template with member and friend function having the same name

[Vladimir Yelistratov <yelistratov at gmd dot de>]

Hi !

Is it really true that member and friend functions of a template can not
have the same name?

It is very convient to have operator* as member function and as friend
function
in template dealing with multiplication of vectors and scalars.

This topic is also mentioned in
http://gcc.gnu.org/ml/gcc-bugs/1999-09n/msg00320.html

Will be the bug for theoretically correct definition
friend A<T> ::operator- <> (const T, const A<T>&);
fixed in next releases of gcc?

Thanks,
Vladimir

**** from http://gcc.gnu.org/ml/gcc-bugs/1999-09n/msg00320.html:
> EGCS 1.1.1 compiles this, I'm pretty sure it's correct.

Yet, it is not. If you have a copy of the standard, please have a look
at the example in 14.5.3/2. It contains the case in question:

  class ... {
    template <class T> void i(T);
    friend void i<>(int); // ill­formed ± A::i
  };

> template <class T> struct A {
>   A<T> operator- (const A<T>& ) const;
>
>   friend A<T> operator- <> (const T, const A<T>&);
> };

The problem is that operator- in the friend declaration here refers to
A::operator-, instead of the global one, so your code is ill-formed.

In theory, you could solve this by writing

   friend A<T> ::operator- <> (const T, const A<T>&);

but g++ does not accept this - this is a known bug.

The work-around is to write

  template<class X>
  friend A<X> operator- (const X, const A<X>&);

instead. Yes, this does have a slightly different semantics.

Hope this helps,
Maritn