On Wed, 18 Dec 2002, Yitzhak Sapir wrote:
> I think it is possible to implement a reflection framework using the
> preprocessor library. In much the same way that one of the links
> previously mentioned
> ( http://www.garret.ru/~knizhnik/cppreflection/docs/reflect.html ), a
> description system could be made. The difference is that unlike the
> example definition of the above link:
<snip>
> We would have something more like the following:
>
> #define A_DESCRIPTION DESCRIBE_STRUCT(6, \
> (RTTI_(int, i, RTTI_FLD_PUBLIC), \
> RTTI_(char*, pc, RTTI_FLD_PUBLIC), \
> RTTI_(double, d, RTTI_FLD_PUBLIC), \
> RTTI_ARRAY(long, 10, larr, RTTI_FLD_PROTECTED), \
> RTTI_(A**, ppa, RTTI_FLD_PROTECTED), \
> RTTI_(A*, pa, RTTI_FLD_PROTECTED)));
>
> class A
> {
> IMPLEMENT_DESCRIPTION(A_DESCRIPTION);
> int x; // some member variable which is not described
> };
>
> This has the advantage that you don't to respecify everything twice, and
> yet, most of the serialization functions could be generated automatically
> from A_DESCRIPTION.
I agree. I don't like redundant data. We could provide a macro interface
as well as provide tools to automagically generate the metadata. A macro
interface could also be useful to customize the fields, as Jeremy
Maitin-Shephard mentioned (but so could source-to-source translation
tools).
However, IMHO, using the PP is much more difficult to program, debug, and
maintain. It's also less powerful than having access to the IL (eg: how
will this handle free functions and namespaces? how could it support
existing code bases?). We could keep it around as a lowest common
denominator option tho.
--craig
_______________________________________________
Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
