Vaclav Vesely wrote:

...

Fernando Cacciola wrote:

...

Vaclav Vesely wrote:

...

I would like to do somethink like this:

class Base {...} class Derived: public Base {...}

optional<Base> x = Derived(); y->call_virtual_functions()

It looks a lot like Boost.Any. Have you looked at it?

boost::any doesn't support derived classes as well. For example:

any x = Derived(); any_cast<Base>(x).virtual_function(); // throws bad_any_cast

And moreover boost::any is too general. I want to store only values

of Base

...

and its derived classes.

OK. Still, Boost.Any is a lot closer to that than Boost.Optional The fundamental issue is that Boost.Optional stores copies of the values in its own storage. That's a key feature of optional<>. In your design, as in Boost.Any's, the values are stored in a separate storage and the wrapper is merely a proxy to it. So this is more the job of a form of smart pointer (that manages external resources).

IIRC, a few years ago Peter Dimov sketched a polymorphic variant class in line with your idea (unless I got it all wrong). I can't recall where I saw it though. Peter?

In any event, I think what you need is a new beast rather an extension to optional<> because of the separate storage needed to hold the polymorphic value.

I agree that a new beast would be more appropriate. Last year I wrote a class similar to boost.variant. It took as template parameters a base class and a list of derived classes (in an MPL container) and created a variant that could hold, on the stack, any of the derived objects, but provide virtual function dispatch through the base class. By virtue of providing stack storage, it meant you could return polymorphic objects on the stack - which was the main usage pattern I was after. If anyone is interested, I may be able to dig out the work and submit it for some consideration... Dave Handley