filtering on an EmptyQuerySet raises an Exception

[anonymous]

Trying to filter an instance of EmptyQuerySet causes an Exception to be raised:

>>> from django.db.models.query import *
>>> q = EmptyQuerySet()
>>> q.filter(x=10)
Traceback (most recent call last):
  File "<console>", line 1, in <module>
  File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/django/db/models/query.py", line 370, in filter
    return self._filter_or_exclude(False, *args, **kwargs)
  File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/django/db/models/query.py", line 388, in _filter_or_exclude
    clone.query.add_q(Q(*args, **kwargs))
  File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/django/db/models/sql/query.py", line 1068, in add_q
    can_reuse=used_aliases)
  File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/django/db/models/sql/query.py", line 954, in add_filter
    opts = self.get_meta()
  File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/django/db/models/sql/query.py", line 127, in get_meta
    return self.model._meta
AttributeError: 'NoneType' object has no attribute '_meta'
>>> 

I would think that no matter what you pass in as arguments to filter(), it should always return EmptyQuerySet.

[John-Scott Atlakson]

I can confirm that the above error occurs with Django version 0.97-pre-SVN-7477

With Django version 0.97-pre-SVN-7476 I get the expected behavior:

>>> from django.db.models.query import EmptyQuerySet
>>> q = EmptyQuerySet()
>>> q.filter(x=10)
[]

[Johannes Dollinger]

Do QuerySets without an associated Model make sense?

Solution proposals:

# Keep EmptyQuerySet, require the model argument: nothing to do.
# Keep EmptyQuerySet, make the model argument optional: overwrite all methods that return a derived QuerySet and return self (since QuerySets are immutable). This would be uncomfortable to maintain.
# Remove EmptyQuerySet, add an is_empty flag to Query: given that Query is already 1.5kloc ... (allthough enabling something like query.mark_empty() in add_to_query() methods would be nice).

Additionally: If EmptyQuerySet doesn't require the model argument, there might as well be a single instance. Otherwise one instance per model would do.

I'd give this a try, as long as the tempting third option isn't choosen (I'm not comfortable with the Query code).

[Malcolm Tredinnick]

This isn't a valid problem, since creating an EmptyQuerySet directly is not something you ever do. It's an instance that is created (correctly) using the none() method on other querysets or model managers. Internal bits of code can sometimes create QuerySet classes and subclasses without immediately setting the model, which is why it's permitted, but this bug report is just a case of "if you use the code incorrectly, you get nonsense results." There is nothing in the documentation to suggest that what you tried here should work, whereas working with the result of a none() call does (or should) work correctly.

[Adam Gomaa]

I believe this should be reopened because of the AnonymousUser use case.

django.contrib.auth.models.AnonymousUser provides a
pseudo-related-manager under AnonymousUser.groups. This makes a call
like request.user.groups.filter(name__in=group_names) unsafe - if
you get an AnonymousUser object instead of a User, you'll get an
exception instead of what I think would be the correct behavior - an
EmptyQuerySet. It may be that django.contrib.auth is using
EmptyQuerySet incorrectly - as a public API instead as a temporary
pre-real-QuerySet as Malcom indicates. But, I don't think there's a
way to fix that in a backwards-compatible manner; after all,
AnonymousUser is not a model, and switching to a real QuerySet
is not an option.

The other alternative I can see is for programmers to either typecheck
request.user or check request.user.is_anonymous() before
attempting to filter on request.user.groups or
request.user.user_permissions. I don't think that this is either
correct or friendly.

[Malcolm Tredinnick]

In light of AdamG's observation, which I happened to notice going past (start a thread on django-dev next time, rather than commenting on a closed ticket, please), reopening. That's a valid case.

The fix is to make every queryset operation on EmptyQuerySet pretty much just return self.

[Jacob]

Punting from 1.1: this is annoying, but there's an easy (if ugly) workaround for now.

[aljosa]

EmptyQuerySet methods all, filter, exclude, complex_filter, select_related, annotate, order_by, distinct, extra, reverse, defer and only "return self"

[taylormarshall]

regression tests for 7235

[Adrian Holovaty]

(In [12147]) Fixed #7235 -- EmptyQuerySet no longer raises and exception when it's filter()ed (along with some other QuerySet methods). Thanks, taylormarshall

[Jacob]

Milestone 1.2 deleted