Re: PATCH to remove signature support

[Fredrik Öhrström <d92-foh@nada.kth.se>]

On Wed, 11 Aug 1999, Mark Mitchell wrote:

> This patch removes support for `signature', a g++ extension that is
> little-used and which Jason and I agreed should go.  The reduction in
> complexity elsewhere in the front-end will be a big win.

Signatures fills a similar role as interfaces in Java, but are
more advanced. A signature is in fact compiler support for the
bridge pattern. Now, this IS useful when building distributed
objects. You can bind the signatures to either an implementation or
a proxy and with the help of signatures the inheritance hierarchy can
be different in both cases. 

I also like signatures because they streamline the design of an
interface.

Anyway, I was trying to add inheritance to signatures, but I can just
as well stop that work now... :-)

So now I have a question about efficiency:
It seems to me that virtual abstract base classes is needed if you
want to separate the implementation classes from the interfaces
without the help of signatures.

struct AlfaInterface
{
        virtual int a () = 0;
};

struct BetaInterface : public virtual AlfaInterface
{
        virtual int b () = 0;
};

struct AlfaImpl : public virtual AlfaInterface
{
        int a () { return 0; }
};

struct BetaImpl : public virtual BetaInterface,
                  public AlfaImpl
{
        int q;

        BetaImpl (int a) : q(a) {}
        int b () { return q; }
};

int main ()
{
        BetaInterface *b = new BetaImpl(2);
        return b->b();
}

What is the efficiency of the code when using empty abstract virtual
base classes? I looked at the assembler output and it seems have its
share of thunks. Is it possible to make the vtable smarter since
there is no data in the interfaces?

I can understand that not many people uses signatures since it is gcc
extension, but is there nothing in upcoming C++ standards that sort
of resembles interfaces or signatures? 

//Fredrik