def __init__(self, evaluator: IEvaluator, fixed_fields_front: List[str],
fixed_fields_back: List[str]):
self.fsm = FilesystemConfig()
self.summary_file = path.join(
self.fsm.get_result_dir(),
evaluator.get_eval_id() + "_" + ResultAnalyzer.RESULTS_SUMMARY)
print(self.summary_file)
self.evaluator = evaluator
# caching
self.subtasks = evaluator.get_subtask_ids_ordered()
self.interpretations = evaluator.get_subtask_interpretation()
self.required_subtasks = [
subtask for subtask in self.subtasks
if self.interpretations[subtask] == IEvaluator.REQUIRED
]
self.assumed_subtasks = [
subtask for subtask in self.subtasks
if self.interpretations[subtask] == IEvaluator.ASSUMPTION
]
self.dontcare_subtasks = [
subtask for subtask in self.subtasks
if self.interpretations[subtask] == IEvaluator.DONTCARE
]
self.fixed_fields_front = fixed_fields_front
self.fixed_fields_back = fixed_fields_back
self.ordered_fieldnames = self.fixed_fields_front + self.subtasks + self.fixed_fields_back
def __init__(self, group=None, target=None, name: str = "DeviceProcess", args=(), kwargs={}, # process args
control_channel=None, known_subtasks: List[str] = list(), # reporter specific args
analyzer: IResultAnalyzer = None, eval_name: str = '<Unknown Eval>'):
# cache
fsc = FilesystemConfig()
self.reports_dir = fsc.get_report_dir()
self.results_dir = fsc.get_result_dir()
# this worker generates its own queue
queue = Queue(ReportWriter.queue_capacity)
super(ReportWriter, self).__init__(group, target, name, args, kwargs, control_channel, queue, None,
"report_writer")
self.known_subtasks = known_subtasks
self.eval = eval_name
if not analyzer:
raise AssertionError('No analyzer provided.')
self.analyzer = analyzer
# unless this is set, the reporter keeps running even though the queue is currently empty
self.exit_after_empty_queue = False
# csv
self.csv_keys = [ReportWriter.KEY_PKG, ReportWriter.KEY_CATS] + self.known_subtasks + [ReportWriter.KEY_SUCC] \
+ [ReportWriter.KEY_WORKER] + [ReportWriter.KEY_TIMESTAMP]
with open(self.get_summary_path(), 'a+') as csv_summary:
csv_summary.write(('#' * 3) + ' ' + str(datetime.now().isoformat()) + '\n') # log date
header_writer = DictWriter(csv_summary, self.csv_keys, delimiter=';', quotechar='"')
header_writer.writeheader()
# values for statistics
tested, outs, failures, successes = self.analyzer.get_all()
self.tested = len(tested)
self.outs = len(outs)
self.fails = len(failures)
self.successes = len(successes)
def main() -> None:
parser = create_parser()
args = parser.parse_args()
evaluation = args.evaluation
task = args.task
out_overwrite = args.out_folder
print('eval: ' + evaluation)
print('task: ' + task)
print('out overwrite: ' + str(out_overwrite))
fixed_fields_front = [ReportWriter.KEY_PKG, ReportWriter.KEY_CATS]
fixed_fields_back = [
ReportWriter.KEY_SUCC, ReportWriter.KEY_WORKER,
ReportWriter.KEY_TIMESTAMP
]
if evaluation not in Evaluations.MAP.keys():
print('Evaluation not found: ' + evaluation)
exit(-1)
evaluator = Evaluations.MAP[evaluation]
# setup singleton
fsm_args = dict()
if out_overwrite is not None:
fsm_args['out'] = out_overwrite
FilesystemConfig(**fsm_args)
analyzer = evaluator.get_analyzer(fixed_fields_front=fixed_fields_front,
fixed_fields_back=fixed_fields_back)
api = analyzer.get_command_api()
for command, method in api.items():
if command == task:
method()
exit(0)
print('Command ' + task + ' not found for evaluation ' + evaluator.EVAL_ID)
def read_apps(packages_file: str) -> Tuple[Dict[str, List[str]], int]:
"""
Helper methods to read package names from list files.
:param packages_file: the name of the list file. Search path is the current list directory.
:return: a tuple of the dictionary mapping from package names to categories, and the amount of unique apps
"""
# realtive to search dir
lists_dir = FilesystemConfig().get_lists_dir()
packages_path = join(lists_dir, packages_file)
# mapping: app -> list(category)
app_dictionary = dict()
# mock category for apps without a real category
current_category = "<NO_CATEGORY>"
# number of unique apps
unique_count = 0
with open(packages_path) as file:
for line in file:
stripped = str(line).strip()
# drop lines that only contain whitespace and/or line breaks
if not stripped:
continue
if stripped.startswith('~'):
continue
if stripped.startswith('#'):
current_category = stripped.replace('#', '').strip()
else:
package_name = stripped
# app not encountered yet
if not package_name in app_dictionary.keys():
unique_count += 1
app_dictionary[package_name] = list()
app_dictionary[package_name].append(current_category)
return app_dictionary, unique_count
def main() -> None:
# parsing general arguments
parser = create_parser()
args = parser.parse_args()
evaluation_name = args.evaluation
apk = args.apk_folder
out = args.out_folder
tmp = args.tmp_folder
lists = args.list_folder
# initialize singleton
fsm_args = dict()
if apk is not None:
fsm_args['apk'] = apk
if out is not None:
fsm_args['out'] = out
if tmp is not None:
fsm_args['tmp'] = tmp
if lists is not None:
fsm_args['lists'] = lists
fsm = FilesystemConfig(**fsm_args)
# set dir values to the updated value
# apk_dir = fsm.get_apk_dir()
out_dir = fsm.get_out_dir()
# tmp_dir = fsm.get_tmp_dir()
report_dir = fsm.get_report_dir()
result_dir = fsm.get_result_dir()
# lists = fsm.get_lists_dir()
# finding the right evaluation
if len(argv) < 2:
print('Did not provide evaluation name.')
exit(-1)
# ~ strategy pattern
try:
evaluator = Evaluations.MAP[evaluation_name]
if evaluator is None:
raise KeyError
except KeyError as no_such_evaluator:
print('No such evaluator: ' + evaluation_name)
exit(-1)
return # ide workaround
# initialization (e.g. eval-specific input parsing)
evaluator.init()
# preparing the queue
analyzer = evaluator.get_analyzer(
[ReportWriter.KEY_PKG, ReportWriter.KEY_CATS], [
ReportWriter.KEY_SUCC, ReportWriter.KEY_WORKER,
ReportWriter.KEY_TIMESTAMP
])
tested = [
app_row[ReportWriter.KEY_PKG] for app_row in analyzer.get_tested()
]
# if unfinished_runs_exist(evaluation_name):
# cont = input("Evaluation was finished prematurely the last time. Do you want to proceed? (y/n)").lower()
# if cont == 'y' or cont == 'yes':
# # restore state if present (do not count from 0)
# (tested, succeeded) = read_progress(get_save_path(evaluation_name))
# else:
# if cont != 'n' and cont != 'no':
# print('No valid answer given. Treated as no.')
# print('Deleting saved progress.')
# remove(get_save_path(evaluation_name))
skip = []
if len(tested) > 0:
sure = False
while not sure:
cont = input(
"Evaluation was finished prematurely the last time. Do you want to proceed? (y/n)"
).lower()
# continue
if cont == 'y' or cont == 'yes':
print('Proceeding with evaluation.')
print('Current state:')
analyzer.api_summary()
skip = tested
sure = True
# do not continue
else:
if cont != 'n' and cont != 'no':
print('No valid answer given. Treated as no.')
print('Do not proceed with evaluation.')
delete = input(
'Are you sure you want to delete all persisted data from the last evaluation? (y/n)'
).lower()
if delete == 'y' or delete == 'yes':
# delete data from last evaluation
shutil.rmtree(out_dir)
sure = True
# ensure 'out' directories exist
makedirs(report_dir, exist_ok=True)
makedirs(result_dir, exist_ok=True)
tasks = evaluator.create_task_queue(skip)
# should be reliable since no one touched the queue yet
# task_num = tasks.qsize()
# preparing the devices
devices = shellutils.list_devices()
if devices is None:
print('No devices available.')
exit(0)
device_workers = list()
device_worker_connections = dict()
helper_workers = list()
helper_worker_connections = dict()
# preparing the reporter process
reporter_pipe_worker, reporter_pipe_main = Pipe(True)
reporter = ReportWriter(name='ReportWriter',
control_channel=reporter_pipe_worker,
known_subtasks=evaluator.get_subtask_ids_ordered(),
analyzer=analyzer,
eval_name=evaluation_name)
report_queue = reporter.get_task_queue()
helper_workers.append(reporter)
helper_worker_connections[reporter] = reporter_pipe_main
reporter.start()
# try: handle interrupts and errors
try:
# preparing and starting the device workers
# we have an early bail-out in case there are no devices
# noinspection PyTypeChecker
for device in devices:
recv, send = Pipe(False)
worker = evaluator.create_device_worker(recv, tasks, device,
report_queue)
device_workers.append(worker)
device_worker_connections[worker] = send
worker.start()
print('started ' + device)
# wait for workers to finish
waited_rounds = 0
while not tasks.empty():
waited_rounds += 1
sleep(5)
if waited_rounds % 10 == 0:
print('main process: queue not empty, sleeping')
# signalling the reporter is should stop the next time its queue gets empty
# give the workers a chance to finish the last task and send a report before telling the
# reporter to finish the report queue and exit
wait_for_workers(device_workers)
# now signal the reporter to stop after finishing the current queue
reporter_pipe_main.send(ReportWriter.msg_producers_done)
# and wait for it to finish
wait_for_workers([reporter])
print('Evaluation completed.')
except KeyboardInterrupt as abort:
print('Keyboard interrupt.')
except Exception as e:
print('Encountered exception ' + str(e))
# fallthrough to termination
# depending on whether we aborted or succeeded, device workers might be running or not,
# so we just terminate all of them
print('Terminating worker processes that are possibly still running.')
# merge worker lists and connection dicts
all_workers = device_workers + helper_workers
all_connections = device_worker_connections.copy()
all_connections.update(helper_worker_connections)
for worker in all_workers:
if worker.is_alive():
all_connections[worker].send(DeviceWorker.msg_terminate)
wait_for_workers(all_workers)
print('Evaluation finished.')
analyzer.api_summary()
# some threads (e.g. daemon threads of queue) need some time to finish before we finish the main process.
sleep(2)
def instrument(self, app: str) -> bool:
"""
Trigger on-device recompilation of an application using an installed ARTist version.
:param app: the app to be instrumented
:return: success or failure
"""
activity = self.artist_package + '/.' + self.artist_activity
self.log('Starting ARTistGUI to instrument ' + app)
(start_success, start_out) = adb_shell(
'am start -n ' + activity + ' --es INTENT_EXTRA_PACKAGE ' + app,
device=self.device_id)
self.log(start_out)
if not start_success:
self.log('Could not start ARTistGUI')
return False
# path where ARTistGUI stores the instrumentation result file to
result_path = self.instrumentation_result_path(app)
# roughly how long to wait before considering it a fail
wait_seconds = 1800 # 30 minutes
instrumentation_success = False
# busy wait
self.log('Starting busy wait: checking ' + result_path +
' until results appear or timeout occurs.')
ls = 'ls ' + result_path
for i in range(0, wait_seconds):
# check if file exists
(exists, ls_out) = adb_shell(ls, device=self.device_id)
# not yet done
if (not exists) and 'No such file or directory' in ls_out:
if i % 10 == 0:
self.log(
str(i).zfill(4) + ' - ' + app +
' - waiting for instrumentation to finish.')
sleep(1)
# unexpected error, should only occur during debugging (wrong permission etc)
elif not exists:
self.log('ls command failed. Abort.')
instrumentation_success = False # making it explicit
self.log(ls_out)
break
else:
self.log('Found instrumentation result file')
tmp_result = path.join(FilesystemConfig().get_tmp_dir(),
'ARTIST_RESULT')
# (pulled, pulled_out) = shell('adb pull ' + result_path + ' ' + tmp_result)
(pulled, pulled_out) = adb_pull(result_path,
tmp_result,
device=self.device_id)
self.log('Pulling instrumentation result file from device ' +
('succeeded' if pulled else 'failed') + ':')
self.log(pulled_out)
if not pulled:
instrumentation_success = False
break
with open(tmp_result, 'r') as result:
result_line = result.readline().strip()
self.log('Read "' + result_line + '" from result file.')
if result_line == 'true':
self.log('Result file: compilation succeeded!')
instrumentation_success = True
else:
self.log('Result file: compilation failed!')
instrumentation_success = False
# cleanup result file but do not abort if it does not work, just log
(cleanup, cleanup_out) = shell('rm ' + tmp_result)
self.log('Cleaning the result file ' +
('succeeded' if cleanup else 'failed') + '.')
self.log(cleanup_out)
break
self.log('Stopping ARTistGUI')
(stop_success,
stop_log) = adb_shell('am force-stop ' + self.artist_package,
device=self.device_id)
self.log(stop_log)
if not stop_success:
self.log('Stopping ARTistGUI failed.')
# TODO reinstall? what if reinstall fails??
return instrumentation_success
class ResultAnalyzer(IResultAnalyzer):
RESULTS_SUMMARY = 'summary.csv'
CMD_SUMMARY = 'summary'
CMD_SUCC = 'successes'
CMD_FAILS = 'fails'
CMD_OUTS = 'outs'
CMD_CHECK = 'check'
LOG_TAG = "ResultAnalyzer"
def __init__(self, evaluator: IEvaluator, fixed_fields_front: List[str],
fixed_fields_back: List[str]):
self.fsm = FilesystemConfig()
self.summary_file = path.join(
self.fsm.get_result_dir(),
evaluator.get_eval_id() + "_" + ResultAnalyzer.RESULTS_SUMMARY)
print(self.summary_file)
self.evaluator = evaluator
# caching
self.subtasks = evaluator.get_subtask_ids_ordered()
self.interpretations = evaluator.get_subtask_interpretation()
self.required_subtasks = [
subtask for subtask in self.subtasks
if self.interpretations[subtask] == IEvaluator.REQUIRED
]
self.assumed_subtasks = [
subtask for subtask in self.subtasks
if self.interpretations[subtask] == IEvaluator.ASSUMPTION
]
self.dontcare_subtasks = [
subtask for subtask in self.subtasks
if self.interpretations[subtask] == IEvaluator.DONTCARE
]
self.fixed_fields_front = fixed_fields_front
self.fixed_fields_back = fixed_fields_back
self.ordered_fieldnames = self.fixed_fields_front + self.subtasks + self.fixed_fields_back
### interface
def is_success(self, summary_row: Dict[str, str]) -> bool:
return self.interpret(summary_row) == IResultAnalyzer.SUCCESS
def is_out(self, summary_row: Dict[str, str]) -> bool:
return self.interpret(summary_row) == IResultAnalyzer.OUT
def is_failure(self, summary_row: Dict[str, str]) -> bool:
return self.interpret(summary_row) == IResultAnalyzer.FAIL
def interpret(self, summary_row: Dict[str, str]) -> Union[str, None]:
for subtask in self.subtasks:
# sanity check
if subtask not in self.interpretations.keys():
self.log('Error! No interpretation available for subtask ' +
subtask)
exit(-1)
interpretation = self.interpretations[subtask]
subtask_result = summary_row[subtask]
if not self.is_app_row(summary_row):
# invalid
return None
subtask_success = (subtask_result == 'True')
# sanity check
if not subtask_success and subtask_result != 'False':
self.log('Unexpected value for subtask ' + subtask + ': ' +
subtask_result)
self.log(str(summary_row))
exit(-1)
if interpretation == IEvaluator.DONTCARE:
continue
if not subtask_success:
if interpretation == IEvaluator.REQUIRED:
# count as fail
return IResultAnalyzer.FAIL
elif interpretation == IEvaluator.ASSUMPTION:
# remove from counting
return IResultAnalyzer.OUT
else:
self.log('Error! Unexpected interpretation for subtask ' +
subtask + ': ' + str(interpretation))
exit(-1)
# no fails (that we care about) occurred
return IResultAnalyzer.SUCCESS
#
def get_all(
self
) -> Tuple[List[Dict[str, str]], List[Dict[str, str]], List[Dict[
str, str]], List[Dict[str, str]]]:
# if this method ever gets too slow because the data is too big: implement in one interation
tested = self.get_tested()
outs = self.get_outs(csv_rows=tested)
fails = self.get_fails(csv_rows=tested)
successes = self.get_successes(csv_rows=tested)
return tested, outs, fails, successes
def get_tested(self) -> List[Dict[str, str]]:
# filter out non-data rows (comments, headlines, ...)
return self.find_matching_entries(
lambda row: self.interpret(row) is not None)
def get_outs(
self,
csv_rows: Union[List[Dict[str, str]], None] = None
) -> List[Dict[str, str]]:
if not csv_rows:
csv_rows = self.get_app_rows()
return self.find_matching_entries(lambda x: self.is_out(x),
csv_rows=csv_rows)
def get_fails(
self,
csv_rows: Union[List[Dict[str,
str]]] = None) -> List[Dict[str, str]]:
if not csv_rows:
csv_rows = self.get_app_rows()
return self.find_matching_entries(lambda x: self.is_failure(x),
csv_rows=csv_rows)
# returns summary rows from apps that succeeded
def get_successes(
self,
csv_rows: Union[List[Dict[str, str]], None] = None
) -> List[Dict[str, str]]:
if not csv_rows:
csv_rows = self.get_app_rows()
return self.find_matching_entries(lambda x: self.is_success(x),
csv_rows=csv_rows)
def get_command_api(self) -> Dict[str, Callable[[], None]]:
return {
ResultAnalyzer.CMD_SUMMARY: self.api_summary,
ResultAnalyzer.CMD_CHECK: self.api_check,
ResultAnalyzer.CMD_SUCC: self.get_successes,
ResultAnalyzer.CMD_FAILS: self.api_failures,
ResultAnalyzer.CMD_OUTS: self.get_outs
}
### API implementation ###
def api_summary(self) -> None:
"""
API method to print a summary of the current evaluation results.
"""
# category -> (tests, outs, fails, successes)
results = dict()
# use list since tuples do not support item assignment
overall = [0, 0, 0, 0]
for row in self.get_app_rows():
# app = row[ReportWriter.KEY_PKG]
categories = row[ReportWriter.KEY_CATS].strip().split(
ReportWriter.CSV_IN_CELL_SEPARATOR)
for cat in categories:
if cat not in results.keys():
# use list since tuples do not support item assignment
results[cat] = [0, 0, 0, 0]
# increase number of tests
results[cat][0] += 1
overall[0] += 1
# None not possible since we iterate over checked app rows
interpretation = self.interpret(row)
if interpretation == IResultAnalyzer.OUT:
results[cat][1] += 1
overall[1] += 1
continue
if interpretation == IResultAnalyzer.FAIL:
results[cat][2] += 1
overall[2] += 1
continue
if interpretation == IResultAnalyzer.SUCCESS:
results[cat][3] += 1
overall[3] += 1
continue
raise AssertionError('Unknown interpretation: ' +
interpretation)
for cat in sorted(results.keys()):
tests, outs, fails, successes = results[cat]
included = tests - outs
percentage = (successes / included) * 100 if included > 0 else 0
self.log('Category ' + cat + ': Tested: ' + str(tests) +
', removed: ' + str(outs) + ', success: ' +
str(successes) + '/' + str(included) + ' = ' +
str(percentage) + '%')
overall_tests = overall[0]
overall_outs = overall[1]
overall_successes = overall[3]
overall_included = overall_tests - overall_outs
overall_percentage = (overall_successes / overall_included
) * 100 if overall_included > 0 else 0
self.log('Overall: Tested: ' + str(overall_tests) + ', removed: ' +
str(overall_outs) + ', success: ' + str(overall_successes) +
'/' + str(overall_included) + ' = ' +
str(overall_percentage) + '%')
def api_successes(self, dump: bool = False) -> None:
"""
API method to display all apps that were successfully tested.
"""
successes = self.get_successes()
if dump:
for row in successes:
self.log(row[ReportWriter.KEY_PKG])
self.log('Found ' + str(len(successes)) + ' successes.')
def api_failures(self, dump: bool = False) -> None:
"""
API method to display all apps that were successfully tested.
"""
failures = self.get_fails()
if dump:
for row in failures:
self.log(row[ReportWriter.KEY_PKG])
self.log('Found ' + str(len(failures)) + ' failures.')
def api_outs(self, dump: bool = False) -> None:
"""
API method to display all apps that were removed from the evaluation because they failed for reasons beyond our
control.
"""
outs = self.get_outs()
if dump:
for row in outs:
self.log(row[ReportWriter.KEY_PKG])
self.log('Found ' + str(len(outs)) + ' outs.')
# API for command CMD_CHECK
# check results for inconsistencies
def api_check(self) -> None:
"""
API method to check the current evaluation resilts for inconsistencies, such as duplicate app packages.
"""
row_dict = self.get_app_rows()
# count occurences of package names
packages_counter = dict() # mapping: package -> occurence
for row in row_dict:
for fieldname in row.keys():
# check for unknown fieldname
if not fieldname in self.fixed_fields_front + self.subtasks + self.fixed_fields_back:
self.log('Warning: Found unknown field: ' + fieldname)
# remove header lines and comments
if row[ReportWriter.KEY_TIMESTAMP] is None or row[
ReportWriter.KEY_TIMESTAMP] == ReportWriter.KEY_TIMESTAMP:
# self.log('Skipping entry: ' + str(row))
continue
package_name = row[ReportWriter.KEY_PKG]
if package_name in packages_counter.keys():
packages_counter[package_name] += 1
else:
packages_counter[package_name] = 1
# find and print duplicates
duplicates = [(package, count)
for (package, count) in packages_counter.items()
if count > 1]
if len(duplicates) > 0:
self.log(str(len(duplicates)) + ' duplicates:')
for pkg, count in duplicates:
self.log(pkg + ': ' + str(count))
else:
self.log('No duplicates.')
### helper methods ###
def log(self, s: str) -> None:
print(ResultAnalyzer.LOG_TAG + ": " + str(s))
def read_csv_dict(self) -> List[Dict[str, str]]:
"""
Reads a csv into a list of row dictionaries. The keys are the ordered fieldnames (package, subtask, ...)saved
in the analyzer. Also contains entries such as header lines and empty lines.
:return: list of dictionaries mapping from fieldnames to values
"""
result = []
if path.isfile(self.summary_file):
with open(self.summary_file, 'r') as result_csv:
csv_reader = DictReader(result_csv,
delimiter=';',
fieldnames=self.ordered_fieldnames)
for row in csv_reader:
result.append(row)
return result
def find_matching_entries(
self,
condition: Callable[[Dict[str, str]], bool],
csv_rows: Union[List[Dict[str, str]], None] = None
) -> List[Dict[str, str]]:
"""
Scans the provided csv_rows for entries matching the provided condition.
:param condition: function mapping rows to booleans
:param csv_rows: the rows of a read csv result file. Falls back to the current results if no rows are provided.
:return: list of row dicts meeting the provided condition
"""
results = []
if csv_rows is None:
csv_rows = self.read_csv_dict()
# noinspection PyTypeChecker
for row in csv_rows:
if condition(row):
results.append(row)
return results
# get all app entries from the summary csv, omitting categories and empty lines
def get_app_rows(self) -> List[Dict[str, str]]:
"""
Reads all app rows from the summary csv, omitting categories and empty lines.
:return: list of row dicts
"""
return self.find_matching_entries(ResultAnalyzer.is_app_row)
# True if a given row is an app testing result, False otherwise (categories, header lines)
@staticmethod
def is_app_row(row: Dict[str, str]) -> bool:
"""
Checks whether a csv row is an app entry.
:param row: the csv row dict to check
:return: true if app row, false if other (e.g., category, empty)
"""
return row[ReportWriter.KEY_TIMESTAMP] is not None \
and row[ReportWriter.KEY_TIMESTAMP] != ReportWriter.KEY_TIMESTAMP
def __init__(self):
self.root = FilesystemConfig().get_apk_dir()