[boost] Help with deficient compilers?

From: Daryle Walker (darylew_at_[hidden])
Date: 2008-07-26 17:27:29

I need help with fixing code in BOOST_ROOT/libs/integer/test/ 
integer_test.cpp.  It works fine on my computer, but other ones are  
giving errors.  I have some guesses on what's wrong, but I need  
suggestions on what the fixes should be.

1.  I have a custom meta-function like:

//=========================
typedef boost::mpl::vector<unsigned char, unsigned short, unsigned  
int, unsigned long long>  distinct_unsigned_types;

typedef boost::mpl::transform<
     distinct_unsigned_types,
     boost::make_signed< boost::mpl::_1 >
 >::type  distinct_signed_types;

// List the digit counts for each integral type
template < typename T >
struct digits_of
     : boost::mpl::int_< std::numeric_limits<T>::digits >
{
};

typedef boost::mpl::transform<
     distinct_unsigned_types,
     digits_of< boost::mpl::_1 >
 >::type  distinct_integral_bit_counts;
//=========================

I noticed that supplied meta-functions, like boost.make_unsigned, use  
macros in their implementations, instead of the straight  
implementation I use in digits_of.  Now MPL says that its machinery  
will make sure that straight implementations work.  However, I don't  
think it's successful on deficient compilers.  Maybe I should  
structure digits_of with all the work-around macros.  How?  (I don't  
want to just look at the headers and guess how they work.)

2.  I don't think I have SFINAE selection correct:

//=========================
#ifndef BOOST_NO_SFINAE
template < typename ValueT, template<ValueT> class Tmpl, ValueT Value >
bool
print_out_template( Tmpl<Value> const &, ValueT setting, char const
  *template_pre_name, char const *template_post_name, typename  
Tmpl<Value>::type
  *unused = 0 )
{
     // Too bad the type-id expression couldn't use the compact form  
"*unused",
     // but type-ids of dereferenced null pointers throw by order of C 
++ 2003,
     // sect. 5.2.8, para. 2 (although the result is not conceptually  
needed).
     PRIVATE_SHOW_MESSAGE( "This is " << template_pre_name << setting
      << template_post_name << " specialization, with type '" <<  
typeid(typename
      Tmpl<Value>::type).name() << "'." );
     return true;
}

template < typename ValueT, typename T >
bool
print_out_template( T const &, ValueT setting, char const  
*template_pre_name,
  char const *template_post_name )
{
     PRIVATE_SHOW_MESSAGE( "Looking for " << template_pre_name <<  
setting
      << template_post_name << " specialization?  It doesn't exist." );
     return false;
}
#else
#error "These tests cannot work without Substitution-Failure-Is-Not- 
An-Error"
#endif
//=========================

The BOOST_NO_SFINAE guards didn't help, compilers still choked on  
this code.  (I guess I'll remove those guards later.)  I think that  
any type that doesn't have an inner "type" will only choose the false- 
return case.  On my computer, types with an inner "type" will choose  
the true return case, with that case being given priority.  The  
reports are giving errors where, when given a type with an inner  
"type", the two cases are considered equal, leading to an ambiguity  
error.  How do I restructure this code?

-- 
Daryle Walker
Mac, Internet, and Video Game Junkie
darylew AT hotmail DOT com