zephyr/scripts/pylib/twister/twisterlib/harness.py

396 lines
14 KiB
Python
Raw Normal View History

# SPDX-License-Identifier: Apache-2.0
from asyncio.log import logger
import re
import os
import subprocess
import shlex
from collections import OrderedDict
import xml.etree.ElementTree as ET
import logging
logger = logging.getLogger('twister')
logger.setLevel(logging.DEBUG)
# pylint: disable=anomalous-backslash-in-string
result_re = re.compile(".*(PASS|FAIL|SKIP) - (test_)?(.*) in (\\d*[.,]?\\d*) seconds")
class Harness:
GCOV_START = "GCOV_COVERAGE_DUMP_START"
GCOV_END = "GCOV_COVERAGE_DUMP_END"
FAULT = "ZEPHYR FATAL ERROR"
sanitycheck: Complete overhaul and job handling rework A complete overhaul of the sanitycheck script and how we build and run tests. This new version of sanitycheck uses python for job distribution and drop use of Make. In addition to the move to python threading library, the following has been changed: - All handlers now run in parallel, meaning that any simulator will run in parallel and when testing on multiple devices (using --device-testing) the tests are run in parallel. - Lexicial filtering (using the filter keyword in yaml files) is now evaluated at runtime and is no long being pre-processed. This will allow us to immediately start executing tests and skip the wait time that was needed for filtering. - Device testing now supports multiple devices connected at the same time and is managed using a hardware map that needs to be generated and maintained for every test environment. (using --generate-hardware-map option). - Reports are not long stored in the Zephyr tree and instead stored in the output directory where all build artifacts are generated. - Each tested target now has a junit report in the output directory. - Recording option for performance data and other metrics is now available. This will allow us to record the output from the console and store the data for later processing. For example benchmark data can be captured and uploaded to a tracking server. - Test configurations (or instances) are no longer being sorted, this will help with balancing the load when we run sanitycheck on multiple hosts (as we do in CI). And many other cleanups and improvements... Signed-off-by: Anas Nashif <anas.nashif@intel.com>
2019-06-22 11:04:10 -04:00
RUN_PASSED = "PROJECT EXECUTION SUCCESSFUL"
RUN_FAILED = "PROJECT EXECUTION FAILED"
run_id_pattern = r"RunID: (?P<run_id>.*)"
ztest_to_status = {
'PASS': 'passed',
'SKIP': 'skipped',
'BLOCK': 'blocked',
'FAIL': 'failed'
}
def __init__(self):
self.state = None
self.type = None
self.regex = []
self.matches = OrderedDict()
self.ordered = True
self.repeat = 1
self.testcases = []
self.id = None
self.fail_on_fault = True
self.fault = False
self.capture_coverage = False
self.next_pattern = 0
sanitycheck: Complete overhaul and job handling rework A complete overhaul of the sanitycheck script and how we build and run tests. This new version of sanitycheck uses python for job distribution and drop use of Make. In addition to the move to python threading library, the following has been changed: - All handlers now run in parallel, meaning that any simulator will run in parallel and when testing on multiple devices (using --device-testing) the tests are run in parallel. - Lexicial filtering (using the filter keyword in yaml files) is now evaluated at runtime and is no long being pre-processed. This will allow us to immediately start executing tests and skip the wait time that was needed for filtering. - Device testing now supports multiple devices connected at the same time and is managed using a hardware map that needs to be generated and maintained for every test environment. (using --generate-hardware-map option). - Reports are not long stored in the Zephyr tree and instead stored in the output directory where all build artifacts are generated. - Each tested target now has a junit report in the output directory. - Recording option for performance data and other metrics is now available. This will allow us to record the output from the console and store the data for later processing. For example benchmark data can be captured and uploaded to a tracking server. - Test configurations (or instances) are no longer being sorted, this will help with balancing the load when we run sanitycheck on multiple hosts (as we do in CI). And many other cleanups and improvements... Signed-off-by: Anas Nashif <anas.nashif@intel.com>
2019-06-22 11:04:10 -04:00
self.record = None
self.recording = []
self.fieldnames = []
self.ztest = False
self.is_pytest = False
self.detected_suite_names = []
self.run_id = None
self.matched_run_id = False
self.run_id_exists = False
self.instance = None
self.testcase_output = ""
self._match = False
def configure(self, instance):
self.instance = instance
config = instance.testsuite.harness_config
self.id = instance.testsuite.id
self.run_id = instance.run_id
if instance.testsuite.ignore_faults:
self.fail_on_fault = False
if config:
self.type = config.get('type', None)
self.regex = config.get('regex', [])
self.repeat = config.get('repeat', 1)
self.ordered = config.get('ordered', True)
sanitycheck: Complete overhaul and job handling rework A complete overhaul of the sanitycheck script and how we build and run tests. This new version of sanitycheck uses python for job distribution and drop use of Make. In addition to the move to python threading library, the following has been changed: - All handlers now run in parallel, meaning that any simulator will run in parallel and when testing on multiple devices (using --device-testing) the tests are run in parallel. - Lexicial filtering (using the filter keyword in yaml files) is now evaluated at runtime and is no long being pre-processed. This will allow us to immediately start executing tests and skip the wait time that was needed for filtering. - Device testing now supports multiple devices connected at the same time and is managed using a hardware map that needs to be generated and maintained for every test environment. (using --generate-hardware-map option). - Reports are not long stored in the Zephyr tree and instead stored in the output directory where all build artifacts are generated. - Each tested target now has a junit report in the output directory. - Recording option for performance data and other metrics is now available. This will allow us to record the output from the console and store the data for later processing. For example benchmark data can be captured and uploaded to a tracking server. - Test configurations (or instances) are no longer being sorted, this will help with balancing the load when we run sanitycheck on multiple hosts (as we do in CI). And many other cleanups and improvements... Signed-off-by: Anas Nashif <anas.nashif@intel.com>
2019-06-22 11:04:10 -04:00
self.record = config.get('record', {})
def process_test(self, line):
runid_match = re.search(self.run_id_pattern, line)
if runid_match:
run_id = runid_match.group("run_id")
self.run_id_exists = True
if run_id == str(self.run_id):
self.matched_run_id = True
sanitycheck: Complete overhaul and job handling rework A complete overhaul of the sanitycheck script and how we build and run tests. This new version of sanitycheck uses python for job distribution and drop use of Make. In addition to the move to python threading library, the following has been changed: - All handlers now run in parallel, meaning that any simulator will run in parallel and when testing on multiple devices (using --device-testing) the tests are run in parallel. - Lexicial filtering (using the filter keyword in yaml files) is now evaluated at runtime and is no long being pre-processed. This will allow us to immediately start executing tests and skip the wait time that was needed for filtering. - Device testing now supports multiple devices connected at the same time and is managed using a hardware map that needs to be generated and maintained for every test environment. (using --generate-hardware-map option). - Reports are not long stored in the Zephyr tree and instead stored in the output directory where all build artifacts are generated. - Each tested target now has a junit report in the output directory. - Recording option for performance data and other metrics is now available. This will allow us to record the output from the console and store the data for later processing. For example benchmark data can be captured and uploaded to a tracking server. - Test configurations (or instances) are no longer being sorted, this will help with balancing the load when we run sanitycheck on multiple hosts (as we do in CI). And many other cleanups and improvements... Signed-off-by: Anas Nashif <anas.nashif@intel.com>
2019-06-22 11:04:10 -04:00
if self.RUN_PASSED in line:
if self.fault:
self.state = "failed"
else:
self.state = "passed"
if self.RUN_FAILED in line:
self.state = "failed"
if self.fail_on_fault:
if self.FAULT == line:
self.fault = True
if self.GCOV_START in line:
self.capture_coverage = True
elif self.GCOV_END in line:
self.capture_coverage = False
class Console(Harness):
def configure(self, instance):
super(Console, self).configure(instance)
if self.type == "one_line":
self.pattern = re.compile(self.regex[0])
elif self.type == "multi_line":
self.patterns = []
for r in self.regex:
self.patterns.append(re.compile(r))
def handle(self, line):
if self.type == "one_line":
if self.pattern.search(line):
self.state = "passed"
elif self.type == "multi_line" and self.ordered:
if (self.next_pattern < len(self.patterns) and
self.patterns[self.next_pattern].search(line)):
self.next_pattern += 1
if self.next_pattern >= len(self.patterns):
self.state = "passed"
elif self.type == "multi_line" and not self.ordered:
for i, pattern in enumerate(self.patterns):
r = self.regex[i]
if pattern.search(line) and not r in self.matches:
self.matches[r] = line
if len(self.matches) == len(self.regex):
self.state = "passed"
else:
logger.error("Unknown harness_config type")
if self.fail_on_fault:
if self.FAULT in line:
self.fault = True
if self.GCOV_START in line:
self.capture_coverage = True
elif self.GCOV_END in line:
self.capture_coverage = False
sanitycheck: Complete overhaul and job handling rework A complete overhaul of the sanitycheck script and how we build and run tests. This new version of sanitycheck uses python for job distribution and drop use of Make. In addition to the move to python threading library, the following has been changed: - All handlers now run in parallel, meaning that any simulator will run in parallel and when testing on multiple devices (using --device-testing) the tests are run in parallel. - Lexicial filtering (using the filter keyword in yaml files) is now evaluated at runtime and is no long being pre-processed. This will allow us to immediately start executing tests and skip the wait time that was needed for filtering. - Device testing now supports multiple devices connected at the same time and is managed using a hardware map that needs to be generated and maintained for every test environment. (using --generate-hardware-map option). - Reports are not long stored in the Zephyr tree and instead stored in the output directory where all build artifacts are generated. - Each tested target now has a junit report in the output directory. - Recording option for performance data and other metrics is now available. This will allow us to record the output from the console and store the data for later processing. For example benchmark data can be captured and uploaded to a tracking server. - Test configurations (or instances) are no longer being sorted, this will help with balancing the load when we run sanitycheck on multiple hosts (as we do in CI). And many other cleanups and improvements... Signed-off-by: Anas Nashif <anas.nashif@intel.com>
2019-06-22 11:04:10 -04:00
if self.record:
pattern = re.compile(self.record.get("regex", ""))
match = pattern.search(line)
if match:
csv = []
if not self.fieldnames:
for k,v in match.groupdict().items():
self.fieldnames.append(k)
for k,v in match.groupdict().items():
csv.append(v.strip())
self.recording.append(csv)
self.process_test(line)
tc = self.instance.get_case_or_create(self.id)
if self.state == "passed":
tc.status = "passed"
else:
tc.status = "failed"
class Pytest(Harness):
def configure(self, instance):
super(Pytest, self).configure(instance)
self.running_dir = instance.build_dir
self.source_dir = instance.testsuite.source_dir
self.pytest_root = 'pytest'
self.pytest_args = []
self.is_pytest = True
config = instance.testsuite.harness_config
if config:
self.pytest_root = config.get('pytest_root', 'pytest')
self.pytest_args = config.get('pytest_args', [])
def handle(self, line):
''' Test cases that make use of pytest more care about results given
by pytest tool which is called in pytest_run(), so works of this
handle is trying to give a PASS or FAIL to avoid timeout, nothing
is writen into handler.log
'''
self.state = "passed"
tc = self.instance.get_case_or_create(self.id)
tc.status = "passed"
def pytest_run(self, log_file):
''' To keep artifacts of pytest in self.running_dir, pass this directory
by "--cmdopt". On pytest end, add a command line option and provide
the cmdopt through a fixture function
If pytest harness report failure, twister will direct user to see
handler.log, this method writes test result in handler.log
'''
cmd = [
'pytest',
'-s',
os.path.join(self.source_dir, self.pytest_root),
'--cmdopt',
self.running_dir,
'--junit-xml',
os.path.join(self.running_dir, 'report.xml'),
'-q'
]
for arg in self.pytest_args:
cmd.append(arg)
log = open(log_file, "a")
outs = []
errs = []
logger.debug(
"Running pytest command: %s",
" ".join(shlex.quote(a) for a in cmd))
with subprocess.Popen(cmd,
stdout = subprocess.PIPE,
stderr = subprocess.PIPE) as proc:
try:
outs, errs = proc.communicate()
tree = ET.parse(os.path.join(self.running_dir, "report.xml"))
root = tree.getroot()
for child in root:
if child.tag == 'testsuite':
if child.attrib['failures'] != '0':
self.state = "failed"
elif child.attrib['skipped'] != '0':
self.state = "skipped"
elif child.attrib['errors'] != '0':
self.state = "errors"
else:
self.state = "passed"
except subprocess.TimeoutExpired:
proc.kill()
self.state = "failed"
except ET.ParseError:
self.state = "failed"
except IOError:
log.write("Can't access report.xml\n")
self.state = "failed"
tc = self.instance.get_case_or_create(self.id)
if self.state == "passed":
tc.status = "passed"
log.write("Pytest cases passed\n")
elif self.state == "skipped":
tc.status = "skipped"
log.write("Pytest cases skipped\n")
log.write("Please refer report.xml for detail")
else:
tc.status = "failed"
log.write("Pytest cases failed\n")
log.write("\nOutput from pytest:\n")
log.write(outs.decode('UTF-8'))
log.write(errs.decode('UTF-8'))
log.close()
class Gtest(Harness):
ANSI_ESCAPE = re.compile(r'\x1B(?:[@-Z\\-_]|\[[0-?]*[ -/]*[@-~])')
TEST_START_PATTERN = r"\[ RUN \] (?P<suite_name>.*)\.(?P<test_name>.*)$"
TEST_PASS_PATTERN = r"\[ OK \] (?P<suite_name>.*)\.(?P<test_name>.*)$"
TEST_FAIL_PATTERN = r"\[ FAILED \] (?P<suite_name>.*)\.(?P<test_name>.*)$"
FINISHED_PATTERN = r"\[==========\] Done running all tests\.$"
has_failures = False
tc = None
def handle(self, line):
# Strip the ANSI characters, they mess up the patterns
non_ansi_line = self.ANSI_ESCAPE.sub('', line)
# Check if we started running a new test
test_start_match = re.search(self.TEST_START_PATTERN, non_ansi_line)
if test_start_match:
# Add the suite name
suite_name = test_start_match.group("suite_name")
if suite_name not in self.detected_suite_names:
self.detected_suite_names.append(suite_name)
# Generate the internal name of the test
name = "{}.{}.{}".format(self.id, suite_name, test_start_match.group("test_name"))
# Assert that we don't already have a running test
assert (
self.tc is None
), "gTest error, {} didn't finish".format(self.tc)
# Check that the instance doesn't exist yet (prevents re-running)
tc = self.instance.get_case_by_name(name)
assert tc is None, "gTest error, {} running twice".format(tc)
# Create the test instance and set the context
tc = self.instance.get_case_or_create(name)
self.tc = tc
self.tc.status = "started"
self.testcase_output += line + "\n"
self._match = True
# Check if the test run finished
finished_match = re.search(self.FINISHED_PATTERN, non_ansi_line)
if finished_match:
tc = self.instance.get_case_or_create(self.id)
if self.has_failures or self.tc is not None:
self.state = "failed"
tc.status = "failed"
else:
self.state = "passed"
tc.status = "passed"
return
# Check if the individual test finished
state, name = self._check_result(non_ansi_line)
if state is None or name is None:
# Nothing finished, keep processing lines
return
# Get the matching test and make sure it's the same as the current context
tc = self.instance.get_case_by_name(name)
assert (
tc is not None and tc == self.tc
), "gTest error, mismatched tests. Expected {} but got {}".format(self.tc, tc)
# Test finished, clear the context
self.tc = None
# Update the status of the test
tc.status = state
if tc.status == "failed":
self.has_failures = True
tc.output = self.testcase_output
self.testcase_output = ""
self._match = False
def _check_result(self, line):
test_pass_match = re.search(self.TEST_PASS_PATTERN, line)
if test_pass_match:
return "passed", "{}.{}.{}".format(self.id, test_pass_match.group("suite_name"), test_pass_match.group("test_name"))
test_fail_match = re.search(self.TEST_FAIL_PATTERN, line)
if test_fail_match:
return "failed", "{}.{}.{}".format(self.id, test_fail_match.group("suite_name"), test_fail_match.group("test_name"))
return None, None
class Test(Harness):
RUN_PASSED = "PROJECT EXECUTION SUCCESSFUL"
RUN_FAILED = "PROJECT EXECUTION FAILED"
test_suite_start_pattern = r"Running TESTSUITE (?P<suite_name>.*)"
ZTEST_START_PATTERN = r"START - (test_)?(.*)"
def handle(self, line):
test_suite_match = re.search(self.test_suite_start_pattern, line)
if test_suite_match:
suite_name = test_suite_match.group("suite_name")
self.detected_suite_names.append(suite_name)
testcase_match = re.search(self.ZTEST_START_PATTERN, line)
if testcase_match:
name = "{}.{}".format(self.id, testcase_match.group(2))
tc = self.instance.get_case_or_create(name)
# Mark the test as started, if something happens here, it is mostly
# due to this tests, for example timeout. This should in this case
# be marked as failed and not blocked (not run).
tc.status = "started"
if testcase_match or self._match:
self.testcase_output += line + "\n"
self._match = True
result_match = result_re.match(line)
if result_match and result_match.group(2):
matched_status = result_match.group(1)
name = "{}.{}".format(self.id, result_match.group(3))
tc = self.instance.get_case_or_create(name)
tc.status = self.ztest_to_status[matched_status]
if tc.status == "skipped":
tc.reason = "ztest skip"
tc.duration = float(result_match.group(4))
if tc.status == "failed":
tc.output = self.testcase_output
self.testcase_output = ""
self._match = False
self.ztest = True
self.process_test(line)
if not self.ztest and self.state:
logger.debug(f"not a ztest and no state for {self.id}")
tc = self.instance.get_case_or_create(self.id)
if self.state == "passed":
tc.status = "passed"
else:
tc.status = "failed"
class Ztest(Test):
pass