first() adds id field to group by clause on aggregation queries

[John Gresty]

When evaluating an aggregation query, for example:

MyModel.objects.values('group').annotate(value=Sum('value'))

works as expected. However adding first() to the end will add the id field into the resulting queryset and as such the annotated value will be the value of the first row, instead of the sum of the group.

>>> MyModel.objects.create(value=10)
<MyModel: MyModel object>
>>> MyModel.objects.create(value=20)
<MyModel: MyModel object>
>>> MyModel.objects.values('group').annotate(value=Sum('value'))
<QuerySet [{'group': 0, 'value': 30}]>
>>> MyModel.objects.values('group').annotate(value=Sum('value')).first()
{'group': 0, 'value': 10}

[Simon Charette]

The first() method requires an ordering to be specified else it would return non-deterministic results hence why it's using order_by('pk') when the queryset isn't ordered.

This is a bit similar to #10574 except the implicit ordering is added by first() and not _meta.ordering so I'd suggest ​we link to the aggregation section mentioning interaction with default ordering or adjust the documention to account for that.

IMHO it would have been better for first() to raise an exception when called on an unordered queryset instead of implicitly choosing pk but in your case I assume you want to either order_by('group') or 'value' before calling first().

[Botond Béres]

I've added a small patch that links from the first() definition to the ​aggregation section which mentions interaction with default ordering

[Tim Graham <timograham@…>]

In 95a14cf:

Fixed #28395 -- Doc'd that QuerySet.first() can affect aggregation queries.

[Tim Graham <timograham@…>]

In 854aec48:

[2.0.x] Fixed #28395 -- Doc'd that QuerySet.first() can affect aggregation queries.

Backport of 95a14cfc47de5762ddb1400e6e5152f9e3172657 from master