Handling errors inside BOOST_PYTHON_MODULE()

[db4]

I have a module that may fail to initialize due to missing dependencies. So I do

BOOST_PYTHON_MODULE(mymodule)
{
...
    if (missing_dependency) {
        throw std::exception("Missing dependency");
    }
...
}

The exception is correctly caught and PyErr_SetString(PyExc_RuntimeError, x.what()) is called, but python reports

    import mymodule
SystemError: initialization of mymodule raised unreported exception

Why unreported exception? My platform is Windows 10/Python 3.7.4 x64/MSVC 2017 if that matters. python_boost version is 1.69.0.

[stefanseefeld]

What happens if you replace std::exception by std::runtime_error or some other more specific exception type ?

[db4]

No, switching to std::runtime_error does not help, I still have unreported exception. And looks like the exception type has nothing to do with the issue: I set a debug breakpoint in

python/src/errors.cpp

Lines 48 to 51 in ac62db1

	catch(const std::exception& x)
	{
	PyErr_SetString(PyExc_RuntimeError, x.what());
	}

and verified that this catch case was executed.

[stefanseefeld]

OK, I see. So the Python error has correctly been set by Boost.Python, but Python's own runtime then flags this as an "unreported exception". In other words, you could strip your code off all Boost.Python references, only using Python's own C API, and you'd still observe the issue.
Can you report this to the Python bug tracker ?

[db4]

Are you sure that's a Python problem? Maybe

python/src/module.cpp

Line 24 in 1727b67

handle_exception(init_function);

should be replaced with

if (handle_exception(init_function)) return NULL;

?

[stefanseefeld]

Oh, indeed, that looks sensible. Can you write a (minimal) test case to demonstrate the problem as well as that your proposed change fixes it ? That would make a great PR !

[db4]

I just tried to fix it locally this way and it solved the problem. Well, I could try to create a test case and PR, but I'm quite new to boost libs testing infrastructure...

[stefanseefeld]

OK, great. I'll see whether I can write a quick test.

[db4]

Hi @stefanseefeld
Any news on the issue? Are you still planning to create a test and finally commit the fix?

[tdelame]

Had the same issue, boost::python was hiding the real exception with raised unreported exception. I applied the patch if (handle_exception(init_function)) return NULL; mentioned by @db4 and now the real exception is thrown.

[twvd]

Also just ran into this problem.. Just bumping this issue to show it's still relevant and affects users.

[db4]

Also just ran into this problem.. Just bumping this issue to show it's still relevant and affects users.

I finally decided to stop waiting for something and create PR myself