In every C++ program (or library, or object file), all non-static functions are represented in the binary file as symbols. These symbols are special text strings that uniquely identify a function in the program, library, or object file.

In C, the symbol name is the same as the function name: the symbol of strcpy will be strcpy, and so on. This is possible because in C no two non-static functions can have the same name.

Because C++ allows overloading (different functions with the same name but different arguments) and has many features C does not — like classes, member functions, exception specifications — it is not possible to simply use the function name as the symbol name. To solve that, C++ uses so-called name mangling, which transforms the function name and all the necessary information (like the number and size of the arguments) into some weird-looking string which only the compiler knows about. The mangled name of foo might look like foo@4%6^, for example. Or it might not even contain the word “foo”.

One of the problems with name mangling is that the C++ standard (currently [ISO14882]) does not define how names have to be mangled; thus every compiler mangles names in its own way. Some compilers even change their name mangling algorithm between different versions (notably g++ 2.x and 3.x). Even if you worked out how your particular compiler mangles names (and would thus be able to load functions via dlsym), this would most probably work with your compiler only, and might already be broken with the next version.

Another problem with the dlopen API is the fact that it only supports loading functions. But in C++ a library often exposes a class which you would like to use in your program. Obviously, to use that class you need to create an instance of it, but that cannot be easily done.