[C++] My unreviewed patches

[Kriang Lerdsuwanakij]

Jason Merrill wrote:

>Ah, yes.  But then you don't handle
>
>template <class T> template <class U> template <class V> 
>  friend void A<T>::B<U>::f(V);
>
>which would have friend_depth == 3.  Why is that too many template headers?
>Similarly, why do you want to reject
>
>template <class T> template <class U>
>  friend void A<T>::B<U>::f();
>  
>
We can support those.  It's just a tweak to deal with more 
complex cases in is_friend (or a function called by it). 
But I am not sure if the standard intends to allow arbitrary
number of nested class templates like the above cases.  The 
feature is only described in two sentences and an example.  
Do you think we should handle this?

--Kriang