URLPathVersioning should maybe remove the version_param kwarg from subsequent view dispatch?

[shabble]

I've been playing around with getting URLPathVersioning working properly, and have discovered what might be a bug, or maybe just undocumented (or obvious and expected) behaviour.

urlconf contains:

from rest_framework.authtoken.views import obtain_auth_token
# ...

api_patterns = [
    url(r'^', include(router.urls), name='api-root'),
    url(r'^auth-token/', obtain_auth_token, name='api-auth-token'),
]
urlpatterns = [
    url(r'^api/(?P<version>v\d{1,})/', include(api_patterns)),
    # ...
]

When testing, with something like:

url = reverse('api-auth-token', kwargs=dict(version='v1'))
self.client.post(url, data, format='json')

I get an exception that ultimately ends in:

  File "/usr/local/lib/python2.7/dist-packages/rest_framework/views.py", line 463, in dispatch
    response = handler(request, *args, **kwargs)
TypeError: post() got an unexpected keyword argument 'version'

Investigation shows that the ObtainAuthToken.post() method doesn't accept any kwargs, so it's clearly failing because it's getting version passed in by default from the rest of the versionating mechanism.

Simplest fix would be to just fix the definition of ObtainAuthToken.post(self, request, **kwargs),
but it made me wonder if it'll be an issue for other views that are receiving a somewhat unexpected kwarg, and which is by intent already available to them in request.version.

If there's no good reason to keep it around after extracting it, would it make sense to clean it from the
view args in views.APIView somewhere in the initial pre-dispatch phases, maybe as simple as a
kwargs.pop(request.versioning_scheme.version_param, None) (with some checks around it) in the appropriate place?

I'm currently subclassing authtoken.ObtainAuthToken to do exactly this in post(), but I figured it might make sense to raise a a more general issue.

[xordoquy]

I'd rather fix the AuthToken to accept the extra kwarg.

[xordoquy]

Thanks for reporting.

[tomchristie]

I'd rather fix the AuthToken to accept the extra kwarg.

Seems reasonable.