template <class T>
Clifford3<T> operator * (const Clifford3<T> &arg1,
                         const Clifford3<T> &arg2)
{
//   cout << "Entered multiply with " << arg1 << " * " << arg2 << endl;
   Clifford3<T> result(arg1.r*arg2.r   + arg1.i*arg2.i   + arg1.j*arg2.j   + arg1.k*arg2.k
                     - arg1.bi*arg2.bi - arg1.bj*arg2.bj - arg1.bk*arg2.bk - arg1.p*arg2.p,  //r (1)
                       arg1.r*arg2.i   + arg1.i*arg2.r   + arg1.bi*arg2.j  - arg1.j*arg2.bi
                     - arg1.bk*arg2.k  + arg1.k*arg2.bk  - arg1.p*arg2.bj  - arg1.bj*arg2.p,  //i (s1)
                       arg1.r*arg2.j   + arg1.j*arg2.r   - arg1.bi*arg2.i  + arg1.i*arg2.bi
                     + arg1.bj*arg2.k  - arg1.k*arg2.bj  - arg1.p*arg2.bk  - arg1.bk*arg2.p,  //j (s2)
                       arg1.r*arg2.k   + arg1.k*arg2.r   - arg1.bj*arg2.j  + arg1.j*arg2.bj
                     + arg1.bk*arg2.i  - arg1.i*arg2.bk  - arg1.p*arg2.bi  - arg1.bi*arg2.p,  //k (s3)
                       arg1.r*arg2.bi  + arg1.bi*arg2.r  - arg1.j*arg2.i   + arg1.i*arg2.j
                     + arg1.k*arg2.p   + arg1.p*arg2.k   - arg1.bj*arg2.bk + arg1.bk*arg2.bj, //bi(s12)
                       arg1.r*arg2.bj  + arg1.bj*arg2.r  - arg1.k*arg2.j   + arg1.j*arg2.k
                     + arg1.i*arg2.p   + arg1.p*arg2.i   - arg1.bk*arg2.bi + arg1.bi*arg2.bk, //bj(s23)
                       arg1.r*arg2.bk  + arg1.bk*arg2.r  - arg1.i*arg2.k   + arg1.k*arg2.i
                     + arg1.j*arg2.p   + arg1.p*arg2.j   - arg1.bi*arg2.bj + arg1.bj*arg2.bi, //bk(s31)
                       arg1.r*arg2.p   + arg1.p*arg2.r   + arg1.i*arg2.bj  + arg1.bj*arg2.i
                     + arg1.j*arg2.bk  + arg1.bk*arg2.j  + arg1.k*arg2.bi  + arg1.bi*arg2.k); //p(s123)
   cout << "Exiting multiply with " << result << endl;
   return result;
}


Section of machine code corresponding with end of the declaration and begining
to tidy up (to first call of destructor). NO cout statement. THIS HANGS.

        call    near ptr syscall Clifford3<Sum<double > >::Clifford3<Sum<double > >(Sum<double > const &,Sum<double > const &,Sum<double > const &,Sum<double > const &,Sum<double > const &,Sum<double > const &,Sum<double > const &,Sum<double > const &)
        mov EDX,8[EBP]
        mov -4[EBP],EDX
        lea ECX,-0D0h[EBP]
        call    near ptr syscall Sum<double >::~Sum<double >()

The SAME section of the code when the cout statement is compiled. THIS WORKS.

        call    near ptr syscall Clifford3<Sum<double > >::Clifford3<Sum<double > >(Sum<double > const &,Sum<double > const &,Sum<double > const &,Sum<double > const &,Sum<double > const &,Sum<double > const &,Sum<double > const &,Sum<double > const &)
        push    dword ptr 8[EBP]
        mov EDX,offset DGROUP:ostream_withassign cout
        mov -4[EBP],EDX
        mov ESI,offset DGROUP:__stack[084Fh]
        push    0
        push    ESI
        mov ECX,EDX
        call    near ptr void syscall ostream::outstr(char const *,char const *)
        mov EAX,-4[EBP]
        push    EAX
        call    near ptr ostream &cdecl <<(ostream &,Clifford3<Sum<double > > const &)
        mov EBX,offset ostream &cdecl endl(ostream &)
        push    EAX
        call    EBX
        mov EDX,8[EBP]
        mov -8[EBP],EDX
        lea ECX,-0D4h[EBP]
        call    near ptr syscall Sum<double >::~Sum<double >()

There follow lots of calls to ~Sum<double >()

Without cout:

        call    near ptr syscall Sum<double >::~Sum<double >()
        mov EAX,-4[EBP]
        pop EDI
        pop ESI
        pop EBX
        leave
        ret

With cout:

        call    near ptr syscall Sum<double >::~Sum<double >()
        mov EAX,-8[EBP]
        add ESP,0Ch
        pop EDI
        pop ESI
        pop EBX
        leave
        ret