Provide a "force_login" feature for the test client

[Marc Tamlyn]

At present client.login() takes the normal authentication parameters, runs them through the authentication backends and then logs the user in. A lot of my sites currently require the user to be logged in on every page. Consequently, every unit test which uses the client to check a page creates a user and then logs them in. When I create the user with a helper factory, I have to be careful to record the raw password of that user so I can pass it back through the hashing algorithms.

I'd like to provide client.simple_login(user), which simply sets up the test client's session so that the user in question is logged in, without going through the hashing algorithms. This would include most of the business logic from the current login() method, but without the authenticate() check. This would be a minor performance improvement which probably doesn't justify itself in most cases, the main benefit is ease-of-use.

Related in a way to #15179

[Tim Graham]

Would it make sense to add a parameter to login rather than making it a separate method?

[Marc Tamlyn]

The issue with that is that at present login takes **kwargs which get passed to the auth backend.

[alasdair]

I had a go at this ticket at the Djangoweekend sprints. You can see my work so far in my branch on github: ​https://github.com/alasdairnicol/django/compare/ticket_20916. I'd welcome any comments, either here or on github.

As mjtamlyn says, authentication credentials are passed to the authenticate method, which makes it tricky to add the desired functionality to the existing login method. We could change the function signature to

def login(*args, **kwargs):
    if args:
        user = args[0]
    else:
        # use credentials
        credentials = **kwargs

but it feels hacky, so I decided to try implementing a new method simple_login.

I found that we have to specify user.backend, even though the authentication backend is not used to log the user in. I have documented that backend must be in AUTHENTICATION_BACKENDS (if not, simple_login currently returns True, but any requests to a login protected view will be redirected to the login page). Perhaps I should raise a ValueError if the backend is not in AUTHENTICATION_BACKENDS.

[Marc Tamlyn]

The patch is a good approach, but there is now a large amount of duplication between login and simple_login. Also, the new functionality needs both documentation and release notes.

[Alasdair]

That's a good point about the duplication. I'll factor out the common functionality.

There is already documentation and an entry in the release notes in my patch. I can work on these if they need improvement.

[Tim Graham]

I wonder if it might be nice to complete #20915 (remove django.test.client dependency on django.contrib.auth) before adding this which is more of that.

[Jason Filipe]

I've tried a different approach where a new authentication backend is added when setting up the test environment.

The changes are here: ​https://github.com/jfilipe/django/compare/django:master...simple-login

If you guys like this approach I can add some docs.

[jarshwah]

I really like the approach you've taken here with the custom backend rather than polluting the standard auth system. I'm not sure that you'd want to set that backend as a default backend for the test suite though. Perhaps some kind of context manager could be used or just the standard @override_settings.

[Jason Filipe]

Technically it's not setting it as the default, it's just adding it as another backend the auth system will try and use.

[Julien Phalip]

Thank you for your work on this ticket.

I feel a bit uncomfortable adding an inherently insecure authenticate backend to core. This feature only really makes sense for testing, so I'd personally prefer to add it directly to the test client.

Also, the term "simple" seems a little vague to me. Something like force_login(user), or login(user, force=True) would seem more explicit.

I'd appreciate getting another core dev's opinion about the above.

By the way, could you re-create the PR against the official Django repository on github (instead of your own fork)?

Thanks!

[Jason Filipe]

PR against Django's repo has been created: ​https://github.com/django/django/pull/2570.

That's a valid point, Exposing SimpleLoginBackend in the auth contrib package could open it up to be abused by others.

[alasdair]

jfilipes approach looks promising, so I have unassigned the ticket from myself.

Also, the term "simple" seems a little vague to me. Something like force_login(user), or login(user, force=True) would seem more explicit.

Adding force=True to the login method would be backwards incompatible. As mjtamlyn says earlier in the comments, login currently takes **kwargs which get passed to the auth backend.

[Jon Dufresne]

I started looking into this ticket after the recommendation in ticket #24987. I have started a WIP branch based on the ideas and PR of jfilipes. Early feedback and comments are welcome. If the approach is well received I'll try to continue down that path to completion.

​https://github.com/django/django/pull/4865

Thanks.

[Jon Dufresne]

Received a positive review from mjtamlyn, so removing some flags.

More reviews and feedback is welcome. Thanks.

[Tim Graham <timograham@…>]

In b44dee16:

Fixed #20916 -- Added Client.force_login() to bypass authentication.