This repo contains a minimal use-case for a bug in Sulong's C++ exception handling. When we try to catch a std::runtime_error in the C++ code, we get the following abort message printed to stderr.

libc++abi.dylib: Type error in sulong_eh_canCatch(…)
Syscalls (write) not supported on this OS.
Syscalls (write) not supported on this OS.
Syscalls (write) not supported on this OS.
Syscalls (write) not supported on this OS.
Syscalls (write) not supported on this OS.
Syscalls (write) not supported on this OS.
...

The Syscalls (write) portion runs forever (I know the syscalls message is because I'm on macOS. I haven't run it on a linux vm yet.

I tracked this down to the definition of sulong_eh_canCatch

unsigned int sulong_eh_canCatch(_Unwind_Exception *unwindHeader, std::type_info *catchType) {
    __cxa_exception *ex = cxa_exception_from_exception_unwind_exception(unwindHeader);
    void *p = thrown_object_from_cxa_exception(ex);
    __shim_type_info *et = dynamic_cast<__shim_type_info*>(ex->exceptionType); 
    __shim_type_info *ct = dynamic_cast<__shim_type_info*>(catchType);
    if (et == NULL || ct == NULL) { 
        abort_message("Type error in sulong_eh_canCatch(...).\n");
    }
    if (ct->can_catch(et, p)) {
        ex->adjustedPtr = p;
        return 1;
    } else {
        return 0;
    }
}

In the case that the C++ code throws a std::runtime_error, ex->exceptionType is an St13runtime_error and et becomes null. The exception never gets marshalled over into the Java code that I’ve got, so I’m unable to catch the error. I would have expected it to get wrapped in a RuntimeException so that I could handle it.

I've checked in the bitcode file I compiled on macOS. To get a clean build, just run

mvn clean package && java -jar target/sulong-runtime-exception-1.0-SNAPSHOT.jar