Re: [boost] question regarding vacpp tests and mpl

From: Aleksey Gurtovoy (agurtovoy_at_[hidden])
Date: 2004-07-22 05:23:21

Robert Ramey writes:
> The serialization library is failing all tests with IBM visual age compiler.
> Errors occur within mpl.

Not quite, I am afraid. Considering that the 'apply_if' test itself
compiles without an error, this looks more like a problem with the
particular context where the facility is used:

"/home/tk/boost_regression/boost/boost/serialization/level.hpp", line
44.9: 1540-0038 (S) The name "typename mpl::apply_if<
boost::is_base_and_derived< boost::serialization::basic_traits,
boost::serialization::nvp<A[9]> >,
boost::serialization::implementation_level<
boost::serialization::nvp<A[9]> >::traits_class_level<
boost::serialization::nvp<A[9]> >, boost::mpl::int_<3> >::type" is
not a type.

Looking at the code in "level.hpp", this is one complicated typedef!

    typedef
        BOOST_DEDUCED_TYPENAME mpl::apply_if<
            is_base_and_derived<basic_traits, T>,
            traits_class_level<T>,
        //else
        BOOST_DEDUCED_TYPENAME mpl::apply_if<
            is_fundamental<T>,
            mpl::int_<primitive_type>,
        //else
        BOOST_DEDUCED_TYPENAME mpl::apply_if<
            is_class<T>,
            mpl::int_<object_class_info>,
        //else
        BOOST_DEDUCED_TYPENAME mpl::apply_if<
            is_array<T>,
            #if BOOST_WORKAROUND(__BORLANDC__, BOOST_TESTED_AT(0x564))
                mpl::int_<not_serializable>,
            #else
                mpl::int_<object_serializable>,
            #endif
        //else
        BOOST_DEDUCED_TYPENAME mpl::apply_if<
            is_enum<T>,
            #if BOOST_WORKAROUND(__BORLANDC__, BOOST_TESTED_AT(0x564))
                mpl::int_<not_serializable>,
            #else
                mpl::int_<object_serializable>,
            #endif
        //else
            mpl::int_<not_serializable>
        >::type
        >::type
        >::type
        >::type
        >::type type;

It's inefficient, too (you don't need all these '::type's except the
very last one), but in any case: I think the most practical way to
approach the error is to simplify the above to the bare minimum
(ignoring the semantics for now), and see how the compiler reacts. For
instance,

#if BOOST_WORKAROUND(__IBMCPP__, BOOST_TESTED_AT(600))
    typedef
        BOOST_DEDUCED_TYPENAME mpl::apply_if<
            is_base_and_derived<basic_traits, T>,
            traits_class_level<T>,
            mpl::int_<object_serializable>
            >::type type;
#else
    // original code ...

#endif

>
> a) Can any one give me some hints to try to address this?

I am afraid the only thing we can do here is to tweak the code in a
few different ways and see how it affects the compiler -- unless
somebody else is familiar with the issue.

> b) Is there not a new version of mpl that is going to uploaded.  

Yes...

> Maybe if I just wait things will improve? 

... and no -- I don't think that it's strictly an MPL problem, and
even if it is, the earlier we start searching for a workaround the
more chances we have to fix it before the release.

-- 
Aleksey Gurtovoy
MetaCommunications Engineering