Boost's call_traits_test.cpp

[Paolo Carlini <pcarlini at unitus dot it>]

Hi all,

today I investigated in some more detail this remaining failure (the
other 3 are due to the obnoxious long long in templates problem...
Nathan Sidwell has a patch in progress!).

Basically, in lines 252-255, 4 tests are carried, which, however seems
not supported by 3.0 as was for 2.95.2 :( :

    /* */

   // cv-qualifiers applied to reference types should have no effect
   // declare these here for later use with is_reference and
remove_reference:
   typedef int& r_type;
   typedef const r_type cr_type;

    /* */

#if !(defined(__GNUC__) && (__GNUC__ < 3))
   type_test(int&, boost::call_traits<cr_type>::value_type)
   type_test(int&, boost::call_traits<cr_type>::reference)
   type_test(const int&, boost::call_traits<cr_type>::const_reference)
   type_test(int&, boost::call_traits<cr_type>::param_type)

    /* */


Those 4 tests fail.
The reason is that, apparently, Gcc3.0 (as 2.95.2) as some trouble
dealing with cv-qualified types in template instantiation. For instance,
for a call_traits<const int&>::value_type, Gcc3.0 ends up instantiating
the less specialized (the first one) of the following two templates, as
if lead in error by a not-discarded cv-qualification:

template <typename T>
struct call_traits
{
public:
   typedef T value_type;
   typedef T& reference;
};

template <typename T>
struct call_traits<T&>
{
   typedef T& value_type;
   typedef T& reference;
};


Of course, a reference to reference compilation error systematically
occurs.

Now, the problem seems to me quite clear.

Do you agree with my analysis?
In particular must in fact cv-qualifiers be discarded according to the
standard? In that case, the problem is already known and dealt with by
the Gcc developers???

Thanks,
Paolo.