When following a reverse relationship, the cache where the objects pulled from the database are stored is not using the proper variable name for storing the reverse result. Two models both related to the original model will cause the relation to be overwritten on the second access, assuming that both models are related to the original model by the same name (and therefore are forced by the forward relationship to have a related_name attribute to distinguish them).
The solution, provided by Travis Terry (patch attached), is to cache by using the related_name instead of the field name.
Here's an example:
class Test4(models.Model):
b = models.CharField(max_length=10)
class Test5(models.Model):
c = models.CharField(max_length=10)
f2 = models.OneToOneField(Test4, related_name='first_link')
class Test6(models.Model):
d = models.CharField(max_length=10)
f2 = models.OneToOneField(Test4, related_name='second_link')
# This shows how to trigger the error.
t4 = Test4.objects.create(b='Thing1')
t5 = Test5.objects.create(c='obj1', f2=t4)
t6 = Test6.objects.create(d='obj2', f2=t4)
a = Test4.objects.get(pk=t4.id)
Here's what you get:
>>> print a.first_link
Test5 object
>>> print a.first_link.c
obj1
>>> print a.second_link
Test6 object
>>> print a.second_link.d
Traceback (most recent call last):
...
AttributeError: 'Test5' object has no attribute 'd'
Here's what you should get:
>>> print a.first_link
Test5 object
>>> print a.first_link.c
obj1
>>> print a.second_link
Test6 object
>>> print a.second_link.d
obj2
