Andriy Tylychko writes:

Consider following example:

BOOST_MPL_ASSERT((boost::is_same< mpl::apply::type, boost::add_pointer<_1>, int>::type , int*>));

It should apply "add_pointer" to "int" and receive "int*", but it doesn't work. It returns just "int".

Consider another example:

template struct apply1:

mpl::lambda<F>::type::template apply<X> {};

BOOST_MPL_ASSERT((boost::is_same< mpl::apply/*nothing here*/, boost::add_pointermpl::_1, int>::type , int*>));

Handmade apply1 is used instead of mpl::apply. And this example works. What difference? Usage of nested "::type" of apply metafunction. As result:

BOOST_MPL_ASSERT((boost::is_same< mpl::apply::type, boost::add_pointermpl::_1, int>::type , int*>));

doesn't work. Seems like this change affects time of placeholders substitution. Can anybody explain this in detail?

So, since apply1 is just simplified version of mpl::apply, why we cannot use mpl::apply w/o "::type"? Because it cannot be compiled, at least using VC8. And seems like the main difference is that apply1 is primary template but mpl::apply is partial template specialization. Is such behavior standard? Can anybody explain the logic?

There is not much logic here: it's simply a (known) bug. Please see http://thread.gmane.org/gmane.comp.lib.boost.user/9699 for more details. -- Aleksey Gurtovoy MetaCommunications Engineering