Migrated extra_regress doctests. Thanks to Stephan Jaekel.

git-svn-id: http://code.djangoproject.com/svn/django/trunk@13951 bcc190cf-cafb-0310-a4f2-bffc1f526a37
This commit is contained in:
Russell Keith-Magee 2010-09-28 13:42:09 +00:00
parent d0da38410b
commit 0856c765f5
2 changed files with 314 additions and 177 deletions

View File

@ -4,9 +4,6 @@ import django.utils.copycompat as copy
from django.contrib.auth.models import User
from django.db import models
from django.db.models.query import Q
from django.utils.datastructures import SortedDict
class RevisionableModel(models.Model):
base = models.ForeignKey('self', null=True)
@ -41,177 +38,3 @@ class TestObject(models.Model):
def __unicode__(self):
return u'TestObject: %s,%s,%s' % (self.first,self.second,self.third)
__test__ = {"API_TESTS": """
# Regression tests for #7314 and #7372
>>> rm = RevisionableModel.objects.create(title='First Revision', when=datetime.datetime(2008, 9, 28, 10, 30, 0))
>>> rm.pk, rm.base.pk
(1, 1)
>>> rm2 = rm.new_revision()
>>> rm2.title = "Second Revision"
>>> rm.when = datetime.datetime(2008, 9, 28, 14, 25, 0)
>>> rm2.save()
>>> print u"%s of %s" % (rm2.title, rm2.base.title)
Second Revision of First Revision
>>> rm2.pk, rm2.base.pk
(2, 1)
Queryset to match most recent revision:
>>> qs = RevisionableModel.objects.extra(where=["%(table)s.id IN (SELECT MAX(rev.id) FROM %(table)s rev GROUP BY rev.base_id)" % {'table': RevisionableModel._meta.db_table,}],)
>>> qs
[<RevisionableModel: Second Revision (2, 1)>]
Queryset to search for string in title:
>>> qs2 = RevisionableModel.objects.filter(title__contains="Revision")
>>> qs2
[<RevisionableModel: First Revision (1, 1)>, <RevisionableModel: Second Revision (2, 1)>]
Following queryset should return the most recent revision:
>>> qs & qs2
[<RevisionableModel: Second Revision (2, 1)>]
>>> u = User.objects.create_user(username="fred", password="secret", email="fred@example.com")
# General regression tests: extra select parameters should stay tied to their
# corresponding select portions. Applies when portions are updated or otherwise
# moved around.
>>> qs = User.objects.extra(select=SortedDict((("alpha", "%s"), ("beta", "2"), ("gamma", "%s"))), select_params=(1, 3))
>>> qs = qs.extra(select={"beta": 4})
>>> qs = qs.extra(select={"alpha": "%s"}, select_params=[5])
>>> result = {'alpha': 5, 'beta': 4, 'gamma': 3}
>>> list(qs.filter(id=u.id).values('alpha', 'beta', 'gamma')) == [result]
True
# Regression test for #7957: Combining extra() calls should leave the
# corresponding parameters associated with the right extra() bit. I.e. internal
# dictionary must remain sorted.
>>> User.objects.extra(select={"alpha": "%s"}, select_params=(1,)).extra(select={"beta": "%s"}, select_params=(2,))[0].alpha
1
>>> User.objects.extra(select={"beta": "%s"}, select_params=(1,)).extra(select={"alpha": "%s"}, select_params=(2,))[0].alpha
2
# Regression test for #7961: When not using a portion of an extra(...) in a
# query, remove any corresponding parameters from the query as well.
>>> list(User.objects.extra(select={"alpha": "%s"}, select_params=(-6,)).filter(id=u.id).values_list('id', flat=True)) == [u.id]
True
# Regression test for #8063: limiting a query shouldn't discard any extra()
# bits.
>>> qs = User.objects.all().extra(where=['id=%s'], params=[u.id])
>>> qs
[<User: fred>]
>>> qs[:1]
[<User: fred>]
# Regression test for #8039: Ordering sometimes removed relevant tables from
# extra(). This test is the critical case: ordering uses a table, but then
# removes the reference because of an optimisation. The table should still be
# present because of the extra() call.
>>> Order.objects.extra(where=["username=%s"], params=["fred"], tables=["auth_user"]).order_by('created_by')
[]
# Regression test for #8819: Fields in the extra(select=...) list should be
# available to extra(order_by=...).
>>> User.objects.filter(pk=u.id).extra(select={'extra_field': 1}).distinct()
[<User: fred>]
>>> User.objects.filter(pk=u.id).extra(select={'extra_field': 1}, order_by=['extra_field'])
[<User: fred>]
>>> User.objects.filter(pk=u.id).extra(select={'extra_field': 1}, order_by=['extra_field']).distinct()
[<User: fred>]
# When calling the dates() method on a queryset with extra selection columns,
# we can (and should) ignore those columns. They don't change the result and
# cause incorrect SQL to be produced otherwise.
>>> RevisionableModel.objects.extra(select={"the_answer": 'id'}).dates('when', 'month')
[datetime.datetime(2008, 9, 1, 0, 0)]
# Regression test for #10256... If there is a values() clause, Extra columns are
# only returned if they are explicitly mentioned.
>>> TestObject(first='first', second='second', third='third').save()
>>> list(TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values()) == [{'bar': u'second', 'third': u'third', 'second': u'second', 'whiz': u'third', 'foo': u'first', 'id': 1, 'first': u'first'}]
True
# Extra clauses after an empty values clause are still included
>>> list(TestObject.objects.values().extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third'))))) == [{'bar': u'second', 'third': u'third', 'second': u'second', 'whiz': u'third', 'foo': u'first', 'id': 1, 'first': u'first'}]
True
# Extra columns are ignored if not mentioned in the values() clause
>>> TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values('first', 'second')
[{'second': u'second', 'first': u'first'}]
# Extra columns after a non-empty values() clause are ignored
>>> TestObject.objects.values('first', 'second').extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third'))))
[{'second': u'second', 'first': u'first'}]
# Extra columns can be partially returned
>>> TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values('first', 'second', 'foo')
[{'second': u'second', 'foo': u'first', 'first': u'first'}]
# Also works if only extra columns are included
>>> TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values('foo', 'whiz')
[{'foo': u'first', 'whiz': u'third'}]
# Values list works the same way
# All columns are returned for an empty values_list()
>>> TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values_list()
[(u'first', u'second', u'third', 1, u'first', u'second', u'third')]
# Extra columns after an empty values_list() are still included
>>> TestObject.objects.values_list().extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third'))))
[(u'first', u'second', u'third', 1, u'first', u'second', u'third')]
# Extra columns ignored completely if not mentioned in values_list()
>>> TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values_list('first', 'second')
[(u'first', u'second')]
# Extra columns after a non-empty values_list() clause are ignored completely
>>> TestObject.objects.values_list('first', 'second').extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third'))))
[(u'first', u'second')]
>>> TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values_list('second', flat=True)
[u'second']
# Only the extra columns specified in the values_list() are returned
>>> TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values_list('first', 'second', 'whiz')
[(u'first', u'second', u'third')]
# ...also works if only extra columns are included
>>> TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values_list('foo','whiz')
[(u'first', u'third')]
>>> TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values_list('whiz', flat=True)
[u'third']
# ... and values are returned in the order they are specified
>>> TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values_list('whiz','foo')
[(u'third', u'first')]
>>> TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values_list('first','id')
[(u'first', 1)]
>>> TestObject.objects.extra(select=SortedDict((('foo','first'),('bar','second'),('whiz','third')))).values_list('whiz', 'first', 'bar', 'id')
[(u'third', u'first', u'second', 1)]
# Regression for #10847: the list of extra columns can always be accurately evaluated.
# Using an inner query ensures that as_sql() is producing correct output
# without requiring full evaluation and execution of the inner query.
>>> TestObject.objects.extra(select={'extra': 1}).values('pk')
[{'pk': 1}]
>>> TestObject.objects.filter(pk__in=TestObject.objects.extra(select={'extra': 1}).values('pk'))
[<TestObject: TestObject: first,second,third>]
>>> TestObject.objects.values('pk').extra(select={'extra': 1})
[{'pk': 1}]
>>> TestObject.objects.filter(pk__in=TestObject.objects.values('pk').extra(select={'extra': 1}))
[<TestObject: TestObject: first,second,third>]
>>> pk = TestObject.objects.get().pk
>>> TestObject.objects.filter(pk=pk) | TestObject.objects.extra(where=["id > %s"], params=[pk])
[<TestObject: TestObject: first,second,third>]
"""}

View File

@ -0,0 +1,314 @@
from django.test import TestCase
from django.utils.datastructures import SortedDict
from django.contrib.auth.models import User
from regressiontests.extra_regress.models import TestObject, Order, \
RevisionableModel
import datetime
class ExtraRegressTests(TestCase):
def setUp(self):
self.u = User.objects.create_user(
username="fred",
password="secret",
email="fred@example.com"
)
def test_regression_7314_7372(self):
"""
Regression tests for #7314 and #7372
"""
rm = RevisionableModel.objects.create(
title='First Revision',
when=datetime.datetime(2008, 9, 28, 10, 30, 0)
)
self.assertEqual(rm.pk, rm.base.pk)
rm2 = rm.new_revision()
rm2.title = "Second Revision"
rm.when = datetime.datetime(2008, 9, 28, 14, 25, 0)
rm2.save()
self.assertEqual(rm2.title, 'Second Revision')
self.assertEqual(rm2.base.title, 'First Revision')
self.assertNotEqual(rm2.pk, rm.pk)
self.assertEqual(rm2.base.pk, rm.pk)
# Queryset to match most recent revision:
qs = RevisionableModel.objects.extra(
where=["%(table)s.id IN (SELECT MAX(rev.id) FROM %(table)s rev GROUP BY rev.base_id)" % {
'table': RevisionableModel._meta.db_table,
}]
)
self.assertQuerysetEqual(qs,
[('Second Revision', 'First Revision')],
transform=lambda r: (r.title, r.base.title)
)
# Queryset to search for string in title:
qs2 = RevisionableModel.objects.filter(title__contains="Revision")
self.assertQuerysetEqual(qs2,
[
('First Revision', 'First Revision'),
('Second Revision', 'First Revision'),
],
transform=lambda r: (r.title, r.base.title)
)
# Following queryset should return the most recent revision:
self.assertQuerysetEqual(qs & qs2,
[('Second Revision', 'First Revision')],
transform=lambda r: (r.title, r.base.title)
)
def test_extra_stay_tied(self):
# Extra select parameters should stay tied to their corresponding
# select portions. Applies when portions are updated or otherwise
# moved around.
qs = User.objects.extra(
select=SortedDict((("alpha", "%s"), ("beta", "2"), ("gamma", "%s"))),
select_params=(1, 3)
)
qs = qs.extra(select={"beta": 4})
qs = qs.extra(select={"alpha": "%s"}, select_params=[5])
self.assertEqual(
list(qs.filter(id=self.u.id).values('alpha', 'beta', 'gamma')),
[{'alpha': 5, 'beta': 4, 'gamma': 3}]
)
def test_regression_7957(self):
"""
Regression test for #7957: Combining extra() calls should leave the
corresponding parameters associated with the right extra() bit. I.e.
internal dictionary must remain sorted.
"""
self.assertEqual(
User.objects.extra(select={"alpha": "%s"}, select_params=(1,)
).extra(select={"beta": "%s"}, select_params=(2,))[0].alpha,
1)
self.assertEqual(
User.objects.extra(select={"beta": "%s"}, select_params=(1,)
).extra(select={"alpha": "%s"}, select_params=(2,))[0].alpha,
2)
def test_regression_7961(self):
"""
Regression test for #7961: When not using a portion of an
extra(...) in a query, remove any corresponding parameters from the
query as well.
"""
self.assertEqual(
list(User.objects.extra(select={"alpha": "%s"}, select_params=(-6,)
).filter(id=self.u.id).values_list('id', flat=True)),
[self.u.id]
)
def test_regression_8063(self):
"""
Regression test for #8063: limiting a query shouldn't discard any
extra() bits.
"""
qs = User.objects.all().extra(where=['id=%s'], params=[self.u.id])
self.assertQuerysetEqual(qs, ['<User: fred>'])
self.assertQuerysetEqual(qs[:1], ['<User: fred>'])
def test_regression_8039(self):
"""
Regression test for #8039: Ordering sometimes removed relevant tables
from extra(). This test is the critical case: ordering uses a table,
but then removes the reference because of an optimisation. The table
should still be present because of the extra() call.
"""
self.assertQuerysetEqual(
Order.objects.extra(where=["username=%s"],
params=["fred"],
tables=["auth_user"]
).order_by('created_by'),
[]
)
def test_regression_8819(self):
"""
Regression test for #8819: Fields in the extra(select=...) list
should be available to extra(order_by=...).
"""
self.assertQuerysetEqual(
User.objects.filter(pk=self.u.id).extra(select={'extra_field': 1}).distinct(),
['<User: fred>']
)
self.assertQuerysetEqual(
User.objects.filter(pk=self.u.id).extra(select={'extra_field': 1}, order_by=['extra_field']),
['<User: fred>']
)
self.assertQuerysetEqual(
User.objects.filter(pk=self.u.id).extra(select={'extra_field': 1}, order_by=['extra_field']).distinct(),
['<User: fred>']
)
def test_dates_query(self):
"""
When calling the dates() method on a queryset with extra selection
columns, we can (and should) ignore those columns. They don't change
the result and cause incorrect SQL to be produced otherwise.
"""
rm = RevisionableModel.objects.create(
title='First Revision',
when=datetime.datetime(2008, 9, 28, 10, 30, 0)
)
self.assertQuerysetEqual(
RevisionableModel.objects.extra(select={"the_answer": 'id'}).dates('when', 'month'),
['datetime.datetime(2008, 9, 1, 0, 0)']
)
def test_values_with_extra(self):
"""
Regression test for #10256... If there is a values() clause, Extra
columns are only returned if they are explicitly mentioned.
"""
obj = TestObject(first='first', second='second', third='third')
obj.save()
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values()),
[{'bar': u'second', 'third': u'third', 'second': u'second', 'whiz': u'third', 'foo': u'first', 'id': obj.pk, 'first': u'first'}]
)
# Extra clauses after an empty values clause are still included
self.assertEqual(
list(TestObject.objects.values().extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third'))))),
[{'bar': u'second', 'third': u'third', 'second': u'second', 'whiz': u'third', 'foo': u'first', 'id': obj.pk, 'first': u'first'}]
)
# Extra columns are ignored if not mentioned in the values() clause
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values('first', 'second')),
[{'second': u'second', 'first': u'first'}]
)
# Extra columns after a non-empty values() clause are ignored
self.assertEqual(
list(TestObject.objects.values('first', 'second').extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third'))))),
[{'second': u'second', 'first': u'first'}]
)
# Extra columns can be partially returned
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values('first', 'second', 'foo')),
[{'second': u'second', 'foo': u'first', 'first': u'first'}]
)
# Also works if only extra columns are included
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values('foo', 'whiz')),
[{'foo': u'first', 'whiz': u'third'}]
)
# Values list works the same way
# All columns are returned for an empty values_list()
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values_list()),
[(u'first', u'second', u'third', obj.pk, u'first', u'second', u'third')]
)
# Extra columns after an empty values_list() are still included
self.assertEqual(
list(TestObject.objects.values_list().extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third'))))),
[(u'first', u'second', u'third', obj.pk, u'first', u'second', u'third')]
)
# Extra columns ignored completely if not mentioned in values_list()
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values_list('first', 'second')),
[(u'first', u'second')]
)
# Extra columns after a non-empty values_list() clause are ignored completely
self.assertEqual(
list(TestObject.objects.values_list('first', 'second').extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third'))))),
[(u'first', u'second')]
)
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values_list('second', flat=True)),
[u'second']
)
# Only the extra columns specified in the values_list() are returned
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values_list('first', 'second', 'whiz')),
[(u'first', u'second', u'third')]
)
# ...also works if only extra columns are included
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values_list('foo','whiz')),
[(u'first', u'third')]
)
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values_list('whiz', flat=True)),
[u'third']
)
# ... and values are returned in the order they are specified
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values_list('whiz','foo')),
[(u'third', u'first')]
)
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values_list('first','id')),
[(u'first', obj.pk)]
)
self.assertEqual(
list(TestObject.objects.extra(select=SortedDict((('foo','first'), ('bar','second'), ('whiz','third')))).values_list('whiz', 'first', 'bar', 'id')),
[(u'third', u'first', u'second', obj.pk)]
)
def test_regression_10847(self):
"""
Regression for #10847: the list of extra columns can always be
accurately evaluated. Using an inner query ensures that as_sql() is
producing correct output without requiring full evaluation and
execution of the inner query.
"""
obj = TestObject(first='first', second='second', third='third')
obj.save()
self.assertEqual(
list(TestObject.objects.extra(select={'extra': 1}).values('pk')),
[{'pk': obj.pk}]
)
self.assertQuerysetEqual(
TestObject.objects.filter(
pk__in=TestObject.objects.extra(select={'extra': 1}).values('pk')
),
['<TestObject: TestObject: first,second,third>']
)
self.assertEqual(
list(TestObject.objects.values('pk').extra(select={'extra': 1})),
[{'pk': obj.pk}]
)
self.assertQuerysetEqual(
TestObject.objects.filter(
pk__in=TestObject.objects.values('pk').extra(select={'extra': 1})
),
['<TestObject: TestObject: first,second,third>']
)
self.assertQuerysetEqual(
TestObject.objects.filter(pk=obj.pk) |
TestObject.objects.extra(where=["id > %s"], params=[obj.pk]),
['<TestObject: TestObject: first,second,third>']
)