GROUP BY behavior change in Django 2.0 with annotate/extra

[Michael Soderstrom]

In a 1.11 branch I had a complex raw sql snippet added to a query set via annotate.  

When I updated Django to 2.0, the query set behavior changed to include the raw sql in the GROUP BY which destroyed the performance.

The same behavior occurred using annotate and extra.

I was able to hack around it by overwriting the get_group_by_cols on my RawSQL instance to return an empty list, but that seems wrong.

Is the new behavior a bug or a feature and is there a less hacky way of handling this?

[Tim Graham]

I'd use git bisect to find the commit where the behavior changed in Django. That will likely yield some insight.

https://docs.djangoproject.com/en/dev/internals/contributing/triaging-tickets/#bisecting-a-regression