FilterMethod.__call__ calling `method` with value=`<Queryset []>`

[CelestialGuru]

Here's my filterset:

class LockFilterSet(filterset.FilterSet):
    class Meta:
        model = models.Lock
        fields = ()

    grate_widths = filterset.ModelMultipleChoiceFilter(
        field_name="grate_width",
        queryset=configuration_models.SizeClass.objects.all(),
        method="filter_grate_widths",
    )

    def filter_grate_widths(self, queryset, name, value):
        return queryset.filter(Q(**{f"{name}__in": value}) | Q(**{f"{name}__isnull": True}))

If I GET /api/locks/, /api/locks/?, or /api/locks/?grate_widths= my filter_grate_widths method is called. The value is <Queryset []>.

Examining the call stack we see that it came from

# django_filters/filters.py
class FilterMethod:
    def __init__(self, filter_instance):
        self.f = filter_instance

    def __call__(self, qs, value):
        if value in EMPTY_VALUES:
            return qs

        return self.method(qs, self.f.field_name, value)

EMPTY_VALUES is ([], (), {}, "", None).

But of course the issue is that value is <Queryset []>, which is not in EMPTY_VALUES.

I did some digging and this value is generated by the form. The form is django.forms.Form looking at the metaclass code. So I am not sure if this bug is django-filter or django itself.

Yes I can fix this by adding if not value: return queryset, but the source code in FilteMethod suggests that my method should only be called when there is some non-empty value, which my url parameters are (empty).

What am I using

• Python 3.11
• django 4.1.3
• djangorestframework 3.14.0
• django-filter 22.1

[CelestialGuru]

Further study about Django forms:

Setup

from django import forms
from django.contrib.auth import get_user_model

User = get_user_model()

class MyForm(forms.Form):
   users = forms.ModelMultipleChoiceField(queryset=User.objects.all(), required=False)

Empty form

form = MyForm(data={"users":[]})
form.is_valid()
print(form.cleaned_data)  # {'users': <Queryset []>}

Non-empty form

form = MyForm(data={"users":[1]})
form.is_valid()
print(form.cleaned_data)  # {'users': <Queryset [<User: [e-mail]>]>}

It appears that vanilla Django forms return Queryset instances for the ModelMultipleChoiceField. So that leaves the question of why does FilterMethod mentioned earlier return sometimes a Queryset (empty) and other times a list?

The metaclass FilterSetOptions has this line: self.form = getattr(options, "form", forms.Form) which would just be the vanilla Django form class. That leaves django_filters.fields.ModelMultipleChoiceField in question.

Setup, but with django-filter fields

from django import forms
from django.contrib.auth import get_user_model
from django_filters import fields as django_filter_fields

User = get_user_model()

class MyForm(forms.Form):
   users = django_filter_fields.ModelMultipleChoiceField(queryset=User.objects.all(), required=False)

Empty form

form = MyForm(data={"users":[]})
form.is_valid()
print(form.validated_data)  # dict_values([<QuerySet []>])

Non-empty form

form = MyForm(data={"users":[1]})
form.is_valid()
print(form.validated_data)  # dict_values([[<User: [e-mail]>]])

There, I found it. A minimal case demonstrating the difference in django.forms.ModelMultipleChoiceField and django_filters.fields.ModelMultipleChoiceField. Why does the former always return Queryset instances while the latter does not? I do not know yet.

[CelestialGuru]

I think I found out where the problem is. In django.forms.models.py:ModelMultipleChoiceField.clean it has

def clean(self, value):
    value = self.prepare_value(value)
    if self.required and not value:
        raise ValidationError(self.error_messages["required"], code="required")
    elif not self.required and not value:
        return self.queryset.none()
    qs = self._check_values(value)
    self.run_validators(value)
    return qs

The two return paths are either self.queryset.none() or self._check_values(value).

The former is where the empty queryset comes from.

self._check_values(value) calls django_filters.fields.py:ModelMultipleChoiceField._check_values which is

def _check_values(self, value):
    ...
    result = list(super()._check_values(value))  # This is where the list vs queryset comes from
    result += [self.null_value] if null else []
    return result

we can clearly see the list() changing the queryset to a list. I don't pretend to know what result += [self.null_value] if null else [] is for and I can see that is why you needed to change a queryset to a list to be able to append self.null_value if needed. However, like mentioned in the original post, this does not work with EMPTY_VALUES in FilterMethod. value in EMPTY_VALUES is False for the empty queryset. I might suggest changing the logic to if not value: return qs, but I can see how that might not work properly with valid numeric values like 0 or 0.0.

Solution

class ModelMultipleChoiceFilter(QuerySetRequestMixin, MultipleChoiceFilter):
    field_class = ModelMultipleChoiceField

    def clean(self, value):
        # When a 'method' argument is passed, the proxy FilterMethod class is used
        # and first checks if the value is in EMPTY_VALUES, calling 'method' only if it is not.
        # When value is empty, super() returns an empty queryset. `value in EMPTY_VALUES is False`.
        # When value is not empty, super() returns a list. `value in EMPTY_VALUES is True`.
        # 
        # The inconsistency is fixed by calling list() on whatever super() returns. That way
        # FilterMethod will always get a list and `value in EMPTY_VALUES` will work as intended.
        return list(super().clean(value))