mirror of
https://github.com/django/django.git
synced 2024-11-18 07:26:04 +00:00
2c40681805
This was a stealth option only used by the tests, and it isn't useful any more since `atomic` provides nested transactions.
393 lines
32 KiB
Python
393 lines
32 KiB
Python
from __future__ import absolute_import
|
|
|
|
import warnings
|
|
|
|
from django.contrib.sites.models import Site
|
|
from django.core import management
|
|
from django.db import connection, IntegrityError
|
|
from django.test import TestCase, TransactionTestCase, skipUnlessDBFeature
|
|
from django.utils import six
|
|
|
|
from .models import Article, Book, Spy, Tag, Visa
|
|
|
|
|
|
class TestCaseFixtureLoadingTests(TestCase):
|
|
fixtures = ['fixture1.json', 'fixture2.json']
|
|
|
|
def testClassFixtures(self):
|
|
"Check that test case has installed 3 fixture objects"
|
|
self.assertEqual(Article.objects.count(), 3)
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: Django conquers world!>',
|
|
'<Article: Copyright is fine the way it is>',
|
|
'<Article: Poker has no place on ESPN>',
|
|
])
|
|
|
|
|
|
class DumpDataAssertMixin(object):
|
|
|
|
def _dumpdata_assert(self, args, output, format='json', natural_keys=False,
|
|
use_base_manager=False, exclude_list=[], primary_keys=''):
|
|
new_io = six.StringIO()
|
|
management.call_command('dumpdata', *args, **{'format': format,
|
|
'stdout': new_io,
|
|
'stderr': new_io,
|
|
'use_natural_keys': natural_keys,
|
|
'use_base_manager': use_base_manager,
|
|
'exclude': exclude_list,
|
|
'primary_keys': primary_keys})
|
|
command_output = new_io.getvalue().strip()
|
|
if format == "json":
|
|
self.assertJSONEqual(command_output, output)
|
|
elif format == "xml":
|
|
self.assertXMLEqual(command_output, output)
|
|
else:
|
|
self.assertEqual(command_output, output)
|
|
|
|
|
|
class FixtureLoadingTests(DumpDataAssertMixin, TestCase):
|
|
|
|
def test_initial_data(self):
|
|
# syncdb introduces 1 initial data object from initial_data.json.
|
|
self.assertQuerysetEqual(Book.objects.all(), [
|
|
'<Book: Achieving self-awareness of Python programs>'
|
|
])
|
|
|
|
def test_loading_and_dumping(self):
|
|
Site.objects.all().delete()
|
|
# Load fixture 1. Single JSON file, with two objects.
|
|
management.call_command('loaddata', 'fixture1.json', verbosity=0)
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: Time to reform copyright>',
|
|
'<Article: Poker has no place on ESPN>',
|
|
])
|
|
|
|
# Dump the current contents of the database as a JSON fixture
|
|
self._dumpdata_assert(['fixtures'], '[{"pk": 1, "model": "fixtures.category", "fields": {"description": "Latest news stories", "title": "News Stories"}}, {"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Time to reform copyright", "pub_date": "2006-06-16T13:00:00"}}, {"pk": 10, "model": "fixtures.book", "fields": {"name": "Achieving self-awareness of Python programs", "authors": []}}]')
|
|
|
|
# Try just dumping the contents of fixtures.Category
|
|
self._dumpdata_assert(['fixtures.Category'], '[{"pk": 1, "model": "fixtures.category", "fields": {"description": "Latest news stories", "title": "News Stories"}}]')
|
|
|
|
# ...and just fixtures.Article
|
|
self._dumpdata_assert(['fixtures.Article'], '[{"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Time to reform copyright", "pub_date": "2006-06-16T13:00:00"}}]')
|
|
|
|
# ...and both
|
|
self._dumpdata_assert(['fixtures.Category', 'fixtures.Article'], '[{"pk": 1, "model": "fixtures.category", "fields": {"description": "Latest news stories", "title": "News Stories"}}, {"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Time to reform copyright", "pub_date": "2006-06-16T13:00:00"}}]')
|
|
|
|
# Specify a specific model twice
|
|
self._dumpdata_assert(['fixtures.Article', 'fixtures.Article'], '[{"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Time to reform copyright", "pub_date": "2006-06-16T13:00:00"}}]')
|
|
|
|
# Specify a dump that specifies Article both explicitly and implicitly
|
|
self._dumpdata_assert(['fixtures.Article', 'fixtures'], '[{"pk": 1, "model": "fixtures.category", "fields": {"description": "Latest news stories", "title": "News Stories"}}, {"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Time to reform copyright", "pub_date": "2006-06-16T13:00:00"}}, {"pk": 10, "model": "fixtures.book", "fields": {"name": "Achieving self-awareness of Python programs", "authors": []}}]')
|
|
|
|
# Same again, but specify in the reverse order
|
|
self._dumpdata_assert(['fixtures'], '[{"pk": 1, "model": "fixtures.category", "fields": {"description": "Latest news stories", "title": "News Stories"}}, {"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Time to reform copyright", "pub_date": "2006-06-16T13:00:00"}}, {"pk": 10, "model": "fixtures.book", "fields": {"name": "Achieving self-awareness of Python programs", "authors": []}}]')
|
|
|
|
# Specify one model from one application, and an entire other application.
|
|
self._dumpdata_assert(['fixtures.Category', 'sites'], '[{"pk": 1, "model": "fixtures.category", "fields": {"description": "Latest news stories", "title": "News Stories"}}, {"pk": 1, "model": "sites.site", "fields": {"domain": "example.com", "name": "example.com"}}]')
|
|
|
|
# Load fixture 2. JSON file imported by default. Overwrites some existing objects
|
|
management.call_command('loaddata', 'fixture2.json', verbosity=0)
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: Django conquers world!>',
|
|
'<Article: Copyright is fine the way it is>',
|
|
'<Article: Poker has no place on ESPN>',
|
|
])
|
|
|
|
# Load fixture 3, XML format.
|
|
management.call_command('loaddata', 'fixture3.xml', verbosity=0)
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: XML identified as leading cause of cancer>',
|
|
'<Article: Django conquers world!>',
|
|
'<Article: Copyright is fine the way it is>',
|
|
'<Article: Poker on TV is great!>',
|
|
])
|
|
|
|
# Load fixture 6, JSON file with dynamic ContentType fields. Testing ManyToOne.
|
|
management.call_command('loaddata', 'fixture6.json', verbosity=0)
|
|
self.assertQuerysetEqual(Tag.objects.all(), [
|
|
'<Tag: <Article: Copyright is fine the way it is> tagged "copyright">',
|
|
'<Tag: <Article: Copyright is fine the way it is> tagged "law">',
|
|
], ordered=False)
|
|
|
|
# Load fixture 7, XML file with dynamic ContentType fields. Testing ManyToOne.
|
|
management.call_command('loaddata', 'fixture7.xml', verbosity=0)
|
|
self.assertQuerysetEqual(Tag.objects.all(), [
|
|
'<Tag: <Article: Copyright is fine the way it is> tagged "copyright">',
|
|
'<Tag: <Article: Copyright is fine the way it is> tagged "legal">',
|
|
'<Tag: <Article: Django conquers world!> tagged "django">',
|
|
'<Tag: <Article: Django conquers world!> tagged "world domination">',
|
|
], ordered=False)
|
|
|
|
# Load fixture 8, JSON file with dynamic Permission fields. Testing ManyToMany.
|
|
management.call_command('loaddata', 'fixture8.json', verbosity=0)
|
|
self.assertQuerysetEqual(Visa.objects.all(), [
|
|
'<Visa: Django Reinhardt Can add user, Can change user, Can delete user>',
|
|
'<Visa: Stephane Grappelli Can add user>',
|
|
'<Visa: Prince >'
|
|
], ordered=False)
|
|
|
|
# Load fixture 9, XML file with dynamic Permission fields. Testing ManyToMany.
|
|
management.call_command('loaddata', 'fixture9.xml', verbosity=0)
|
|
self.assertQuerysetEqual(Visa.objects.all(), [
|
|
'<Visa: Django Reinhardt Can add user, Can change user, Can delete user>',
|
|
'<Visa: Stephane Grappelli Can add user, Can delete user>',
|
|
'<Visa: Artist formerly known as "Prince" Can change user>'
|
|
], ordered=False)
|
|
|
|
self.assertQuerysetEqual(Book.objects.all(), [
|
|
'<Book: Achieving self-awareness of Python programs>',
|
|
'<Book: Music for all ages by Artist formerly known as "Prince" and Django Reinhardt>'
|
|
])
|
|
|
|
# Loading a fixture that doesn't exist emits a warning
|
|
with warnings.catch_warnings(record=True) as w:
|
|
warnings.simplefilter("always")
|
|
management.call_command('loaddata', 'unknown.json', verbosity=0)
|
|
self.assertEqual(len(w), 1)
|
|
self.assertTrue(w[0].message, "No fixture named 'unknown' found.")
|
|
|
|
# An attempt to load a nonexistent 'initial_data' fixture isn't an error
|
|
with warnings.catch_warnings(record=True) as w:
|
|
management.call_command('loaddata', 'initial_data.json', verbosity=0)
|
|
self.assertEqual(len(w), 0)
|
|
|
|
# object list is unaffected
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: XML identified as leading cause of cancer>',
|
|
'<Article: Django conquers world!>',
|
|
'<Article: Copyright is fine the way it is>',
|
|
'<Article: Poker on TV is great!>',
|
|
])
|
|
|
|
# By default, you get raw keys on dumpdata
|
|
self._dumpdata_assert(['fixtures.book'], '[{"pk": 1, "model": "fixtures.book", "fields": {"name": "Music for all ages", "authors": [3, 1]}}, {"pk": 10, "model": "fixtures.book", "fields": {"name": "Achieving self-awareness of Python programs", "authors": []}}]')
|
|
|
|
# But you can get natural keys if you ask for them and they are available
|
|
self._dumpdata_assert(['fixtures.book'], '[{"pk": 1, "model": "fixtures.book", "fields": {"name": "Music for all ages", "authors": [["Artist formerly known as \\"Prince\\""], ["Django Reinhardt"]]}}, {"pk": 10, "model": "fixtures.book", "fields": {"name": "Achieving self-awareness of Python programs", "authors": []}}]', natural_keys=True)
|
|
|
|
# Dump the current contents of the database as a JSON fixture
|
|
self._dumpdata_assert(['fixtures'], '[{"pk": 1, "model": "fixtures.category", "fields": {"description": "Latest news stories", "title": "News Stories"}}, {"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker on TV is great!", "pub_date": "2006-06-16T11:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Copyright is fine the way it is", "pub_date": "2006-06-16T14:00:00"}}, {"pk": 4, "model": "fixtures.article", "fields": {"headline": "Django conquers world!", "pub_date": "2006-06-16T15:00:00"}}, {"pk": 5, "model": "fixtures.article", "fields": {"headline": "XML identified as leading cause of cancer", "pub_date": "2006-06-16T16:00:00"}}, {"pk": 1, "model": "fixtures.tag", "fields": {"tagged_type": ["fixtures", "article"], "name": "copyright", "tagged_id": 3}}, {"pk": 2, "model": "fixtures.tag", "fields": {"tagged_type": ["fixtures", "article"], "name": "legal", "tagged_id": 3}}, {"pk": 3, "model": "fixtures.tag", "fields": {"tagged_type": ["fixtures", "article"], "name": "django", "tagged_id": 4}}, {"pk": 4, "model": "fixtures.tag", "fields": {"tagged_type": ["fixtures", "article"], "name": "world domination", "tagged_id": 4}}, {"pk": 1, "model": "fixtures.person", "fields": {"name": "Django Reinhardt"}}, {"pk": 2, "model": "fixtures.person", "fields": {"name": "Stephane Grappelli"}}, {"pk": 3, "model": "fixtures.person", "fields": {"name": "Artist formerly known as \\"Prince\\""}}, {"pk": 1, "model": "fixtures.visa", "fields": {"person": ["Django Reinhardt"], "permissions": [["add_user", "auth", "user"], ["change_user", "auth", "user"], ["delete_user", "auth", "user"]]}}, {"pk": 2, "model": "fixtures.visa", "fields": {"person": ["Stephane Grappelli"], "permissions": [["add_user", "auth", "user"], ["delete_user", "auth", "user"]]}}, {"pk": 3, "model": "fixtures.visa", "fields": {"person": ["Artist formerly known as \\"Prince\\""], "permissions": [["change_user", "auth", "user"]]}}, {"pk": 1, "model": "fixtures.book", "fields": {"name": "Music for all ages", "authors": [["Artist formerly known as \\"Prince\\""], ["Django Reinhardt"]]}}, {"pk": 10, "model": "fixtures.book", "fields": {"name": "Achieving self-awareness of Python programs", "authors": []}}]', natural_keys=True)
|
|
|
|
# Dump the current contents of the database as an XML fixture
|
|
self._dumpdata_assert(['fixtures'], """<?xml version="1.0" encoding="utf-8"?>
|
|
<django-objects version="1.0"><object pk="1" model="fixtures.category"><field type="CharField" name="title">News Stories</field><field type="TextField" name="description">Latest news stories</field></object><object pk="2" model="fixtures.article"><field type="CharField" name="headline">Poker on TV is great!</field><field type="DateTimeField" name="pub_date">2006-06-16T11:00:00</field></object><object pk="3" model="fixtures.article"><field type="CharField" name="headline">Copyright is fine the way it is</field><field type="DateTimeField" name="pub_date">2006-06-16T14:00:00</field></object><object pk="4" model="fixtures.article"><field type="CharField" name="headline">Django conquers world!</field><field type="DateTimeField" name="pub_date">2006-06-16T15:00:00</field></object><object pk="5" model="fixtures.article"><field type="CharField" name="headline">XML identified as leading cause of cancer</field><field type="DateTimeField" name="pub_date">2006-06-16T16:00:00</field></object><object pk="1" model="fixtures.tag"><field type="CharField" name="name">copyright</field><field to="contenttypes.contenttype" name="tagged_type" rel="ManyToOneRel"><natural>fixtures</natural><natural>article</natural></field><field type="PositiveIntegerField" name="tagged_id">3</field></object><object pk="2" model="fixtures.tag"><field type="CharField" name="name">legal</field><field to="contenttypes.contenttype" name="tagged_type" rel="ManyToOneRel"><natural>fixtures</natural><natural>article</natural></field><field type="PositiveIntegerField" name="tagged_id">3</field></object><object pk="3" model="fixtures.tag"><field type="CharField" name="name">django</field><field to="contenttypes.contenttype" name="tagged_type" rel="ManyToOneRel"><natural>fixtures</natural><natural>article</natural></field><field type="PositiveIntegerField" name="tagged_id">4</field></object><object pk="4" model="fixtures.tag"><field type="CharField" name="name">world domination</field><field to="contenttypes.contenttype" name="tagged_type" rel="ManyToOneRel"><natural>fixtures</natural><natural>article</natural></field><field type="PositiveIntegerField" name="tagged_id">4</field></object><object pk="1" model="fixtures.person"><field type="CharField" name="name">Django Reinhardt</field></object><object pk="2" model="fixtures.person"><field type="CharField" name="name">Stephane Grappelli</field></object><object pk="3" model="fixtures.person"><field type="CharField" name="name">Artist formerly known as "Prince"</field></object><object pk="1" model="fixtures.visa"><field to="fixtures.person" name="person" rel="ManyToOneRel"><natural>Django Reinhardt</natural></field><field to="auth.permission" name="permissions" rel="ManyToManyRel"><object><natural>add_user</natural><natural>auth</natural><natural>user</natural></object><object><natural>change_user</natural><natural>auth</natural><natural>user</natural></object><object><natural>delete_user</natural><natural>auth</natural><natural>user</natural></object></field></object><object pk="2" model="fixtures.visa"><field to="fixtures.person" name="person" rel="ManyToOneRel"><natural>Stephane Grappelli</natural></field><field to="auth.permission" name="permissions" rel="ManyToManyRel"><object><natural>add_user</natural><natural>auth</natural><natural>user</natural></object><object><natural>delete_user</natural><natural>auth</natural><natural>user</natural></object></field></object><object pk="3" model="fixtures.visa"><field to="fixtures.person" name="person" rel="ManyToOneRel"><natural>Artist formerly known as "Prince"</natural></field><field to="auth.permission" name="permissions" rel="ManyToManyRel"><object><natural>change_user</natural><natural>auth</natural><natural>user</natural></object></field></object><object pk="1" model="fixtures.book"><field type="CharField" name="name">Music for all ages</field><field to="fixtures.person" name="authors" rel="ManyToManyRel"><object><natural>Artist formerly known as "Prince"</natural></object><object><natural>Django Reinhardt</natural></object></field></object><object pk="10" model="fixtures.book"><field type="CharField" name="name">Achieving self-awareness of Python programs</field><field to="fixtures.person" name="authors" rel="ManyToManyRel"></field></object></django-objects>""", format='xml', natural_keys=True)
|
|
|
|
def test_dumpdata_with_excludes(self):
|
|
# Load fixture1 which has a site, two articles, and a category
|
|
Site.objects.all().delete()
|
|
management.call_command('loaddata', 'fixture1.json', verbosity=0)
|
|
|
|
# Excluding fixtures app should only leave sites
|
|
self._dumpdata_assert(
|
|
['sites', 'fixtures'],
|
|
'[{"pk": 1, "model": "sites.site", "fields": {"domain": "example.com", "name": "example.com"}}]',
|
|
exclude_list=['fixtures'])
|
|
|
|
# Excluding fixtures.Article/Book should leave fixtures.Category
|
|
self._dumpdata_assert(
|
|
['sites', 'fixtures'],
|
|
'[{"pk": 1, "model": "sites.site", "fields": {"domain": "example.com", "name": "example.com"}}, {"pk": 1, "model": "fixtures.category", "fields": {"description": "Latest news stories", "title": "News Stories"}}]',
|
|
exclude_list=['fixtures.Article', 'fixtures.Book'])
|
|
|
|
# Excluding fixtures and fixtures.Article/Book should be a no-op
|
|
self._dumpdata_assert(
|
|
['sites', 'fixtures'],
|
|
'[{"pk": 1, "model": "sites.site", "fields": {"domain": "example.com", "name": "example.com"}}, {"pk": 1, "model": "fixtures.category", "fields": {"description": "Latest news stories", "title": "News Stories"}}]',
|
|
exclude_list=['fixtures.Article', 'fixtures.Book'])
|
|
|
|
# Excluding sites and fixtures.Article/Book should only leave fixtures.Category
|
|
self._dumpdata_assert(
|
|
['sites', 'fixtures'],
|
|
'[{"pk": 1, "model": "fixtures.category", "fields": {"description": "Latest news stories", "title": "News Stories"}}]',
|
|
exclude_list=['fixtures.Article', 'fixtures.Book', 'sites'])
|
|
|
|
# Excluding a bogus app should throw an error
|
|
with six.assertRaisesRegex(self, management.CommandError,
|
|
"Unknown app in excludes: foo_app"):
|
|
self._dumpdata_assert(['fixtures', 'sites'], '', exclude_list=['foo_app'])
|
|
|
|
# Excluding a bogus model should throw an error
|
|
with six.assertRaisesRegex(self, management.CommandError,
|
|
"Unknown model in excludes: fixtures.FooModel"):
|
|
self._dumpdata_assert(['fixtures', 'sites'], '', exclude_list=['fixtures.FooModel'])
|
|
|
|
def test_dumpdata_with_filtering_manager(self):
|
|
spy1 = Spy.objects.create(name='Paul')
|
|
spy2 = Spy.objects.create(name='Alex', cover_blown=True)
|
|
self.assertQuerysetEqual(Spy.objects.all(),
|
|
['<Spy: Paul>'])
|
|
# Use the default manager
|
|
self._dumpdata_assert(['fixtures.Spy'], '[{"pk": %d, "model": "fixtures.spy", "fields": {"cover_blown": false}}]' % spy1.pk)
|
|
# Dump using Django's base manager. Should return all objects,
|
|
# even those normally filtered by the manager
|
|
self._dumpdata_assert(['fixtures.Spy'], '[{"pk": %d, "model": "fixtures.spy", "fields": {"cover_blown": true}}, {"pk": %d, "model": "fixtures.spy", "fields": {"cover_blown": false}}]' % (spy2.pk, spy1.pk), use_base_manager=True)
|
|
|
|
def test_dumpdata_with_pks(self):
|
|
management.call_command('loaddata', 'fixture1.json', verbosity=0)
|
|
management.call_command('loaddata', 'fixture2.json', verbosity=0)
|
|
self._dumpdata_assert(
|
|
['fixtures.Article'],
|
|
'[{"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Copyright is fine the way it is", "pub_date": "2006-06-16T14:00:00"}}]',
|
|
primary_keys='2,3'
|
|
)
|
|
|
|
self._dumpdata_assert(
|
|
['fixtures.Article'],
|
|
'[{"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}]',
|
|
primary_keys='2'
|
|
)
|
|
|
|
with six.assertRaisesRegex(self, management.CommandError,
|
|
"You can only use --pks option with one model"):
|
|
self._dumpdata_assert(
|
|
['fixtures'],
|
|
'[{"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Copyright is fine the way it is", "pub_date": "2006-06-16T14:00:00"}}]',
|
|
primary_keys='2,3'
|
|
)
|
|
|
|
with six.assertRaisesRegex(self, management.CommandError,
|
|
"You can only use --pks option with one model"):
|
|
self._dumpdata_assert(
|
|
'',
|
|
'[{"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Copyright is fine the way it is", "pub_date": "2006-06-16T14:00:00"}}]',
|
|
primary_keys='2,3'
|
|
)
|
|
|
|
with six.assertRaisesRegex(self, management.CommandError,
|
|
"You can only use --pks option with one model"):
|
|
self._dumpdata_assert(
|
|
['fixtures.Article', 'fixtures.category'],
|
|
'[{"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Copyright is fine the way it is", "pub_date": "2006-06-16T14:00:00"}}]',
|
|
primary_keys='2,3'
|
|
)
|
|
|
|
def test_compress_format_loading(self):
|
|
# Load fixture 4 (compressed), using format specification
|
|
management.call_command('loaddata', 'fixture4.json', verbosity=0)
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: Django pets kitten>',
|
|
])
|
|
|
|
def test_compressed_specified_loading(self):
|
|
# Load fixture 5 (compressed), using format *and* compression specification
|
|
management.call_command('loaddata', 'fixture5.json.zip', verbosity=0)
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: WoW subscribers now outnumber readers>',
|
|
])
|
|
|
|
def test_compressed_loading(self):
|
|
# Load fixture 5 (compressed), only compression specification
|
|
management.call_command('loaddata', 'fixture5.zip', verbosity=0)
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: WoW subscribers now outnumber readers>',
|
|
])
|
|
|
|
def test_ambiguous_compressed_fixture(self):
|
|
# The name "fixture5" is ambigous, so loading it will raise an error
|
|
with self.assertRaises(management.CommandError) as cm:
|
|
management.call_command('loaddata', 'fixture5', verbosity=0)
|
|
self.assertIn("Multiple fixtures named 'fixture5'", cm.exception.args[0])
|
|
|
|
def test_db_loading(self):
|
|
# Load db fixtures 1 and 2. These will load using the 'default' database identifier implicitly
|
|
management.call_command('loaddata', 'db_fixture_1', verbosity=0)
|
|
management.call_command('loaddata', 'db_fixture_2', verbosity=0)
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: Who needs more than one database?>',
|
|
'<Article: Who needs to use compressed data?>',
|
|
])
|
|
|
|
def test_loaddata_error_message(self):
|
|
"""
|
|
Verifies that loading a fixture which contains an invalid object
|
|
outputs an error message which contains the pk of the object
|
|
that triggered the error.
|
|
"""
|
|
# MySQL needs a little prodding to reject invalid data.
|
|
# This won't affect other tests because the database connection
|
|
# is closed at the end of each test.
|
|
if connection.vendor == 'mysql':
|
|
connection.cursor().execute("SET sql_mode = 'TRADITIONAL'")
|
|
with self.assertRaises(IntegrityError) as cm:
|
|
management.call_command('loaddata', 'invalid.json', verbosity=0)
|
|
self.assertIn("Could not load fixtures.Article(pk=1):", cm.exception.args[0])
|
|
|
|
def test_loading_using(self):
|
|
# Load db fixtures 1 and 2. These will load using the 'default' database identifier explicitly
|
|
management.call_command('loaddata', 'db_fixture_1', verbosity=0, using='default')
|
|
management.call_command('loaddata', 'db_fixture_2', verbosity=0, using='default')
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: Who needs more than one database?>',
|
|
'<Article: Who needs to use compressed data?>',
|
|
])
|
|
|
|
def test_unmatched_identifier_loading(self):
|
|
# Try to load db fixture 3. This won't load because the database identifier doesn't match
|
|
with warnings.catch_warnings(record=True):
|
|
management.call_command('loaddata', 'db_fixture_3', verbosity=0)
|
|
|
|
with warnings.catch_warnings(record=True):
|
|
management.call_command('loaddata', 'db_fixture_3', verbosity=0, using='default')
|
|
self.assertQuerysetEqual(Article.objects.all(), [])
|
|
|
|
def test_output_formats(self):
|
|
# Load back in fixture 1, we need the articles from it
|
|
management.call_command('loaddata', 'fixture1', verbosity=0)
|
|
|
|
# Try to load fixture 6 using format discovery
|
|
management.call_command('loaddata', 'fixture6', verbosity=0)
|
|
self.assertQuerysetEqual(Tag.objects.all(), [
|
|
'<Tag: <Article: Time to reform copyright> tagged "copyright">',
|
|
'<Tag: <Article: Time to reform copyright> tagged "law">'
|
|
], ordered=False)
|
|
|
|
# Dump the current contents of the database as a JSON fixture
|
|
self._dumpdata_assert(['fixtures'], '[{"pk": 1, "model": "fixtures.category", "fields": {"description": "Latest news stories", "title": "News Stories"}}, {"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Time to reform copyright", "pub_date": "2006-06-16T13:00:00"}}, {"pk": 1, "model": "fixtures.tag", "fields": {"tagged_type": ["fixtures", "article"], "name": "copyright", "tagged_id": 3}}, {"pk": 2, "model": "fixtures.tag", "fields": {"tagged_type": ["fixtures", "article"], "name": "law", "tagged_id": 3}}, {"pk": 1, "model": "fixtures.person", "fields": {"name": "Django Reinhardt"}}, {"pk": 2, "model": "fixtures.person", "fields": {"name": "Stephane Grappelli"}}, {"pk": 3, "model": "fixtures.person", "fields": {"name": "Prince"}}, {"pk": 10, "model": "fixtures.book", "fields": {"name": "Achieving self-awareness of Python programs", "authors": []}}]', natural_keys=True)
|
|
|
|
# Dump the current contents of the database as an XML fixture
|
|
self._dumpdata_assert(['fixtures'], """<?xml version="1.0" encoding="utf-8"?>
|
|
<django-objects version="1.0"><object pk="1" model="fixtures.category"><field type="CharField" name="title">News Stories</field><field type="TextField" name="description">Latest news stories</field></object><object pk="2" model="fixtures.article"><field type="CharField" name="headline">Poker has no place on ESPN</field><field type="DateTimeField" name="pub_date">2006-06-16T12:00:00</field></object><object pk="3" model="fixtures.article"><field type="CharField" name="headline">Time to reform copyright</field><field type="DateTimeField" name="pub_date">2006-06-16T13:00:00</field></object><object pk="1" model="fixtures.tag"><field type="CharField" name="name">copyright</field><field to="contenttypes.contenttype" name="tagged_type" rel="ManyToOneRel"><natural>fixtures</natural><natural>article</natural></field><field type="PositiveIntegerField" name="tagged_id">3</field></object><object pk="2" model="fixtures.tag"><field type="CharField" name="name">law</field><field to="contenttypes.contenttype" name="tagged_type" rel="ManyToOneRel"><natural>fixtures</natural><natural>article</natural></field><field type="PositiveIntegerField" name="tagged_id">3</field></object><object pk="1" model="fixtures.person"><field type="CharField" name="name">Django Reinhardt</field></object><object pk="2" model="fixtures.person"><field type="CharField" name="name">Stephane Grappelli</field></object><object pk="3" model="fixtures.person"><field type="CharField" name="name">Prince</field></object><object pk="10" model="fixtures.book"><field type="CharField" name="name">Achieving self-awareness of Python programs</field><field to="fixtures.person" name="authors" rel="ManyToManyRel"></field></object></django-objects>""", format='xml', natural_keys=True)
|
|
|
|
|
|
class FixtureTransactionTests(DumpDataAssertMixin, TransactionTestCase):
|
|
|
|
available_apps = [
|
|
'fixtures',
|
|
'django.contrib.contenttypes',
|
|
'django.contrib.auth',
|
|
'django.contrib.sites',
|
|
]
|
|
|
|
@skipUnlessDBFeature('supports_forward_references')
|
|
def test_format_discovery(self):
|
|
# Load fixture 1 again, using format discovery
|
|
management.call_command('loaddata', 'fixture1', verbosity=0)
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: Time to reform copyright>',
|
|
'<Article: Poker has no place on ESPN>',
|
|
])
|
|
|
|
# Try to load fixture 2 using format discovery; this will fail
|
|
# because there are two fixture2's in the fixtures directory
|
|
with self.assertRaises(management.CommandError) as cm:
|
|
management.call_command('loaddata', 'fixture2', verbosity=0)
|
|
self.assertIn("Multiple fixtures named 'fixture2'", cm.exception.args[0])
|
|
|
|
# object list is unaffected
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: Time to reform copyright>',
|
|
'<Article: Poker has no place on ESPN>',
|
|
])
|
|
|
|
# Dump the current contents of the database as a JSON fixture
|
|
self._dumpdata_assert(['fixtures'], '[{"pk": 1, "model": "fixtures.category", "fields": {"description": "Latest news stories", "title": "News Stories"}}, {"pk": 2, "model": "fixtures.article", "fields": {"headline": "Poker has no place on ESPN", "pub_date": "2006-06-16T12:00:00"}}, {"pk": 3, "model": "fixtures.article", "fields": {"headline": "Time to reform copyright", "pub_date": "2006-06-16T13:00:00"}}, {"pk": 10, "model": "fixtures.book", "fields": {"name": "Achieving self-awareness of Python programs", "authors": []}}]')
|
|
|
|
# Load fixture 4 (compressed), using format discovery
|
|
management.call_command('loaddata', 'fixture4', verbosity=0)
|
|
self.assertQuerysetEqual(Article.objects.all(), [
|
|
'<Article: Django pets kitten>',
|
|
'<Article: Time to reform copyright>',
|
|
'<Article: Poker has no place on ESPN>',
|
|
])
|