Testing

System Development with Python

Git repository:

https://github.com/UWPCE-PythonCert/SystemDevelopment

What is testing?

Code which runs your application in as close to a real environment as feasible and validates its behavior

Terminology of testing

  • Unit tests
  • Integration tests
  • High level system tests
  • Acceptance tests
  • Black box / White box testing

“V” model and tests levels

_images/test_v_model.png

Unit testing

  • Test smallest discrete units of source code
  • Tests should be independent of each other
  • Can separate tests from required resources through fixtures and mocking
  • Automatable
  • Integrates with development process

What should be tested?

The percentage of code which gets run in a test is known as the coverage.

100% coverage is an ideal to strive for. But the decision on when and what to test should take into account the volatility of the project.

Unit-testing tools

About Unit-testing

  1. Tests should be independent.
  2. Tests do not run in order, which shouldn’t matter, see point 1.
  3. Test fixtures are available to do any setup/teardown needed for tests.
  4. Test behavior not implementation
  5. Mocking is available to fake stuff you may not want to run in your tests.

unittest.TestCase anatomy

  • create a new subclass of unittest.TestCase
  • name test methods test_foo so the test runner finds them
  • make calls to the self.assert* family of methods to validate results
import unittest
class TestTest(unittest.TestCase):

    def setUp(self):
        self.x = 2

    def test_add(self):
        self.assertEqual(self.x+2, 4)

    def test_len(self):
        self.assertEqual(len('foo'), 3)

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

Assert Methods

TestCase contains a number of methods named assert* which can be used for validation, here are a few common ones:

assertEqual(first, second, msg=None)
assertNotEqual(first, second, msg=None)
assertTrue(expr, msg=None)
assertFalse(expr, msg=None)
assertIn(first, second)
assertRaises(exc, fun, msg=None, \*args, \*\*kwargs)

See a full list at:

http://docs.python.org/3/library/unittest.html#assert-methods or

dir(unittest.TestCase) or to get really fancy

[print(i) for i in dir(unittest.TestCase) if i.startswith('assert')]

Fixtures: Setting up your tests for success

(or failure!)

Test fixtures are a fixed baseline for tests to run from consistently, also known as test context

Fixtures can be set up fresh before each test, once before each test case, or before an entire test suite

unittest provides fixture support via these methods:

  • setUp / tearDown - these are run before and after each test method
  • setUpClass / tearDownClass - these are run before/after each TestCase
  • setUpModule / tearDownModule - run before/after each TestSuite
  • (new in Python 2.7) addCleanup / doCleanups - called after tearDown, in case a test throws an exception

Testing floating point values

Why can’t we just test if .5 == .5 ?

In [1]: 3 * .15 == .45
Out[1]: False

In [2]: 3 * .15
Out[2]: 0.44999999999999996

In [3]: 3 * .15 * 10 / 10  == .45
Out[3]: True

There are an infinite number of floating point numbers, so they are stored as an approximation in computing hardware.

https://docs.python.org/3/tutorial/floatingpoint.html

levels of precision of floating point

Floating point numbers are stored in IEEE 754 64-bit double precision format, so 1 bit for the sign, 11 bits for the exponent, and the remaining 52 for the fraction

So we can count on up to 16 digits of precision in decimal:

In [39]: len(str(2**52))
Out[39]: 16

In [40]: .1+.2
Out[40]: 0.30000000000000004

In [41]: len('3000000000000000')
Out[41]: 16

# with repeated operations, the errors eventually build up:
# here's multiplying by '1' 10 million times:
In [64]: x=1
In [69]: for i in range(10000000): x *= (.1 + .2)/.3
Out [69]: 1.000000002220446

assertAlmostEqual

Verifies that two floating point values are close enough to each other. Add a places keyword argument to specify the number of significant digits.

import unittest

class TestAlmostEqual(unittest.TestCase):

    def setUp(self):
        pass

    def test_floating_point(self):
        self.assertEqual(3*.15, .45)

    def test_almost_equal(self):
        self.assertAlmostEqual(3*.15, .45, places=7)

What is close?

Warning

assertAlmostEqual lets you specify decimal places, i.e. the number of digits after the decimal point.

This works great for numbers that are about magnitude 1.0 (as above)

But what if you have numbers that are very large? (or small):

  • 1.0e22
  • 1.0000000000001e22

are they almost equal?

Remember that python floating point numbers store the exponent and up to 16 decimal digits.

So those two are almost as close as you can get. But:

In [30]: x = 1e22

In [31]: y = 1.0000000000001e22

In [32]: '%g'%(y - x)
Out[32]: '1.00034e+09'

They are different by about a billion!

In general, we don’t want to compare floating point numbers to within a certain number of decimal places.

Anyone remember “significant figures” from science classes?

isclose()

Python 3.5 introduced the isclose() function in the math module:

https://www.python.org/dev/peps/pep-0485/

In [39]: import math

In [40]: x
Out[40]: 1e+22

In [41]: y
Out[41]: 1.0000000000001e+22

In [42]: math.isclose(x,y)
Out[42]: True

So this works for any magnitude number.

is_close(a, b, *, rel_tol=1e-09, abs_tol=0.0) -> bool

Determine whether two floating point numbers are close in value.

   rel_tol
       maximum difference for being considered "close", relative to the
       magnitude of the input values
    abs_tol
       maximum difference for being considered "close", regardless of the
       magnitude of the input values

Return True if a is close in value to b, and False otherwise.

rel_tol essentially specifies how many significant figures you want: 1e-09 is 9 significant figures: about half of what floats can store.

abs_tol is required for comparisons to zero – nothing is “relatively close” to zero

Using isclose() with unittest

Ideally, TestCase would have an assertIsClose method. But you can use:

import unittest
from math import isclose

class TestAlmostEqual(unittest.TestCase):

    def test_floating_point(self):
        self.assertEqual(3*.15, .45)

    def test_almost_equal(self):
        self.assertTrue( isclose( 3*.15, .45, rel_tol=7) )

Running your tests

How do you actually run your tests?

running tests in a single module

Call unittest.main() right in your module

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

# or from the command line:
python -m unittest test_my_module  # with or without .py on end
python -m unittest test_my_module.TestClass  # particular class in a module
python -m unittest test_my_module.TestClass.test_method  # particular test

If it gets cumbersome with many TestCases, organize the tests into a test suite

Test Suites

Test suites group test cases into a single testable unit

import unittest

from calculator_test import TestCalculatorFunctions

suite = unittest.TestLoader().loadTestsFromTestCase(TestCalculatorFunctions)

unittest.TextTestRunner(verbosity=2).run(suite)

Tests can also be organized into suites in the

if __name__ == "__main__":

block

nose2

Nose2 is the new nose. Nose is barely being maintained, and directs users to nose2.

A test runner which autodiscovers test cases

Nose2 will find tests for you so you can focus on writing tests, not maintaining test suites

To find tests, nose2 looks for modules (such as python files) whose names start with ‘test’. In those modules, nose2 will load tests from all unittest.TestCase subclasses, as well as functions whose names start with ‘test’.

Running your tests is as easy as

$ nose2

http://nose2.readthedocs.org/en/latest/getting_started.html#running-tests

nose2 plugins

Many plugins exist for nose2, such as code coverage: Some plugins, such as coverage, must be additionally installed

$ pip install cov-core
# now it can be used
$ nose2 --with-coverage

Some of many useful plugins installed by default:

running coverage

Install with pip. Written by Ned Batchelder

To run coverage on your test suite:

coverage run my_program.py arg1 arg2

This generates a .coverage file. To analyze it on the console:

coverage report

Else generate an HTML report in the current directory:

coverage html

To find out coverage across the standard library, add -L:

-L, --pylib           Measure coverage even inside the Python installed
                      library, which isn't done by default.

branch coverage

consider the following code:

x = False  # 1
if x:      # 2
    print("in branch")  # 3
print("out of branch")  # 4

We want to make sure the branch is being bypassed correctly in the False case

Track which branch destinations were not visited with the –branch option to run

coverage run --branch myprog.py

http://nedbatchelder.com/code/coverage/branch.html

Doctests

Tests placed in docstrings to demonstrate usage of a component to a human in a machine testable way

def square(x):
    """
    Squares x.

    >>> square(2)
    4
    >>> square(-2)
    4
    """
    return x * x

python -m doctest -v example.py

Now generate documentation, using epydoc for example:

$ epydoc example.py

http://docs.python.org/3/library/doctest.html

http://www.python.org/dev/peps/pep-0257/

http://epydoc.sourceforge.net/

http://sphinx-doc.org/

http://www.doxygen.org

Test Driven Development (TDD)

In TDD, the tests are written the meet the requirements before the code exists.

Once the collection of tests passes, the requirement is considered met.

We don’t always want to run the entire test suite. In order to run a single test with nose:

nose2 test_calculator.TestCalculatorFunctions.test_add

Exercises

  • Add unit tests for each method in calculator_functions.py
  • Add fixtures via setUp/tearDown methods and setUpClass/tearDownClass class methods. Are they behaving how you expect?
  • Add additional unit tests for floating point calculations
  • Fix any failures in the code
  • Add doctests to calculator_functions.py

Context managers

One more Python feature before getting back to testing...

the with statement

Context managers via the “with” statement

If you’ve been opening files using “with” (and you probably should be), you have been using context managers:

with open("file.txt", "w") as f:
    f.write("foo")

A context manager is just a class with __enter__ and __exit__ methods defined to handle setting up and tearing down the context

Provides generalizable execution contexts in which setup and teardown of context are executed no matter what happens

This allows us to do things like setup/teardown and separate out exception handling code

Writing a context manager

Define __enter__(self) and __exit__(self, type, value, traceback) on a class

If __exit__ returns a true value, a caught exception is not re-raised

For example:

import os, random, shutil, time

class TemporaryDirectory(object):
    """A context manager for creating a temporary directory
       which gets destroyed on context exit"""
    def __init__(self,directory):
        self.base_directory = directory

    def __enter__(self):
        self.directory = os.path.join(self.base_directory, str(random.random()))
        return os.makedirs(self.directory)

    def __exit__(self, type, value, traceback):
        shutil.rmtree(self.directory)

with TemporaryDirectory("/tmp/foo") as dir:
    with open(os.path.join(dir, "foo.txt"), 'wb') as f:
        f.write("foo")
    time.sleep(5)

http://www.python.org/dev/peps/pep-0343/

Context Manager exercise

Create a context manager which prints information on all exceptions which occur in the context and continues execution

with YourExceptionHandler():
    print("do some stuff here")
    1/0

print("should still reach this point")

Also see the contextlib module

Why might using a context manager be better than implementing this with try..except..finally ?

For entire code block, see https://www.python.org/dev/peps/pep-0343/ (Specification)

with EXPR as VAR:
    BLOCK
# vs.
mgr = (EXPR)
exit = type(mgr).__exit__  # Not calling it yet
value = type(mgr).__enter__(mgr)
exc = True
try:
    try:
        VAR = value  # Only if "as VAR" is present
        BLOCK
    except:
        # The exceptional case is handled here
        exc = False
        if not exit(mgr, *sys.exc_info()):
            raise
        # The exception is swallowed if exit() returns true
finally:
    # The normal and non-local-goto cases are handled here
    if exc:
        exit(mgr, None, None, None)

Now we’ve got the tools to really test

Consider the application in the examples/wikidef directory. Give the command line utility a subject, and it will return a definition.

./define.py Robot

How can we test our application code without abusing (and waiting for) Wikipedia?

Using Mock objects

Using Mock objects to test an application with service dependencies

Mock objects replace real objects in your code at runtime during test

This allows you to test code which calls these objects without having their actual code run

Useful for testing objects which depend on unimplemented code, resources which are expensive, or resources which are unavailable during test execution

http://www.voidspace.org.uk/python/mock

Mocks

The MagickMock class will keep track of calls to it so we can verify that the class is being called correctly, without having to execute the code underneath

import mock

mock_object = mock.MagicMock()
mock_object.foo.return_value = "foo return"
print(mock_object.foo.call_count)
print(mock_object.foo())
print(mock_object.foo.call_count)
# raise an exception by assigning to the side_effect attribute
mock_object.foo.side_effect = Exception
mock_object.foo()

Easy mocking with mock.patch

patch acts as a function decorator, class decorator, or a context manager

Inside the body of the function or with statement, the target is patched with a new object. When the function/with statement exits the patch is undone

Using patch

# patch with a decorator
@patch.object(Wikipedia, 'article')
def test_article_success_decorator_mocked(self, mock_method):
    article = Definitions.article("Robot")
    mock_method.assert_called_once_with("Robot")

# patch with a context manager
def test_article_success_context_manager_mocked(self):
    with patch.object(Wikipedia, 'article') as mock_method:
        article = Definitions.article("Robot")
        mock_method.assert_called_once_with("Robot")

http://www.voidspace.org.uk/python/mock/patch.html

Exercises

When define.py is given the name of a non-existant article, an exception is thrown. This exception causes another exception to occur, and the whole thing is not very readable. Why does this happen?

Use what you learned last week about exceptions to throw a better exception, and then add a new test that confirms this behavior. Use mock for your test, so you are not hammering Wikipedia.