Re: [orbitcpp-list] problem with the type_container approach...

[Andreas K. <ak...@ix...>]

Why not send the type_container to hell then and have an
_orbitcpp::stubs namespace instead? 

what do you think?

cya andy


Phil Dawes wrote:
> 
> Hi Andreas, Hi all,
> 
> Not really a problem, just an oversight which means the servant may be
> forced to link with the stubs.
> 
> With the current approach, any types declared in the idl interface:
> 
> e.g.
> 
> interface foo {
>     exception bah {
>     };
> };
> 
> go in the type_container class, i.e.:
> 
> foo-cpp-common.hh
> -----------------
> namespace _orbitcpp { namespace type_container {
> class Foo {
>     class bahException : public UserException {
>         ...
>     };
>     ...
> };
> }}
> 
> foo-cpp-stubs.hh
> ----------------
> class Foo : public CORBA::Object,
>             public _orbitcpp::type_container::Foo {
> }
> 
> This means that the poa servant cannot reference the exception using the
> fully qualified ::Foo::BahException  without linking with the stubs,
> since otherwise the compiler doesn't know that Foo inherits from
> _orbitcpp::type_container::Foo. If I understand the spec correctly, the
> servant doesn't have to inherit from the interface class, and so
> referenceing the exception using its fully-qualified name is the only
> way to portably reference it.
> 
> Instead I propose that we do the following to fix the above problem:
> 
> foo-cpp-common.hh
> -----------------
> class Foo {
>     class BahException : public UserException {
>         ...
>     };
>     ...
> };
> 
> and in foo-cpp-stubs.hh
> -----------------------
> class Foo_stub : public CORBA::Object,
>                  public Foo {
> }
> 
> i.e. the type_container class becomes the named corba interface class,
> and a seperate stub class is used for the stub.
> 
> I wrote the code to do this on the train back from London today, but
> I'll need to merge it with your checked in code (and get the inheritance
> stuff working with the new scheme). I thought I'd wait and see if you
> have any thoughts before checking it in.
> 
> Sorry if the above is a little hard to follow - I wrote this in a hurry,
> 
> Cheers,
> 
> Phil.
> 
> _______________________________________________
> orbitcpp-list mailing list
> orb...@li...
> http://lists.sourceforge.net/mailman/listinfo/orbitcpp-list