Defining an own memmove() here is silly. It breaks compilation against
MacOSX 10.6 SDK where memmove is a macro. If we really wanted to avoid
the system memmove() here, surely we should then use
rtl_moveMemory(). But since when is idlcpp performance critical?