django.db.models.query.Row is not pickleable.

[Vitaliy Yelnik]

The new named parameter of QuerySet.values_list() was released In Django 2.0 (#15648).
But resulted namedtuple-s can't be pickled:

class ModelA(models.Model):
    value = models.CharField(max_length=12)

In [12]: row = ModelA.objects.values_list('id', 'value', named=True).first()

In [14]: type(row)                                                                                                                                                                                                                           
Out[14]: django.db.models.query.Row

In [16]: pickle.dumps(row)                                                                                                                                                                                                                   
PicklingError: Can't pickle <class 'django.db.models.query.Row'>: attribute lookup Row on django.db.models.query failed

In particular, as a result, such requests do not work with cacheops package.

[Davit Tovmasyan]

In addition to the data contained in the instances, pickle also stores a string reference to the original class. This means that Row tuple class should be in query module globals so pickle can find the reference. But besides that, it should have different class names for each model and for each list of values.

[Vitaliy Yelnik]

​PR

[Mariusz Felisiak]

You shouldn't mark your own PRs as ready for checkin.

[Mariusz Felisiak <felisiak.mariusz@…>]

In 981a072d:

Fixed #31843 -- Fixed pickling named values from QuerySet.values_list().