bind1st fails to compile when argument type is a constant reference

[Jonathan Wakely]

2011/11/19 Jonathan Wakely:
> 2011/11/19 Krzysztof Żelechowski:
>>
>> I did read the issue.  It is still open, the fix is a suggested fix, there
>> are arguments against the fix on reflector which I do not have access to so
>> I do not know whether the arguments are the same as the mine.  The fix is
>> required by Stroustrup's book, not by the standard.  If removing it would
>> break existing code, it is possible to guard it as an optional fragment with
>> a macro.

The post on the reflector is quoted below. It doesn't argue for
removing the overload, and it wasn't done and what we've implemented
is what was voted into the standard.

c++std-lib-9127:

Consider this code:

#include <functional>
#include <iostream>
#include <vector>

using namespace std;

struct X : binary_function<int, int, void>
{
     void operator () (int arg, int val)
     {
         cout << arg << " " << val << endl;
     }
};

int main()
{
     vector<int> v(5, 2);
     for_each(v.begin(), v.end(), bind2nd(X(), 3));
     return 0;
}

This fails to compile under the current resolution of 109.  The compiler
attempts to call the new binder2nd::operator() that is overloaded on
int& (as opposed to const int&), but since this method is const, it
fails to call X::operator() because /that/ method is non-const.  I
believe that the intent of issue 109 is for code like this to work.  I
propose that the const be removed from the new operator() overloads in
issue 109.