Another fix for the Borland hash array code, and get rid of my attempt at working around Borland problems with infinity and NaN in the hash float functions. It seems impossible to do anything with them.

[SVN r28286]

include/boost/functional/hash/hash.hpp

@@ -211,8 +211,7 @@ namespace boost
 
         for(; first != last; ++first)
         {
-            T& x = *first;
-            hash_combine(seed, x);
+            hash_combine(seed, first[0]);
         }
 
         return seed;
@@ -223,8 +222,7 @@ namespace boost
     {
         for(; first != last; ++first)
         {
-            T const& x = *first;
-            hash_combine(seed, x);
+            hash_combine(seed, first[0]);
         }
     }
 #endif
@@ -259,15 +257,6 @@ namespace boost
         template <class T>
         inline std::size_t float_hash_value(T v)
         {
-#if defined(__BORLAND__)
-            if(v == std::numeric_limits<double>::quiet_NaN())
-                return 0;
-            if(v == std::numeric_limits<double>::infinity())
-                return 1;
-            if(v == -std::numeric_limits<double>::infinity())
-                return 2;
-#endif
-
             int exp;
             v = boost::hash_detail::call_frexp(v, &exp);