tank: def __init__(self): self.image = pygame.image.load("tank.png") self.rect = self.image.get_rect() Results:
~: python TestTAnk.3.py .. ---------------------------------------------Ran 2 tests in 0.000s OK

unittest/TDD: eighth step

all tests pass so:

refactor code ensure tests still pass refactor tests:

import unittest from Tank import Tank class TestTank(unittest.TestCase): -> def testTankHasImage(self): tank = Tank() assert tank.image != None, "tank image is None" def testTankHasRectangle(self): tank = Tank() assert tank.rect != None, "tank rectangle is None" __name__=="__main__": unittest.main()

-> if

unittest/TDD: ninth step

factor out duplicated code and place in setUp function:
import unittest import Tank class TestTank(unittest.TestCase): def setUp(self): self.tank = Tank.Tank() def testTankHasImage(self): self.assertNotEqual( self.tank.image, None, "tank does not have an image") def testTankHasRect(self): self.assertNotEqual( self.tank.rect, None, "tank does not have a rectangle") if __name__=="__main__": unittest.main()

unittest/TDD: tenth step

verify that all tests still pass:
c:/cygwin/bin/sh.exe -c "python TestTank.py" .. -------------------------------------------Ran 2 tests in 0.000s OK

PyMock

by Jeff Younker mock-objects:

most objects collaborate with others mock object is a test-oriented replacement (object) for a collaborator javas EasyMock website verifies method calls/parameters (vs. stub)

simulates objects (function calls, properties, functions, and generators) that interact with your object under test and verify the calls record playback verify (demo 2)

PyMock - sample
TestTankDraw.py: import unittest import pymock import pygame from Tank import Tank class TestTankDraw(unittest.TestCase): def testDrawTank(self): tankGroup = pygame.sprite.Group() tank = Tank() controller = pymock.Controller() 1 2 3 surface = controller.mock() surface.blit( tank.image, tank.rect) controller.replay() tankGroup.add(tank) tankGroup.draw(surface) 4 controller.verify()

if __name__=="__main__": unittest.main()

PyMock

can temporarily override parts of existing objects!!

import pymock import os class TestOverride(pymock.PyMockTestCase): def testOverride(self): 1 2 3 self.override(os, 'listdir') # object, method name self.expectAndReturn(os.listdir(), ['file-one', 'file-two']) self.replay() result = os.listdir() self.assertEqual( result, ['file-one', 'file-two'], "wrong files returned") self.verify()

if __name__=="__main__": import unittest unittest.main()

Nose

by Jason Pellerin discovers and runs unit tests

fast tests while gathering no required superclass even stand-alone functions

anything that matches regex is a test

@with_setup( setup, teardown ) decorator

plugins:

tagging and running by tags doctest

Nose simple functions

TestNoseSimpleFunction.py:
import Tank def testDriveLeft(): tank = Tank.Tank(100,100) tank.driveLeft() assert tank.velocity == (-1,0), "tank velocity wrong when driving left"

Nose functions with setup/teardown

simple functions can have setup/teardown:
from nose import with_setup import Tank tank = None def setupFunction(): print "setup" global tank tank = Tank.Tank(100, 100) def teardownFunction(): print "tear down" @with_setup(setupFunction) def testDriveUp(): tank.driveUp() assert tank.velocity == (0,-1), "tank velocity wrong when driving up" @with_setup(setupFunction, teardownFunction) def testDriveDown(): tank.driveDown() assert tank.velocity == (0,1), "tank velocity wrong when driving down"

Nose functions with setup/teardown

results:
~/setupTests: nosetests -v s TestNoseSetup.py TestNoseSetup.testDriveUp ... setup ok TestNoseSetup.testDriveDown ... setup tear down ok

------------------------------------------------------------------Ran 2 tests in 0.234s

Nose

supports test generators

test function yields a function and parameters to apply to that function will automatically call the function for each set returned by the generator

Nose: generator demo

TestNoseGenerator.py:
def test_tankXPositioning(): for x in [ 0, 160, 161]: yield checkTankPosition, x def checkTankPosition(x): tank = Tank.Tank( x, 100) assert tank.rect.left==x

Results:
c:/cygwin/bin/sh.exe -c "nosetests -v" TestNoseGenerator.test_tankXPositioning:(0,) ... ok TestNoseGenerator.test_tankXPositioning:(160,) ... ok TestNoseGenerator.test_tankXPositioning:(161,) ... ok -----------------------------------------------------Ran 3 tests in 0.219s

Nose: code coverage

uses Ned Batchelders Coverage.py

just download and place on path optionally run coverage for tests can check source under tree not imported by tests specify particular package

features:

run: nosetests demo

--with-coverage

Nose: code coverage report

21 ~/proj/superTank2: nosetests --with-coverage ..................................................................... Name Stmts Exec Cover Missing -----------------------------------------------Border 10 10 100% ... JoystickDriver 76 48 63% 25, 37, 52-53, 60, 77-110 KeyboardDriver 28 28 100% MockPyGame 18 18 100% MockSurface 10 10 100% Shot 57 52 91% 16, 69, 72, 78, 81 SuperTank 120 105 87% 35-38, 47-53, 94-97 Tank 93 92 98% 23 TankExplosion 21 20 95% 19 Tree 8 8 100% -----------------------------------------------TOTAL 2985 1260 42% ---------------------------------------------------------------------Ran 94 tests in 0.891s

Nose: test profiling

uses hotshot profiler very configurable requires some studying best practice:

save to file with profile-stats-file post process with python scripts for reporting

example:
nosetests --with-profile --profile-statsfile=stats.dat

Nose: profiling report

with reporting script: from hotshot import stats s = stats.load("stats.dat") s.sort_stats("time").print_stats()

26409 function calls (26158 primitive calls) in 1.997 CPU seconds Ordered by: internal time ncalls tottime percall cumtime percall filename:lineno(function) 1 0.422 0.422 0.651 0.651 c:\python24\lib\sitepackages\numeric\numeric.py:85(?) 1 0.238 0.238 0.896 0.896 c:\python24\lib\site-packages\pygame\__init__.py:25(?) 1 0.159 0.159 0.166 0.166 c:\python24\lib\pickle.py:25(?) 56 0.131 0.002 0.135 0.002 c:\kpd\proj\codemashpresentation\supertank2\src\tank.py:11(__init__) 1 0.095 0.095 1.051 1.051 c:\kpd\proj\codemashpresentation\supertank2\src\testborder.py:1(?)

Nose: profiling limited report

with reporting script:

from hotshot import stats s = stats.load("stats.dat") s.sort_stats("time").print_stats('src.test', .1)

26409 function calls (26158 primitive calls) in 1.997 CPU seconds Ordered by: internal time List reduced from 484 to 164 due to restriction <'src.test'> List reduced from 164 to 16 due to restriction <0.10000000000000001> ncalls tottime percall cumtime percall filename:lineno(function) 1 0.095 0.095 1.051 1.051 c:\kpd\proj\codemashpresentation\supertank2\src\testborder.py:1(?) 3 0.011 0.004 0.018 0.006 c:\kpd\proj\codemashpresentation\supertank2\src\testtankimagecoloring.py:6(verifyTankC oloring) 1 0.007 0.007 0.018 0.018 c:\kpd\proj\codemashpresentation\supertank2\src\testdrawtankexplosion.py:1(?)

Nose debugging support

drop into python debugger

on error

nostests --pdb nosetests pdb-failures

on failure

Nosy

started as blog post by Jeff Winkler automatic continuous testing

similar to Cruise Control run in a window while you work save file -> view unit test output in nosy window mod in comments adds directory recursion

runs nose when changes detected

scans directory for changes in python source code

live demo

py.test

by Holger Krekel (and others) nose is very similar no api model -> test_ funcs, TestXXX class collects tests module, class, and member-function setup/teardown test generators test ordering test for deprecation warnings

py.test.deprecated_call(func, *args, **kwargs)

test selection by keyword -k <prefix>

Case Study - Super Tank in Pygame

history:

late 70s coin-op 1982 TI-99/4a ext. basic 1992 C w/Mode 13 VGA 2007 pygame!

goal: rigorous TDD game development current status:

tanks, shots, borders, key/joystick input, movement, collision detection, explosions 92 tests run in 0.6s 773 lines in 28 files

471 lines game code in 10 files (37%) 813 lines test code in 18 files (63%)

Case-Study: Super Tank Game

first few cycles went like this: 2. test pygame initialization (testInit) 3. test game runs until escape pressed

mocked out frameworks get_event function

4.

test drawing tank on the screen

may be too big a chunk of testing at once drove need for Tank class drove need for Tank to be pygame sprite tested draw by mocking out pygame drawing surface and checking for image blit

Case-Study: Super Tank Game

1. ensure tank drawn from game loop
game loop is hard to unit test so this forced updateScreen function
now could test updateScreen blits tank also test game loop calls updateScreen tests progressed through tank colors, input event registration and callbacks, tank movement, tank direction changes (and respective tank images), border blocks on screen, tank collision with blocks, joystick driver, interior walls, shots, explosions

(Case Study) TDD Generated Classes

(Case Study) Lessons Learned

was able to TDD everything except:

background paint when sprite moves image conversion to current screen format tank movement delay / FPS timing

progress starts off slow then accelerates the resulting app is very extensible

adding joystick tank driver

working with a safety net is *really* nice unit tests (can) run fast

(Case Study) Lessons Learned

mock objects are almost indispensable when dealing with interacting libraries PyMock is amazing ratio of source to test code stayed fairly constant at about 1:2 ( 1:1.9 -> 1:1.7) nosetest / nosy combination is quite handy thinking about how to test for a behavior drives nicer implementations tests first no shortcuts

Resources

Python Testing Tools Taxonomy:

http://pycheesecake.org/wiki/PythonTestingToolsTaxonomy http://docs.python.org/lib/module-unittest.html http://theblobshop.com/pymock/ http://code.google.com/p/python-nose/ code coverage plugin:

unittest module

Pymock:

Nose:

http://www.nedbatchelder.com/code/modules/coverage.html Nosy:

http://jeffwinkler.net/2006/04/27/keeping-your-nose-green/ http://codespeak.net/py/current/doc/test.html

py.test