Inconsitent behaviour of filter() with multi-valued relations.

[Łukasz Rekucki]

This is probably a duplicate as I recall seeing a similar issue, but I couldn't find it, so here it goes:

We have 2 models (lets say, Company and Person), connected thru a M2M table (Membership) with additional attribute "active".

The following queries work as documented:

base = Company.objects.all()

# 1) select all companies in which person 1 is active
base.filter(membership_set__active=True, membership_set__person=1)

# 2) select all companies in which some membership is active or person '1' is a member
base.filter(membership_set__active=True).filter(membership_set__person=1)

But if you apply the filter() directly to the manager:

Company.objects.filter(membership_set__active=True, membership_set__person=1)

The behaviour is the same as query 2), while it should work like 1).

[Christopher Medrela]

I cannot reproduce it. Could you attach your models.py and example set of Company, Person and Membership_set objects please?

[Łukasz Rekucki]

Ugh, my bad for not testing this with a different set of model. Now that I did, it turns out to be fault of "eav-django" (​https://bitbucket.org/neithere/eav-django/src/df6fac3e18be/eav/managers.py#cl-53), which redefines filter() on the manager (but not the QuerySet, wth?). Thanks for the sanity check and sorry for the noise.