Looking at the code, I see no reason why you cannot trap the error.

Have you tried to see what's the error code that you get back from system when you use the workaround? Is the child process terminated due to some signal, and if yes by which one? See system for figuring out how to check this.