CLR exceptions in mixed-mode modules

[davidaue]

Description

In a mixed-mode (native c++ & c++/cli) module native code calls back to the CLR and a CLR exception is thrown. With the right combination of switches this will be caught by catch(...). In this circumstance std::current_exception returns nullptr. The translators rethrow the nullptr causing a bad_exception. My fix is shown below. I believe this applies to all Windows structured exceptions but I'm a bit rusty on the details.

 std::string ExceptionTranslatorRegistry::translateActiveException() const {
    try {
#ifdef __OBJC__
        // In Objective-C try objective-c exceptions first
        @try {
            return tryTranslators();
        }
        @catch (NSException *exception) {
            return Catch::Detail::stringify( [exception description] );
        }
#else
        // Windows SEH exceptions caught in native code will have no current_exception
        if (std::current_exception() == nullptr)
            return "Non C++ exception";

        return tryTranslators();
#endif
    }

• Catch version: 2.0.1
• Operating System: Windows 10
• Compiler+version: MSVC 15.5.2

[horenmar]

That is an interesting use-case.

Can you provide the set of switches to reproduce this? I want to see what happens inside our SE handlers first.

[davidaue]

Causing any structured exception should be exactly the same and is easier to reproduce than my specific case which involves mixed-mode. This works with the just /EHa switch which allows SEH exceptions to be caught with try/catch.

try {
    auto i = *((int*)nullptr); // cause exception
}
catch (...) {
    auto ex = std::current_exception();  // ex will be nullptr
}

Catch2 is doing the catch(...) and then uses the nullptr returned from current_exception() in the filtering code, causing a second exception (if I remember correctly).

[horenmar]

I got around experimenting with this and, at least for SEH, std::current_exception() == nullptr will evaluate to false when SEH exception was caught inside catch (...) block. Even the rethrow mostly works*, but the Console reporter goes crazy afterwards.

[davidaue]

You are right. I was fooled by how the debugger displays std::exception_ptr. It's good we found another case though! I attached an example project to show CLR exceptions caught in native code, which really do evaluate == nullptr.

CatchSEH.zip

[horenmar]

I was fooled by how the debugger displays std::exception_ptr.

Yep, spent some time there as well.