0
0
mirror of https://github.com/django/django.git synced 2024-11-29 14:46:18 +01:00
django/tests/test_runner/test_discover_runner.py

662 lines
27 KiB
Python

import logging
import multiprocessing
import os
import unittest.loader
from argparse import ArgumentParser
from contextlib import contextmanager
from importlib import import_module
from unittest import TestSuite, TextTestRunner, defaultTestLoader, mock
from django.db import connections
from django.test import SimpleTestCase
from django.test.runner import DiscoverRunner
from django.test.utils import (
NullTimeKeeper, TimeKeeper, captured_stderr, captured_stdout,
)
@contextmanager
def change_cwd(directory):
current_dir = os.path.abspath(os.path.dirname(__file__))
new_dir = os.path.join(current_dir, directory)
old_cwd = os.getcwd()
os.chdir(new_dir)
try:
yield
finally:
os.chdir(old_cwd)
@contextmanager
def change_loader_patterns(patterns):
original_patterns = DiscoverRunner.test_loader.testNamePatterns
DiscoverRunner.test_loader.testNamePatterns = patterns
try:
yield
finally:
DiscoverRunner.test_loader.testNamePatterns = original_patterns
# Isolate from the real environment.
@mock.patch.dict(os.environ, {}, clear=True)
@mock.patch.object(multiprocessing, 'cpu_count', return_value=12)
# Python 3.8 on macOS defaults to 'spawn' mode.
@mock.patch.object(multiprocessing, 'get_start_method', return_value='fork')
class DiscoverRunnerParallelArgumentTests(SimpleTestCase):
def get_parser(self):
parser = ArgumentParser()
DiscoverRunner.add_arguments(parser)
return parser
def test_parallel_default(self, *mocked_objects):
result = self.get_parser().parse_args([])
self.assertEqual(result.parallel, 0)
def test_parallel_flag(self, *mocked_objects):
result = self.get_parser().parse_args(['--parallel'])
self.assertEqual(result.parallel, 12)
def test_parallel_auto(self, *mocked_objects):
result = self.get_parser().parse_args(['--parallel', 'auto'])
self.assertEqual(result.parallel, 12)
def test_parallel_count(self, *mocked_objects):
result = self.get_parser().parse_args(['--parallel', '17'])
self.assertEqual(result.parallel, 17)
def test_parallel_invalid(self, *mocked_objects):
with self.assertRaises(SystemExit), captured_stderr() as stderr:
self.get_parser().parse_args(['--parallel', 'unaccepted'])
msg = "argument --parallel: 'unaccepted' is not an integer or the string 'auto'"
self.assertIn(msg, stderr.getvalue())
@mock.patch.dict(os.environ, {'DJANGO_TEST_PROCESSES': '7'})
def test_parallel_env_var(self, *mocked_objects):
result = self.get_parser().parse_args([])
self.assertEqual(result.parallel, 7)
@mock.patch.dict(os.environ, {'DJANGO_TEST_PROCESSES': 'typo'})
def test_parallel_env_var_non_int(self, *mocked_objects):
with self.assertRaises(ValueError):
self.get_parser().parse_args([])
def test_parallel_spawn(self, mocked_get_start_method, mocked_cpu_count):
mocked_get_start_method.return_value = 'spawn'
result = self.get_parser().parse_args(['--parallel'])
self.assertEqual(result.parallel, 1)
class DiscoverRunnerTests(SimpleTestCase):
@staticmethod
def get_test_methods_names(suite):
return [
t.__class__.__name__ + '.' + t._testMethodName
for t in suite._tests
]
def test_init_debug_mode(self):
runner = DiscoverRunner()
self.assertFalse(runner.debug_mode)
def test_add_arguments_shuffle(self):
parser = ArgumentParser()
DiscoverRunner.add_arguments(parser)
ns = parser.parse_args([])
self.assertIs(ns.shuffle, False)
ns = parser.parse_args(['--shuffle'])
self.assertIsNone(ns.shuffle)
ns = parser.parse_args(['--shuffle', '5'])
self.assertEqual(ns.shuffle, 5)
def test_add_arguments_debug_mode(self):
parser = ArgumentParser()
DiscoverRunner.add_arguments(parser)
ns = parser.parse_args([])
self.assertFalse(ns.debug_mode)
ns = parser.parse_args(["--debug-mode"])
self.assertTrue(ns.debug_mode)
def test_setup_shuffler_no_shuffle_argument(self):
runner = DiscoverRunner()
self.assertIs(runner.shuffle, False)
runner.setup_shuffler()
self.assertIsNone(runner.shuffle_seed)
def test_setup_shuffler_shuffle_none(self):
runner = DiscoverRunner(shuffle=None)
self.assertIsNone(runner.shuffle)
with mock.patch('random.randint', return_value=1):
with captured_stdout() as stdout:
runner.setup_shuffler()
self.assertEqual(stdout.getvalue(), 'Using shuffle seed: 1 (generated)\n')
self.assertEqual(runner.shuffle_seed, 1)
def test_setup_shuffler_shuffle_int(self):
runner = DiscoverRunner(shuffle=2)
self.assertEqual(runner.shuffle, 2)
with captured_stdout() as stdout:
runner.setup_shuffler()
expected_out = 'Using shuffle seed: 2 (given)\n'
self.assertEqual(stdout.getvalue(), expected_out)
self.assertEqual(runner.shuffle_seed, 2)
def test_load_tests_for_label_file_path(self):
with change_cwd('.'):
msg = (
"One of the test labels is a path to a file: "
"'test_discover_runner.py', which is not supported. Use a "
"dotted module name or path to a directory instead."
)
with self.assertRaisesMessage(RuntimeError, msg):
DiscoverRunner().load_tests_for_label('test_discover_runner.py', {})
def test_dotted_test_module(self):
count = DiscoverRunner(verbosity=0).build_suite(
['test_runner_apps.sample.tests_sample'],
).countTestCases()
self.assertEqual(count, 4)
def test_dotted_test_class_vanilla_unittest(self):
count = DiscoverRunner(verbosity=0).build_suite(
['test_runner_apps.sample.tests_sample.TestVanillaUnittest'],
).countTestCases()
self.assertEqual(count, 1)
def test_dotted_test_class_django_testcase(self):
count = DiscoverRunner(verbosity=0).build_suite(
['test_runner_apps.sample.tests_sample.TestDjangoTestCase'],
).countTestCases()
self.assertEqual(count, 1)
def test_dotted_test_method_django_testcase(self):
count = DiscoverRunner(verbosity=0).build_suite(
['test_runner_apps.sample.tests_sample.TestDjangoTestCase.test_sample'],
).countTestCases()
self.assertEqual(count, 1)
def test_pattern(self):
count = DiscoverRunner(
pattern="*_tests.py",
verbosity=0,
).build_suite(['test_runner_apps.sample']).countTestCases()
self.assertEqual(count, 1)
def test_name_patterns(self):
all_test_1 = [
'DjangoCase1.test_1', 'DjangoCase2.test_1',
'SimpleCase1.test_1', 'SimpleCase2.test_1',
'UnittestCase1.test_1', 'UnittestCase2.test_1',
]
all_test_2 = [
'DjangoCase1.test_2', 'DjangoCase2.test_2',
'SimpleCase1.test_2', 'SimpleCase2.test_2',
'UnittestCase1.test_2', 'UnittestCase2.test_2',
]
all_tests = sorted([*all_test_1, *all_test_2, 'UnittestCase2.test_3_test'])
for pattern, expected in [
[['test_1'], all_test_1],
[['UnittestCase1'], ['UnittestCase1.test_1', 'UnittestCase1.test_2']],
[['*test'], ['UnittestCase2.test_3_test']],
[['test*'], all_tests],
[['test'], all_tests],
[['test_1', 'test_2'], sorted([*all_test_1, *all_test_2])],
[['test*1'], all_test_1],
]:
with self.subTest(pattern):
suite = DiscoverRunner(
test_name_patterns=pattern,
verbosity=0,
).build_suite(['test_runner_apps.simple'])
self.assertEqual(expected, self.get_test_methods_names(suite))
def test_loader_patterns_not_mutated(self):
runner = DiscoverRunner(test_name_patterns=['test_sample'], verbosity=0)
tests = [
('test_runner_apps.sample.tests', 1),
('test_runner_apps.sample.tests.Test.test_sample', 1),
('test_runner_apps.sample.empty', 0),
('test_runner_apps.sample.tests_sample.EmptyTestCase', 0),
]
for test_labels, tests_count in tests:
with self.subTest(test_labels=test_labels):
with change_loader_patterns(['UnittestCase1']):
count = runner.build_suite([test_labels]).countTestCases()
self.assertEqual(count, tests_count)
self.assertEqual(runner.test_loader.testNamePatterns, ['UnittestCase1'])
def test_loader_patterns_not_mutated_when_test_label_is_file_path(self):
runner = DiscoverRunner(test_name_patterns=['test_sample'], verbosity=0)
with change_cwd('.'), change_loader_patterns(['UnittestCase1']):
with self.assertRaises(RuntimeError):
runner.build_suite(['test_discover_runner.py'])
self.assertEqual(runner.test_loader.testNamePatterns, ['UnittestCase1'])
def test_file_path(self):
with change_cwd(".."):
count = DiscoverRunner(verbosity=0).build_suite(
['test_runner_apps/sample/'],
).countTestCases()
self.assertEqual(count, 5)
def test_empty_label(self):
"""
If the test label is empty, discovery should happen on the current
working directory.
"""
with change_cwd("."):
suite = DiscoverRunner(verbosity=0).build_suite([])
self.assertEqual(
suite._tests[0].id().split(".")[0],
os.path.basename(os.getcwd()),
)
def test_empty_test_case(self):
count = DiscoverRunner(verbosity=0).build_suite(
['test_runner_apps.sample.tests_sample.EmptyTestCase'],
).countTestCases()
self.assertEqual(count, 0)
def test_discovery_on_package(self):
count = DiscoverRunner(verbosity=0).build_suite(
['test_runner_apps.sample.tests'],
).countTestCases()
self.assertEqual(count, 1)
def test_ignore_adjacent(self):
"""
When given a dotted path to a module, unittest discovery searches
not just the module, but also the directory containing the module.
This results in tests from adjacent modules being run when they
should not. The discover runner avoids this behavior.
"""
count = DiscoverRunner(verbosity=0).build_suite(
['test_runner_apps.sample.empty'],
).countTestCases()
self.assertEqual(count, 0)
def test_testcase_ordering(self):
with change_cwd(".."):
suite = DiscoverRunner(verbosity=0).build_suite(['test_runner_apps/sample/'])
self.assertEqual(
suite._tests[0].__class__.__name__,
'TestDjangoTestCase',
msg="TestDjangoTestCase should be the first test case")
self.assertEqual(
suite._tests[1].__class__.__name__,
'TestZimpleTestCase',
msg="TestZimpleTestCase should be the second test case")
# All others can follow in unspecified order, including doctests
self.assertIn('DocTestCase', [t.__class__.__name__ for t in suite._tests[2:]])
def test_duplicates_ignored(self):
"""
Tests shouldn't be discovered twice when discovering on overlapping paths.
"""
base_app = 'forms_tests'
sub_app = 'forms_tests.field_tests'
runner = DiscoverRunner(verbosity=0)
with self.modify_settings(INSTALLED_APPS={'append': sub_app}):
single = runner.build_suite([base_app]).countTestCases()
dups = runner.build_suite([base_app, sub_app]).countTestCases()
self.assertEqual(single, dups)
def test_reverse(self):
"""
Reverse should reorder tests while maintaining the grouping specified
by ``DiscoverRunner.reorder_by``.
"""
runner = DiscoverRunner(reverse=True, verbosity=0)
suite = runner.build_suite(
test_labels=('test_runner_apps.sample', 'test_runner_apps.simple'))
self.assertIn('test_runner_apps.simple', next(iter(suite)).id(),
msg="Test labels should be reversed.")
suite = runner.build_suite(test_labels=('test_runner_apps.simple',))
suite = tuple(suite)
self.assertIn('DjangoCase', suite[0].id(),
msg="Test groups should not be reversed.")
self.assertIn('SimpleCase', suite[4].id(),
msg="Test groups order should be preserved.")
self.assertIn('DjangoCase2', suite[0].id(),
msg="Django test cases should be reversed.")
self.assertIn('SimpleCase2', suite[4].id(),
msg="Simple test cases should be reversed.")
self.assertIn('UnittestCase2', suite[8].id(),
msg="Unittest test cases should be reversed.")
self.assertIn('test_2', suite[0].id(),
msg="Methods of Django cases should be reversed.")
self.assertIn('test_2', suite[4].id(),
msg="Methods of simple cases should be reversed.")
self.assertIn('test_2', suite[9].id(),
msg="Methods of unittest cases should be reversed.")
def test_build_suite_failed_tests_first(self):
# The "doesnotexist" label results in a _FailedTest instance.
suite = DiscoverRunner(verbosity=0).build_suite(
test_labels=['test_runner_apps.sample', 'doesnotexist'],
)
tests = list(suite)
self.assertIsInstance(tests[0], unittest.loader._FailedTest)
self.assertNotIsInstance(tests[-1], unittest.loader._FailedTest)
def test_build_suite_shuffling(self):
# These will result in unittest.loader._FailedTest instances rather
# than TestCase objects, but they are sufficient for testing.
labels = ['label1', 'label2', 'label3', 'label4']
cases = [
({}, ['label1', 'label2', 'label3', 'label4']),
({'reverse': True}, ['label4', 'label3', 'label2', 'label1']),
({'shuffle': 8}, ['label4', 'label1', 'label3', 'label2']),
({'shuffle': 8, 'reverse': True}, ['label2', 'label3', 'label1', 'label4']),
]
for kwargs, expected in cases:
with self.subTest(kwargs=kwargs):
# Prevent writing the seed to stdout.
runner = DiscoverRunner(**kwargs, verbosity=0)
tests = runner.build_suite(test_labels=labels)
# The ids have the form "unittest.loader._FailedTest.label1".
names = [test.id().split('.')[-1] for test in tests]
self.assertEqual(names, expected)
def test_overridable_get_test_runner_kwargs(self):
self.assertIsInstance(DiscoverRunner().get_test_runner_kwargs(), dict)
def test_overridable_test_suite(self):
self.assertEqual(DiscoverRunner().test_suite, TestSuite)
def test_overridable_test_runner(self):
self.assertEqual(DiscoverRunner().test_runner, TextTestRunner)
def test_overridable_test_loader(self):
self.assertEqual(DiscoverRunner().test_loader, defaultTestLoader)
def test_tags(self):
runner = DiscoverRunner(tags=['core'], verbosity=0)
self.assertEqual(runner.build_suite(['test_runner_apps.tagged.tests']).countTestCases(), 1)
runner = DiscoverRunner(tags=['fast'], verbosity=0)
self.assertEqual(runner.build_suite(['test_runner_apps.tagged.tests']).countTestCases(), 2)
runner = DiscoverRunner(tags=['slow'], verbosity=0)
self.assertEqual(runner.build_suite(['test_runner_apps.tagged.tests']).countTestCases(), 2)
def test_exclude_tags(self):
runner = DiscoverRunner(tags=['fast'], exclude_tags=['core'], verbosity=0)
self.assertEqual(runner.build_suite(['test_runner_apps.tagged.tests']).countTestCases(), 1)
runner = DiscoverRunner(tags=['fast'], exclude_tags=['slow'], verbosity=0)
self.assertEqual(runner.build_suite(['test_runner_apps.tagged.tests']).countTestCases(), 0)
runner = DiscoverRunner(exclude_tags=['slow'], verbosity=0)
self.assertEqual(runner.build_suite(['test_runner_apps.tagged.tests']).countTestCases(), 0)
def test_tag_inheritance(self):
def count_tests(**kwargs):
kwargs.setdefault('verbosity', 0)
suite = DiscoverRunner(**kwargs).build_suite(['test_runner_apps.tagged.tests_inheritance'])
return suite.countTestCases()
self.assertEqual(count_tests(tags=['foo']), 4)
self.assertEqual(count_tests(tags=['bar']), 2)
self.assertEqual(count_tests(tags=['baz']), 2)
self.assertEqual(count_tests(tags=['foo'], exclude_tags=['bar']), 2)
self.assertEqual(count_tests(tags=['foo'], exclude_tags=['bar', 'baz']), 1)
self.assertEqual(count_tests(exclude_tags=['foo']), 0)
def test_tag_fail_to_load(self):
with self.assertRaises(SyntaxError):
import_module('test_runner_apps.tagged.tests_syntax_error')
runner = DiscoverRunner(tags=['syntax_error'], verbosity=0)
# A label that doesn't exist or cannot be loaded due to syntax errors
# is always considered matching.
suite = runner.build_suite(['doesnotexist', 'test_runner_apps.tagged'])
self.assertEqual([test.id() for test in suite], [
'unittest.loader._FailedTest.doesnotexist',
'unittest.loader._FailedTest.test_runner_apps.tagged.tests_syntax_error',
])
def test_included_tags_displayed(self):
runner = DiscoverRunner(tags=['foo', 'bar'], verbosity=2)
with captured_stdout() as stdout:
runner.build_suite(['test_runner_apps.tagged.tests'])
self.assertIn('Including test tag(s): bar, foo.\n', stdout.getvalue())
def test_excluded_tags_displayed(self):
runner = DiscoverRunner(exclude_tags=['foo', 'bar'], verbosity=3)
with captured_stdout() as stdout:
runner.build_suite(['test_runner_apps.tagged.tests'])
self.assertIn('Excluding test tag(s): bar, foo.\n', stdout.getvalue())
def test_number_of_tests_found_displayed(self):
runner = DiscoverRunner()
with captured_stdout() as stdout:
runner.build_suite([
'test_runner_apps.sample.tests_sample.TestDjangoTestCase',
'test_runner_apps.simple',
])
self.assertIn('Found 14 test(s).\n', stdout.getvalue())
def test_pdb_with_parallel(self):
msg = (
'You cannot use --pdb with parallel tests; pass --parallel=1 to '
'use it.'
)
with self.assertRaisesMessage(ValueError, msg):
DiscoverRunner(pdb=True, parallel=2)
def test_number_of_parallel_workers(self):
"""Number of processes doesn't exceed the number of TestCases."""
runner = DiscoverRunner(parallel=5, verbosity=0)
suite = runner.build_suite(['test_runner_apps.tagged'])
self.assertEqual(suite.processes, len(suite.subsuites))
def test_buffer_mode_test_pass(self):
runner = DiscoverRunner(buffer=True, verbosity=0)
with captured_stdout() as stdout, captured_stderr() as stderr:
suite = runner.build_suite([
'test_runner_apps.buffer.tests_buffer.WriteToStdoutStderrTestCase.test_pass',
])
runner.run_suite(suite)
self.assertNotIn('Write to stderr.', stderr.getvalue())
self.assertNotIn('Write to stdout.', stdout.getvalue())
def test_buffer_mode_test_fail(self):
runner = DiscoverRunner(buffer=True, verbosity=0)
with captured_stdout() as stdout, captured_stderr() as stderr:
suite = runner.build_suite([
'test_runner_apps.buffer.tests_buffer.WriteToStdoutStderrTestCase.test_fail',
])
runner.run_suite(suite)
self.assertIn('Write to stderr.', stderr.getvalue())
self.assertIn('Write to stdout.', stdout.getvalue())
def run_suite_with_runner(self, runner_class, **kwargs):
class MyRunner(DiscoverRunner):
def test_runner(self, *args, **kwargs):
return runner_class()
runner = MyRunner(**kwargs)
# Suppress logging "Using shuffle seed" to the console.
with captured_stdout():
runner.setup_shuffler()
with captured_stdout() as stdout:
try:
result = runner.run_suite(None)
except RuntimeError as exc:
result = str(exc)
output = stdout.getvalue()
return result, output
def test_run_suite_logs_seed(self):
class TestRunner:
def run(self, suite):
return '<fake-result>'
expected_prefix = 'Used shuffle seed'
# Test with and without shuffling enabled.
result, output = self.run_suite_with_runner(TestRunner)
self.assertEqual(result, '<fake-result>')
self.assertNotIn(expected_prefix, output)
result, output = self.run_suite_with_runner(TestRunner, shuffle=2)
self.assertEqual(result, '<fake-result>')
expected_output = f'{expected_prefix}: 2 (given)\n'
self.assertEqual(output, expected_output)
def test_run_suite_logs_seed_exception(self):
"""
run_suite() logs the seed when TestRunner.run() raises an exception.
"""
class TestRunner:
def run(self, suite):
raise RuntimeError('my exception')
result, output = self.run_suite_with_runner(TestRunner, shuffle=2)
self.assertEqual(result, 'my exception')
expected_output = 'Used shuffle seed: 2 (given)\n'
self.assertEqual(output, expected_output)
@mock.patch('faulthandler.enable')
def test_faulthandler_enabled(self, mocked_enable):
with mock.patch('faulthandler.is_enabled', return_value=False):
DiscoverRunner(enable_faulthandler=True)
mocked_enable.assert_called()
@mock.patch('faulthandler.enable')
def test_faulthandler_already_enabled(self, mocked_enable):
with mock.patch('faulthandler.is_enabled', return_value=True):
DiscoverRunner(enable_faulthandler=True)
mocked_enable.assert_not_called()
@mock.patch('faulthandler.enable')
def test_faulthandler_enabled_fileno(self, mocked_enable):
# sys.stderr that is not an actual file.
with mock.patch('faulthandler.is_enabled', return_value=False), captured_stderr():
DiscoverRunner(enable_faulthandler=True)
mocked_enable.assert_called()
@mock.patch('faulthandler.enable')
def test_faulthandler_disabled(self, mocked_enable):
with mock.patch('faulthandler.is_enabled', return_value=False):
DiscoverRunner(enable_faulthandler=False)
mocked_enable.assert_not_called()
def test_timings_not_captured(self):
runner = DiscoverRunner(timing=False)
with captured_stderr() as stderr:
with runner.time_keeper.timed('test'):
pass
runner.time_keeper.print_results()
self.assertIsInstance(runner.time_keeper, NullTimeKeeper)
self.assertNotIn('test', stderr.getvalue())
def test_timings_captured(self):
runner = DiscoverRunner(timing=True)
with captured_stderr() as stderr:
with runner.time_keeper.timed('test'):
pass
runner.time_keeper.print_results()
self.assertIsInstance(runner.time_keeper, TimeKeeper)
self.assertIn('test', stderr.getvalue())
def test_log(self):
custom_low_level = 5
custom_high_level = 45
msg = 'logging message'
cases = [
(0, None, False),
(0, custom_low_level, False),
(0, logging.DEBUG, False),
(0, logging.INFO, False),
(0, logging.WARNING, False),
(0, custom_high_level, False),
(1, None, True),
(1, custom_low_level, False),
(1, logging.DEBUG, False),
(1, logging.INFO, True),
(1, logging.WARNING, True),
(1, custom_high_level, True),
(2, None, True),
(2, custom_low_level, True),
(2, logging.DEBUG, True),
(2, logging.INFO, True),
(2, logging.WARNING, True),
(2, custom_high_level, True),
(3, None, True),
(3, custom_low_level, True),
(3, logging.DEBUG, True),
(3, logging.INFO, True),
(3, logging.WARNING, True),
(3, custom_high_level, True),
]
for verbosity, level, output in cases:
with self.subTest(verbosity=verbosity, level=level):
with captured_stdout() as stdout:
runner = DiscoverRunner(verbosity=verbosity)
runner.log(msg, level)
self.assertEqual(stdout.getvalue(), f'{msg}\n' if output else '')
class DiscoverRunnerGetDatabasesTests(SimpleTestCase):
runner = DiscoverRunner(verbosity=2)
skip_msg = 'Skipping setup of unused database(s): '
def get_databases(self, test_labels):
with captured_stdout() as stdout:
suite = self.runner.build_suite(test_labels)
databases = self.runner.get_databases(suite)
return databases, stdout.getvalue()
def assertSkippedDatabases(self, test_labels, expected_databases):
databases, output = self.get_databases(test_labels)
self.assertEqual(databases, expected_databases)
skipped_databases = set(connections) - set(expected_databases)
if skipped_databases:
self.assertIn(self.skip_msg + ', '.join(sorted(skipped_databases)), output)
else:
self.assertNotIn(self.skip_msg, output)
def test_mixed(self):
databases, output = self.get_databases(['test_runner_apps.databases.tests'])
self.assertEqual(databases, {'default': True, 'other': False})
self.assertNotIn(self.skip_msg, output)
def test_all(self):
databases, output = self.get_databases(['test_runner_apps.databases.tests.AllDatabasesTests'])
self.assertEqual(databases, {alias: False for alias in connections})
self.assertNotIn(self.skip_msg, output)
def test_default_and_other(self):
self.assertSkippedDatabases([
'test_runner_apps.databases.tests.DefaultDatabaseTests',
'test_runner_apps.databases.tests.OtherDatabaseTests',
], {'default': False, 'other': False})
def test_default_only(self):
self.assertSkippedDatabases([
'test_runner_apps.databases.tests.DefaultDatabaseTests',
], {'default': False})
def test_other_only(self):
self.assertSkippedDatabases([
'test_runner_apps.databases.tests.OtherDatabaseTests'
], {'other': False})
def test_no_databases_required(self):
self.assertSkippedDatabases([
'test_runner_apps.databases.tests.NoDatabaseTests'
], {})
def test_serialize(self):
databases, _ = self.get_databases([
'test_runner_apps.databases.tests.DefaultDatabaseSerializedTests'
])
self.assertEqual(databases, {'default': True})