On Wed, Oct 13, 2010 at 2:42 PM, Steinar Bang <s...@dod.no> wrote: >>>>>> "Hyrum K. Wright" <hyrum_wri...@mail.utexas.edu>: > >> Refcounting smartpointers are used to avoid duplication of the >> underlying C structures, but they are completely private. I've got >> serious reservations about exposing that complexity in the external >> object interfaces. Given those reservations, the API would just >> return a vanilla pointer, which leaves the caller the responsibility >> of freeing it. > > Here I'm not following you. You would hide the complexity of smart > pointers, and would return plain pointers?
I would let the callers worry about the complexity of smart pointers externally, rather than make an eternal interface decision based upon them. > Hm... the first I wouldn't worry about, and on the second I would go to > great lenghts to make it hard to get at the plain pointers. At this point, it's my turn to get confused. I'm speaking specifically of cases where we return STL classes. Brane's earlier suggestion to create an implicit bool conversion works fine for our custom classes. The problem I'm trying to solve is "how does one return the equivalent of '(const char *) NULL' in an std::string?" It's a this point that pointers start coming into play, because a NULL std::string * is quite feasible, whereas a NULL std::string isn't. These strings, and their contents are copied from the internal values used by the wrapper classes, so the actual C data would be completely obscured. This is evidenced by the current (aforementioned) design. -Hyrum
