blob: f8bb23f344c697fbd7f1097327c4758a7e14a628 [file] [log] [blame]
# Copyright (C) 2016 Intel Corporation
# Released under the MIT license (see COPYING.MIT)
import os
import time
import unittest
import logging
import re
import json
from unittest import TextTestResult as _TestResult
from unittest import TextTestRunner as _TestRunner
class OEStreamLogger(object):
def __init__(self, logger):
self.logger = logger
self.buffer = ""
def write(self, msg):
if len(msg) > 1 and msg[0] != '\n':
if '...' in msg:
self.buffer += msg
elif self.buffer:
self.buffer += msg
self.logger.log(logging.INFO, self.buffer)
self.buffer = ""
else:
self.logger.log(logging.INFO, msg)
def flush(self):
for handler in self.logger.handlers:
handler.flush()
class OETestResult(_TestResult):
def __init__(self, tc, *args, **kwargs):
super(OETestResult, self).__init__(*args, **kwargs)
self.successes = []
# Inject into tc so that TestDepends decorator can see results
tc.results = self
self.tc = tc
def startTest(self, test):
# Allow us to trigger the testcase buffer mode on a per test basis
# so stdout/stderr are only printed upon failure. Enables debugging
# but clean output
if hasattr(test, "buffer"):
self.buffer = test.buffer
super(OETestResult, self).startTest(test)
def logSummary(self, component, context_msg=''):
elapsed_time = self.tc._run_end_time - self.tc._run_start_time
self.tc.logger.info("SUMMARY:")
self.tc.logger.info("%s (%s) - Ran %d test%s in %.3fs" % (component,
context_msg, self.testsRun, self.testsRun != 1 and "s" or "",
elapsed_time))
if self.wasSuccessful():
msg = "%s - OK - All required tests passed" % component
else:
msg = "%s - FAIL - Required tests failed" % component
msg += " (successes=%d, skipped=%d, failures=%d, errors=%d)" % (len(self.successes), len(self.skipped), len(self.failures), len(self.errors))
self.tc.logger.info(msg)
def _getTestResultDetails(self, case):
result_types = {'failures': 'FAILED', 'errors': 'ERROR', 'skipped': 'SKIPPED',
'expectedFailures': 'EXPECTEDFAIL', 'successes': 'PASSED'}
for rtype in result_types:
found = False
for (scase, msg) in getattr(self, rtype):
if case.id() == scase.id():
found = True
break
scase_str = str(scase.id())
# When fails at module or class level the class name is passed as string
# so figure out to see if match
m = re.search("^setUpModule \((?P<module_name>.*)\)$", scase_str)
if m:
if case.__class__.__module__ == m.group('module_name'):
found = True
break
m = re.search("^setUpClass \((?P<class_name>.*)\)$", scase_str)
if m:
class_name = "%s.%s" % (case.__class__.__module__,
case.__class__.__name__)
if class_name == m.group('class_name'):
found = True
break
if found:
return result_types[rtype], msg
return 'UNKNOWN', None
def addSuccess(self, test):
#Added so we can keep track of successes too
self.successes.append((test, None))
super(OETestResult, self).addSuccess(test)
def logDetails(self, json_file_dir=None, configuration=None, result_id=None):
self.tc.logger.info("RESULTS:")
result = {}
logs = {}
if hasattr(self.tc, "extraresults"):
result = self.tc.extraresults
for case_name in self.tc._registry['cases']:
case = self.tc._registry['cases'][case_name]
(status, log) = self._getTestResultDetails(case)
oeid = -1
if hasattr(case, 'decorators'):
for d in case.decorators:
if hasattr(d, 'oeid'):
oeid = d.oeid
if status not in logs:
logs[status] = []
logs[status].append("RESULTS - %s - Testcase %s: %s" % (case.id(), oeid, status))
if log:
result[case.id()] = {'status': status, 'log': log}
else:
result[case.id()] = {'status': status}
for i in ['PASSED', 'SKIPPED', 'EXPECTEDFAIL', 'ERROR', 'FAILED', 'UNKNOWN']:
if i not in logs:
continue
for l in logs[i]:
self.tc.logger.info(l)
if json_file_dir:
tresultjsonhelper = OETestResultJSONHelper()
tresultjsonhelper.dump_testresult_file(json_file_dir, configuration, result_id, result)
def wasSuccessful(self):
# Override as we unexpected successes aren't failures for us
return (len(self.failures) == len(self.errors) == 0)
class OEListTestsResult(object):
def wasSuccessful(self):
return True
class OETestRunner(_TestRunner):
streamLoggerClass = OEStreamLogger
def __init__(self, tc, *args, **kwargs):
kwargs['stream'] = self.streamLoggerClass(tc.logger)
super(OETestRunner, self).__init__(*args, **kwargs)
self.tc = tc
self.resultclass = OETestResult
def _makeResult(self):
return self.resultclass(self.tc, self.stream, self.descriptions,
self.verbosity)
def _walk_suite(self, suite, func):
for obj in suite:
if isinstance(obj, unittest.suite.TestSuite):
if len(obj._tests):
self._walk_suite(obj, func)
elif isinstance(obj, unittest.case.TestCase):
func(self.tc.logger, obj)
self._walked_cases = self._walked_cases + 1
def _list_tests_name(self, suite):
from oeqa.core.decorator.oeid import OETestID
from oeqa.core.decorator.oetag import OETestTag
self._walked_cases = 0
def _list_cases_without_id(logger, case):
found_id = False
if hasattr(case, 'decorators'):
for d in case.decorators:
if isinstance(d, OETestID):
found_id = True
if not found_id:
logger.info('oeid missing for %s' % case.id())
def _list_cases(logger, case):
oeid = None
oetag = None
if hasattr(case, 'decorators'):
for d in case.decorators:
if isinstance(d, OETestID):
oeid = d.oeid
elif isinstance(d, OETestTag):
oetag = d.oetag
logger.info("%s\t%s\t\t%s" % (oeid, oetag, case.id()))
self.tc.logger.info("Listing test cases that don't have oeid ...")
self._walk_suite(suite, _list_cases_without_id)
self.tc.logger.info("-" * 80)
self.tc.logger.info("Listing all available tests:")
self._walked_cases = 0
self.tc.logger.info("id\ttag\t\ttest")
self.tc.logger.info("-" * 80)
self._walk_suite(suite, _list_cases)
self.tc.logger.info("-" * 80)
self.tc.logger.info("Total found:\t%s" % self._walked_cases)
def _list_tests_class(self, suite):
self._walked_cases = 0
curr = {}
def _list_classes(logger, case):
if not 'module' in curr or curr['module'] != case.__module__:
curr['module'] = case.__module__
logger.info(curr['module'])
if not 'class' in curr or curr['class'] != \
case.__class__.__name__:
curr['class'] = case.__class__.__name__
logger.info(" -- %s" % curr['class'])
logger.info(" -- -- %s" % case._testMethodName)
self.tc.logger.info("Listing all available test classes:")
self._walk_suite(suite, _list_classes)
def _list_tests_module(self, suite):
self._walked_cases = 0
listed = []
def _list_modules(logger, case):
if not case.__module__ in listed:
if case.__module__.startswith('_'):
logger.info("%s (hidden)" % case.__module__)
else:
logger.info(case.__module__)
listed.append(case.__module__)
self.tc.logger.info("Listing all available test modules:")
self._walk_suite(suite, _list_modules)
def list_tests(self, suite, display_type):
if display_type == 'name':
self._list_tests_name(suite)
elif display_type == 'class':
self._list_tests_class(suite)
elif display_type == 'module':
self._list_tests_module(suite)
return OEListTestsResult()
class OETestResultJSONHelper(object):
testresult_filename = 'testresults.json'
def _get_existing_testresults_if_available(self, write_dir):
testresults = {}
file = os.path.join(write_dir, self.testresult_filename)
if os.path.exists(file):
with open(file, "r") as f:
testresults = json.load(f)
return testresults
def _write_file(self, write_dir, file_name, file_content):
file_path = os.path.join(write_dir, file_name)
with open(file_path, 'w') as the_file:
the_file.write(file_content)
def dump_testresult_file(self, write_dir, configuration, result_id, test_result):
bb.utils.mkdirhier(write_dir)
lf = bb.utils.lockfile(os.path.join(write_dir, 'jsontestresult.lock'))
test_results = self._get_existing_testresults_if_available(write_dir)
test_results[result_id] = {'configuration': configuration, 'result': test_result}
json_testresults = json.dumps(test_results, sort_keys=True, indent=4)
self._write_file(write_dir, self.testresult_filename, json_testresults)
bb.utils.unlockfile(lf)