Thanks Daniel Wiesmann for spotting the issue.

Thanks Jacob Rief for the report and Tim Graham for the review.

This issue was fixed by refs #24464.

This may cause some backwards compatibility issues, but may also
resolve security issues in third party projects that fail to heed warnings
in our documentation.

Thanks Markus Holtermann for help with tests and docs.

The query used a construct of qs.annotate().values().aggregate() where
the first annotate used an F-object reference and the values() and
aggregate() calls referenced that F-object.

Also made sure the inner query's select clause is as simple as possible,
and made sure .values().distinct().aggreate() works correctly.

Changed --noinput option in makemigrations to suppress all user prompts,
not just when combined with --merge.

Added explanation on why build_absolute_uri always enforces the
request's scheme.

Set apps.ready to False when rendering multiple models. This prevents
that the cache on Model._meta is expired on all models after each time a
single model is rendered. Prevented that Apps.clear_cache() refills the
cache on Apps.get_models(), so that the wrong value cannot be cached
when cloning a StateApps.

Fixed #24456 -- Update modelforms.txt table of Form/Model refs

Added links to the Form/Model table to the field reference docs to
point to their individual field references

When altering from e.g. an IntegerField to a ForeignKey, Django didn't
add a constraint.

Refs #22267.
Thanks Shai Berger for spotting the issue and Tim Graham for the
initial patch.

This test failed because a different error code is raised on Windows when
opening a directory. Since the particular message isn't as important to this
test as the fact that the original IOError is reraised, this fixes the test
by making the assertion more generic.

Fixes commit 87144036.