Re: Q. about GDR's addition of complex-accessor overloads.

[Gabriel Dos Reis <gdr at integrable-solutions dot net>]

Jan van Dijk <jan@etpmod.phys.tue.nl> writes:

| 	L.S,
| 
| With patch http://gcc.gnu.org/ml/gcc-patches/2004-01/msg00866.html Gabriel Dos 
| Reis changed the real() and imag() members of std::complex<T>. These were 
| changed to return const references, while non-const overloads were added to 
| return unqualified references.
| 
| This causes problem with the following (reduced) program:
| 
| #include <complex>
| #include <functional>
| 
| typedef std::complex<double> complex;
| 
| void foo()
| {
|         std::mem_fun_ref(&complex::real);
| }
| 
| which gives me, with today's trunk:
| 
| ./cmplx_acc_overload.cpp: In function ‘void foo()’:
| ./cmplx_acc_overload.cpp:10: error: call of overloaded 
| ‘mem_fun_ref(<unresolved overloaded function type>)’ is ambiguous
| /home/jan/local/gcc-head/lib/gcc/i686-pc-linux-gnu/4.1.0/../../../../include/c++/4.1.0/bits/stl_function.h:723: 
| note: candidates are: std::mem_fun_ref_t<_Ret, _Tp> std::mem_fun_ref(_Ret 
| (_Tp::*)()) [with _Ret = double&, _Tp = std::complex<double>]
| /home/jan/local/gcc-head/lib/gcc/i686-pc-linux-gnu/4.1.0/../../../../include/c++/4.1.0/bits/stl_function.h:728: 
| note:                 std::const_mem_fun_ref_t<_Ret, _Tp> 
| std::mem_fun_ref(_Ret (_Tp::*)()const) [with _Ret = const double&, _Tp = 
| std::complex<double>]
| 
| Translated to English this means: should I create a const_mem_fun_ref_t with 
| complex::real() const, or a a mem_fun_ref_t with the non-const real() 
| overload?
| 
| Since the standard (26.2.2, or more particular: 26.2.3) mentions only one 
| real() and imag(), it is my belief that the above test program is correct, 
| and gcc's libstdc++ is off-specs.
| 
| Any thoughts?
| 
| In case this issue has already been discussed, I apologize for not googling 
| well enough and a pointer to the previous discourse would be appreciated.

The changes were made subsequent to a DR I reported and LWG agreed upon
the suggested resolutions. 

Then a distraction happened.

The issue you're reporting is a fundamental flaw with mem_fun_ref
rather than with complex::real being overloaded on const. 

BTW, as a general rule, it is not a portable code to take the address
of a non-virtual member function of a standard class -- the compiler
may or may not accept it.

-- Gaby