Repository navigation
SIGSEGV on shutdown of tests #1830
Description
Activity
One can avoid the SIGSEGV fault by transforming the
test_python_logger.pyeither:- Transform all of the test cases in
pytest.mark.parmetrizeto not use .Net types.@pytest.mark.parametrize( "int_dotnet_level,expected_level", [ pytest.param(5, logging.CRITICAL), pytest.param(4, logging.ERROR), pytest.param(3, logging.WARNING), pytest.param(2, logging.INFO), pytest.param(1, logging.DEBUG), pytest.param(0, logging.DEBUG - 1) ] ) def test_get_python_level(int_dotnet_level, expected_level) -> None: # Setup dotnet_level = LogLevel(int_dotnet_level) # SUT result = get_python_level(dotnet_level) # Verify assert( result == expected_level)
- Do not use
pytest.mark.parametrizeand instead write multiple tests for each test case.clr.AddReference(r"Microsoft.Extensions.Logging.Abstractions") from Microsoft.Extensions.Logging import LogLevel def test_get_python_level_critical() -> None: _test_get_python_level(LogLevel.Critical, logging.CRITICAL) def test_get_python_level_error() -> None: _test_get_python_level(LogLevel.Error, logging.ERROR) # ... def _test_get_python_level(dotnet_level, expected_level) -> None: result = get_python_level(dotnet_level) # Verify assert( result == expected_level)
- Transform all of the test cases in
Rewriting
src/python_logger.pyso as to not use a chained-if, but to use a dictionary may cause the SIGSEGV fault:__dotnet_to_python = { getattr(LogLevel, "None"): logging.NOTSET, # Python has no trace level, so we return the level right below debug as trace. LogLevel.Trace: logging.DEBUG - 1, LogLevel.Debug: logging.DEBUG, LogLevel.Information: logging.INFO, LogLevel.Warning: logging.WARN, LogLevel.Error: logging.ERROR, } def get_python_level(log_level: LogLevel) -> int: global __dotnet_to_python return __dotnet_to_python.get(log_level, logging.CRITICAL)
One can work around this issue by adding an
atexitfunction to deallocate the lookup dictionary during shutdown.def __at_exit(): global __dotnet_to_python __dotnet_to_python = None atexit.register(__at_exit)
Using the second modification from the first comment to avoid the issue in the test file, and the first modification from the second comment to have the issue be caused from the code, I then added to the test file a main so that the tests could be run without pytest.
if __name__ == '__main__': print("START") test_get_python_level_critical() test_get_python_level_error() test_get_python_level_warning() test_get_python_level_information() test_get_python_level_debug() test_get_python_level_trace() print("END")
Directly running this also generated the SIGSEGV fault, thus demonstrating that pytest is not a requirements to reproduce the issue.
As soon as I finished the above write-up I thought of a simpler case:
tests/test_simple.py:'''Very simple test case.''' import clr # type: ignore from System import Object __bad_magic: Object = Object() def test_something(): assert True
3.0.0a2is rather old. We had a similar (crash on exit) issue fixed after it was released. Can you try Installing frommasterinstead?
Environment
Details
Describe what you were trying to get done.
In several new projects, unit tests running fine under Windows, but when run under Ubuntu would would fail. On examining closer found that the unit tests all ran and passed, but there would be a SIGSEGV fault during shutdown. In chasing down possibly causes of the issue, the most common trigger appeared to a having a variable at the module level of .Net Type or of a Python container type which contained a .Net type.
What commands did you run to trigger this issue? If you can provide a
Minimal, Complete, and Verifiable example
this will help us understand the issue.
Starting with one of the simpler projects, reducing it down a simple test case that reproduces the code.
src/acmedlls/netstandard2.0Microsoft.Extensions.Logging.Abstractions.dllis needed.python_logger_working.pyteststest_python_logger.pyWorkarounds and addition scenarios follow in comments.