zephyr/scripts/tests/twister_blackbox/test_runner.py

679 lines
24 KiB
Python

#!/usr/bin/env python3
# Copyright (c) 2023 Intel Corporation
#
# SPDX-License-Identifier: Apache-2.0
"""
Blackbox tests for twister's command line functions
"""
# pylint: disable=duplicate-code
import importlib
import mock
import os
import pytest
import re
import sys
import time
from conftest import TEST_DATA, ZEPHYR_BASE, testsuite_filename_mock, clear_log_in_test
from twisterlib.testplan import TestPlan
@mock.patch.object(TestPlan, 'TESTSUITE_FILENAME', testsuite_filename_mock)
class TestRunner:
TESTDATA_1 = [
(
os.path.join(TEST_DATA, 'tests', 'dummy', 'agnostic'),
['qemu_x86', 'qemu_x86_64', 'intel_adl_crb'],
{
'executed_on_platform': 0,
'only_built': 6
}
),
(
os.path.join(TEST_DATA, 'tests', 'dummy', 'device'),
['qemu_x86', 'qemu_x86_64', 'intel_adl_crb'],
{
'executed_on_platform': 0,
'only_built': 1
}
),
]
TESTDATA_2 = [
(
os.path.join(TEST_DATA, 'tests', 'dummy', 'agnostic'),
['qemu_x86', 'qemu_x86_64', 'intel_adl_crb'],
{
'selected_test_scenarios': 3,
'selected_test_instances': 6,
'skipped_configurations': 0,
'skipped_by_static_filter': 0,
'skipped_at_runtime': 0,
'passed_configurations': 4,
'failed_configurations': 0,
'errored_configurations': 0,
'executed_test_cases': 8,
'skipped_test_cases': 0,
'platform_count': 0,
'executed_on_platform': 4,
'only_built': 2
}
)
]
TESTDATA_3 = [
(
os.path.join(TEST_DATA, 'tests', 'dummy', 'agnostic'),
['qemu_x86'],
),
]
TESTDATA_4 = [
(
os.path.join(TEST_DATA, 'tests', 'dummy', 'agnostic'),
['qemu_x86', 'qemu_x86_64'],
{
'passed_configurations': 6,
'selected_test_instances': 6,
'executed_on_platform': 0,
'only_built': 6,
}
),
]
TESTDATA_5 = [
(
os.path.join(TEST_DATA, 'tests', 'dummy', 'agnostic'),
['qemu_x86'],
os.path.join(TEST_DATA, "pre_script.sh")
),
]
TESTDATA_6 = [
(
os.path.join(TEST_DATA, 'tests', 'always_fail', 'dummy'),
['qemu_x86_64'],
'1',
),
(
os.path.join(TEST_DATA, 'tests', 'always_fail', 'dummy'),
['qemu_x86'],
'2',
),
]
TESTDATA_7 = [
(
os.path.join(TEST_DATA, 'tests', 'always_fail', 'dummy'),
['qemu_x86'],
'15',
),
(
os.path.join(TEST_DATA, 'tests', 'always_fail', 'dummy'),
['qemu_x86'],
'30',
),
]
TESTDATA_8 = [
(
os.path.join(TEST_DATA, 'tests', 'always_timeout', 'dummy'),
['qemu_x86'],
'2',
),
(
os.path.join(TEST_DATA, 'tests', 'always_timeout', 'dummy'),
['qemu_x86'],
'0.5',
),
]
TESTDATA_9 = [
(
os.path.join(TEST_DATA, 'tests', 'dummy'),
['qemu_x86'],
['device'],
['dummy.agnostic.group2 SKIPPED: Command line testsuite tag filter',
'dummy.agnostic.group1.subgroup2 SKIPPED: Command line testsuite tag filter',
'dummy.agnostic.group1.subgroup1 SKIPPED: Command line testsuite tag filter',
r'0 of 4 test configurations passed \(0.00%\), 0 failed, 0 errored, 4 skipped'
]
),
(
os.path.join(TEST_DATA, 'tests', 'dummy'),
['qemu_x86'],
['subgrouped'],
['dummy.agnostic.group2 SKIPPED: Command line testsuite tag filter',
r'2 of 4 test configurations passed \(100.00%\), 0 failed, 0 errored, 2 skipped'
]
),
(
os.path.join(TEST_DATA, 'tests', 'dummy'),
['qemu_x86'],
['agnostic', 'device'],
[r'3 of 4 test configurations passed \(100.00%\), 0 failed, 0 errored, 1 skipped']
),
]
TESTDATA_10 = [
(
os.path.join(TEST_DATA, 'tests', 'one_fail_one_pass'),
['qemu_x86'],
{
'selected_test_instances': 2,
'skipped_configurations': 0,
'passed_configurations': 0,
'failed_configurations': 1,
'errored_configurations': 0,
}
)
]
TESTDATA_11 = [
(
os.path.join(TEST_DATA, 'tests', 'always_build_error'),
['qemu_x86_64'],
'1',
),
(
os.path.join(TEST_DATA, 'tests', 'always_build_error'),
['qemu_x86'],
'4',
),
]
@classmethod
def setup_class(cls):
apath = os.path.join(ZEPHYR_BASE, 'scripts', 'twister')
cls.loader = importlib.machinery.SourceFileLoader('__main__', apath)
cls.spec = importlib.util.spec_from_loader(cls.loader.name, cls.loader)
cls.twister_module = importlib.util.module_from_spec(cls.spec)
@classmethod
def teardown_class(cls):
pass
@pytest.mark.parametrize(
'test_path, test_platforms, expected',
TESTDATA_1,
ids=[
'build_only tests/dummy/agnostic',
'build_only tests/dummy/device',
],
)
@pytest.mark.parametrize(
'flag',
['--build-only', '-b']
)
def test_build_only(self, capfd, out_path, test_path, test_platforms, expected, flag):
args = ['-i', '--outdir', out_path, '-T', test_path, flag] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
built_regex = r'^INFO - (?P<executed_on_platform>[0-9]+)' \
r' test configurations executed on platforms, (?P<only_built>[0-9]+)' \
r' test configurations were only built.$'
out, err = capfd.readouterr()
sys.stdout.write(out)
sys.stderr.write(err)
built_search = re.search(built_regex, err, re.MULTILINE)
assert built_search
assert int(built_search.group('executed_on_platform')) == \
expected['executed_on_platform']
assert int(built_search.group('only_built')) == \
expected['only_built']
assert str(sys_exit.value) == '0'
@pytest.mark.parametrize(
'test_path, test_platforms, expected',
TESTDATA_2,
ids=[
'test_only'
],
)
def test_runtest_only(self, capfd, out_path, test_path, test_platforms, expected):
args = ['--outdir', out_path,'-i', '-T', test_path, '--build-only'] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
capfd.readouterr()
clear_log_in_test()
args = ['--outdir', out_path,'-i', '-T', test_path, '--test-only'] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
select_regex = r'^INFO - (?P<test_scenarios>[0-9]+) test scenarios' \
r' \((?P<test_instances>[0-9]+) test instances\) selected,' \
r' (?P<skipped_configurations>[0-9]+) configurations skipped' \
r' \((?P<skipped_by_static_filter>[0-9]+) by static filter,' \
r' (?P<skipped_at_runtime>[0-9]+) at runtime\)\.$'
pass_regex = r'^INFO - (?P<passed_configurations>[0-9]+) of' \
r' (?P<test_instances>[0-9]+) test configurations passed' \
r' \([0-9]+\.[0-9]+%\), (?P<failed_configurations>[0-9]+) failed,' \
r' (?P<errored_configurations>[0-9]+) errored,' \
r' (?P<skipped_configurations>[0-9]+) skipped with' \
r' [0-9]+ warnings in [0-9]+\.[0-9]+ seconds$'
case_regex = r'^INFO - In total (?P<executed_test_cases>[0-9]+)' \
r' test cases were executed, (?P<skipped_test_cases>[0-9]+) skipped' \
r' on (?P<platform_count>[0-9]+) out of total [0-9]+ platforms' \
r' \([0-9]+\.[0-9]+%\)$'
built_regex = r'^INFO - (?P<executed_on_platform>[0-9]+)' \
r' test configurations executed on platforms, (?P<only_built>[0-9]+)' \
r' test configurations were only built.$'
out, err = capfd.readouterr()
sys.stdout.write(out)
sys.stderr.write(err)
select_search = re.search(select_regex, err, re.MULTILINE)
assert select_search
assert int(select_search.group('test_scenarios')) == \
expected['selected_test_scenarios']
assert int(select_search.group('test_instances')) == \
expected['selected_test_instances']
assert int(select_search.group('skipped_configurations')) == \
expected['skipped_configurations']
assert int(select_search.group('skipped_by_static_filter')) == \
expected['skipped_by_static_filter']
assert int(select_search.group('skipped_at_runtime')) == \
expected['skipped_at_runtime']
pass_search = re.search(pass_regex, err, re.MULTILINE)
assert pass_search
assert int(pass_search.group('passed_configurations')) == \
expected['passed_configurations']
assert int(pass_search.group('test_instances')) == \
expected['selected_test_instances']
assert int(pass_search.group('failed_configurations')) == \
expected['failed_configurations']
assert int(pass_search.group('errored_configurations')) == \
expected['errored_configurations']
assert int(pass_search.group('skipped_configurations')) == \
expected['skipped_configurations']
case_search = re.search(case_regex, err, re.MULTILINE)
assert case_search
assert int(case_search.group('executed_test_cases')) == \
expected['executed_test_cases']
assert int(case_search.group('skipped_test_cases')) == \
expected['skipped_test_cases']
assert int(case_search.group('platform_count')) == \
expected['platform_count']
built_search = re.search(built_regex, err, re.MULTILINE)
assert built_search
assert int(built_search.group('executed_on_platform')) == \
expected['executed_on_platform']
assert int(built_search.group('only_built')) == \
expected['only_built']
assert str(sys_exit.value) == '0'
@pytest.mark.parametrize(
'test_path, test_platforms',
TESTDATA_3,
ids=[
'dry_run',
],
)
@pytest.mark.parametrize(
'flag',
['--dry-run', '-y']
)
def test_dry_run(self, capfd, out_path, test_path, test_platforms, flag):
args = ['--outdir', out_path, '-T', test_path, flag] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
out, err = capfd.readouterr()
sys.stdout.write(out)
sys.stderr.write(err)
assert str(sys_exit.value) == '0'
@pytest.mark.parametrize(
'test_path, test_platforms, expected',
TESTDATA_4,
ids=[
'cmake_only',
],
)
def test_cmake_only(self, capfd, out_path, test_path, test_platforms, expected):
args = ['--outdir', out_path, '-T', test_path, '--cmake-only'] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
out, err = capfd.readouterr()
sys.stdout.write(out)
sys.stderr.write(err)
pass_regex = r'^INFO - (?P<passed_configurations>[0-9]+) of' \
r' (?P<test_instances>[0-9]+) test configurations passed'
built_regex = r'^INFO - (?P<executed_on_platform>[0-9]+)' \
r' test configurations executed on platforms, (?P<only_built>[0-9]+)' \
r' test configurations were only built.$'
pass_search = re.search(pass_regex, err, re.MULTILINE)
assert pass_search
assert int(pass_search.group('passed_configurations')) == \
expected['passed_configurations']
assert int(pass_search.group('test_instances')) == \
expected['selected_test_instances']
built_search = re.search(built_regex, err, re.MULTILINE)
assert built_search
assert int(built_search.group('executed_on_platform')) == \
expected['executed_on_platform']
assert int(built_search.group('only_built')) == \
expected['only_built']
assert str(sys_exit.value) == '0'
@pytest.mark.parametrize(
'test_path, test_platforms, file_name',
TESTDATA_5,
ids=[
'pre_script',
],
)
def test_pre_script(self, capfd, out_path, test_path, test_platforms, file_name):
args = ['--outdir', out_path, '-T', test_path, '--pre-script', file_name] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
out, err = capfd.readouterr()
sys.stdout.write(out)
sys.stderr.write(err)
assert str(sys_exit.value) == '0'
@pytest.mark.parametrize(
'test_path, test_platforms',
TESTDATA_3,
ids=[
'device_flash_timeout',
],
)
def test_device_flash_timeout(self, capfd, out_path, test_path, test_platforms):
args = ['--outdir', out_path, '-T', test_path, '--device-flash-timeout', "240"] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
out, err = capfd.readouterr()
sys.stdout.write(out)
sys.stderr.write(err)
assert str(sys_exit.value) == '0'
@pytest.mark.parametrize(
'test_path, test_platforms, iterations',
TESTDATA_6,
ids=[
'retry 2',
'retry 3'
],
)
def test_retry(self, capfd, out_path, test_path, test_platforms, iterations):
args = ['--outdir', out_path, '-T', test_path, '--retry-failed', iterations, '--retry-interval', '1'] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
out, err = capfd.readouterr()
sys.stdout.write(out)
sys.stderr.write(err)
pattern = re.compile(r'INFO\s+-\s+(\d+)\s+Iteration:[\s\S]*?ERROR\s+-\s+(\w+)')
matches = pattern.findall(err)
if matches:
last_iteration = max(int(match[0]) for match in matches)
last_match = next(match for match in matches if int(match[0]) == last_iteration)
iteration_number, platform_name = int(last_match[0]), last_match[1]
assert int(iteration_number) == int(iterations) + 1
assert [platform_name] == test_platforms
else:
assert 'Pattern not found in the output'
assert str(sys_exit.value) == '1'
@pytest.mark.parametrize(
'test_path, test_platforms, interval',
TESTDATA_7,
ids=[
'retry interval 15',
'retry interval 30'
],
)
def test_retry_interval(self, capfd, out_path, test_path, test_platforms, interval):
args = ['--outdir', out_path, '-T', test_path, '--retry-failed', '1', '--retry-interval', interval] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
start_time = time.time()
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
out, err = capfd.readouterr()
sys.stdout.write(out)
sys.stderr.write(err)
end_time = time.time()
elapsed_time = end_time - start_time
if elapsed_time < int(interval):
assert 'interval was too short'
assert str(sys_exit.value) == '1'
@pytest.mark.parametrize(
'test_path, test_platforms, timeout',
TESTDATA_8,
ids=[
'timeout-multiplier 2 - 20s',
'timeout-multiplier 0.5 - 5s'
],
)
def test_timeout_multiplier(self, capfd, out_path, test_path, test_platforms, timeout):
args = ['--outdir', out_path, '-T', test_path, '--timeout-multiplier', timeout, '-v'] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
tolerance = 1.0
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
out, err = capfd.readouterr()
sys.stdout.write(out)
sys.stderr.write(err)
elapsed_time = float(re.search(r'Timeout \(qemu (\d+\.\d+)s\)', err).group(1))
assert abs(
elapsed_time - float(timeout) * 10) <= tolerance, f"Time is different from expected"
assert str(sys_exit.value) == '1'
@pytest.mark.parametrize(
'test_path, test_platforms, tags, expected',
TESTDATA_9,
ids=[
'tags device',
'tags subgruped',
'tag agnostic and device'
],
)
def test_tag(self, capfd, out_path, test_path, test_platforms, tags, expected):
args = ['--outdir', out_path, '-T', test_path, '-vv', '-ll', 'DEBUG'] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair] + \
[val for pairs in zip(
['-t'] * len(tags), tags
) for val in pairs]
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
out, err = capfd.readouterr()
sys.stdout.write(out)
sys.stderr.write(err)
for line in expected:
assert re.search(line, err)
assert str(sys_exit.value) == '0'
@pytest.mark.parametrize(
'test_path, test_platforms, expected',
TESTDATA_10,
ids=[
'only_failed'
],
)
def test_only_failed(self, capfd, out_path, test_path, test_platforms, expected):
args = ['--outdir', out_path,'-i', '-T', test_path, '-v'] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
capfd.readouterr()
clear_log_in_test()
args = ['--outdir', out_path,'-i', '-T', test_path, '--only-failed'] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
pass_regex = r'^INFO - (?P<passed_configurations>[0-9]+) of' \
r' (?P<test_instances>[0-9]+) test configurations passed' \
r' \([0-9]+\.[0-9]+%\), (?P<failed_configurations>[0-9]+) failed,' \
r' (?P<errored_configurations>[0-9]+) errored,' \
r' (?P<skipped_configurations>[0-9]+) skipped with' \
r' [0-9]+ warnings in [0-9]+\.[0-9]+ seconds$'
out, err = capfd.readouterr()
sys.stdout.write(out)
sys.stderr.write(err)
assert re.search(
r'one_fail_one_pass.agnostic.group1.subgroup2 on qemu_x86 failed \(.*\)', err)
pass_search = re.search(pass_regex, err, re.MULTILINE)
assert pass_search
assert int(pass_search.group('passed_configurations')) == \
expected['passed_configurations']
assert int(pass_search.group('test_instances')) == \
expected['selected_test_instances']
assert int(pass_search.group('failed_configurations')) == \
expected['failed_configurations']
assert int(pass_search.group('errored_configurations')) == \
expected['errored_configurations']
assert int(pass_search.group('skipped_configurations')) == \
expected['skipped_configurations']
assert str(sys_exit.value) == '1'
@pytest.mark.parametrize(
'test_path, test_platforms, iterations',
TESTDATA_11,
ids=[
'retry 2',
'retry 3'
],
)
def test_retry_build_errors(self, capfd, out_path, test_path, test_platforms, iterations):
args = ['--outdir', out_path, '-T', test_path, '--retry-build-errors', '--retry-failed', iterations,
'--retry-interval', '10'] + \
[val for pair in zip(
['-p'] * len(test_platforms), test_platforms
) for val in pair]
with mock.patch.object(sys, 'argv', [sys.argv[0]] + args), \
pytest.raises(SystemExit) as sys_exit:
self.loader.exec_module(self.twister_module)
out, err = capfd.readouterr()
sys.stdout.write(out)
sys.stderr.write(err)
pattern = re.compile(r'INFO\s+-\s+(\d+)\s+Iteration:[\s\S]*?ERROR\s+-\s+(\w+)')
matches = pattern.findall(err)
if matches:
last_iteration = max(int(match[0]) for match in matches)
last_match = next(match for match in matches if int(match[0]) == last_iteration)
iteration_number, platform_name = int(last_match[0]), last_match[1]
assert int(iteration_number) == int(iterations) + 1
assert [platform_name] == test_platforms
else:
assert 'Pattern not found in the output'
assert str(sys_exit.value) == '1'