Re: [lmi] still MSVC compilation problems

[Greg Chicares]

On 2008-06-22 01:22Z, Vadim Zeitlin wrote:
> 
>  I'm about ready to just accept the defeat in my struggle to make LMI code
> compilable with MSVC. I spent quite a lot of time on this already but
> unfortunately it seems that there is just something fundamentally wrong in
> (all versions, including 2008 one, of) the compiler which prevents it from
> compiling the code in any_member.hpp. More precisely, it's the
> any_member<ClassType>::exact_cast() function which can't be compiled
> because it insists on fully instantiating holder<ClassType,pmd_type> (even
> though normally MSVC 7+ does understand the concept of SFINAE).
> 
>  Would you have any ideas about how could this code possibly be changed to
> accommodate this compiler? I'm afraid I ran out of ideas...

At first I was going to ask whether we could use this function instead
of exact_cast(), conditionally, for msvc:

template<typename MemberType, typename ClassType>
MemberType* member_cast(any_member<ClassType>& member)
{
    MemberType* z = (member.type() == typeid(MemberType ClassType::*))
        ? member.template exact_cast<MemberType>()
        : reconstitutor<MemberType,ClassType>::reconstitute(member)
        ;
    [...throw if z is NULL]

but the first '?:' branch calls exact_cast(), and the alternative...
calls exact_cast().

I guess that if we can get this example to work, then we've solved
the problem, because other uses of member_cast() are so similar:

datum_base const* Input::DoBaseDatumPointer
    (std::string const& name
    ) const
{
#if !defined APPROPRIATE_MSVC_MACRO
    return member_cast<datum_base>(operator[](name));
#else  // defined APPROPRIATE_MSVC_MACRO
    return ?????????
#endif // defined APPROPRIATE_MSVC_MACRO
}

But IIRC that function returns a datum_base* that must point to an
object of a (deduced) derived class, so
  (datum_base*)(&something)
won't work: what we need is like this:
  (datum_base*)(deduced_class*)(&something)
But that's just what exact_cast() does, so we're back to the original
problem; let's look for another approach....

> it insists on fully instantiating holder<ClassType,pmd_type>

Is there some way to instantiate it? Is this one of the problems shown
in this
  http://lists.nongnu.org/archive/html/lmi/2008-01/msg00010.html
message? For instance, if it's this:

error C2259: 'datum_base' : cannot instantiate abstract class
1>        due to following members:
1>        'std::istream &datum_base::read(std::istream &)' : is
abstract

then could we just do the following?

class LMI_SO datum_base
{
...
#if !defined APPROPRIATE_MSVC_MACRO
    virtual std::istream& read (std::istream&)       = 0;
#else  // defined APPROPRIATE_MSVC_MACRO
    virtual std::istream& read (std::istream&);
#endif // defined APPROPRIATE_MSVC_MACRO