Table Of Contents

Testing Your Application

Test-driven development (TDD) is a powerful (and increasingly popular) development strategy. In TDD, you write tests before you write your code. This helps you to solidify in your mind what the code should be doing and ensures it will be structured as you want. Test driven development also encourages good architecture decisions such as modularization because you can’t test what you can’t access!

Even if you don’t write tests first, having automated tests can be a lifesaver when the inevitable change request comes and you have to refactor your program to fit the new requirements.

When you install TurboGears, you also get Nose for free [1]. Nose (written by Jason Pellerin) is a powerful and convenient extension to the standard library module unittest and comes with its own discovery-based test runner.


You need to request the “testtools” option when installing TurboGears to get Nose: TurboGears[testtools]

Executing the test suite

When using Nose, all you have to do to run your test suite is open your console and execute following command in your project directory:


The nosetests command will search you project for test cases (any class or method that has a test/Test/TEST prefix) in your project directory, execute all those test cases and give you a report on the outcome.

By running your tests before you commit to source control repository (you are using source control, right?), you can catch unintended consequences of your edit before they become an issue in production. Early detection also makes tracking down the bug considerably simpler.

When a tests produces an error, nose will try to reload you test module, thereby causing errors because the model will get loaded twice. If you have trouble finding out whcih test case is the culprit, run nose with

nosetests -x

and it will stop after the first test which fails or produces an error. nosetests has many other useful command line options. Run

nosetests --help

to view the help message and find out more about them.


You can put test-specific configuration into the test.cfg file in your project directory. This allows you, for example, to have a special logging configuration when running tests or, the most common case, to define a different dburi for test, so your test will work with a database specially set up for tests or an in-memory database that will get wiped before/after each test.

For more information see the Configuration page and the section on testing your model below.

Testing Basics

A simple test demonstration:

# This method is a demo to be tested
def getsum(a, b):
    return a + b

# Test case are start with :doc:`/test`
def test_getsum():
    assert 3 == getsum(1, 2)
    assert 4 != getsum(1 ,2)

As mentioned, Nose looks for modules that start with test. In the above example, it will find the test_getsum() function and the report from nosetests will show that one (1) test passed. By convention, test modules go into separate packages called “tests” located beneath the package they are testing, though Nose does not require this (as the above example demonstrates).

Nose tries to be as unobtrusive as possible. For example you can write your tests using the standard Python assert statement, which raises an error if an expression returns False, or equivalent. The assert statement also takes an optional second argument, which is a human-readable description of the test case, e.g.:

assert 4 is getsum(1, 2), "assert that 1 + 2 == 4"

This is a failing test which, when run with nosetests -d, will output:

File "/path/to/", line XX, in test_getsum:
  assert 4 is getsum(1, 2), "assert that 1 + 2 == 4"
  AssertionError: assert 4 is getsum(1, 2)
        >>  assert 4 is getsum(1, 2), "assert that 1 + 2 == 4"

The last line is the nosetests assert introspection, which will replace the variables on the line (there aren’t any here) with the values.

The only portion of assertion testing that gets a bit ugly occurs when testing for exceptions:

    test_int = int('five')
    assert False, "Should have raised ValueError"
except ValueError, e:
    assert "invalid literal" in str(e)

TurboGears-specific Testing: the testutil Module

Testing web applications isn’t as easy as testing other environments. There’s the request dispatching and handling as well as the database setup and teardown. The turbogears.testutil module was written to facilitate testing for the TurboGears framework itself, but it’s in the framework because it’s useful for testing your applications as well.

Testing Your View

Here’s the sample file that will be used for our examples:

from turbogears import expose

class Root:
    def index(self, value="0"):
        value = int(value)
        return dict(newvalue=value*2)

Here is the test module for the above controller:

from turbogears import testutil
from projectname.controllers import Root
import cherrypy

##The template contains
# The new value is ${newvalue}.

# to test template
def test_withtemplate():
    "Tests the output passed through a template"
    cherrypy.root = Root()
    assert "The new value is 54." in cherrypy.response.body[0]

Here we’re not using any of the Nose-provided setup and teardown functionality, we’re doing everything in our test function. The docstring provides nicer output for failing tests in the testrunner.

In order to test out CherryPy, we need an active root object, this is created by setting cherrypy.root:

cherrypy.root = Root()

With the root created, the testutil.create_request creates a fake request that passes through all the CherryPy url traversal, the decorators for our function, template processing, etc. The request is processed and results put in cherrypy.response, but the response isn’t sent anywhere.

With the request processed, we test for correctness by using the in operator to find our modified substring in the body (cherrypy.response.body[0]).

Testing Your Controller

Continuing with our previous example, we’ll write a test that calls the index() method directly, bypassing CherryPy and our templates:

from turbogears import testutil
from projectname.controllers import Root
import cherrypy

# to test controller
def test_directcall():
    "Tests the output of the method without the template"
    root = Root()
    d =, "5")
    assert d["newvalue"] == 10

Despite not going through CherryPy, we do still need an instance of our project Root. The method is then used to call our function, which returns an object. The first argument to is a reference to the method, the remainder are *args or **kwargs. The return value for the controller is returned by call and can be tested as shown.

If your model makes use of the values in cherrypy.request or you’d like to check cherrypy.response in addition to the dictionary output, you can use the testutil.call_with_request(). call_with_request() takes both a method reference and a request as parameters and returns a tuple of the method output and the response object. If you don’t have a request object handy, use testutil.DummyRequest to create a fake one.

Testing Your Model

Testing your model is a thorny problem for unit testing, since the output is usually very dependent on the state of your database. This means that you’ll probably need a separate database for testing. To make this simpler, TurboGears provides the testutil.DBTest class.

If you inherit from this class, and you are using SQLObject, TG will provide setUp() and tearDown() methods that create and drop all the tables in your model for each method. In the example below, test_model_reset() is working on a completely empty database despite coming after test_name(), thanks to the setUp() and tearDown() methods inherited from testutil.DBTest.

from turbogears import testutil
## from turbogears import database
## database.set_db_uri("sqlite:///:memory:") #this is the default

from projectname import model

class TestMyURL(testutil.DBTest):
    model = model

    def test_name(self):
        entry = model.MyUrl(name="TurboGears",
              description="cool python web framework")

    def test_model_reset(self):
        entry = list(
        assert len(entry) is 0

If you want to define your own setUp() and/or tearDown() make sure that you call those methods from the parent class or you’ll get OperationalError: no such table: ... exceptions.

from turbogears import testutil
from projectname import model

class TestMyURL(testutil.DBTest):
    model = model

    def setUp(self):
        """Pre-test setup.

        Use the parent class setUp() method to create database tables,
        then ...
        super(TestMyURL, self).setUp()
        # Additional set-up code

    def tearDown(self):
        """Post-test tear-down.

        Use parent class tearDown() method to reset database
        before next test.
        super(TestMyURL, self).tearDown()
        # Additional tear-down code


For another alternative, you can use testutil’s BrowsingSession class to test your application using a browser metaphor. Simply create an instance and use the goto method to navigate your site. The response attribute will contain the body of your page. If the page sets a cookie, it will be available under the cookie attribute:

bs1 = testutil.BrowsingSession()
bs2 = testutil.BrowsingSession()
assert 'emma' in bs1.response
assert 'paul' in bs2.response

If your application sets an encoding in the ‘Content-Type’ header, the BrowsingSession instance will have a unicode_response attribute assigned as well.

Other Functionality

Since testutil was developed to test the TurboGears framework, there are a number of other methods that are generally less useful outside the framework but are listed here for completeness:

Category is the name passed to logging.getLogger() (e.g 'projectname.controllers' is the default controller logger). You must call print_log() or get_log() to reset the logger when you’re done.
Prints the captured log to stdout and resets the log.
Returns the captured list of log messages and resets the log.
catch_validation_errors(widget, value)
Tries to create widget using value, returns a tuple of the widget and the dict of Invalid instances.
Completely resets all sqlalchemy functionality in TG.
Sets cherrypy.root to None

Issues With CherryPy Filters

When testing your TurboGears application you should be aware of some issues which can bite you. The main problem is that, ...) won’t trigger any CherryPy filters. These filters modify the originally requested object or the request parameters. The main difference between filters and function decorators is that filters will be executed on all exposed methods while decorators are specific for the decorated functionality.

Some important functionality in TurboGears is implemented as a CherryPy filter, especially visit tracking and identity authentication. Therefore you can not test your login page easily. In order to test other methods which rely on identity for user authorization, there is a special method set_identity_user in turbogears.testutil which can be used to set a user before using

Another catch may be the NestedVariablesFilter, which is implemented as a filter too. You can either pass the nested dictionary to you method under test or use NestedVariables.to_python(...) manually before actually calling the real method.

from formencode.variabledecode import NestedVariables
from turbogears import testutil

# first method
args = {'phone': [{'nr': '12345', 'area_code': '042'},
                  {'nr': '6789', 'area_code': '021'}]
        'name': 'Foo'}
result =, **args)

# second method
args = {'': '12345', 'phone-1.area_code': '042',
        '': '6789', 'phone-2.area_code': '021',
        'name': 'Foo'}
args = NestedVariables.to_python(args)
result =, **args)

Note you can avoid these problems by using testutil.create_request(...) because this will create a faked request which goes through all stages of CherryPy’s request processing as mentioned above. The disadvantage is that you will get only the generated output from your template engine so you can’t easily extract all variables passed to the template.