when including <cmath>, the standard says that in particular T sin (T) where T =
float, double or long double will go to namespace std. However, for T = double,
sin will go into :: and std::, which causes trouble for people who implement
their own sin at ::

to reproduce
------------

#include <iostream>
#include <cmath>

typedef float T;              // OK.
//typedef double T;           // PROBLEM!
//typedef long double T;      // OK.

namespace foo
{
    T sin (T x)
    {
      std::cout << "custom sin () called" << std::endl;
      return (T) 0;
    }
}

using namespace foo;

int main ()
{
    sin ((T) 0);
    return 0;
}

-------------
the root of problem i guess (from "cmath")


#include <math.h>

namespace std {
...
  using ::sin;

  inline float
  sin(float __x)
  { return __builtin_sinf(__x); }

  inline long double
  sin(long double __x)
  { return __builtin_sinl(__x); }
...
}
-------------

"using ::sin;" drags ::sin into std, but doesn't remove it from ::