testtools for framework folk


In addition to having many features for test authors, testtools also has many bits and pieces that are useful for folk who write testing frameworks.

If you are the author of a test runner, are working on a very large unit-tested project, are trying to get one testing framework to play nicely with another or are hacking away at getting your test suite to run in parallel over a heterogenous cluster of machines, this guide is for you.

This manual is a summary. You can get details by consulting the testtools API docs.

Extensions to TestCase

In addition to the TestCase specific methods, we have extensions for TestSuite that also apply to TestCase (because TestCase and TestSuite follow the Composite pattern).

Custom exception handling

testtools provides a way to control how test exceptions are handled. To do this, add a new exception to self.exception_handlers on a testtools.TestCase. For example:

>>> self.exception_handlers.insert(-1, (ExceptionClass, handler)).

Having done this, if any of setUp, tearDown, or the test method raise ExceptionClass, handler will be called with the test case, test result and the raised exception.

Use this if you want to add a new kind of test result, that is, if you think that addError, addFailure and so forth are not enough for your needs.

Controlling test execution

If you want to control more than just how exceptions are raised, you can provide a custom RunTest to a TestCase. The RunTest object can change everything about how the test executes.

To work with testtools.TestCase, a RunTest must have a factory that takes a test and an optional list of exception handlers and an optional last_resort handler. Instances returned by the factory must have a run() method that takes an optional TestResult object.

The default is testtools.runtest.RunTest, which calls setUp, the test method, tearDown and clean ups (see addCleanup) in the normal, vanilla way that Python’s standard unittest does.

To specify a RunTest for all the tests in a TestCase class, do something like this:

class SomeTests(TestCase):
    run_tests_with = CustomRunTestFactory

To specify a RunTest for a specific test in a TestCase class, do:

class SomeTests(TestCase):
    @run_test_with(CustomRunTestFactory, extra_arg=42, foo='whatever')
    def test_something(self):

In addition, either of these can be overridden by passing a factory in to the TestCase constructor with the optional runTest argument.

Test renaming

testtools.clone_test_with_new_id is a function to copy a test case instance to one with a new name. This is helpful for implementing test parameterization.

Delayed Test Failure

Setting the testtools.TestCase.force_failure instance variable to True will cause testtools.RunTest to fail the test case after the test has finished. This is useful when you want to cause a test to fail, but don’t want to prevent the remainder of the test code from being executed.

Exception formatting

Testtools TestCase instances format their own exceptions. The attribute __testtools_tb_locals__ controls whether to include local variables in the formatted exceptions.

Test placeholders

Sometimes, it’s useful to be able to add things to a test suite that are not actually tests. For example, you might wish to represents import failures that occur during test discovery as tests, so that your test result object doesn’t have to do special work to handle them nicely.

testtools provides two such objects, called “placeholders”: PlaceHolder and ErrorHolder. PlaceHolder takes a test id and an optional description. When it’s run, it succeeds. ErrorHolder takes a test id, and error and an optional short description. When it’s run, it reports that error.

These placeholders are best used to log events that occur outside the test suite proper, but are still very relevant to its results.


>>> suite = TestSuite()
>>> suite.add(PlaceHolder('I record an event'))
>>> suite.run(TextTestResult(verbose=True))
I record an event                                                   [OK]

Test instance decorators


This object calls out to your code when run / __call__ are called and allows the result object that will be used to run the test to be altered. This is very useful when working with a test runner that doesn’t know your test case requirements. For instance, it can be used to inject a unittest2 compatible adapter when someone attempts to run your test suite with a TestResult that does not support addSkip or other unittest2 methods. Similarly it can aid the migration to StreamResult.


>>> suite = TestSuite()
>>> suite = DecorateTestCaseResult(suite, ExtendedToOriginalDecorator)

Extensions to TestResult


StreamResult is a new API for dealing with test case progress that supports concurrent and distributed testing without the various issues that TestResult has such as buffering in multiplexers.

The design has several key principles:

  • Nothing that requires up-front knowledge of all tests.
  • Deal with tests running in concurrent environments, potentially distributed across multiple processes (or even machines). This implies allowing multiple tests to be active at once, supplying time explicitly, being able to differentiate between tests running in different contexts and removing any assumption that tests are necessarily in the same process.
  • Make the API as simple as possible - each aspect should do one thing well.

The TestResult API this is intended to replace has three different clients.

  • Each executing TestCase notifies the TestResult about activity.
  • The testrunner running tests uses the API to find out whether the test run had errors, how many tests ran and so on.
  • Finally, each TestCase queries the TestResult to see whether the test run should be aborted.

With StreamResult we need to be able to provide a TestResult compatible adapter (StreamToExtendedDecorator) to allow incremental migration. However, we don’t need to conflate things long term. So - we define three separate APIs, and merely mix them together to provide the StreamToExtendedDecorator. StreamResult is the first of these APIs - meeting the needs of TestCase clients. It handles events generated by running tests. See the API documentation for testtools.StreamResult for details.


Secondly we define the StreamSummary API which takes responsibility for collating errors, detecting incomplete tests and counting tests. This provides a compatible API with those aspects of TestResult. Again, see the API documentation for testtools.StreamSummary.


Lastly we define the TestControl API which is used to provide the shouldStop and stop elements from TestResult. Again, see the API documentation for testtools.TestControl. TestControl can be paired with a StreamFailFast to trigger aborting a test run when a failure is observed. Aborting multiple workers in a distributed environment requires hooking whatever signalling mechanism the distributed environment has up to a TestControl in each worker process.


A StreamResult filter that adds or removes tags from events:

>>> from testtools import StreamTagger
>>> sink = StreamResult()
>>> result = StreamTagger([sink], set(['add']), set(['discard']))
>>> result.startTestRun()
>>> # Run tests against result here.
>>> result.stopTestRun()


A simplified API for dealing with StreamResult streams. Each test is buffered until it completes and then reported as a trivial dict. This makes writing analysers very easy - you can ignore all the plumbing and just work with the result. e.g.:

>>> from testtools import StreamToDict
>>> def handle_test(test_dict):
...     print(test_dict['id'])
>>> result = StreamToDict(handle_test)
>>> result.startTestRun()
>>> # Run tests against result here.
>>> # At stopTestRun() any incomplete buffered tests are announced.
>>> result.stopTestRun()


This is a hybrid object that combines both the Extended and Stream TestResult APIs into one class, but only emits StreamResult events. This is useful when a StreamResult stream is desired, but you cannot be sure that the tests which will run have been updated to the StreamResult API.


This is a simple converter that emits the ExtendedTestResult API in response to events from the StreamResult API. Useful when outputting StreamResult events from a TestCase but the supplied TestResult does not support the status and file methods.


This is a StreamResult decorator for reporting tests from multiple threads at once. Each method submits an event to a supplied Queue object as a simple dict. See ConcurrentStreamTestSuite for a convenient way to use this.


This is a StreamResult decorator for adding timestamps to events that lack them. This allows writing the simplest possible generators of events and passing the events via this decorator to get timestamped data. As long as no buffering/queueing or blocking happen before the timestamper sees the event the timestamp will be as accurate as if the original event had it.


This is a StreamResult which forwards events to an arbitrary set of target StreamResult objects. Events that have no forwarding rule are passed onto an fallback StreamResult for processing. The mapping can be changed at runtime, allowing great flexibility and responsiveness to changes. Because The mapping can change dynamically and there could be the same recipient for two different maps, startTestRun and stopTestRun handling is fine grained and up to the user.

If no fallback has been supplied, an unroutable event will raise an exception.

For instance:

>>> router = StreamResultRouter()
>>> sink = doubles.StreamResult()
>>> router.add_rule(sink, 'route_code_prefix', route_prefix='0',
...     consume_route=True)
>>> router.status(test_id='foo', route_code='0/1', test_status='uxsuccess')

Would remove the 0/ from the route_code and forward the event like so:

>>> sink.status('test_id=foo', route_code='1', test_status='uxsuccess')

See pydoc testtools.StreamResultRouter for details.


This method is called on result objects when a test skips. The testtools.TestResult class records skips in its skip_reasons instance dict. The can be reported on in much the same way as succesful tests.


This method controls the time used by a TestResult, permitting accurate timing of test results gathered on different machines or in different threads. See pydoc testtools.TestResult.time for more details.


A TestResult which forwards activity to another test result, but synchronises on a semaphore to ensure that all the activity for a single test arrives in a batch. This allows simple TestResults which do not expect concurrent test reporting to be fed the activity from multiple test threads, or processes.

Note that when you provide multiple errors for a single test, the target sees each error as a distinct complete test.


A test result that dispatches its events to many test results. Use this to combine multiple different test result objects into one test result object that can be passed to TestCase.run() or similar. For example:

a = TestResult()
b = TestResult()
combined = MultiTestResult(a, b)
combined.startTestRun()  # Calls a.startTestRun() and b.startTestRun()

Each of the methods on MultiTestResult will return a tuple of whatever the component test results return.


Not strictly a TestResult, but something that implements the extended TestResult interface of testtools. It can be subclassed to create objects that wrap TestResults.


A TestResult that provides a text UI very similar to the Python standard library UI. Key differences are that its supports the extended outcomes and details API, and is completely encapsulated into the result object, permitting it to be used without a ‘TestRunner’ object. Not all the Python 2.7 outcomes are displayed (yet). It is also a ‘quiet’ result with no dots or verbose mode. These limitations will be corrected soon.


Adapts legacy TestResult objects, such as those found in older Pythons, to meet the testtools TestResult API.

Test Doubles

In testtools.testresult.doubles there are three test doubles that testtools uses for its own testing: Python26TestResult, Python27TestResult, ExtendedTestResult. These TestResult objects implement a single variation of the TestResult API each, and log activity to a list self._events. These are made available for the convenience of people writing their own extensions.

startTestRun and stopTestRun

Python 2.7 added hooks startTestRun and stopTestRun which are called before and after the entire test run. ‘stopTestRun’ is particularly useful for test results that wish to produce summary output.

testtools.TestResult provides default startTestRun and stopTestRun methods, and he default testtools runner will call these methods appropriately.

The startTestRun method will reset any errors, failures and so forth on the result, making the result object look as if no tests have been run.

Extensions to TestSuite


A TestSuite for parallel testing. This is used in conjuction with a helper that runs a single suite in some parallel fashion (for instance, forking, handing off to a subprocess, to a compute cloud, or simple threads). ConcurrentTestSuite uses the helper to get a number of separate runnable objects with a run(result), runs them all in threads using the ThreadsafeForwardingResult to coalesce their activity.


A variant of ConcurrentTestSuite that uses the new StreamResult API instead of the TestResult API. ConcurrentStreamTestSuite coordinates running some number of test/suites concurrently, with one StreamToQueue per test/suite.

Each test/suite gets given its own ExtendedToStreamDecorator + TimestampingStreamResult wrapped StreamToQueue instance, forwarding onto the StreamResult that ConcurrentStreamTestSuite.run was called with.

ConcurrentStreamTestSuite is a thin shim and it is easy to implement your own specialised form if that is needed.


A test suite that sets up a fixture before running any tests, and then tears it down after all of the tests are run. The fixture is not made available to any of the tests due to there being no standard channel for suites to pass information to the tests they contain (and we don’t have enough data on what such a channel would need to achieve to design a good one yet - or even decide if it is a good idea).


Given the composite structure of TestSuite / TestCase, sorting tests is problematic - you can’t tell what functionality is embedded into custom Suite implementations. In order to deliver consistent test orders when using test discovery (see http://bugs.python.org/issue16709), testtools flattens and sorts tests that have the standard TestSuite, and defines a new method sort_tests, which can be used by non-standard TestSuites to know when they should sort their tests. An example implementation can be seen at FixtureSuite.sorted_tests.

If there are duplicate test ids in a suite, ValueError will be raised.


Similarly to sorted_tests running a subset of tests is problematic - the standard run interface provides no way to limit what runs. Rather than confounding the two problems (selection and execution) we defined a method that filters the tests in a suite (or a case) by their unique test id. If you a writing custom wrapping suites, consider implementing filter_by_ids to support this (though most wrappers that subclass unittest.TestSuite will work just fine [see testtools.testsuite.filter_by_ids for details.]

Extensions to TestRunner

To facilitate custom listing of tests, testtools.run.TestProgram attempts to call list on the TestRunner, falling back to a generic implementation if it is not present.