MSVC++ Tuple

From: William Kempf (sirwillard_at_[hidden])
Date: 2000-09-22 12:49:14

Wow!  After some very ugly hacking I actually was able to implement 
tuples using MSVC++!

I was inspired by the post by Dave Abrahams a while back about a poor 
man's partial template specialization technique.  I decided I wanted 
to try the technique out and figured tuples would be a good test of 
the technique.  So I started hacking away and came up with this first 
attempt at part of the cons template (the back bone of the tuple):

struct nil { };

template <typename T>
struct cons_select
{
  template <typename H, typename T>
  struct cons_type
  {
    typedef H head_type;
    typedef T tail_type;
    H head;
    T tail;
  };
};

template <>
struct cons_select<nil>
{
  template <typename H, typename T>
  struct cons_type
  {
    typedef H head_type;
    H head;
  };
};

template <typename H, typename T>
struct cons : cons_select<T>::cons_type<H, T>
{
};

Looked good, but didn't compile.  I received the following error with 
this code:

d:\hold\boost\tuple\tuple.h(26) : error C2516: 'cons_type<`template-
parameter257',`template-parameter258'>' : is not a legal base class
        d:\hold\boost\tuple\tuple.h(29) : see reference to class 
template instantiation 'cons<H,T>' being compiled

I couldn't see anything wrong with the code so I spent a full day 
hacking away with variations on the idea, just to discover VC++ has a 
parsing bug (go figure) that can be worked around simply by changing 
the order of template parameters for cons<> (template <typename T, 
typename H>)!  Makes no sense to me, but it works.

I've uploaded a complete tuple implementation that compiles under 
VC++ (ver 6, SP 4) to the lambda folder (in a subfolder) so you can 
take a look at what was required to do this.  The code is based off 
of the article "Tuples and multiple return values in C++", and so 
isn't completely compatible with the implementation in LL (yet), but 
it compiles cleanly and passes preliminary test code.  I'm not sure 
about the efficiency of the code produced, but at least I've got a 
working tuple!

Bill Kempf