#23589 closed Bug (invalid)
Django 1.7 filter Q m2m bug
Reported by: | Grafumbly | Owned by: | nobody |
---|---|---|---|
Component: | Database layer (models, ORM) | Version: | 1.7 |
Severity: | Normal | Keywords: | |
Cc: | Triage Stage: | Unreviewed | |
Has patch: | no | Needs documentation: | no |
Needs tests: | no | Patch needs improvement: | no |
Easy pickings: | no | UI/UX: | no |
Description (last modified by )
The behavior of Q objects has changed in an undocumented and breaking way for m2m relationships in Django 1.7 from the behavior in 1.6.7, returning different query results depending on the Django version.
In the case where you want to select a primary object that has related objects of one type, but none of another, you could use a Q like this:
q = Q(Q(relatedobject__someparam=True), ~Q(relatedobject__someparam=False))
results = PrimaryObject.objects.filter(q)
Using this Q in a filter in 1.6.7 would give you all primary objects that have relatedobjects with someparam=True and omit any primary objects that also had a relatedobject with someparam=False. In 1.7 these objects are returned despite the ~Q(relatedobject__someparam=False)
.
This gist provides a sample to illustrate the problem:
https://gist.github.com/scottsexton/375f7869839a98593695
Change History (4)
comment:1 by , 10 years ago
Description: | modified (diff) |
---|
comment:2 by , 10 years ago
Resolution: | → invalid |
---|---|
Status: | new → closed |
The interpretation of the query is that PrimaryObject must have a *single* relatedobject row for which (someparam=True AND NOT someparam=False). Obviously if someparam=True, then someparam is also not False for that row. The 1.7 results seem correct to me. (See https://docs.djangoproject.com/en/1.7/topics/db/queries/#spanning-multi-valued-relationships for how Django handles filters to multivalued relationships).
I believe you will get correct results by using .filter(Q(relatedobject__someparam=True)).filter(~Q(relatedobject__someparam=False))
.
comment:3 by , 10 years ago
The documentation hasn't changed but the behavior has. If this is working as intended in 1.7, is it a bug in 1.6 or does the documentation just need to be updated?
comment:4 by , 10 years ago
Yes, this was a bug in 1.6. This is an unfortunate bug, as you might have gotten correct results. That is, the query worked, but didn't match the documentation. The query produced results for .filter(Q(relatedobject__someparam=True)).filter(~Q(relatedobject__someparam=False))
.
It would be helpful if you could create a test for Django's test suite that demonstrates the issue (ideally reusing existing models if you can find appropriate ones) and bisect to the commit that introduced the regression.