foreign key _id optimisation leads to incorrect descriptor value

[Will Hardy]

A common optimisation (also recommended in the docs: ​https://docs.djangoproject.com/en/1.8/topics/db/optimization/#use-foreign-key-values-directly) is to use MyModel.relatedfield_id = 5 for setting a foreign key, when you want to avoid querying the database.

However, if the instance is then later used (accidentally or occasionally, that would defeat the point of the optimisation), the value of MyModel.relatedfield can be incorrect. I think inefficient would be a better default than incorrect, because the mistake is subtle.

I imagine the ForwardManyToOneDescriptor could check the primary key of the cache value being returned against the primary key on the model instance and if there is a mismatch, then the cache value should be deleted and a new query should be made.

>>> instance = MyModel(relatedfield_id=1)
>>> print(instance.relatedfield.pk)
1
>>> instance.relatedfield_id = 2
>>> print(instance.relatedfield.pk)
1
>>> instance.save()
>>> print(instance.relatedfield.pk)
1

[Tim Graham]

I thought there might be another ticket where this was proposed and possibly rejected as "too magical", but I'm not going to do a search now. The idea seems reasonable at first glance.

[Paulo]

PR: ​https://github.com/django/django/pull/9122

[Tim Graham <timograham@…>]

In ee493061:

Fixed #27710 -- Made Model.save() invalidate cached, stale relations after a primary key assignment.