Cannot fill parent model instance in cache

[Vlastimil Zíma]

Continue of #14371
I tried again in django trunk (tested also in 1.2.4)

I have following models (in SQLite3)

class Parent(models.Model):
    parent_int = models.IntegerField()

class Child(Parent):
    child_int = models.IntegerField()

In [1]: from django.db import connection

In [2]: from testmodule.models import Parent, Child

In [3]: Child.objects.create(parent_int = 1, child_int = 2)
Out[3]: <Child: Child object>

In [4]: Child.objects.get(parent_int = 1, child_int = 2).parent_ptr
Out[4]: <Parent: Parent object>

In [5]: connection.queries
Out[5]: 
[{'sql': u'INSERT INTO "testmodule_parent" ("parent_int") VALUES (1)',
  'time': '0.001'},
 {'sql': u'INSERT INTO "testmodule_child" ("parent_ptr_id", "child_int") VALUES (2, 2)',
  'time': '0.000'},
 {'sql': u'SELECT "testmodule_parent"."id", "testmodule_parent"."parent_int", "testmodule_child"."parent_ptr_id", "testmodule_child"."child_int" FROM "testmodule_child" INNER JOIN "testmodule_parent" ON ("testmodule_child"."parent_ptr_id" = "testmodule_parent"."id") WHERE ("testmodule_parent"."parent_int" = 1  AND "testmodule_child"."child_int" = 2 )',
  'time': '0.000'},
 {'sql': u'SELECT "testmodule_parent"."id", "testmodule_parent"."parent_int" FROM "testmodule_parent" WHERE "testmodule_parent"."id" = 2 ',
  'time': '0.000'}]

In [6]: Child.objects.select_related('parent').get(parent_int = 1, child_int = 2).parent_ptr
Out[6]: <Parent: Parent object>

In [7]: connection.queries
Out[7]: 
[{'sql': u'INSERT INTO "testmodule_parent" ("parent_int") VALUES (1)',
  'time': '0.001'},
 {'sql': u'INSERT INTO "testmodule_child" ("parent_ptr_id", "child_int") VALUES (2, 2)',
  'time': '0.000'},
 {'sql': u'SELECT "testmodule_parent"."id", "testmodule_parent"."parent_int", "testmodule_child"."parent_ptr_id", "testmodule_child"."child_int" FROM "testmodule_child" INNER JOIN "testmodule_parent" ON ("testmodule_child"."parent_ptr_id" = "testmodule_parent"."id") WHERE ("testmodule_parent"."parent_int" = 1  AND "testmodule_child"."child_int" = 2 )',
  'time': '0.000'},
 {'sql': u'SELECT "testmodule_parent"."id", "testmodule_parent"."parent_int" FROM "testmodule_parent" WHERE "testmodule_parent"."id" = 2 ',
  'time': '0.000'},
 {'sql': u'SELECT "testmodule_parent"."id", "testmodule_parent"."parent_int", "testmodule_child"."parent_ptr_id", "testmodule_child"."child_int" FROM "testmodule_child" INNER JOIN "testmodule_parent" ON ("testmodule_child"."parent_ptr_id" = "testmodule_parent"."id") WHERE ("testmodule_parent"."parent_int" = 1  AND "testmodule_child"."child_int" = 2 )',
  'time': '0.000'},
 {'sql': u'SELECT "testmodule_parent"."id", "testmodule_parent"."parent_int" FROM "testmodule_parent" WHERE "testmodule_parent"."id" = 2 ',
  'time': '0.000'}]

You can see that query is correct for selecting parent, but it's instance is not created in cache as usual and that leads to another select.

[Russell Keith-Magee]

Ok - thanks for the extra debug information. Looks like there is something going wrong here. As I indicated on #14371, the select_related() isn't necessary, but you've clearly shown that the parent object isn't being prepopulated as I said, either.

As a matter of procedure -- this is the sort of situation where it would have been ok to reopen the old ticket. It's the same problem, but by providing more debugging information you can demonstrate that the assertion I made was wrong. If that's the case, its perfectly acceptable to reopen an old ticket.

[Aymeric Augustin]

See #16043 for a similar bug that has a test case.

[Tim Graham]

I closed #26523 as a duplicate (but it should be confirmed when solving this ticket).

[Tim Graham]

​PR with some comments for improvement.

[Tim Graham <timograham@…>]

In 38575b00:

Fixed #15250 -- Avoided extra query on some multi-table inheritance queries.

Thanks marekw2143 for the initial patch and carljm for support.

[Tim Graham <timograham@…>]

In 3e9f769:

Refs #15250 -- Added docstring for ForwardOneToOneDescriptor.

[Simon Charette <charette.s@…>]

In 0bbab97:

Refs #15250 -- Removed an obsolete comment regarding MTI cascade deletion.