1
0
mirror of https://github.com/django/django.git synced 2024-11-18 15:34:16 +00:00
django/tests/lookup/tests.py
Jakub Roztocil 7131e14d00 Fixed #22489 -- missing implemenation for search lookup
When custom lookups were added, converting the search lookup to use
the new Lookup infrastructure wasn't done.

Some changes were needed to the added test, main change done by
committer was ensuring the test works on MySQL versions prior to 5.6.
2014-05-05 14:36:15 +03:00

739 lines
35 KiB
Python

from __future__ import unicode_literals
from datetime import datetime
from operator import attrgetter
from unittest import skipUnless
from django.core.exceptions import FieldError
from django.db import connection
from django.test import TestCase, TransactionTestCase, skipUnlessDBFeature
from .models import Author, Article, Tag, Game, Season, Player, MyISAMArticle
class LookupTests(TestCase):
def setUp(self):
# Create a few Authors.
self.au1 = Author(name='Author 1')
self.au1.save()
self.au2 = Author(name='Author 2')
self.au2.save()
# Create a couple of Articles.
self.a1 = Article(headline='Article 1', pub_date=datetime(2005, 7, 26), author=self.au1)
self.a1.save()
self.a2 = Article(headline='Article 2', pub_date=datetime(2005, 7, 27), author=self.au1)
self.a2.save()
self.a3 = Article(headline='Article 3', pub_date=datetime(2005, 7, 27), author=self.au1)
self.a3.save()
self.a4 = Article(headline='Article 4', pub_date=datetime(2005, 7, 28), author=self.au1)
self.a4.save()
self.a5 = Article(headline='Article 5', pub_date=datetime(2005, 8, 1, 9, 0), author=self.au2)
self.a5.save()
self.a6 = Article(headline='Article 6', pub_date=datetime(2005, 8, 1, 8, 0), author=self.au2)
self.a6.save()
self.a7 = Article(headline='Article 7', pub_date=datetime(2005, 7, 27), author=self.au2)
self.a7.save()
# Create a few Tags.
self.t1 = Tag(name='Tag 1')
self.t1.save()
self.t1.articles.add(self.a1, self.a2, self.a3)
self.t2 = Tag(name='Tag 2')
self.t2.save()
self.t2.articles.add(self.a3, self.a4, self.a5)
self.t3 = Tag(name='Tag 3')
self.t3.save()
self.t3.articles.add(self.a5, self.a6, self.a7)
def test_exists(self):
# We can use .exists() to check that there are some
self.assertTrue(Article.objects.exists())
for a in Article.objects.all():
a.delete()
# There should be none now!
self.assertFalse(Article.objects.exists())
def test_lookup_int_as_str(self):
# Integer value can be queried using string
self.assertQuerysetEqual(Article.objects.filter(id__iexact=str(self.a1.id)),
['<Article: Article 1>'])
@skipUnlessDBFeature('supports_date_lookup_using_string')
def test_lookup_date_as_str(self):
# A date lookup can be performed using a string search
self.assertQuerysetEqual(Article.objects.filter(pub_date__startswith='2005'),
[
'<Article: Article 5>',
'<Article: Article 6>',
'<Article: Article 4>',
'<Article: Article 2>',
'<Article: Article 3>',
'<Article: Article 7>',
'<Article: Article 1>',
])
def test_iterator(self):
# Each QuerySet gets iterator(), which is a generator that "lazily"
# returns results using database-level iteration.
self.assertQuerysetEqual(Article.objects.iterator(),
[
'Article 5',
'Article 6',
'Article 4',
'Article 2',
'Article 3',
'Article 7',
'Article 1',
],
transform=attrgetter('headline'))
# iterator() can be used on any QuerySet.
self.assertQuerysetEqual(
Article.objects.filter(headline__endswith='4').iterator(),
['Article 4'],
transform=attrgetter('headline'))
def test_count(self):
# count() returns the number of objects matching search criteria.
self.assertEqual(Article.objects.count(), 7)
self.assertEqual(Article.objects.filter(pub_date__exact=datetime(2005, 7, 27)).count(), 3)
self.assertEqual(Article.objects.filter(headline__startswith='Blah blah').count(), 0)
# count() should respect sliced query sets.
articles = Article.objects.all()
self.assertEqual(articles.count(), 7)
self.assertEqual(articles[:4].count(), 4)
self.assertEqual(articles[1:100].count(), 6)
self.assertEqual(articles[10:100].count(), 0)
# Date and date/time lookups can also be done with strings.
self.assertEqual(Article.objects.filter(pub_date__exact='2005-07-27 00:00:00').count(), 3)
def test_in_bulk(self):
# in_bulk() takes a list of IDs and returns a dictionary mapping IDs to objects.
arts = Article.objects.in_bulk([self.a1.id, self.a2.id])
self.assertEqual(arts[self.a1.id], self.a1)
self.assertEqual(arts[self.a2.id], self.a2)
self.assertEqual(Article.objects.in_bulk([self.a3.id]), {self.a3.id: self.a3})
self.assertEqual(Article.objects.in_bulk(set([self.a3.id])), {self.a3.id: self.a3})
self.assertEqual(Article.objects.in_bulk(frozenset([self.a3.id])), {self.a3.id: self.a3})
self.assertEqual(Article.objects.in_bulk((self.a3.id,)), {self.a3.id: self.a3})
self.assertEqual(Article.objects.in_bulk([1000]), {})
self.assertEqual(Article.objects.in_bulk([]), {})
self.assertEqual(Article.objects.in_bulk(iter([self.a1.id])), {self.a1.id: self.a1})
self.assertEqual(Article.objects.in_bulk(iter([])), {})
self.assertRaises(TypeError, Article.objects.in_bulk)
self.assertRaises(TypeError, Article.objects.in_bulk, headline__startswith='Blah')
def test_values(self):
# values() returns a list of dictionaries instead of object instances --
# and you can specify which fields you want to retrieve.
identity = lambda x: x
self.assertQuerysetEqual(Article.objects.values('headline'),
[
{'headline': 'Article 5'},
{'headline': 'Article 6'},
{'headline': 'Article 4'},
{'headline': 'Article 2'},
{'headline': 'Article 3'},
{'headline': 'Article 7'},
{'headline': 'Article 1'},
],
transform=identity)
self.assertQuerysetEqual(
Article.objects.filter(pub_date__exact=datetime(2005, 7, 27)).values('id'),
[{'id': self.a2.id}, {'id': self.a3.id}, {'id': self.a7.id}],
transform=identity)
self.assertQuerysetEqual(Article.objects.values('id', 'headline'),
[
{'id': self.a5.id, 'headline': 'Article 5'},
{'id': self.a6.id, 'headline': 'Article 6'},
{'id': self.a4.id, 'headline': 'Article 4'},
{'id': self.a2.id, 'headline': 'Article 2'},
{'id': self.a3.id, 'headline': 'Article 3'},
{'id': self.a7.id, 'headline': 'Article 7'},
{'id': self.a1.id, 'headline': 'Article 1'},
],
transform=identity)
# You can use values() with iterator() for memory savings,
# because iterator() uses database-level iteration.
self.assertQuerysetEqual(Article.objects.values('id', 'headline').iterator(),
[
{'headline': 'Article 5', 'id': self.a5.id},
{'headline': 'Article 6', 'id': self.a6.id},
{'headline': 'Article 4', 'id': self.a4.id},
{'headline': 'Article 2', 'id': self.a2.id},
{'headline': 'Article 3', 'id': self.a3.id},
{'headline': 'Article 7', 'id': self.a7.id},
{'headline': 'Article 1', 'id': self.a1.id},
],
transform=identity)
# The values() method works with "extra" fields specified in extra(select).
self.assertQuerysetEqual(
Article.objects.extra(select={'id_plus_one': 'id + 1'}).values('id', 'id_plus_one'),
[
{'id': self.a5.id, 'id_plus_one': self.a5.id + 1},
{'id': self.a6.id, 'id_plus_one': self.a6.id + 1},
{'id': self.a4.id, 'id_plus_one': self.a4.id + 1},
{'id': self.a2.id, 'id_plus_one': self.a2.id + 1},
{'id': self.a3.id, 'id_plus_one': self.a3.id + 1},
{'id': self.a7.id, 'id_plus_one': self.a7.id + 1},
{'id': self.a1.id, 'id_plus_one': self.a1.id + 1},
],
transform=identity)
data = {
'id_plus_one': 'id+1',
'id_plus_two': 'id+2',
'id_plus_three': 'id+3',
'id_plus_four': 'id+4',
'id_plus_five': 'id+5',
'id_plus_six': 'id+6',
'id_plus_seven': 'id+7',
'id_plus_eight': 'id+8',
}
self.assertQuerysetEqual(
Article.objects.filter(id=self.a1.id).extra(select=data).values(*data.keys()),
[{
'id_plus_one': self.a1.id + 1,
'id_plus_two': self.a1.id + 2,
'id_plus_three': self.a1.id + 3,
'id_plus_four': self.a1.id + 4,
'id_plus_five': self.a1.id + 5,
'id_plus_six': self.a1.id + 6,
'id_plus_seven': self.a1.id + 7,
'id_plus_eight': self.a1.id + 8,
}], transform=identity)
# You can specify fields from forward and reverse relations, just like filter().
self.assertQuerysetEqual(
Article.objects.values('headline', 'author__name'),
[
{'headline': self.a5.headline, 'author__name': self.au2.name},
{'headline': self.a6.headline, 'author__name': self.au2.name},
{'headline': self.a4.headline, 'author__name': self.au1.name},
{'headline': self.a2.headline, 'author__name': self.au1.name},
{'headline': self.a3.headline, 'author__name': self.au1.name},
{'headline': self.a7.headline, 'author__name': self.au2.name},
{'headline': self.a1.headline, 'author__name': self.au1.name},
], transform=identity)
self.assertQuerysetEqual(
Author.objects.values('name', 'article__headline').order_by('name', 'article__headline'),
[
{'name': self.au1.name, 'article__headline': self.a1.headline},
{'name': self.au1.name, 'article__headline': self.a2.headline},
{'name': self.au1.name, 'article__headline': self.a3.headline},
{'name': self.au1.name, 'article__headline': self.a4.headline},
{'name': self.au2.name, 'article__headline': self.a5.headline},
{'name': self.au2.name, 'article__headline': self.a6.headline},
{'name': self.au2.name, 'article__headline': self.a7.headline},
], transform=identity)
self.assertQuerysetEqual(
Author.objects.values('name', 'article__headline', 'article__tag__name').order_by('name', 'article__headline', 'article__tag__name'),
[
{'name': self.au1.name, 'article__headline': self.a1.headline, 'article__tag__name': self.t1.name},
{'name': self.au1.name, 'article__headline': self.a2.headline, 'article__tag__name': self.t1.name},
{'name': self.au1.name, 'article__headline': self.a3.headline, 'article__tag__name': self.t1.name},
{'name': self.au1.name, 'article__headline': self.a3.headline, 'article__tag__name': self.t2.name},
{'name': self.au1.name, 'article__headline': self.a4.headline, 'article__tag__name': self.t2.name},
{'name': self.au2.name, 'article__headline': self.a5.headline, 'article__tag__name': self.t2.name},
{'name': self.au2.name, 'article__headline': self.a5.headline, 'article__tag__name': self.t3.name},
{'name': self.au2.name, 'article__headline': self.a6.headline, 'article__tag__name': self.t3.name},
{'name': self.au2.name, 'article__headline': self.a7.headline, 'article__tag__name': self.t3.name},
], transform=identity)
# However, an exception FieldDoesNotExist will be thrown if you specify
# a non-existent field name in values() (a field that is neither in the
# model nor in extra(select)).
self.assertRaises(FieldError,
Article.objects.extra(select={'id_plus_one': 'id + 1'}).values,
'id', 'id_plus_two')
# If you don't specify field names to values(), all are returned.
self.assertQuerysetEqual(Article.objects.filter(id=self.a5.id).values(),
[{
'id': self.a5.id,
'author_id': self.au2.id,
'headline': 'Article 5',
'pub_date': datetime(2005, 8, 1, 9, 0)
}], transform=identity)
def test_values_list(self):
# values_list() is similar to values(), except that the results are
# returned as a list of tuples, rather than a list of dictionaries.
# Within each tuple, the order of the elements is the same as the order
# of fields in the values_list() call.
identity = lambda x: x
self.assertQuerysetEqual(Article.objects.values_list('headline'),
[
('Article 5',),
('Article 6',),
('Article 4',),
('Article 2',),
('Article 3',),
('Article 7',),
('Article 1',),
], transform=identity)
self.assertQuerysetEqual(Article.objects.values_list('id').order_by('id'),
[(self.a1.id,), (self.a2.id,), (self.a3.id,), (self.a4.id,), (self.a5.id,), (self.a6.id,), (self.a7.id,)],
transform=identity)
self.assertQuerysetEqual(
Article.objects.values_list('id', flat=True).order_by('id'),
[self.a1.id, self.a2.id, self.a3.id, self.a4.id, self.a5.id, self.a6.id, self.a7.id],
transform=identity)
self.assertQuerysetEqual(
Article.objects.extra(select={'id_plus_one': 'id+1'})
.order_by('id').values_list('id'),
[(self.a1.id,), (self.a2.id,), (self.a3.id,), (self.a4.id,), (self.a5.id,), (self.a6.id,), (self.a7.id,)],
transform=identity)
self.assertQuerysetEqual(
Article.objects.extra(select={'id_plus_one': 'id+1'})
.order_by('id').values_list('id_plus_one', 'id'),
[
(self.a1.id + 1, self.a1.id),
(self.a2.id + 1, self.a2.id),
(self.a3.id + 1, self.a3.id),
(self.a4.id + 1, self.a4.id),
(self.a5.id + 1, self.a5.id),
(self.a6.id + 1, self.a6.id),
(self.a7.id + 1, self.a7.id)
],
transform=identity)
self.assertQuerysetEqual(
Article.objects.extra(select={'id_plus_one': 'id+1'})
.order_by('id').values_list('id', 'id_plus_one'),
[
(self.a1.id, self.a1.id + 1),
(self.a2.id, self.a2.id + 1),
(self.a3.id, self.a3.id + 1),
(self.a4.id, self.a4.id + 1),
(self.a5.id, self.a5.id + 1),
(self.a6.id, self.a6.id + 1),
(self.a7.id, self.a7.id + 1)
],
transform=identity)
self.assertQuerysetEqual(
Author.objects.values_list('name', 'article__headline', 'article__tag__name').order_by('name', 'article__headline', 'article__tag__name'),
[
(self.au1.name, self.a1.headline, self.t1.name),
(self.au1.name, self.a2.headline, self.t1.name),
(self.au1.name, self.a3.headline, self.t1.name),
(self.au1.name, self.a3.headline, self.t2.name),
(self.au1.name, self.a4.headline, self.t2.name),
(self.au2.name, self.a5.headline, self.t2.name),
(self.au2.name, self.a5.headline, self.t3.name),
(self.au2.name, self.a6.headline, self.t3.name),
(self.au2.name, self.a7.headline, self.t3.name),
], transform=identity)
self.assertRaises(TypeError, Article.objects.values_list, 'id', 'headline', flat=True)
def test_get_next_previous_by(self):
# Every DateField and DateTimeField creates get_next_by_FOO() and
# get_previous_by_FOO() methods. In the case of identical date values,
# these methods will use the ID as a fallback check. This guarantees
# that no records are skipped or duplicated.
self.assertEqual(repr(self.a1.get_next_by_pub_date()),
'<Article: Article 2>')
self.assertEqual(repr(self.a2.get_next_by_pub_date()),
'<Article: Article 3>')
self.assertEqual(repr(self.a2.get_next_by_pub_date(headline__endswith='6')),
'<Article: Article 6>')
self.assertEqual(repr(self.a3.get_next_by_pub_date()),
'<Article: Article 7>')
self.assertEqual(repr(self.a4.get_next_by_pub_date()),
'<Article: Article 6>')
self.assertRaises(Article.DoesNotExist, self.a5.get_next_by_pub_date)
self.assertEqual(repr(self.a6.get_next_by_pub_date()),
'<Article: Article 5>')
self.assertEqual(repr(self.a7.get_next_by_pub_date()),
'<Article: Article 4>')
self.assertEqual(repr(self.a7.get_previous_by_pub_date()),
'<Article: Article 3>')
self.assertEqual(repr(self.a6.get_previous_by_pub_date()),
'<Article: Article 4>')
self.assertEqual(repr(self.a5.get_previous_by_pub_date()),
'<Article: Article 6>')
self.assertEqual(repr(self.a4.get_previous_by_pub_date()),
'<Article: Article 7>')
self.assertEqual(repr(self.a3.get_previous_by_pub_date()),
'<Article: Article 2>')
self.assertEqual(repr(self.a2.get_previous_by_pub_date()),
'<Article: Article 1>')
def test_escaping(self):
# Underscores, percent signs and backslashes have special meaning in the
# underlying SQL code, but Django handles the quoting of them automatically.
a8 = Article(headline='Article_ with underscore', pub_date=datetime(2005, 11, 20))
a8.save()
self.assertQuerysetEqual(Article.objects.filter(headline__startswith='Article'),
[
'<Article: Article_ with underscore>',
'<Article: Article 5>',
'<Article: Article 6>',
'<Article: Article 4>',
'<Article: Article 2>',
'<Article: Article 3>',
'<Article: Article 7>',
'<Article: Article 1>',
])
self.assertQuerysetEqual(Article.objects.filter(headline__startswith='Article_'),
['<Article: Article_ with underscore>'])
a9 = Article(headline='Article% with percent sign', pub_date=datetime(2005, 11, 21))
a9.save()
self.assertQuerysetEqual(Article.objects.filter(headline__startswith='Article'),
[
'<Article: Article% with percent sign>',
'<Article: Article_ with underscore>',
'<Article: Article 5>',
'<Article: Article 6>',
'<Article: Article 4>',
'<Article: Article 2>',
'<Article: Article 3>',
'<Article: Article 7>',
'<Article: Article 1>',
])
self.assertQuerysetEqual(Article.objects.filter(headline__startswith='Article%'),
['<Article: Article% with percent sign>'])
a10 = Article(headline='Article with \\ backslash', pub_date=datetime(2005, 11, 22))
a10.save()
self.assertQuerysetEqual(Article.objects.filter(headline__contains='\\'),
['<Article: Article with \ backslash>'])
def test_exclude(self):
Article.objects.create(headline='Article_ with underscore', pub_date=datetime(2005, 11, 20))
Article.objects.create(headline='Article% with percent sign', pub_date=datetime(2005, 11, 21))
Article.objects.create(headline='Article with \\ backslash', pub_date=datetime(2005, 11, 22))
# exclude() is the opposite of filter() when doing lookups:
self.assertQuerysetEqual(
Article.objects.filter(headline__contains='Article').exclude(headline__contains='with'),
[
'<Article: Article 5>',
'<Article: Article 6>',
'<Article: Article 4>',
'<Article: Article 2>',
'<Article: Article 3>',
'<Article: Article 7>',
'<Article: Article 1>',
])
self.assertQuerysetEqual(Article.objects.exclude(headline__startswith="Article_"),
[
'<Article: Article with \\ backslash>',
'<Article: Article% with percent sign>',
'<Article: Article 5>',
'<Article: Article 6>',
'<Article: Article 4>',
'<Article: Article 2>',
'<Article: Article 3>',
'<Article: Article 7>',
'<Article: Article 1>',
])
self.assertQuerysetEqual(Article.objects.exclude(headline="Article 7"),
[
'<Article: Article with \\ backslash>',
'<Article: Article% with percent sign>',
'<Article: Article_ with underscore>',
'<Article: Article 5>',
'<Article: Article 6>',
'<Article: Article 4>',
'<Article: Article 2>',
'<Article: Article 3>',
'<Article: Article 1>',
])
def test_none(self):
# none() returns a QuerySet that behaves like any other QuerySet object
self.assertQuerysetEqual(Article.objects.none(), [])
self.assertQuerysetEqual(
Article.objects.none().filter(headline__startswith='Article'), [])
self.assertQuerysetEqual(
Article.objects.filter(headline__startswith='Article').none(), [])
self.assertEqual(Article.objects.none().count(), 0)
self.assertEqual(
Article.objects.none().update(headline="This should not take effect"), 0)
self.assertQuerysetEqual(
[article for article in Article.objects.none().iterator()],
[])
def test_in(self):
# using __in with an empty list should return an empty query set
self.assertQuerysetEqual(Article.objects.filter(id__in=[]), [])
self.assertQuerysetEqual(Article.objects.exclude(id__in=[]),
[
'<Article: Article 5>',
'<Article: Article 6>',
'<Article: Article 4>',
'<Article: Article 2>',
'<Article: Article 3>',
'<Article: Article 7>',
'<Article: Article 1>',
])
def test_error_messages(self):
# Programming errors are pointed out with nice error messages
try:
Article.objects.filter(pub_date_year='2005').count()
self.fail('FieldError not raised')
except FieldError as ex:
self.assertEqual(str(ex), "Cannot resolve keyword 'pub_date_year' "
"into field. Choices are: author, author_id, headline, "
"id, pub_date, tag")
try:
Article.objects.filter(headline__starts='Article')
self.fail('FieldError not raised')
except FieldError as ex:
self.assertEqual(
str(ex), "Unsupported lookup 'starts' for CharField "
"or join on the field not permitted.")
def test_regex(self):
# Create some articles with a bit more interesting headlines for testing field lookups:
for a in Article.objects.all():
a.delete()
now = datetime.now()
a1 = Article(pub_date=now, headline='f')
a1.save()
a2 = Article(pub_date=now, headline='fo')
a2.save()
a3 = Article(pub_date=now, headline='foo')
a3.save()
a4 = Article(pub_date=now, headline='fooo')
a4.save()
a5 = Article(pub_date=now, headline='hey-Foo')
a5.save()
a6 = Article(pub_date=now, headline='bar')
a6.save()
a7 = Article(pub_date=now, headline='AbBa')
a7.save()
a8 = Article(pub_date=now, headline='baz')
a8.save()
a9 = Article(pub_date=now, headline='baxZ')
a9.save()
# zero-or-more
self.assertQuerysetEqual(Article.objects.filter(headline__regex=r'fo*'),
['<Article: f>', '<Article: fo>', '<Article: foo>', '<Article: fooo>'])
self.assertQuerysetEqual(Article.objects.filter(headline__iregex=r'fo*'),
[
'<Article: f>',
'<Article: fo>',
'<Article: foo>',
'<Article: fooo>',
'<Article: hey-Foo>',
])
# one-or-more
self.assertQuerysetEqual(Article.objects.filter(headline__regex=r'fo+'),
['<Article: fo>', '<Article: foo>', '<Article: fooo>'])
# wildcard
self.assertQuerysetEqual(Article.objects.filter(headline__regex=r'fooo?'),
['<Article: foo>', '<Article: fooo>'])
# leading anchor
self.assertQuerysetEqual(Article.objects.filter(headline__regex=r'^b'),
['<Article: bar>', '<Article: baxZ>', '<Article: baz>'])
self.assertQuerysetEqual(Article.objects.filter(headline__iregex=r'^a'),
['<Article: AbBa>'])
# trailing anchor
self.assertQuerysetEqual(Article.objects.filter(headline__regex=r'z$'),
['<Article: baz>'])
self.assertQuerysetEqual(Article.objects.filter(headline__iregex=r'z$'),
['<Article: baxZ>', '<Article: baz>'])
# character sets
self.assertQuerysetEqual(Article.objects.filter(headline__regex=r'ba[rz]'),
['<Article: bar>', '<Article: baz>'])
self.assertQuerysetEqual(Article.objects.filter(headline__regex=r'ba.[RxZ]'),
['<Article: baxZ>'])
self.assertQuerysetEqual(Article.objects.filter(headline__iregex=r'ba[RxZ]'),
['<Article: bar>', '<Article: baxZ>', '<Article: baz>'])
# and more articles:
a10 = Article(pub_date=now, headline='foobar')
a10.save()
a11 = Article(pub_date=now, headline='foobaz')
a11.save()
a12 = Article(pub_date=now, headline='ooF')
a12.save()
a13 = Article(pub_date=now, headline='foobarbaz')
a13.save()
a14 = Article(pub_date=now, headline='zoocarfaz')
a14.save()
a15 = Article(pub_date=now, headline='barfoobaz')
a15.save()
a16 = Article(pub_date=now, headline='bazbaRFOO')
a16.save()
# alternation
self.assertQuerysetEqual(Article.objects.filter(headline__regex=r'oo(f|b)'),
[
'<Article: barfoobaz>',
'<Article: foobar>',
'<Article: foobarbaz>',
'<Article: foobaz>',
])
self.assertQuerysetEqual(Article.objects.filter(headline__iregex=r'oo(f|b)'),
[
'<Article: barfoobaz>',
'<Article: foobar>',
'<Article: foobarbaz>',
'<Article: foobaz>',
'<Article: ooF>',
])
self.assertQuerysetEqual(Article.objects.filter(headline__regex=r'^foo(f|b)'),
['<Article: foobar>', '<Article: foobarbaz>', '<Article: foobaz>'])
# greedy matching
self.assertQuerysetEqual(Article.objects.filter(headline__regex=r'b.*az'),
[
'<Article: barfoobaz>',
'<Article: baz>',
'<Article: bazbaRFOO>',
'<Article: foobarbaz>',
'<Article: foobaz>',
])
self.assertQuerysetEqual(Article.objects.filter(headline__iregex=r'b.*ar'),
[
'<Article: bar>',
'<Article: barfoobaz>',
'<Article: bazbaRFOO>',
'<Article: foobar>',
'<Article: foobarbaz>',
])
@skipUnlessDBFeature('supports_regex_backreferencing')
def test_regex_backreferencing(self):
# grouping and backreferences
now = datetime.now()
a10 = Article(pub_date=now, headline='foobar')
a10.save()
a11 = Article(pub_date=now, headline='foobaz')
a11.save()
a12 = Article(pub_date=now, headline='ooF')
a12.save()
a13 = Article(pub_date=now, headline='foobarbaz')
a13.save()
a14 = Article(pub_date=now, headline='zoocarfaz')
a14.save()
a15 = Article(pub_date=now, headline='barfoobaz')
a15.save()
a16 = Article(pub_date=now, headline='bazbaRFOO')
a16.save()
self.assertQuerysetEqual(Article.objects.filter(headline__regex=r'b(.).*b\1'),
['<Article: barfoobaz>', '<Article: bazbaRFOO>', '<Article: foobarbaz>'])
def test_regex_null(self):
"""
Ensure that a regex lookup does not fail on null/None values
"""
Season.objects.create(year=2012, gt=None)
self.assertQuerysetEqual(Season.objects.filter(gt__regex=r'^$'), [])
def test_regex_non_string(self):
"""
Ensure that a regex lookup does not fail on non-string fields
"""
Season.objects.create(year=2013, gt=444)
self.assertQuerysetEqual(Season.objects.filter(gt__regex=r'^444$'),
['<Season: 2013>'])
def test_regex_non_ascii(self):
"""
Ensure that a regex lookup does not trip on non-ASCII characters.
"""
Player.objects.create(name='\u2660')
Player.objects.get(name__regex='\u2660')
def test_nonfield_lookups(self):
"""
Ensure that a lookup query containing non-fields raises the proper
exception.
"""
with self.assertRaises(FieldError):
Article.objects.filter(headline__blahblah=99)
with self.assertRaises(FieldError):
Article.objects.filter(headline__blahblah__exact=99)
with self.assertRaises(FieldError):
Article.objects.filter(blahblah=99)
def test_lookup_collision(self):
"""
Ensure that genuine field names don't collide with built-in lookup
types ('year', 'gt', 'range', 'in' etc.).
Refs #11670.
"""
# Here we're using 'gt' as a code number for the year, e.g. 111=>2009.
season_2009 = Season.objects.create(year=2009, gt=111)
season_2009.games.create(home="Houston Astros", away="St. Louis Cardinals")
season_2010 = Season.objects.create(year=2010, gt=222)
season_2010.games.create(home="Houston Astros", away="Chicago Cubs")
season_2010.games.create(home="Houston Astros", away="Milwaukee Brewers")
season_2010.games.create(home="Houston Astros", away="St. Louis Cardinals")
season_2011 = Season.objects.create(year=2011, gt=333)
season_2011.games.create(home="Houston Astros", away="St. Louis Cardinals")
season_2011.games.create(home="Houston Astros", away="Milwaukee Brewers")
hunter_pence = Player.objects.create(name="Hunter Pence")
hunter_pence.games = Game.objects.filter(season__year__in=[2009, 2010])
pudge = Player.objects.create(name="Ivan Rodriquez")
pudge.games = Game.objects.filter(season__year=2009)
pedro_feliz = Player.objects.create(name="Pedro Feliz")
pedro_feliz.games = Game.objects.filter(season__year__in=[2011])
johnson = Player.objects.create(name="Johnson")
johnson.games = Game.objects.filter(season__year__in=[2011])
# Games in 2010
self.assertEqual(Game.objects.filter(season__year=2010).count(), 3)
self.assertEqual(Game.objects.filter(season__year__exact=2010).count(), 3)
self.assertEqual(Game.objects.filter(season__gt=222).count(), 3)
self.assertEqual(Game.objects.filter(season__gt__exact=222).count(), 3)
# Games in 2011
self.assertEqual(Game.objects.filter(season__year=2011).count(), 2)
self.assertEqual(Game.objects.filter(season__year__exact=2011).count(), 2)
self.assertEqual(Game.objects.filter(season__gt=333).count(), 2)
self.assertEqual(Game.objects.filter(season__gt__exact=333).count(), 2)
self.assertEqual(Game.objects.filter(season__year__gt=2010).count(), 2)
self.assertEqual(Game.objects.filter(season__gt__gt=222).count(), 2)
# Games played in 2010 and 2011
self.assertEqual(Game.objects.filter(season__year__in=[2010, 2011]).count(), 5)
self.assertEqual(Game.objects.filter(season__year__gt=2009).count(), 5)
self.assertEqual(Game.objects.filter(season__gt__in=[222, 333]).count(), 5)
self.assertEqual(Game.objects.filter(season__gt__gt=111).count(), 5)
# Players who played in 2009
self.assertEqual(Player.objects.filter(games__season__year=2009).distinct().count(), 2)
self.assertEqual(Player.objects.filter(games__season__year__exact=2009).distinct().count(), 2)
self.assertEqual(Player.objects.filter(games__season__gt=111).distinct().count(), 2)
self.assertEqual(Player.objects.filter(games__season__gt__exact=111).distinct().count(), 2)
# Players who played in 2010
self.assertEqual(Player.objects.filter(games__season__year=2010).distinct().count(), 1)
self.assertEqual(Player.objects.filter(games__season__year__exact=2010).distinct().count(), 1)
self.assertEqual(Player.objects.filter(games__season__gt=222).distinct().count(), 1)
self.assertEqual(Player.objects.filter(games__season__gt__exact=222).distinct().count(), 1)
# Players who played in 2011
self.assertEqual(Player.objects.filter(games__season__year=2011).distinct().count(), 2)
self.assertEqual(Player.objects.filter(games__season__year__exact=2011).distinct().count(), 2)
self.assertEqual(Player.objects.filter(games__season__gt=333).distinct().count(), 2)
self.assertEqual(Player.objects.filter(games__season__year__gt=2010).distinct().count(), 2)
self.assertEqual(Player.objects.filter(games__season__gt__gt=222).distinct().count(), 2)
class LookupTransactionTests(TransactionTestCase):
available_apps = ['lookup']
@skipUnless(connection.vendor == 'mysql', 'requires MySQL')
def test_mysql_lookup_search(self):
# To use fulltext indexes on MySQL either version 5.6 is needed, or one must use
# MyISAM tables. Neither of these combinations is currently available on CI, so
# lets manually create a MyISAM table for Article model.
with connection.cursor() as cursor:
cursor.execute(
"CREATE TEMPORARY TABLE myisam_article ("
" id INTEGER PRIMARY KEY AUTO_INCREMENT, "
" headline VARCHAR(100) NOT NULL "
") ENGINE MYISAM")
dr = MyISAMArticle.objects.create(headline='Django Reinhardt')
MyISAMArticle.objects.create(headline='Ringo Star')
# NOTE: Needs to be created after the article has been saved.
cursor.execute(
'CREATE FULLTEXT INDEX myisam_article_ft ON myisam_article (headline)')
self.assertQuerysetEqual(
MyISAMArticle.objects.filter(headline__search='Reinhardt'),
[dr], lambda x: x)