inconsistent values('col1', 'col2').distinct().count()

[David S. <davidschein@…>]

A ValueQuerySet with selected columns and using distinct returns a count that is accurate only if _result_cache exists--because it can get its len().
If _result_cache does not exists then the SQL to fetch to count is inaccurate because it relies on the PK rather than the selected columns.

For instance:

>>> cs = Course.objects.filter(subject__code="ACC", course_num=207).values('subject', 'course_num')
>>> cs
[{'course_num': '207', 'subject': 'ACC'}, {'course_num': '207', 'subject': 'ACC'}]
>>> cs.count()
2
>>> cs.distinct().count()
2L
>>> csdistinct = cs.distinct()
>>> csdistinct.count()
2L
>>> csdistinct
[{'course_num': '207', 'subject': 'ACC'}]
>>> csdistinct.count()
1

The attached patch has been tested and works on Postgres and Oracle and, though the documentation explicitly says it should work, it seems to fail with SQLite (at least testing in its shell.)

[David S. <davidschein@…>]

patch

[Chris Beaven]

Duplicate of #2939. My patch there is pretty much the same as this (but abstracted) and includes tests