Cache back-references on OneToOne fields

[Shai Berger]

Today, back-references on OneToOneField's are not cached. In more detail, assume:

    class A(model):
        pass

    class B(model):
        a = OneToOneField(A)

    a1 = A.objects.get(...) # assume a1 has a related B record
    b1 = B.objects.get(...)

Today, each of the expressions a1.b.a and b1.a.b generates two database queries, and as a result, both (a1.b.a is a1) and (b1.a.b is b1) are false. This is surprising and can be both a performance issue and a correctness issue -- therefore classified as "Bug", although it is borderline "optimization".

(Of course, nobody in their right mind actually writes code like a1.b.a; the issue arises e.g. by function f taking a1 and calling g(a1.b), where g does something with b.a).

Attached is a patch with tests that failed on trunk r17377.

[Shai Berger]

Patch adding regression tests for OneToOneField back-reference caching

[Shai Berger]

And here is the fix. This makes the tests attached above pass, and doesn't seem to break any others. Also, a similar solution has been exercised with no problems in a Django-1.2 based solution.

[Shai Berger]

I've just checked: both patches also apply and work against branch 1.3.X@17370. They don't break any tests (one test, regressiontests.generic_views.dates.WeekArchiveViewTests.test_week_view_allow_future, breaks for me on the branch, with or without the patch).

[Aymeric Augustin]

Discussed on the mailing list here: ​http://groups.google.com/group/django-developers/browse_thread/thread/b8dd06eb13ccc8c2

See comments by Adrian and Carl in particular.

[Shai Berger]

Hi,

Just saw this mentioned in the release notes as one of the new features of Django 1.5. I haven't tested it myself, but I guess this bug can be closed now.

[Aymeric Augustin]

Indeed. This was reported in several duplicate tickets and I missed this one when I fixed the issue. Thanks for noticing!