TestCase can't define an empty list of fixtures

[Lorenzo Gil Sanchez]

This could be useful if you have a TestCase base class that define some very basic fixtures that are used across all your tests but then, there is a few particular tests that do not want such fixtures.

Let's say you have something like:

class BaseTestCase(TestCase):

  fixtures = ['basic_data.json']


class NoFixturesTestCase(BaseTestCase):

  fixtures = []

If you do this you will get this error:

Traceback (most recent call last):
  File "/home/lgs/proyectos/personal/django/django/test/testcases.py", line 173, in __call__
    self._pre_setup()
  File "/home/lgs/proyectos/personal/django/django/test/testcases.py", line 716, in _pre_setup
    self._fixture_setup()
  File "/home/lgs/proyectos/personal/django/django/test/testcases.py", line 848, in _fixture_setup
    'skip_validation': True,
  File "/home/lgs/proyectos/personal/django/django/core/management/__init__.py", line 159, in call_command
    return klass.execute(*args, **defaults)
  File "/home/lgs/proyectos/personal/django/django/core/management/base.py", line 289, in execute
    output = self.handle(*args, **options)
  File "/home/lgs/proyectos/personal/django/django/core/management/commands/loaddata.py", line 49, in handle
    "No database fixture specified. Please provide the path "
CommandError: No database fixture specified. Please provide the path of at least one fixture in the command line.

You can use this hack as a workaround:

class NoFixturesTestCase(BaseTestCase):

  fixtures = ['non-existing-file.json']

And now you will get a warning instead of an error.

I'll create a Github branch with a test showing this behaviour and a fix for it.

[Marc Tamlyn]

Is this problem solved by setting fixtures = None?

[Lorenzo Gil Sanchez]

No, it is not solved by setting fixtures = None

The Github branch with the testcase to reproduce this error and the fix is located at: ​https://github.com/lorenzogil/django/compare/ticket_21089

[Baptiste Mispelon]

This looks useful, but I have some questions about the patch.

Is there a particular reason that we're using getattr instead of simply defining a fixtures attribute on TransactionTestCase (see attached diff)?

[Lorenzo Gil Sanchez]

I don't know about the reasons behind getattr. That code was there before, I just added an extra check. Your patch would work too so I can update my branch if that option is preferred.

[Baptiste Mispelon <bmispelon@…>]

In abb10db06fb2ecb3e897462ec72417d10b39b8a4:

Fixed #21089 -- Allow TransactionTestcase subclasses to define an empty list of fixtures.

Thanks to lgs for the report and initial patch.

[Lorenzo Gil Sanchez]

Thanks for fixing this so fast :-)