зеркало из https://github.com/mozilla/gecko-dev.git
815 строки
31 KiB
Python
815 строки
31 KiB
Python
import gc
|
|
import mock
|
|
import os
|
|
import re
|
|
import types
|
|
import unittest
|
|
PYWIN32 = False
|
|
if os.name == 'nt':
|
|
try:
|
|
import win32file
|
|
PYWIN32 = True
|
|
except:
|
|
pass
|
|
|
|
|
|
import mozharness.base.errors as errors
|
|
import mozharness.base.log as log
|
|
from mozharness.base.log import DEBUG, INFO, WARNING, ERROR, CRITICAL, FATAL, IGNORE
|
|
import mozharness.base.script as script
|
|
from mozharness.base.config import parse_config_file
|
|
|
|
test_string = '''foo
|
|
bar
|
|
baz'''
|
|
|
|
|
|
class CleanupObj(script.ScriptMixin, log.LogMixin):
|
|
def __init__(self):
|
|
super(CleanupObj, self).__init__()
|
|
self.log_obj = None
|
|
self.config = {'log_level': ERROR}
|
|
|
|
|
|
def cleanup():
|
|
gc.collect()
|
|
c = CleanupObj()
|
|
for f in ('test_logs', 'test_dir', 'tmpfile_stdout', 'tmpfile_stderr'):
|
|
c.rmtree(f)
|
|
|
|
|
|
def get_debug_script_obj():
|
|
s = script.BaseScript(config={'log_type': 'multi',
|
|
'log_level': DEBUG},
|
|
initial_config_file='test/test.json')
|
|
return s
|
|
|
|
|
|
def _post_fatal(self, **kwargs):
|
|
fh = open('tmpfile_stdout', 'w')
|
|
print >>fh, test_string
|
|
fh.close()
|
|
|
|
|
|
# TestScript {{{1
|
|
class TestScript(unittest.TestCase):
|
|
def setUp(self):
|
|
cleanup()
|
|
self.s = None
|
|
|
|
def tearDown(self):
|
|
# Close the logfile handles, or windows can't remove the logs
|
|
if hasattr(self, 's') and isinstance(self.s, object):
|
|
del(self.s)
|
|
cleanup()
|
|
|
|
# test _dump_config_hierarchy() when --dump-config-hierarchy is passed
|
|
def test_dump_config_hierarchy_valid_files_len(self):
|
|
try:
|
|
self.s = script.BaseScript(
|
|
initial_config_file='test/test.json',
|
|
option_args=['--cfg', 'test/test_override.py,test/test_override2.py'],
|
|
config={'dump_config_hierarchy': True}
|
|
)
|
|
except SystemExit:
|
|
local_cfg_files = parse_config_file('test_logs/localconfigfiles.json')
|
|
# first let's see if the correct number of config files were
|
|
# realized
|
|
self.assertEqual(
|
|
len(local_cfg_files), 4,
|
|
msg="--dump-config-hierarchy dumped wrong number of config files"
|
|
)
|
|
|
|
def test_dump_config_hierarchy_keys_unique_and_valid(self):
|
|
try:
|
|
self.s = script.BaseScript(
|
|
initial_config_file='test/test.json',
|
|
option_args=['--cfg', 'test/test_override.py,test/test_override2.py'],
|
|
config={'dump_config_hierarchy': True}
|
|
)
|
|
except SystemExit:
|
|
local_cfg_files = parse_config_file('test_logs/localconfigfiles.json')
|
|
# now let's see if only unique items were added from each config
|
|
t_override = local_cfg_files.get('test/test_override.py', {})
|
|
self.assertTrue(
|
|
t_override.get('keep_string') == "don't change me" and len(t_override.keys()) == 1,
|
|
msg="--dump-config-hierarchy dumped wrong keys/value for "
|
|
"`test/test_override.py`. There should only be one "
|
|
"item and it should be unique to all the other "
|
|
"items in test_log/localconfigfiles.json."
|
|
)
|
|
|
|
def test_dump_config_hierarchy_matches_self_config(self):
|
|
try:
|
|
######
|
|
# we need temp_cfg because self.s will be gcollected (NoneType) by
|
|
# the time we get to SystemExit exception
|
|
# temp_cfg will differ from self.s.config because of
|
|
# 'dump_config_hierarchy'. we have to make a deepcopy because
|
|
# config is a locked dict
|
|
temp_s = script.BaseScript(
|
|
initial_config_file='test/test.json',
|
|
option_args=['--cfg', 'test/test_override.py,test/test_override2.py'],
|
|
)
|
|
from copy import deepcopy
|
|
temp_cfg = deepcopy(temp_s.config)
|
|
temp_cfg.update({'dump_config_hierarchy': True})
|
|
######
|
|
self.s = script.BaseScript(
|
|
initial_config_file='test/test.json',
|
|
option_args=['--cfg', 'test/test_override.py,test/test_override2.py'],
|
|
config={'dump_config_hierarchy': True}
|
|
)
|
|
except SystemExit:
|
|
local_cfg_files = parse_config_file('test_logs/localconfigfiles.json')
|
|
# finally let's just make sure that all the items added up, equals
|
|
# what we started with: self.config
|
|
target_cfg = {}
|
|
for cfg_file in local_cfg_files:
|
|
target_cfg.update(local_cfg_files[cfg_file])
|
|
self.assertEqual(
|
|
target_cfg, temp_cfg,
|
|
msg="all of the items (combined) in each cfg file dumped via "
|
|
"--dump-config-hierarchy does not equal self.config "
|
|
)
|
|
|
|
# test _dump_config() when --dump-config is passed
|
|
def test_dump_config_equals_self_config(self):
|
|
try:
|
|
######
|
|
# we need temp_cfg because self.s will be gcollected (NoneType) by
|
|
# the time we get to SystemExit exception
|
|
# temp_cfg will differ from self.s.config because of
|
|
# 'dump_config_hierarchy'. we have to make a deepcopy because
|
|
# config is a locked dict
|
|
temp_s = script.BaseScript(
|
|
initial_config_file='test/test.json',
|
|
option_args=['--cfg', 'test/test_override.py,test/test_override2.py'],
|
|
)
|
|
from copy import deepcopy
|
|
temp_cfg = deepcopy(temp_s.config)
|
|
temp_cfg.update({'dump_config': True})
|
|
######
|
|
self.s = script.BaseScript(
|
|
initial_config_file='test/test.json',
|
|
option_args=['--cfg', 'test/test_override.py,test/test_override2.py'],
|
|
config={'dump_config': True}
|
|
)
|
|
except SystemExit:
|
|
target_cfg = parse_config_file('test_logs/localconfig.json')
|
|
self.assertEqual(
|
|
target_cfg, temp_cfg,
|
|
msg="all of the items (combined) in each cfg file dumped via "
|
|
"--dump-config does not equal self.config "
|
|
)
|
|
|
|
def test_nonexistent_mkdir_p(self):
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
self.s.mkdir_p('test_dir/foo/bar/baz')
|
|
self.assertTrue(os.path.isdir('test_dir/foo/bar/baz'),
|
|
msg="mkdir_p error")
|
|
|
|
def test_existing_mkdir_p(self):
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
os.makedirs('test_dir/foo/bar/baz')
|
|
self.s.mkdir_p('test_dir/foo/bar/baz')
|
|
self.assertTrue(os.path.isdir('test_dir/foo/bar/baz'),
|
|
msg="mkdir_p error when dir exists")
|
|
|
|
def test_chdir(self):
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
cwd = os.getcwd()
|
|
self.s.chdir('test_logs')
|
|
self.assertEqual(os.path.join(cwd, "test_logs"), os.getcwd(),
|
|
msg="chdir error")
|
|
self.s.chdir(cwd)
|
|
|
|
def _test_log_helper(self, obj):
|
|
obj.debug("Testing DEBUG")
|
|
obj.warning("Testing WARNING")
|
|
obj.error("Testing ERROR")
|
|
obj.critical("Testing CRITICAL")
|
|
try:
|
|
obj.fatal("Testing FATAL")
|
|
except SystemExit:
|
|
pass
|
|
else:
|
|
self.assertTrue(False, msg="fatal() didn't SystemExit!")
|
|
|
|
def test_log(self):
|
|
self.s = get_debug_script_obj()
|
|
self.s.log_obj = None
|
|
self._test_log_helper(self.s)
|
|
del(self.s)
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
self._test_log_helper(self.s)
|
|
|
|
def test_run_nonexistent_command(self):
|
|
self.s = get_debug_script_obj()
|
|
self.s.run_command(command="this_cmd_should_not_exist --help",
|
|
env={'GARBLE': 'FARG'},
|
|
error_list=errors.PythonErrorList)
|
|
error_logsize = os.path.getsize("test_logs/test_error.log")
|
|
self.assertTrue(error_logsize > 0,
|
|
msg="command not found error not hit")
|
|
|
|
def test_run_command_in_bad_dir(self):
|
|
self.s = get_debug_script_obj()
|
|
self.s.run_command(command="ls",
|
|
cwd='/this_dir_should_not_exist',
|
|
error_list=errors.PythonErrorList)
|
|
error_logsize = os.path.getsize("test_logs/test_error.log")
|
|
self.assertTrue(error_logsize > 0,
|
|
msg="bad dir error not hit")
|
|
|
|
def test_get_output_from_command_in_bad_dir(self):
|
|
self.s = get_debug_script_obj()
|
|
self.s.get_output_from_command(command="ls", cwd='/this_dir_should_not_exist')
|
|
error_logsize = os.path.getsize("test_logs/test_error.log")
|
|
self.assertTrue(error_logsize > 0,
|
|
msg="bad dir error not hit")
|
|
|
|
def test_get_output_from_command_with_missing_file(self):
|
|
self.s = get_debug_script_obj()
|
|
self.s.get_output_from_command(command="ls /this_file_should_not_exist")
|
|
error_logsize = os.path.getsize("test_logs/test_error.log")
|
|
self.assertTrue(error_logsize > 0,
|
|
msg="bad file error not hit")
|
|
|
|
def test_get_output_from_command_with_missing_file2(self):
|
|
self.s = get_debug_script_obj()
|
|
self.s.run_command(
|
|
command="cat mozharness/base/errors.py",
|
|
error_list=[{
|
|
'substr': "error", 'level': ERROR
|
|
}, {
|
|
'regex': re.compile(',$'), 'level': IGNORE,
|
|
}, {
|
|
'substr': ']$', 'level': WARNING,
|
|
}])
|
|
error_logsize = os.path.getsize("test_logs/test_error.log")
|
|
self.assertTrue(error_logsize > 0,
|
|
msg="error list not working properly")
|
|
|
|
|
|
# TestHelperFunctions {{{1
|
|
class TestHelperFunctions(unittest.TestCase):
|
|
temp_file = "test_dir/mozilla"
|
|
|
|
def setUp(self):
|
|
cleanup()
|
|
self.s = None
|
|
|
|
def tearDown(self):
|
|
# Close the logfile handles, or windows can't remove the logs
|
|
if hasattr(self, 's') and isinstance(self.s, object):
|
|
del(self.s)
|
|
cleanup()
|
|
|
|
def _create_temp_file(self, contents=test_string):
|
|
os.mkdir('test_dir')
|
|
fh = open(self.temp_file, "w+")
|
|
fh.write(contents)
|
|
fh.close
|
|
|
|
def test_mkdir_p(self):
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
self.s.mkdir_p('test_dir')
|
|
self.assertTrue(os.path.isdir('test_dir'),
|
|
msg="mkdir_p error")
|
|
|
|
def test_get_output_from_command(self):
|
|
self._create_temp_file()
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
contents = self.s.get_output_from_command(["bash", "-c", "cat %s" % self.temp_file])
|
|
self.assertEqual(test_string, contents,
|
|
msg="get_output_from_command('cat file') differs from fh.write")
|
|
|
|
def test_run_command(self):
|
|
self._create_temp_file()
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
temp_file_name = os.path.basename(self.temp_file)
|
|
self.assertEqual(self.s.run_command("cat %s" % temp_file_name,
|
|
cwd="test_dir"), 0,
|
|
msg="run_command('cat file') did not exit 0")
|
|
|
|
def test_move1(self):
|
|
self._create_temp_file()
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
temp_file2 = '%s2' % self.temp_file
|
|
self.s.move(self.temp_file, temp_file2)
|
|
self.assertFalse(os.path.exists(self.temp_file),
|
|
msg="%s still exists after move()" % self.temp_file)
|
|
|
|
def test_move2(self):
|
|
self._create_temp_file()
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
temp_file2 = '%s2' % self.temp_file
|
|
self.s.move(self.temp_file, temp_file2)
|
|
self.assertTrue(os.path.exists(temp_file2),
|
|
msg="%s doesn't exist after move()" % temp_file2)
|
|
|
|
def test_copyfile(self):
|
|
self._create_temp_file()
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
temp_file2 = '%s2' % self.temp_file
|
|
self.s.copyfile(self.temp_file, temp_file2)
|
|
self.assertEqual(os.path.getsize(self.temp_file),
|
|
os.path.getsize(temp_file2),
|
|
msg="%s and %s are different sizes after copyfile()" %
|
|
(self.temp_file, temp_file2))
|
|
|
|
def test_existing_rmtree(self):
|
|
self._create_temp_file()
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
self.s.mkdir_p('test_dir/foo/bar/baz')
|
|
self.s.rmtree('test_dir')
|
|
self.assertFalse(os.path.exists('test_dir'),
|
|
msg="rmtree unsuccessful")
|
|
|
|
def test_nonexistent_rmtree(self):
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
status = self.s.rmtree('test_dir')
|
|
self.assertFalse(status, msg="nonexistent rmtree error")
|
|
|
|
@unittest.skipUnless(PYWIN32, "PyWin32 specific")
|
|
def test_long_dir_rmtree(self):
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
# create a very long path that the command-prompt cannot delete
|
|
# by using unicode format (max path length 32000)
|
|
path = u'\\\\?\\%s\\test_dir' % os.getcwd()
|
|
win32file.CreateDirectoryExW(u'.', path)
|
|
|
|
for x in range(0, 20):
|
|
print("path=%s" % path)
|
|
path = path + u'\\%sxxxxxxxxxxxxxxxxxxxx' % x
|
|
win32file.CreateDirectoryExW(u'.', path)
|
|
self.s.rmtree('test_dir')
|
|
self.assertFalse(os.path.exists('test_dir'),
|
|
msg="rmtree unsuccessful")
|
|
|
|
@unittest.skipUnless(PYWIN32, "PyWin32 specific")
|
|
def test_chmod_rmtree(self):
|
|
self._create_temp_file()
|
|
win32file.SetFileAttributesW(self.temp_file, win32file.FILE_ATTRIBUTE_READONLY)
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
self.s.rmtree('test_dir')
|
|
self.assertFalse(os.path.exists('test_dir'),
|
|
msg="rmtree unsuccessful")
|
|
|
|
@unittest.skipIf(os.name == "nt", "Not for Windows")
|
|
def test_chmod(self):
|
|
self._create_temp_file()
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
self.s.chmod(self.temp_file, 0100700)
|
|
self.assertEqual(os.stat(self.temp_file)[0], 33216,
|
|
msg="chmod unsuccessful")
|
|
|
|
def test_env_normal(self):
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
script_env = self.s.query_env()
|
|
self.assertEqual(script_env, os.environ,
|
|
msg="query_env() != env\n%s\n%s" % (script_env, os.environ))
|
|
|
|
def test_env_normal2(self):
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
self.s.query_env()
|
|
script_env = self.s.query_env()
|
|
self.assertEqual(script_env, os.environ,
|
|
msg="Second query_env() != env\n%s\n%s" % (script_env, os.environ))
|
|
|
|
def test_env_partial(self):
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
script_env = self.s.query_env(partial_env={'foo': 'bar'})
|
|
self.assertTrue('foo' in script_env and script_env['foo'] == 'bar')
|
|
|
|
def test_env_path(self):
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
partial_path = "yaddayadda:%(PATH)s"
|
|
full_path = partial_path % {'PATH': os.environ['PATH']}
|
|
script_env = self.s.query_env(partial_env={'PATH': partial_path})
|
|
self.assertEqual(script_env['PATH'], full_path)
|
|
|
|
def test_query_exe(self):
|
|
self.s = script.BaseScript(
|
|
initial_config_file='test/test.json',
|
|
config={'exes': {'foo': 'bar'}},
|
|
)
|
|
path = self.s.query_exe('foo')
|
|
self.assertEqual(path, 'bar')
|
|
|
|
def test_query_exe_string_replacement(self):
|
|
self.s = script.BaseScript(
|
|
initial_config_file='test/test.json',
|
|
config={
|
|
'base_work_dir': 'foo',
|
|
'work_dir': 'bar',
|
|
'exes': {'foo': os.path.join('%(abs_work_dir)s', 'baz')},
|
|
},
|
|
)
|
|
path = self.s.query_exe('foo')
|
|
self.assertEqual(path, os.path.join('foo', 'bar', 'baz'))
|
|
|
|
def test_read_from_file(self):
|
|
self._create_temp_file()
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
contents = self.s.read_from_file(self.temp_file)
|
|
self.assertEqual(contents, test_string)
|
|
|
|
def test_read_from_nonexistent_file(self):
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
contents = self.s.read_from_file("nonexistent_file!!!")
|
|
self.assertEqual(contents, None)
|
|
|
|
|
|
# TestScriptLogging {{{1
|
|
class TestScriptLogging(unittest.TestCase):
|
|
# I need a log watcher helper function, here and in test_log.
|
|
def setUp(self):
|
|
cleanup()
|
|
self.s = None
|
|
|
|
def tearDown(self):
|
|
# Close the logfile handles, or windows can't remove the logs
|
|
if hasattr(self, 's') and isinstance(self.s, object):
|
|
del(self.s)
|
|
cleanup()
|
|
|
|
def test_info_logsize(self):
|
|
self.s = script.BaseScript(config={'log_type': 'multi'},
|
|
initial_config_file='test/test.json')
|
|
info_logsize = os.path.getsize("test_logs/test_info.log")
|
|
self.assertTrue(info_logsize > 0,
|
|
msg="initial info logfile missing/size 0")
|
|
|
|
def test_add_summary_info(self):
|
|
self.s = script.BaseScript(config={'log_type': 'multi'},
|
|
initial_config_file='test/test.json')
|
|
info_logsize = os.path.getsize("test_logs/test_info.log")
|
|
self.s.add_summary('one')
|
|
info_logsize2 = os.path.getsize("test_logs/test_info.log")
|
|
self.assertTrue(info_logsize < info_logsize2,
|
|
msg="add_summary() info not logged")
|
|
|
|
def test_add_summary_warning(self):
|
|
self.s = script.BaseScript(config={'log_type': 'multi'},
|
|
initial_config_file='test/test.json')
|
|
warning_logsize = os.path.getsize("test_logs/test_warning.log")
|
|
self.s.add_summary('two', level=WARNING)
|
|
warning_logsize2 = os.path.getsize("test_logs/test_warning.log")
|
|
self.assertTrue(warning_logsize < warning_logsize2,
|
|
msg="add_summary(level=%s) not logged in warning log" % WARNING)
|
|
|
|
def test_summary(self):
|
|
self.s = script.BaseScript(config={'log_type': 'multi'},
|
|
initial_config_file='test/test.json')
|
|
self.s.add_summary('one')
|
|
self.s.add_summary('two', level=WARNING)
|
|
info_logsize = os.path.getsize("test_logs/test_info.log")
|
|
warning_logsize = os.path.getsize("test_logs/test_warning.log")
|
|
self.s.summary()
|
|
info_logsize2 = os.path.getsize("test_logs/test_info.log")
|
|
warning_logsize2 = os.path.getsize("test_logs/test_warning.log")
|
|
msg = ""
|
|
if info_logsize >= info_logsize2:
|
|
msg += "summary() didn't log to info!\n"
|
|
if warning_logsize >= warning_logsize2:
|
|
msg += "summary() didn't log to warning!\n"
|
|
self.assertEqual(msg, "", msg=msg)
|
|
|
|
def _test_log_level(self, log_level, log_level_file_list):
|
|
self.s = script.BaseScript(config={'log_type': 'multi'},
|
|
initial_config_file='test/test.json')
|
|
if log_level != FATAL:
|
|
self.s.log('testing', level=log_level)
|
|
else:
|
|
self.s._post_fatal = types.MethodType(_post_fatal, self.s)
|
|
try:
|
|
self.s.fatal('testing')
|
|
except SystemExit:
|
|
contents = None
|
|
if os.path.exists('tmpfile_stdout'):
|
|
fh = open('tmpfile_stdout')
|
|
contents = fh.read()
|
|
fh.close()
|
|
self.assertEqual(contents.rstrip(), test_string, "_post_fatal failed!")
|
|
del(self.s)
|
|
msg = ""
|
|
for level in log_level_file_list:
|
|
log_path = "test_logs/test_%s.log" % level
|
|
if not os.path.exists(log_path):
|
|
msg += "%s doesn't exist!\n" % log_path
|
|
else:
|
|
filesize = os.path.getsize(log_path)
|
|
if not filesize > 0:
|
|
msg += "%s is size 0!\n" % log_path
|
|
self.assertEqual(msg, "", msg=msg)
|
|
|
|
def test_debug(self):
|
|
self._test_log_level(DEBUG, [])
|
|
|
|
def test_ignore(self):
|
|
self._test_log_level(IGNORE, [])
|
|
|
|
def test_info(self):
|
|
self._test_log_level(INFO, [INFO])
|
|
|
|
def test_warning(self):
|
|
self._test_log_level(WARNING, [INFO, WARNING])
|
|
|
|
def test_error(self):
|
|
self._test_log_level(ERROR, [INFO, WARNING, ERROR])
|
|
|
|
def test_critical(self):
|
|
self._test_log_level(CRITICAL, [INFO, WARNING, ERROR, CRITICAL])
|
|
|
|
def test_fatal(self):
|
|
self._test_log_level(FATAL, [INFO, WARNING, ERROR, CRITICAL, FATAL])
|
|
|
|
|
|
# TestRetry {{{1
|
|
class NewError(Exception):
|
|
pass
|
|
|
|
|
|
class OtherError(Exception):
|
|
pass
|
|
|
|
|
|
class TestRetry(unittest.TestCase):
|
|
def setUp(self):
|
|
self.ATTEMPT_N = 1
|
|
self.s = script.BaseScript(initial_config_file='test/test.json')
|
|
|
|
def tearDown(self):
|
|
# Close the logfile handles, or windows can't remove the logs
|
|
if hasattr(self, 's') and isinstance(self.s, object):
|
|
del(self.s)
|
|
cleanup()
|
|
|
|
def _succeedOnSecondAttempt(self, foo=None, exception=Exception):
|
|
if self.ATTEMPT_N == 2:
|
|
self.ATTEMPT_N += 1
|
|
return
|
|
self.ATTEMPT_N += 1
|
|
raise exception("Fail")
|
|
|
|
def _raiseCustomException(self):
|
|
return self._succeedOnSecondAttempt(exception=NewError)
|
|
|
|
def _alwaysPass(self):
|
|
self.ATTEMPT_N += 1
|
|
return True
|
|
|
|
def _mirrorArgs(self, *args, **kwargs):
|
|
return args, kwargs
|
|
|
|
def _alwaysFail(self):
|
|
raise Exception("Fail")
|
|
|
|
def testRetrySucceed(self):
|
|
# Will raise if anything goes wrong
|
|
self.s.retry(self._succeedOnSecondAttempt, attempts=2, sleeptime=0)
|
|
|
|
def testRetryFailWithoutCatching(self):
|
|
self.assertRaises(Exception, self.s.retry, self._alwaysFail, sleeptime=0,
|
|
exceptions=())
|
|
|
|
def testRetryFailEnsureRaisesLastException(self):
|
|
self.assertRaises(SystemExit, self.s.retry, self._alwaysFail, sleeptime=0,
|
|
error_level=FATAL)
|
|
|
|
def testRetrySelectiveExceptionSucceed(self):
|
|
self.s.retry(self._raiseCustomException, attempts=2, sleeptime=0,
|
|
retry_exceptions=(NewError,))
|
|
|
|
def testRetrySelectiveExceptionFail(self):
|
|
self.assertRaises(NewError, self.s.retry, self._raiseCustomException, attempts=2,
|
|
sleeptime=0, retry_exceptions=(OtherError,))
|
|
|
|
# TODO: figure out a way to test that the sleep actually happened
|
|
def testRetryWithSleep(self):
|
|
self.s.retry(self._succeedOnSecondAttempt, attempts=2, sleeptime=1)
|
|
|
|
def testRetryOnlyRunOnce(self):
|
|
"""Tests that retry() doesn't call the action again after success"""
|
|
self.s.retry(self._alwaysPass, attempts=3, sleeptime=0)
|
|
# self.ATTEMPT_N gets increased regardless of pass/fail
|
|
self.assertEquals(2, self.ATTEMPT_N)
|
|
|
|
def testRetryReturns(self):
|
|
ret = self.s.retry(self._alwaysPass, sleeptime=0)
|
|
self.assertEquals(ret, True)
|
|
|
|
def testRetryCleanupIsCalled(self):
|
|
cleanup = mock.Mock()
|
|
self.s.retry(self._succeedOnSecondAttempt, cleanup=cleanup, sleeptime=0)
|
|
self.assertEquals(cleanup.call_count, 1)
|
|
|
|
def testRetryArgsPassed(self):
|
|
args = (1, 'two', 3)
|
|
kwargs = dict(foo='a', bar=7)
|
|
ret = self.s.retry(self._mirrorArgs, args=args, kwargs=kwargs.copy(), sleeptime=0)
|
|
print ret
|
|
self.assertEqual(ret[0], args)
|
|
self.assertEqual(ret[1], kwargs)
|
|
|
|
|
|
class BaseScriptWithDecorators(script.BaseScript):
|
|
def __init__(self, *args, **kwargs):
|
|
super(BaseScriptWithDecorators, self).__init__(*args, **kwargs)
|
|
|
|
self.pre_run_1_args = []
|
|
self.raise_during_pre_run_1 = False
|
|
self.pre_action_1_args = []
|
|
self.raise_during_pre_action_1 = False
|
|
self.pre_action_2_args = []
|
|
self.pre_action_3_args = []
|
|
self.post_action_1_args = []
|
|
self.raise_during_post_action_1 = False
|
|
self.post_action_2_args = []
|
|
self.post_action_3_args = []
|
|
self.post_run_1_args = []
|
|
self.raise_during_post_run_1 = False
|
|
self.post_run_2_args = []
|
|
self.raise_during_build = False
|
|
|
|
@script.PreScriptRun
|
|
def pre_run_1(self, *args, **kwargs):
|
|
self.pre_run_1_args.append((args, kwargs))
|
|
|
|
if self.raise_during_pre_run_1:
|
|
raise Exception(self.raise_during_pre_run_1)
|
|
|
|
@script.PreScriptAction
|
|
def pre_action_1(self, *args, **kwargs):
|
|
self.pre_action_1_args.append((args, kwargs))
|
|
|
|
if self.raise_during_pre_action_1:
|
|
raise Exception(self.raise_during_pre_action_1)
|
|
|
|
@script.PreScriptAction
|
|
def pre_action_2(self, *args, **kwargs):
|
|
self.pre_action_2_args.append((args, kwargs))
|
|
|
|
@script.PreScriptAction('clobber')
|
|
def pre_action_3(self, *args, **kwargs):
|
|
self.pre_action_3_args.append((args, kwargs))
|
|
|
|
@script.PostScriptAction
|
|
def post_action_1(self, *args, **kwargs):
|
|
self.post_action_1_args.append((args, kwargs))
|
|
|
|
if self.raise_during_post_action_1:
|
|
raise Exception(self.raise_during_post_action_1)
|
|
|
|
@script.PostScriptAction
|
|
def post_action_2(self, *args, **kwargs):
|
|
self.post_action_2_args.append((args, kwargs))
|
|
|
|
@script.PostScriptAction('build')
|
|
def post_action_3(self, *args, **kwargs):
|
|
self.post_action_3_args.append((args, kwargs))
|
|
|
|
@script.PostScriptRun
|
|
def post_run_1(self, *args, **kwargs):
|
|
self.post_run_1_args.append((args, kwargs))
|
|
|
|
if self.raise_during_post_run_1:
|
|
raise Exception(self.raise_during_post_run_1)
|
|
|
|
@script.PostScriptRun
|
|
def post_run_2(self, *args, **kwargs):
|
|
self.post_run_2_args.append((args, kwargs))
|
|
|
|
def build(self):
|
|
if self.raise_during_build:
|
|
raise Exception(self.raise_during_build)
|
|
|
|
|
|
class TestScriptDecorators(unittest.TestCase):
|
|
def setUp(self):
|
|
cleanup()
|
|
self.s = None
|
|
|
|
def tearDown(self):
|
|
if hasattr(self, 's') and isinstance(self.s, object):
|
|
del self.s
|
|
|
|
cleanup()
|
|
|
|
def test_decorators_registered(self):
|
|
self.s = BaseScriptWithDecorators(initial_config_file='test/test.json')
|
|
|
|
self.assertEqual(len(self.s._listeners['pre_run']), 1)
|
|
self.assertEqual(len(self.s._listeners['pre_action']), 3)
|
|
self.assertEqual(len(self.s._listeners['post_action']), 3)
|
|
self.assertEqual(len(self.s._listeners['post_run']), 2)
|
|
|
|
def test_pre_post_fired(self):
|
|
self.s = BaseScriptWithDecorators(initial_config_file='test/test.json')
|
|
self.s.run()
|
|
|
|
self.assertEqual(len(self.s.pre_run_1_args), 1)
|
|
self.assertEqual(len(self.s.pre_action_1_args), 2)
|
|
self.assertEqual(len(self.s.pre_action_2_args), 2)
|
|
self.assertEqual(len(self.s.pre_action_3_args), 1)
|
|
self.assertEqual(len(self.s.post_action_1_args), 2)
|
|
self.assertEqual(len(self.s.post_action_2_args), 2)
|
|
self.assertEqual(len(self.s.post_action_3_args), 1)
|
|
self.assertEqual(len(self.s.post_run_1_args), 1)
|
|
|
|
self.assertEqual(self.s.pre_run_1_args[0], ((), {}))
|
|
|
|
self.assertEqual(self.s.pre_action_1_args[0], (('clobber',), {}))
|
|
self.assertEqual(self.s.pre_action_1_args[1], (('build',), {}))
|
|
|
|
# pre_action_3 should only get called for the action it is registered
|
|
# with.
|
|
self.assertEqual(self.s.pre_action_3_args[0], (('clobber',), {}))
|
|
|
|
self.assertEqual(self.s.post_action_1_args[0][0], ('clobber',))
|
|
self.assertEqual(self.s.post_action_1_args[0][1], dict(success=True))
|
|
self.assertEqual(self.s.post_action_1_args[1][0], ('build',))
|
|
self.assertEqual(self.s.post_action_1_args[1][1], dict(success=True))
|
|
|
|
# post_action_3 should only get called for the action it is registered
|
|
# with.
|
|
self.assertEqual(self.s.post_action_3_args[0], (('build',),
|
|
dict(success=True)))
|
|
|
|
self.assertEqual(self.s.post_run_1_args[0], ((), {}))
|
|
|
|
def test_post_always_fired(self):
|
|
self.s = BaseScriptWithDecorators(initial_config_file='test/test.json')
|
|
self.s.raise_during_build = 'Testing post always fired.'
|
|
|
|
with self.assertRaises(SystemExit):
|
|
self.s.run()
|
|
|
|
self.assertEqual(len(self.s.pre_run_1_args), 1)
|
|
self.assertEqual(len(self.s.pre_action_1_args), 2)
|
|
self.assertEqual(len(self.s.post_action_1_args), 2)
|
|
self.assertEqual(len(self.s.post_action_2_args), 2)
|
|
self.assertEqual(len(self.s.post_run_1_args), 1)
|
|
self.assertEqual(len(self.s.post_run_2_args), 1)
|
|
|
|
self.assertEqual(self.s.post_action_1_args[0][1], dict(success=True))
|
|
self.assertEqual(self.s.post_action_1_args[1][1], dict(success=False))
|
|
self.assertEqual(self.s.post_action_2_args[1][1], dict(success=False))
|
|
|
|
def test_pre_run_exception(self):
|
|
self.s = BaseScriptWithDecorators(initial_config_file='test/test.json')
|
|
self.s.raise_during_pre_run_1 = 'Error during pre run 1'
|
|
|
|
with self.assertRaises(SystemExit):
|
|
self.s.run()
|
|
|
|
self.assertEqual(len(self.s.pre_run_1_args), 1)
|
|
self.assertEqual(len(self.s.pre_action_1_args), 0)
|
|
self.assertEqual(len(self.s.post_run_1_args), 1)
|
|
self.assertEqual(len(self.s.post_run_2_args), 1)
|
|
|
|
def test_pre_action_exception(self):
|
|
self.s = BaseScriptWithDecorators(initial_config_file='test/test.json')
|
|
self.s.raise_during_pre_action_1 = 'Error during pre 1'
|
|
|
|
with self.assertRaises(SystemExit):
|
|
self.s.run()
|
|
|
|
self.assertEqual(len(self.s.pre_run_1_args), 1)
|
|
self.assertEqual(len(self.s.pre_action_1_args), 1)
|
|
self.assertEqual(len(self.s.pre_action_2_args), 0)
|
|
self.assertEqual(len(self.s.post_action_1_args), 1)
|
|
self.assertEqual(len(self.s.post_action_2_args), 1)
|
|
self.assertEqual(len(self.s.post_run_1_args), 1)
|
|
self.assertEqual(len(self.s.post_run_2_args), 1)
|
|
|
|
def test_post_action_exception(self):
|
|
self.s = BaseScriptWithDecorators(initial_config_file='test/test.json')
|
|
self.s.raise_during_post_action_1 = 'Error during post 1'
|
|
|
|
with self.assertRaises(SystemExit):
|
|
self.s.run()
|
|
|
|
self.assertEqual(len(self.s.pre_run_1_args), 1)
|
|
self.assertEqual(len(self.s.post_action_1_args), 1)
|
|
self.assertEqual(len(self.s.post_action_2_args), 1)
|
|
self.assertEqual(len(self.s.post_run_1_args), 1)
|
|
self.assertEqual(len(self.s.post_run_2_args), 1)
|
|
|
|
def test_post_run_exception(self):
|
|
self.s = BaseScriptWithDecorators(initial_config_file='test/test.json')
|
|
self.s.raise_during_post_run_1 = 'Error during post run 1'
|
|
|
|
with self.assertRaises(SystemExit):
|
|
self.s.run()
|
|
|
|
self.assertEqual(len(self.s.post_run_1_args), 1)
|
|
self.assertEqual(len(self.s.post_run_2_args), 1)
|
|
|
|
|
|
# main {{{1
|
|
if __name__ == '__main__':
|
|
unittest.main()
|