Mainly, I use templates to implement 'properties'. the syntax I would forsee is something like this:
struct A {
int z;
int getz() const {return z;}
void setz(int k) {z = k;}
void print() {cout << z << endl;}
};
namespace Prop {
struct val {};
struct print {};
struct val2 {};
}
template <> struct Reflect<A> {
typedef Implements<
Prop::val, GetSetPair<A,int,&A::getz,&A::setz>,
Prop::print, Function<void>::MemberFun<A,&A::print>,
//another way of accessing z, without get/setters:
Prop::val2, Member<A,int,&A::z>,
//read only (even could be write-only!)
Prop::val3, GetSet<Getter<A,int,&A::getz> >
};Info;
That's the basic sytnax. This *could* be done using an automated parser, but there's certain things that the system can do that if you do by hand, you can do.
For instance, you can have a virtual get/set pair, and have the proeprty correctly dispatch to the virtual function from a base pointer. You can inherit properties (...Prop::Inherit, InheritsFrom<BaseClass>...). You can also have the type of a property be different from the get/set:
...GetSet<Getter<A,const char*,&A::GetName>,Setter<A,std::string,&A::SetName> >...
Although I'm not sure what use this could be. Similiarly, there is a lot of flexibility with an function reflection definition.
It enforces access specifier (private/public/protected) distinctions, but in a previous version I was able to access the internal private data members (again at compile time). But now I can't because I use member pointers.
There are more interesting things that it can do too, like provide an "interface" / "implementation" distinction (define a subset of the full reflection for say serialization - I know there was some discussion about that). This can be useful, for providing a kind of interface analogy from java. Functions can act on classes generically, wihtout knowing the true implementation name, type, or (get/set) vs. (public member). The code is the same as hand-written though on full optimization. If the types must be converted, an handwritten bit of codee would similiarly have to do that.
As far as I can tell, there are no 'gotcha's in the implementation, and it's fairly clean - the only problem is a lot of boost preprocessor code to avoid PTS. It should work on MSVC6 sp5 (I did not use PTS) and does on MSVC7 and gcc.
However, if there is more interest I will elaborate further, mainly because this email is getting far too long already. Any interest? I know the naming / coding style is certainly not up to boost standards, but I'm willing to make it work :)
Thanks.
Lin Xu.
_________________________________________________________________
The new MSN 8: smart spam protection and 2 months FREE* http://join.msn.com/?page=features/junkmail
_______________________________________________
Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
