http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49311
--- Comment #8 from Jonathan Wakely <redi at gcc dot gnu.org> 2011-06-08 10:27:28 UTC --- (In reply to comment #7) > > Does the way I use an rvalue change it's definition? I don't understand the question, I don't know what you mean by the definition of an rvalue. > So if A and B are rvalue's does > c = x ? A : B > to mutate to lvalues? No of course not. If A and B are both lvalues of the same type then (x ? A : B) is an lvalue. Otherwise (x ? A : B) is an rvalue. This is a simplification, check the rules for yourself in the C++ standard if you want the gory details. > Looking at http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#454 > the compiler should handle static const data members like rvalues wherever > possible. Currently gcc does so nearly all the time, the few exceptions cause > really unexpected troubles The proposed resolution shown in DR 454 was not applied to the draft, it was resolved differently. G++ handles your code correctly, it is invalid C++. Consider if foo is rewritten as: unsigned foo(int direction) { const unsigned char& result = direction==0 ? A::PEN : A::NEN; return result; } This takes a reference to an lvalue, which requires it to have an address. The solution is simple, just fix your code. This is not a bug.
