Taking a lot of inspiration from Ruby’s RSpec, I’ve established a new way to write my tests when doing TDD in Django. Using django-nose, factoryboy, and a few of my own helpers I have greatly simplified and consolidated test writing.

My file structure:

project
  |
  +-- app1
  |
  +-- app2
  |
  +-- project
  |
  +-- tests
        |
        +-- models
        |     |
        |     +-- app1_models.py
        |     |
        |     +-- app2_models.py
        |
        +-- requests
        |     |
        |     +-- app1_pages.py
        |     |
        |     +-- app2_pages.py
        |
        +-- views
        |     |
        |     +-- app1_views.py
        |     |
        |     +-- app2_views.py
        |
        +-- support
        |     |
        |     +-- utilities.py
        |
        +-- factories.py

For brevity I skipped over the __init__.py files. You can see the tree in full here.

The utilities.py file contains some helper code I’ve written that expands on the assertions provided by Nose and provides a few other shortcuts. When all is said and done, we get three basic types of tests: models, requests, and views.

As an example for “app1” model tests, suppose app1 is a blog application:

from tests.support.utilities import *
from tests.factories import *

class BlogModel(TestCase):

  def test_fields(self):
    it = get_model(BlogFactory())

    it.should_respond_to('pub_date')
    it.should_respond_to('last_modified')
    it.should_respond_to('title')
    it.should_respond_to('author')
    it.should_respond_to('content')
    it.should_respond_to('slug')
    it.should_be_valid

  def test_with_blank_title(self):
    it = get_model(BlogFactory.build(title=''))

    it.should_not_be_valid

  def test_with blank_content(self):
    it = get_model(BlogFactory.build(content=''))

    it.should_not_be_valid

And so on. Here is an example of a “requests” test.

from app1.models import Blog
from tests.factories import *
from tests.support.utilities import *

class BlogPages(TestCase):

  def test_index(self):
    for _ in range(3):
      BlogFactory()

    page = visit('blog:index')

    for e in Entry.objects.all()[:3]:
      page.should_have_html('<span>%s</span>' % e.title)

  def test_blog_page(self):
    blog = BlogFactory()

    page = visit(blog)

    page.should_have_html('<title>%s</title>' % blog.title)
    page.should_contain(blog.author.first_name)

And so on. Finally, the “views” tests. Suppose app2 is a contact form application:

from app2.models import Message
from tests.support.utilities import *

class ContactViews(TestCase):

  def test_message_creation_with_invalid_information(self):
    old_message_count = Message.objects.count()
    ajax('post', 'contact', {})
    assert_equal(Message.objects.count(), old_message_count)

  def test_message_creation_with_valid_information(self):
    old_message_count = Message.objects.count()
    message = {
      'name': "Sample Person",
      'email': "[email protected]",
      'message': "This is an example message."}
    ajax('post', 'contact', message)
    assert_equal(Message.objects.count(), old_message_count + 1)

I think having tests arranged this way is much more comprehensive and readable. I’ve been “testing” the tests on this blog using Travis continuous integration and have been very pleased. Feel free to look at the source code for ideas.