query `first` method clears cached queryset for prefetch_related

[Aleck Landgraf]

When a cached queryset is iterated through with prefetch_related, if .first() is called during the iteration, a new DB query is fired off. This is potentially due to the LIMIT 1 [:1] or the order_by added to the queryset with the first method.

​https://github.com/django/django/blob/master/django/db/models/query.py#L518

ex.

from django.db import connection
connection.queries = []

pizzas = Pizza.objects.all().prefetch_related('toppings')
print len(connection.queries)  # 2

for pizza in pizzas:
    first_topping = pizza.toppings.first()

print len(connection.queries)  # 2 + number of pizzas

This fixes in my code at least:

from django.db import connection
connection.queries = []

pizzas = Pizza.objects.all().prefetch_related('toppings')
print len(connection.queries)  # 2

for pizza in pizzas:
    try:
        first_topping = pizza.toppings.all()[0]
    except IndexError:
        first_topping = None

print len(connection.queries)  # 2

[Thomas C]

In the first code example, first() is used on an unordered queryset. As first() is intended to be deterministic, it isn't surprising that it tries to add an order condition (which default to order_by('pk').

Your second code example might work, but it isn't deterministic either as there is no guarantee on the ordering of pizza.toppings.all().

Do you need an arbitrary element or the first one ? If you need the first one, there should be some notion of ordering involved, otherwise your second solution is fine.

Also, you might wanna have a look at ​https://docs.djangoproject.com/en/1.7/ref/models/queries/#django.db.models.Prefetch (I've never used it myself but it should allow more complex prefetch_related queryset, including ORDER clause as described here ​https://docs.djangoproject.com/en/1.7/ref/models/querysets/#prefetch-related).

[Tim Graham]

tchaumeny has explained the issue and I don't see any bug here.