diff options
-rwxr-xr-x | run_project_tests.py | 50 |
1 files changed, 24 insertions, 26 deletions
diff --git a/run_project_tests.py b/run_project_tests.py index f35a0ec..f635f0e 100755 --- a/run_project_tests.py +++ b/run_project_tests.py @@ -267,10 +267,7 @@ uninstall_commands: T.List[str] backend: 'Backend' backend_flags: T.List[str] -stop: bool = False -logfile: T.TextIO -executor: ProcessPoolExecutor -futures: T.List[T.Tuple[str, TestDef, Future[T.Optional[TestResult]]]] +stop: bool = False class StopException(Exception): def __init__(self) -> None: @@ -355,30 +352,23 @@ def validate_install(test: TestDef, installdir: Path, compiler: str, env: enviro ret_msg += f' - {p}\n' return ret_msg -def log_text_file(testdir: Path, stdo: str, stde: str) -> None: - global stop, executor, futures, logfile +def log_text_file(logfile: T.TextIO, testdir: Path, result: TestResult) -> None: logfile.write('%s\nstdout\n\n---\n' % testdir.as_posix()) - logfile.write(stdo) + logfile.write(result.stdo) logfile.write('\n\n---\n\nstderr\n\n---\n') - logfile.write(stde) + logfile.write(result.stde) logfile.write('\n\n---\n\n') if print_debug: try: - print(stdo) + print(result.stdo) except UnicodeError: - sanitized_out = stdo.encode('ascii', errors='replace').decode() + sanitized_out = result.stdo.encode('ascii', errors='replace').decode() print(sanitized_out) try: - print(stde, file=sys.stderr) + print(result.stde, file=sys.stderr) except UnicodeError: - sanitized_err = stde.encode('ascii', errors='replace').decode() + sanitized_err = result.stde.encode('ascii', errors='replace').decode() print(sanitized_err, file=sys.stderr) - if stop: - print("Aborting..") - for f in futures: - f[2].cancel() - executor.shutdown() - raise StopException() def _run_ci_include(args: T.List[str]) -> str: @@ -1081,16 +1071,17 @@ def detect_tests_to_run(only: T.Dict[str, T.List[str]], use_tmp: bool) -> T.List def run_tests(all_tests: T.List[T.Tuple[str, T.List[TestDef], bool]], log_name_base: str, failfast: bool, extra_args: T.List[str], use_tmp: bool) -> T.Tuple[int, int, int]: - global logfile txtname = log_name_base + '.txt' with open(txtname, 'w', encoding='utf-8', errors='ignore') as lf: - logfile = lf - return _run_tests(all_tests, log_name_base, failfast, extra_args, use_tmp) + return _run_tests(all_tests, log_name_base, failfast, extra_args, use_tmp, lf) def _run_tests(all_tests: T.List[T.Tuple[str, T.List[TestDef], bool]], - log_name_base: str, failfast: bool, - extra_args: T.List[str], use_tmp: bool) -> T.Tuple[int, int, int]: - global stop, executor, futures, host_c_compiler + log_name_base: str, + failfast: bool, + extra_args: T.List[str], + use_tmp: bool, + logfile: T.TextIO) -> T.Tuple[int, int, int]: + global stop, host_c_compiler xmlname = log_name_base + '.xml' junit_root = ET.Element('testsuites') conf_time: float = 0 @@ -1125,7 +1116,7 @@ def _run_tests(all_tests: T.List[T.Tuple[str, T.List[TestDef], bool]], else: print(bold('Running %s tests.' % name)) print() - futures = [] + futures: T.List[T.Tuple[str, TestDef, Future[T.Optional[TestResult]]]] = [] for t in test_cases: # Jenkins screws us over by automatically sorting test cases by name # and getting it wrong by not doing logical number sorting. @@ -1190,7 +1181,7 @@ def _run_tests(all_tests: T.List[T.Tuple[str, T.List[TestDef], bool]], build_time += result.buildtime test_time += result.testtime total_time = conf_time + build_time + test_time - log_text_file(t.path, result.stdo, result.stde) + log_text_file(logfile, t.path, result) current_test = ET.SubElement(current_suite, 'testcase', {'name': testname, 'classname': name, 'time': '%.3f' % total_time}) @@ -1201,6 +1192,13 @@ def _run_tests(all_tests: T.List[T.Tuple[str, T.List[TestDef], bool]], stdeel = ET.SubElement(current_test, 'system-err') stdeel.text = result.stde + if stop: + print("Aborting..") + for f in futures: + f[2].cancel() + executor.shutdown() + raise StopException() + if failfast and failing_tests > 0: break |