I've been spending some time today doing research on how to best use django's Q objects. Then I did a bit of testing myself and decided to put this whole thing together for future's sake, you know, for the time when I'll find myself lost and in need some coding recipe.
First things first: the official documentation.
Now... let's get started by resurrecting the simple model used in the django Tutorial, Polls, and adding a few instances to play with:
# ==> in models.py
from django.db import models
import datetime
class Poll(models.Model):
question = models.CharField(max_length=200)
pub_date = models.DateTimeField('date published')
# ==> add some instances from the shell
>>> p = Poll(question='what shall I make for dinner', pub_date=datetime.date.today())
>>> p.save()
>>> p = Poll(question='what is your favourite meal?', pub_date=datetime.datetime(2009, 7, 19, 0, 0)) #one year ago
>>> p.save()
>>> p = Poll(question='how do you make omelettes?', pub_date=datetime.date.today())
>>> p.save()
In a nutshell, the main reason why you might want to use them is because you need to do complex queries, for example involving OR and AND statements a the same time.
Imagine that you want to retrieve all polls whose question contains the word 'dinner' OR 'meal'. The usual 'filter' option lets you do the following:
Poll.objects.filter(question__contains='dinner').filter(question__contains='meal')
But this is not really what we want, is it? We just got a concatenation of AND statements (= all constraints need to be satisfied), while we wanted an OR (=at least one of the constraints needs to be satisfied). This is where Q objects become useful:
>>> from django.db.models import Q
# a simple query
>>> Poll.objects.filter(Q(question__contains='dinner'))
[<Poll: what shall I make **for** dinner>]
# two constraints linked by AND
>>> Poll.objects.filter(Q(question__contains='dinner') & Q(question__contains='meal'))
[]
# two constraints linked by OR - that's it!
>>> Poll.objects.filter(Q(question__contains='dinner') | Q(question__contains='meal'))
[<Poll: what shall I make **for** dinner>, <Poll: what **is** your favourite meal?>]
Note that if you don't specify any logical connector the Q sequence is implicitly interpreted as AND:
# no logical connector is interpreted as AND
>>> Poll.objects.filter(Q(question__contains='dinner'), Q(question__contains='meal'))
[]
Things start getting more interesting when creating complex queries:
# eg (A OR B) AND C:
>>> Poll.objects.filter((Q(question__contains='dinner') | Q(question__contains='meal')) & Q(pub_date=datetime.date.today()))
[<Poll: what shall I make **for** dinner>]
Now, it is likely that you want to build up queries like the ones above dynamically. This is another place where Q objects can save lots of time....for example, when constructing search engines or faceted browsers where the interface lets a user accumulate search filters in an incremental manner.
One way to handle this situation is by creating lists of Q objects, and then combining them together using python's operator and reduce methods:
>>> import operator
# create a list of Q objects
>>> mylist = [Q(question__contains='dinner'), Q(question__contains='meal')]
# OR
>>> Poll.objects.filter(reduce(operator.or_, mylist))
[<Poll: what shall I make **for** dinner>, <Poll: what **is** your favourite meal?>]
# AND
>>> Poll.objects.filter(reduce(operator.and_, mylist))
[]
Now, if you're building the query dynamically you probably won't know in advance which are the filters you have to use. It is likely instead that you find yourself having to generate a list of Q objects programatically from a list of strings representing queries to your models:
# string representation of our queries
>>> predicates = [('question__contains', 'dinner'), ('question__contains', 'meal')]
# create the list of Q objects and run the queries as above..
>>> q_list = [Q(x) **for** x **in** predicates]
>>> Poll.objects.filter(reduce(operator.or_, q_list))
[<Poll: what shall I make **for** dinner>, <Poll: what **is** your favourite meal?>]
>>> Poll.objects.filter(reduce(operator.and_, q_list))
[]
# now let's append another filter to the query-strings..
>>> predicates.append(('pub_date', datetime.date.today()))
>>> predicates
[('question__contains', 'dinner'), ('question__contains', 'meal'), ('pub_date', datetime.date(2010, 7, 19))]
# .. and the results will change too
>>> q_list = [Q(x) **for** x **in** predicates]
>>> Poll.objects.filter(reduce(operator.or_, q_list))
[<Poll: what shall I make **for** dinner>, <Poll: what **is** your favourite meal?>, <Poll: how do you make omelettes?>]
>>> Poll.objects.filter(reduce(operator.and_, q_list))
[]
Using query expansion syntax you could normally do things like this:
>>> mydict = {'question__contains': 'omelette', 'pub_date' : datetime.date.today()}
>>> Poll.objects.filter(**mydict)
[<Poll: how do you make omelettes?>]
Here you are basically sequencing AND statements which are built dynamically from some string values.
A really cool feature of django ORM is that you can keep doing this even while using Q objects. In other words, you can delegate all the 'normal' stuff to the query expansion mechanism, while instead resorting to Q objects whenever you need more complex queries, such an OR statement.
For example (remember to put the dictionary always in second position):****
# OR plus query expansion
>>> Poll.objects.filter(reduce(operator.or_, q_list), **mydict)
[<Poll: how do you make omelettes?>]
# AND plus query expansion
>>> Poll.objects.filter(reduce(operator.and_, q_list), **mydict)
[]
In the first case, we have one result only because, although the OR constraint on q_list is would match more stuff by itself, _mydict_ is exploded as a series of constraints connected by AND, thus making [Poll: how do you make omelettes?] the only matchable object. In the second** case instead there are no results at all simply because we are asking for a Poll instance that satisfies simultaneously all the constraints (which doesn't exist)!
That's all for now! As I said, I found quite a few very good posts about the subject online and I strongly advise you to check them out too in order to have a clearer picture of how Q objects work.
Cite this blog post:
Comments via Github:
2009