Sum() doesn't seem to respect .distinct() in the queryset filter before .aggregate() when filtering by m2m.

[Mark]

Hello,
I setup a minimal project here: ​https://github.com/mcagl/minimal_django_sum_test to demonstrate the problem.

I'm working with Django 1.6.x and I noticed something that I don't understand.

Edit: I checked also with the last stable version, 1.7.4 downloaded right from the website, and the problem is still there.

As you can see from the github repository, I have a Tag model and a Row model with a m2m towards Tag and a DecimalField called amount.

If I filter Row objects for more than one Tag, and there is/are Row objects that have more than one Tag among the one filtered by, Sum('amount') counts it/them once per Tag, even if I use distinct().

Please note also that I assert, in the test, that the filtered queryset is composed by three Row objects, as expected, but the next assert fails, with 40 != 30.

I added a test that instead of aggregate(Sum('amount')) does sum([x.amount for x in rows]) which passes.

Is this a bug in Sum() or am I missing something?

Kind regards,
Mark