Services Blog Fran├žais

django-dbdiff 0.7.3 release

| by jpic | django-dbdiff python django

I’m pretty lazy when it comes to writing tests for existing code, however, I’m even lazier when it comes to repetitive manual testing action, catching up with untested legacy code when I want a lot of coverage now, at least to ease upstream dependency upgrade testing.

PHPUnit has had a DBUnit component since at least 10 years, and Junit too, I ported the concept to Django, first in several apps. Extracted it in the package to de-duplicating the data import tests from django-representatives and django-representatives-votes which is re-used in django-cities-light.

But nowadays, we even use it in Open Source projects such as @ La Quadrature du Net or @ BetaGouv.

Database state assertion

A nice way to test a data import script is to create a source data fixture with a subset of data, ie. with only 10 cities instead of 28K or only 3 european parliament representatives instead of 3600, feed the import function with that and then compare the database state with a django fixture. This looks like what I was used to do:

  • use such a command to create a small data extract shuf -n3 cities15000.txt > cities_light/tests/cities_test_fixture.txt,
  • use it against the import script on a clean database,
  • verify the database manually, and run django-admin dumpdata --indent=4 cities_light > cities_light/tests/cities_test_expected.txt
  • then, make a test case that calls the import script against the fixture,
  • write and maintain some funny (fuzzy ?) repetitive test code to ensure that the database is in the expected state.

When a bug is fixed, just add the case to the fixture and repeat the process to create new expected data dumps, use coverage to ensure no case is missed.

With django-dbdiff, I just need to maintain to initial data extract, and test it with Fixture('appname/path/to/fixture', models=[YourModelToTest]).assertNoDiff() in a django.test.TransactionTestCase which has reset_sequences=True:

  • if the fixture in question doesn’t exist, it’ll be automatically created on with dumpdata for the concerned models on the first run, raising “FixtureCreated” exception to fail the test and inform of the path of the created fixture, so that it doesn’t mislead the user in thinking the test passed with an existing fixture,
  • if the fixture exists, it’ll run dumpdata on the models concerned and GNU diff it against the fixture, if there’s any output it’ll be raised in the “DiffFound” exception, failing the test and printing the diff.



    from django import TransactionTestCase
    from dbdiff.fixture import Fixture

    class YourImportTest(test.TransactionTestCase):
        reset_sequences = True

        def test_your_import(self):


The first time, it will raise a FixtureCreated exception, and the test will fail. This is to inform the user that the test didn’t really run. On the next run though, it will pass.

If any difference is found between the database and the test fixture, then diff() will return the diff as outputed by GNU diff.

See tests and docstrings for crunchy details.


MySQL, SQLite and PostgreSQL, Python 2.7 and 3.4 are supported along with Django 1.7 to 1.10 - it’s always better to support django’s master so that we can upgrade easily when it is released, which is one of the selling points for having 100% coverage.


Install django-dbdiff with pip and add dbdiff to INSTALLED_APPS.

Django model observer

It is interresting to note that a related, perhaps sort-of similar app exists:

They trust us