testtools for test authors

If you are writing tests for a Python project and you (rather wisely) want to use testtools to do so, this is the manual for you.

We assume that you already know Python and that you know something about automated testing already.

If you are a test author of an unusually large or unusually unusual test suite, you might be interested in testtools for framework folk.

You might also be interested in the testtools API docs.


testtools is a set of extensions to Python’s standard unittest module. Writing tests with testtools is very much like writing tests with standard Python, or with Twisted’s “trial”, or nose, except a little bit easier and more enjoyable.

Below, we’ll try to give some examples of how to use testtools in its most basic way, as well as a sort of feature-by-feature breakdown of the cool bits that you could easily miss.

The basics

Here’s what a basic testtools unit tests look like:

from testtools import TestCase
from myproject import silly

class TestSillySquare(TestCase):
    """Tests for silly square function."""

    def test_square(self):
        # 'square' takes a number and multiplies it by itself.
        result = silly.square(7)
        self.assertEqual(result, 49)

    def test_square_bad_input(self):
        # 'square' raises a TypeError if it's given bad input, say a
        # string.
        self.assertRaises(TypeError, silly.square, "orange")

Here you have a class that inherits from testtools.TestCase and bundles together a bunch of related tests. The tests themselves are methods on that class that begin with test_.

Running your tests

You can run these tests in many ways. testtools provides a very basic mechanism for doing so:

$ python -m testtools.run exampletest
Tests running...
Ran 2 tests in 0.000s


where ‘exampletest’ is a module that contains unit tests. By default, testtools.run will not recursively search the module or package for unit tests. To do this, you will need to either have the discover module installed or have Python 2.7 or later, and then run:

$ python -m testtools.run discover packagecontainingtests

For more information see the Python unittest documentation, and:

python -m testtools.run --help

which describes the options available to testtools.run.

As your testing needs grow and evolve, you will probably want to use a more sophisticated test runner. There are many of these for Python, and almost all of them will happily run testtools tests. In particular:

From now on, we’ll assume that you know how to run your tests.

Running test with Distutils

If you are using Distutils to build your Python project, you can use the testtools Distutils command to integrate testtools into your Distutils workflow:

from distutils.core import setup
from testtools import TestCommand
    cmdclass={'test': TestCommand}

You can then run:

$ python setup.py test -m exampletest
Tests running...
Ran 2 tests in 0.000s


For more information about the capabilities of the TestCommand command see:

$ python setup.py test --help

You can use the setup configuration to specify the default behavior of the TestCommand command.


The core of automated testing is making assertions about the way things are, and getting a nice, helpful, informative error message when things are not as they ought to be.

All of the assertions that you can find in Python standard unittest can be found in testtools (remember, testtools extends unittest). testtools changes the behaviour of some of those assertions slightly and adds some new assertions that you will almost certainly find useful.

Improved assertRaises

TestCase.assertRaises returns the caught exception. This is useful for asserting more things about the exception than just the type:

def test_square_bad_input(self):
    # 'square' raises a TypeError if it's given bad input, say a
    # string.
    e = self.assertRaises(TypeError, silly.square, "orange")
    self.assertEqual("orange", e.bad_value)
    self.assertEqual("Cannot square 'orange', not a number.", str(e))

Note that this is incompatible with the assertRaises in unittest2 and Python2.7.


If you are using a version of Python that supports the with context manager syntax, you might prefer to use that syntax to ensure that code raises particular errors. ExpectedException does just that. For example:

def test_square_root_bad_input_2(self):
    # 'square' raises a TypeError if it's given bad input.
    with ExpectedException(TypeError, "Cannot square.*"):

The first argument to ExpectedException is the type of exception you expect to see raised. The second argument is optional, and can be either a regular expression or a matcher. If it is a regular expression, the str() of the raised exception must match the regular expression. If it is a matcher, then the raised exception object must match it. The optional third argument msg will cause the raised error to be annotated with that message.

assertIn, assertNotIn

These two assertions check whether a value is in a sequence and whether a value is not in a sequence. They are “assert” versions of the in and not in operators. For example:

def test_assert_in_example(self):
    self.assertIn('a', 'cat')
    self.assertNotIn('o', 'cat')
    self.assertIn(5, list_of_primes_under_ten)
    self.assertNotIn(12, list_of_primes_under_ten)

assertIs, assertIsNot

These two assertions check whether values are identical to one another. This is sometimes useful when you want to test something more strict than mere equality. For example:

def test_assert_is_example(self):
    foo = [None]
    foo_alias = foo
    bar = [None]
    self.assertIs(foo, foo_alias)
    self.assertIsNot(foo, bar)
    self.assertEqual(foo, bar) # They are equal, but not identical


As much as we love duck-typing and polymorphism, sometimes you need to check whether or not a value is of a given type. This method does that. For example:

def test_assert_is_instance_example(self):
    now = datetime.now()
    self.assertIsInstance(now, datetime)

Note that there is no assertIsNotInstance in testtools currently.


Sometimes it’s useful to write tests that fail. For example, you might want to turn a bug report into a unit test, but you don’t know how to fix the bug yet. Or perhaps you want to document a known, temporary deficiency in a dependency.

testtools gives you the TestCase.expectFailure to help with this. You use it to say that you expect this assertion to fail. When the test runs and the assertion fails, testtools will report it as an “expected failure”.

Here’s an example:

def test_expect_failure_example(self):
        "cats should be dogs", self.assertEqual, 'cats', 'dogs')

As long as ‘cats’ is not equal to ‘dogs’, the test will be reported as an expected failure.

If ever by some miracle ‘cats’ becomes ‘dogs’, then testtools will report an “unexpected success”. Unlike standard unittest, testtools treats this as something that fails the test suite, like an error or a failure.


The built-in assertion methods are very useful, they are the bread and butter of writing tests. However, soon enough you will probably want to write your own assertions. Perhaps there are domain specific things that you want to check (e.g. assert that two widgets are aligned parallel to the flux grid), or perhaps you want to check something that could almost but not quite be found in some other standard library (e.g. assert that two paths point to the same file).

When you are in such situations, you could either make a base class for your project that inherits from testtools.TestCase and make sure that all of your tests derive from that, or you could use the testtools Matcher system.

Using Matchers

Here’s a really basic example using stock matchers found in testtools:

import testtools
from testtools.matchers import Equals

class TestSquare(TestCase):
    def test_square(self):
       result = square(7)
       self.assertThat(result, Equals(49))

The line self.assertThat(result, Equals(49)) is equivalent to self.assertEqual(result, 49) and means “assert that result equals 49”. The difference is that assertThat is a more general method that takes some kind of observed value (in this case, result) and any matcher object (here, Equals(49)).

The matcher object could be absolutely anything that implements the Matcher protocol. This means that you can make more complex matchers by combining existing ones:

def test_square_silly(self):
    result = square(7)
    self.assertThat(result, Not(Equals(50)))

Which is roughly equivalent to:

def test_square_silly(self):
    result = square(7)
    self.assertNotEqual(result, 50)

assert_that Function

In addition to self.assertThat, testtools also provides the assert_that function in testtools.assertions This behaves like the method version does:

class TestSquare(TestCase):

    def test_square():
        result = square(7)
        assert_that(result, Equals(49))

    def test_square_silly():
        result = square(7)
        assert_that(result, Not(Equals(50)))

Delayed Assertions

A failure in the self.assertThat method will immediately fail the test: No more test code will be run after the assertion failure.

The expectThat method behaves the same as assertThat with one exception: when failing the test it does so at the end of the test code rather than when the mismatch is detected. For example:

import subprocess

from testtools import TestCase
from testtools.matchers import Equals

class SomeProcessTests(TestCase):

    def test_process_output(self):
        process = subprocess.Popen(
          ["my-app", "/some/path"],

        stdout, stderrr = process.communicate()

        self.expectThat(process.returncode, Equals(0))
        self.expectThat(stdout, Equals("Expected Output"))
        self.expectThat(stderr, Equals(""))

In this example, should the expectThat call fail, the failure will be recorded in the test result, but the test will continue as normal. If all three assertions fail, the test result will have three failures recorded, and the failure details for each failed assertion will be attached to the test result.

Stock matchers

testtools comes with many matchers built in. They can all be found in and imported from the testtools.matchers module.


Matches if two items are equal. For example:

def test_equals_example(self):
    self.assertThat([42], Equals([42]))


Matches if two items are identical. For example:

def test_is_example(self):
    foo = object()
    self.assertThat(foo, Is(foo))


Adapts isinstance() to use as a matcher. For example:

def test_isinstance_example(self):
    class MyClass:pass
    self.assertThat(MyClass(), IsInstance(MyClass))
    self.assertThat(MyClass(), IsInstance(MyClass, str))

The raises helper

Matches if a callable raises a particular type of exception. For example:

def test_raises_example(self):
    self.assertThat(lambda: 1/0, raises(ZeroDivisionError))

This is actually a convenience function that combines two other matchers: Raises and MatchesException.


Matches a string as if it were the output of a doctest example. Very useful for making assertions about large chunks of text. For example:

import doctest

def test_doctest_example(self):
    output = "Colorless green ideas"
        DocTestMatches("Colorless ... ideas", doctest.ELLIPSIS))

We highly recommend using the following flags:



Matches if the given thing is greater than the thing in the matcher. For example:

def test_greater_than_example(self):
    self.assertThat(3, GreaterThan(2))


Matches if the given thing is less than the thing in the matcher. For example:

def test_less_than_example(self):
    self.assertThat(2, LessThan(3))

StartsWith, EndsWith

These matchers check to see if a string starts with or ends with a particular substring. For example:

def test_starts_and_ends_with_example(self):
    self.assertThat('underground', StartsWith('und'))
    self.assertThat('underground', EndsWith('und'))


This matcher checks to see if the given thing contains the thing in the matcher. For example:

def test_contains_example(self):
    self.assertThat('abc', Contains('b'))


Matches an exc_info tuple if the exception is of the correct type. For example:

def test_matches_exception_example(self):
        raise RuntimeError('foo')
    except RuntimeError:
        exc_info = sys.exc_info()
    self.assertThat(exc_info, MatchesException(RuntimeError))
    self.assertThat(exc_info, MatchesException(RuntimeError('bar')))

Most of the time, you will want to uses The raises helper instead.


Matches if something is not equal to something else. Note that this is subtly different to Not(Equals(x)). NotEquals(x) will match if y != x, Not(Equals(x)) will match if not y == x.

You only need to worry about this distinction if you are testing code that relies on badly written overloaded equality operators.


Matches if the keys of one dict are equal to the keys of another dict. For example:

def test_keys_equal(self):
    x = {'a': 1, 'b': 2}
    y = {'a': 2, 'b': 3}
    self.assertThat(x, KeysEqual(y))


Matches a string against a regular expression, which is a wonderful thing to be able to do, if you think about it:

def test_matches_regex_example(self):
    self.assertThat('foo', MatchesRegex('fo+'))


Check the length of a collection. The following assertion will fail:

self.assertThat([1, 2, 3], HasLength(2))

But this one won’t:

self.assertThat([1, 2, 3], HasLength(3))

Combining matchers

One great thing about matchers is that you can readily combine existing matchers to get variations on their behaviour or to quickly build more complex assertions.

Below are a few of the combining matchers that come with testtools.


Negates another matcher. For example:

def test_not_example(self):
    self.assertThat([42], Not(Equals("potato")))
    self.assertThat([42], Not(Is([42])))

If you find yourself using Not frequently, you may wish to create a custom matcher for it. For example:

IsNot = lambda x: Not(Is(x))

def test_not_example_2(self):
    self.assertThat([42], IsNot([42]))


Used to add custom notes to a matcher. For example:

def test_annotate_example(self):
    result = 43
        result, Annotate("Not the answer to the Question!", Equals(42)))

Since the annotation is only ever displayed when there is a mismatch (e.g. when result does not equal 42), it’s a good idea to phrase the note negatively, so that it describes what a mismatch actually means.

As with Not, you may wish to create a custom matcher that describes a common operation. For example:

PoliticallyEquals = lambda x: Annotate("Death to the aristos!", Equals(x))

def test_annotate_example_2(self):
    self.assertThat("orange", PoliticallyEquals("yellow"))

You can have assertThat perform the annotation for you as a convenience:

def test_annotate_example_3(self):
    self.assertThat("orange", Equals("yellow"), "Death to the aristos!")


Used to make a matcher that applies a function to the matched object before matching. This can be used to aid in creating trivial matchers as functions, for example:

def test_after_preprocessing_example(self):
    def PathHasFileContent(content):
        def _read(path):
            return open(path).read()
        return AfterPreprocessing(_read, Equals(content))
    self.assertThat('/tmp/foo.txt', PathHasFileContent("Hello world!"))


Combines many matchers to make a new matcher. The new matcher will only match things that match every single one of the component matchers.

It’s much easier to understand in Python than in English:

def test_matches_all_example(self):
    has_und_at_both_ends = MatchesAll(StartsWith("und"), EndsWith("und"))
    # This will succeed.
    self.assertThat("underground", has_und_at_both_ends)
    # This will fail.
    self.assertThat("found", has_und_at_both_ends)
    # So will this.
    self.assertThat("undead", has_und_at_both_ends)

At this point some people ask themselves, “why bother doing this at all? why not just have two separate assertions?”. It’s a good question.

The first reason is that when a MatchesAll gets a mismatch, the error will include information about all of the bits that mismatched. When you have two separate assertions, as below:

def test_two_separate_assertions(self):
     self.assertThat("foo", StartsWith("und"))
     self.assertThat("foo", EndsWith("und"))

Then you get absolutely no information from the second assertion if the first assertion fails. Tests are largely there to help you debug code, so having more information in error messages is a big help.

The second reason is that it is sometimes useful to give a name to a set of matchers. has_und_at_both_ends is a bit contrived, of course, but it is clear. The FileExists and DirExists matchers included in testtools are perhaps better real examples.

If you want only the first mismatch to be reported, pass first_only=True as a keyword parameter to MatchesAll.


Like MatchesAll, MatchesAny combines many matchers to make a new matcher. The difference is that the new matchers will match a thing if it matches any of the component matchers.

For example:

def test_matches_any_example(self):
    self.assertThat(42, MatchesAny(Equals(5), Not(Equals(6))))


Matches many values against a single matcher. Can be used to make sure that many things all meet the same condition:

def test_all_match_example(self):
    self.assertThat([2, 3, 5, 7], AllMatch(LessThan(10)))

If the match fails, then all of the values that fail to match will be included in the error message.

In some ways, this is the converse of MatchesAll.


Where MatchesAny and MatchesAll combine many matchers to match a single value, MatchesListwise combines many matches to match many values.

For example:

def test_matches_listwise_example(self):
        [1, 2, 3], MatchesListwise(map(Equals, [1, 2, 3])))

This is useful for writing custom, domain-specific matchers.

If you want only the first mismatch to be reported, pass first_only=True to MatchesListwise.


Combines many matchers to match many values, without regard to their order.

Here’s an example:

def test_matches_setwise_example(self):
        [1, 2, 3], MatchesSetwise(Equals(2), Equals(3), Equals(1)))

Much like MatchesListwise, best used for writing custom, domain-specific matchers.


Creates a matcher that matches certain attributes of an object against a pre-defined set of matchers.

It’s much easier to understand in Python than in English:

def test_matches_structure_example(self):
    foo = Foo()
    foo.a = 1
    foo.b = 2
    matcher = MatchesStructure(a=Equals(1), b=Equals(2))
    self.assertThat(foo, matcher)

Since all of the matchers used were Equals, we could also write this using the byEquality helper:

def test_matches_structure_example(self):
    foo = Foo()
    foo.a = 1
    foo.b = 2
    matcher = MatchesStructure.byEquality(a=1, b=2)
    self.assertThat(foo, matcher)

MatchesStructure.fromExample takes an object and a list of attributes and creates a MatchesStructure matcher where each attribute of the matched object must equal each attribute of the example object. For example:

matcher = MatchesStructure.fromExample(foo, 'a', 'b')

is exactly equivalent to matcher in the previous example.


Sometimes, all you want to do is create a matcher that matches if a given function returns True, and mismatches if it returns False.

For example, you might have an is_prime function and want to make a matcher based on it:

def test_prime_numbers(self):
    IsPrime = MatchesPredicate(is_prime, '%s is not prime.')
    self.assertThat(7, IsPrime)
    self.assertThat(1983, IsPrime)
    # This will fail.
    self.assertThat(42, IsPrime)

Which will produce the error message:

Traceback (most recent call last):
  File "...", line ..., in test_prime_numbers
    self.assertThat(42, IsPrime)
MismatchError: 42 is not prime.


Sometimes you can’t use a trivial predicate and instead need to pass in some parameters each time. In that case, MatchesPredicateWithParams is your go-to tool for creating ad hoc matchers. MatchesPredicateWithParams takes a predicate function and message and returns a factory to produce matchers from that. The predicate needs to return a boolean (or any truthy object), and accept the object to match + whatever was passed into the factory.

For example, you might have an divisible function and want to make a matcher based on it:

def test_divisible_numbers(self):
    IsDivisibleBy = MatchesPredicateWithParams(
        divisible, '{0} is not divisible by {1}')
    self.assertThat(7, IsDivisibleBy(1))
    self.assertThat(7, IsDivisibleBy(7))
    self.assertThat(7, IsDivisibleBy(2))
    # This will fail.

Which will produce the error message:

Traceback (most recent call last):
  File "...", line ..., in test_divisible
    self.assertThat(7, IsDivisibleBy(2))
MismatchError: 7 is not divisible by 2.


Takes whatever the callable raises as an exc_info tuple and matches it against whatever matcher it was given. For example, if you want to assert that a callable raises an exception of a given type:

def test_raises_example(self):
        lambda: 1/0, Raises(MatchesException(ZeroDivisionError)))

Although note that this could also be written as:

def test_raises_example_convenient(self):
    self.assertThat(lambda: 1/0, raises(ZeroDivisionError))

See also MatchesException and the raises helper

Writing your own matchers

Combining matchers is fun and can get you a very long way indeed, but sometimes you will have to write your own. Here’s how.

You need to make two closely-linked objects: a Matcher and a Mismatch. The Matcher knows how to actually make the comparison, and the Mismatch knows how to describe a failure to match.

Here’s an example matcher:

class IsDivisibleBy(object):
    """Match if a number is divisible by another number."""
    def __init__(self, divider):
        self.divider = divider
    def __str__(self):
        return 'IsDivisibleBy(%s)' % (self.divider,)
    def match(self, actual):
        remainder = actual % self.divider
        if remainder != 0:
            return IsDivisibleByMismatch(actual, self.divider, remainder)
            return None

The matcher has a constructor that takes parameters that describe what you actually expect, in this case a number that other numbers ought to be divisible by. It has a __str__ method, the result of which is displayed on failure by assertThat and a match method that does the actual matching.

match takes something to match against, here actual, and decides whether or not it matches. If it does match, then match must return None. If it does not match, then match must return a Mismatch object. assertThat will call match and then fail the test if it returns a non-None value. For example:

def test_is_divisible_by_example(self):
    # This succeeds, since IsDivisibleBy(5).match(10) returns None.
    self.assertThat(10, IsDivisibleBy(5))
    # This fails, since IsDivisibleBy(7).match(10) returns a mismatch.
    self.assertThat(10, IsDivisibleBy(7))

The mismatch is responsible for what sort of error message the failing test generates. Here’s an example mismatch:

class IsDivisibleByMismatch(object):
    def __init__(self, number, divider, remainder):
        self.number = number
        self.divider = divider
        self.remainder = remainder

    def describe(self):
        return "%r is not divisible by %r, %r remains" % (
            self.number, self.divider, self.remainder)

    def get_details(self):
        return {}

The mismatch takes information about the mismatch, and provides a describe method that assembles all of that into a nice error message for end users. You can use the get_details method to provide extra, arbitrary data with the mismatch (e.g. the contents of a log file). Most of the time it’s fine to just return an empty dict. You can read more about Details elsewhere in this document.

Sometimes you don’t need to create a custom mismatch class. In particular, if you don’t care when the description is calculated, then you can just do that in the Matcher itself like this:

def match(self, actual):
    remainder = actual % self.divider
    if remainder != 0:
        return Mismatch(
            "%r is not divisible by %r, %r remains" % (
                actual, self.divider, remainder))
        return None

When writing a describe method or constructing a Mismatch object the code should ensure it only emits printable unicode. As this output must be combined with other text and forwarded for presentation, letting through non-ascii bytes of ambiguous encoding or control characters could throw an exception or mangle the display. In most cases simply avoiding the %s format specifier and using %r instead will be enough. For examples of more complex formatting see the testtools.matchers implementatons.


As we may have mentioned once or twice already, one of the great benefits of automated tests is that they help find, isolate and debug errors in your system.

Frequently however, the information provided by a mere assertion failure is not enough. It’s often useful to have other information: the contents of log files; what queries were run; benchmark timing information; what state certain subsystem components are in and so forth.

testtools calls all of these things “details” and provides a single, powerful mechanism for including this information in your test run.

Here’s an example of how to add them:

from testtools import TestCase
from testtools.content import text_content

class TestSomething(TestCase):

    def test_thingy(self):
        self.addDetail('arbitrary-color-name', text_content("blue"))
        1 / 0 # Gratuitous error!

A detail an arbitrary piece of content given a name that’s unique within the test. Here the name is arbitrary-color-name and the content is text_content("blue"). The name can be any text string, and the content can be any testtools.content.Content object.

When the test runs, testtools will show you something like this:

ERROR: exampletest.TestSomething.test_thingy
arbitrary-color-name: {{{blue}}}

Traceback (most recent call last):
  File "exampletest.py", line 8, in test_thingy
    1 / 0 # Gratuitous error!
ZeroDivisionError: integer division or modulo by zero
Ran 1 test in 0.030s

As you can see, the detail is included as an attachment, here saying that our arbitrary-color-name is “blue”.


For the actual content of details, testtools uses its own MIME-based Content object. This allows you to attach any information that you could possibly conceive of to a test, and allows testtools to use or serialize that information.

The basic testtools.content.Content object is constructed from a testtools.content.ContentType and a nullary callable that must return an iterator of chunks of bytes that the content is made from.

So, to make a Content object that is just a simple string of text, you can do:

from testtools.content import Content
from testtools.content_type import ContentType

text = Content(ContentType('text', 'plain'), lambda: ["some text"])

Because adding small bits of text content is very common, there’s also a convenience method:

text = text_content("some text")

To make content out of an image stored on disk, you could do something like:

image = Content(ContentType('image', 'png'), lambda: open('foo.png').read())

Or you could use the convenience function:

image = content_from_file('foo.png', ContentType('image', 'png'))

The lambda helps make sure that the file is opened and the actual bytes read only when they are needed – by default, when the test is finished. This means that tests can construct and add Content objects freely without worrying too much about how they affect run time.

A realistic example

A very common use of details is to add a log file to failing tests. Say your project has a server represented by a class SomeServer that you can start up and shut down in tests, but runs in another process. You want to test interaction with that server, and whenever the interaction fails, you want to see the client-side error and the logs from the server-side. Here’s how you might do it:

from testtools import TestCase
from testtools.content import attach_file, Content
from testtools.content_type import UTF8_TEXT

from myproject import SomeServer

class SomeTestCase(TestCase):

    def setUp(self):
        super(SomeTestCase, self).setUp()
        self.server = SomeServer()
        self.addCleanup(attach_file, self.server.logfile, self)

    def attach_log_file(self):
                    lambda: open(self.server.logfile, 'r').readlines()))

    def test_a_thing(self):
        self.assertEqual("cool", self.server.temperature)

This test will attach the log file of SomeServer to each test that is run. testtools will only display the log file for failing tests, so it’s not such a big deal.

If the act of adding at detail is expensive, you might want to use addOnException so that you only do it when a test actually raises an exception.

Controlling test execution


TestCase.addCleanup is a robust way to arrange for a clean up function to be called before tearDown. This is a powerful and simple alternative to putting clean up logic in a try/finally block or tearDown method. For example:

def test_foo(self):

This is particularly useful if you have some sort of factory in your test:

def make_locked_foo(self):
    foo = Foo()
    return foo

def test_frotz_a_foo(self):
    foo = self.make_locked_foo()
    self.assertEqual(foo.frotz_count, 1)

Any extra arguments or keyword arguments passed to addCleanup are passed to the callable at cleanup time.

Cleanups can also report multiple errors, if appropriate by wrapping them in a testtools.MultipleExceptions object:

raise MultipleExceptions(exc_info1, exc_info2)


Tests often depend on a system being set up in a certain way, or having certain resources available to them. Perhaps a test needs a connection to the database or access to a running external server.

One common way of doing this is to do:

class SomeTest(TestCase):
    def setUp(self):
        super(SomeTest, self).setUp()
        self.server = Server()

testtools provides a more convenient, declarative way to do the same thing:

class SomeTest(TestCase):
    def setUp(self):
        super(SomeTest, self).setUp()
        self.server = self.useFixture(Server())

useFixture(fixture) calls setUp on the fixture, schedules a clean up to clean it up, and schedules a clean up to attach all details held by the fixture to the test case. The fixture object must meet the fixtures.Fixture protocol (version 0.3.4 or newer, see fixtures).

If you have anything beyond the most simple test set up, we recommend that you put this set up into a Fixture class. Once there, the fixture can be easily re-used by other tests and can be combined with other fixtures to make more complex resources.

Skipping tests

Many reasons exist to skip a test: a dependency might be missing; a test might be too expensive and thus should not berun while on battery power; or perhaps the test is testing an incomplete feature.

TestCase.skipTest is a simple way to have a test stop running and be reported as a skipped test, rather than a success, error or failure. For example:

def test_make_symlink(self):
    symlink = getattr(os, 'symlink', None)
    if symlink is None:
        self.skipTest("No symlink support")
    symlink(whatever, something_else)

Using skipTest means that you can make decisions about what tests to run as late as possible, and close to the actual tests. Without it, you might be forced to use convoluted logic during test loading, which is a bit of a mess.

Legacy skip support

If you are using this feature when running your test suite with a legacy TestResult object that is missing the addSkip method, then the addError method will be invoked instead. If you are using a test result from testtools, you do not have to worry about this.

In older versions of testtools, skipTest was known as skip. Since Python 2.7 added skipTest support, the skip name is now deprecated. No warning is emitted yet – some time in the future we may do so.


Sometimes, you might wish to do something only when a test fails. Perhaps you need to run expensive diagnostic routines or some such. TestCase.addOnException allows you to easily do just this. For example:

class SomeTest(TestCase):
    def setUp(self):
        super(SomeTest, self).setUp()
        self.server = self.useFixture(SomeServer())

    def attach_server_diagnostics(self, exc_info):
        self.server.prep_for_diagnostics() # Expensive!
        self.addDetail('server-diagnostics', self.server.get_diagnostics)

    def test_a_thing(self):
        self.assertEqual('cheese', 'chalk')

In this example, attach_server_diagnostics will only be called when a test fails. It is given the exc_info tuple of the error raised by the test, just in case it is needed.

Twisted support

testtools provides highly experimental support for running Twisted tests – tests that return a Deferred and rely on the Twisted reactor. You should not use this feature right now. We reserve the right to change the API and behaviour without telling you first.

However, if you are going to, here’s how you do it:

from testtools import TestCase
from testtools.deferredruntest import AsynchronousDeferredRunTest

class MyTwistedTests(TestCase):

    run_tests_with = AsynchronousDeferredRunTest

    def test_foo(self):
        # ...
        return d

In particular, note that you do not have to use a special base TestCase in order to run Twisted tests.

You can also run individual tests within a test case class using the Twisted test runner:

class MyTestsSomeOfWhichAreTwisted(TestCase):

    def test_normal(self):

    def test_twisted(self):
        # ...
        return d

Here are some tips for converting your Trial tests into testtools tests.

  • Use the AsynchronousDeferredRunTest runner
  • Make sure to upcall to setUp and tearDown
  • Don’t use setUpClass or tearDownClass
  • Don’t expect setting .todo, .timeout or .skip attributes to do anything
  • flushLoggedErrors is testtools.deferredruntest.flush_logged_errors
  • assertFailure is testtools.deferredruntest.assert_fails_with
  • Trial spins the reactor a couple of times before cleaning it up, AsynchronousDeferredRunTest does not. If you rely on this behavior, use AsynchronousDeferredRunTestForBrokenTwisted.


Setting the testtools.TestCase.force_failure instance variable to True will cause the test to be marked as a failure, but won’t stop the test code from running (see Delayed Test Failure).

Test helpers

testtools comes with a few little things that make it a little bit easier to write tests.


patch is a convenient way to monkey-patch a Python object for the duration of your test. It’s especially useful for testing legacy code. e.g.:

def test_foo(self):
    my_stream = StringIO()
    self.patch(sys, 'stderr', my_stream)
    self.assertEqual('', my_stream.getvalue())

The call to patch above masks sys.stderr with my_stream so that anything printed to stderr will be captured in a StringIO variable that can be actually tested. Once the test is done, the real sys.stderr is restored to its rightful place.

Creation methods

Often when writing unit tests, you want to create an object that is a completely normal instance of its type. You don’t want there to be anything special about its properties, because you are testing generic behaviour rather than specific conditions.

A lot of the time, test authors do this by making up silly strings and numbers and passing them to constructors (e.g. 42, ‘foo’, “bar” etc), and that’s fine. However, sometimes it’s useful to be able to create arbitrary objects at will, without having to make up silly sample data.

To help with this, testtools.TestCase implements creation methods called getUniqueString and getUniqueInteger. They return strings and integers that are unique within the context of the test that can be used to assemble more complex objects. Here’s a basic example where getUniqueString is used instead of saying “foo” or “bar” or whatever:

class SomeTest(TestCase):

    def test_full_name(self):
        first_name = self.getUniqueString()
        last_name = self.getUniqueString()
        p = Person(first_name, last_name)
        self.assertEqual(p.full_name, "%s %s" % (first_name, last_name))

And here’s how it could be used to make a complicated test:

class TestCoupleLogic(TestCase):

    def make_arbitrary_person(self):
        return Person(self.getUniqueString(), self.getUniqueString())

    def test_get_invitation(self):
        a = self.make_arbitrary_person()
        b = self.make_arbitrary_person()
        couple = Couple(a, b)
        event_name = self.getUniqueString()
        invitation = couple.get_invitation(event_name)
            "We invite %s and %s to %s" % (
                a.full_name, b.full_name, event_name))

Essentially, creation methods like these are a way of reducing the number of assumptions in your tests and communicating to test readers that the exact details of certain variables don’t actually matter.

See pages 419-423 of xUnit Test Patterns by Gerard Meszaros for a detailed discussion of creation methods.

Test attributes

Inspired by the nosetests attr plugin, testtools provides support for marking up test methods with attributes, which are then exposed in the test id and can be used when filtering tests by id. (e.g. via --load-list):

from testtools.testcase import attr, WithAttributes

class AnnotatedTests(WithAttributes, TestCase):

    def test_one(self):

    @attr('more', 'than', 'one')
    def test_two(self):

    def test_three(self):

General helpers

Conditional imports

Lots of the time we would like to conditionally import modules. testtools uses the small library extras to do this. This used to be part of testtools.

Instead of:

    from twisted.internet import defer
except ImportError:
    defer = None

You can do:

defer = try_import('twisted.internet.defer')

Instead of:

    from StringIO import StringIO
except ImportError:
    from io import StringIO

You can do:

StringIO = try_imports(['StringIO.StringIO', 'io.StringIO'])

Safe attribute testing

hasattr is broken on many versions of Python. The helper safe_hasattr can be used to safely test whether an object has a particular attribute. Like try_import this used to be in testtools but is now in extras.

Nullary callables

Sometimes you want to be able to pass around a function with the arguments already specified. The normal way of doing this in Python is:

nullary = lambda: f(*args, **kwargs)

Which is mostly good enough, but loses a bit of debugging information. If you take the repr() of nullary, you’re only told that it’s a lambda, and you get none of the juicy meaning that you’d get from the repr() of f.

The solution is to use Nullary instead:

nullary = Nullary(f, *args, **kwargs)

Here, repr(nullary) will be the same as repr(f).