From: Peter Dimov (pdimov_at_[hidden])
Date: 2002-09-11 08:43:04
From: "Vladimir Prus" <ghost_at_[hidden]>
> Peter Dimov wrote:
> > From: "Vladimir Prus" <ghost_at_[hidden]>
> >
> >>I have tried hard to explain that while typeid would make archives
> >>unportable between different machines (not really an issue if there's
> >>denamgler for type_info::name() and I think work is underway on it),
> >
> >
> > An issue. You never know whether, or how, type_info::name() is mangled,
or
> > whether there is type_info::name() at all (speaking of portable code.)
>
> You still know that for any given platform.
If you are willing to confine yourself to a given platform/compiler/version,
which I am not, you don't need to demangle type_info::name.
> > The only portable C++-only solution (AFAIK) is to require registration
of
> > all polymorphic types, including an identifier that uniquely describes
the
> > type.
>
> You see, proposed serialization library requires much more than that. It
> requires the you
> 1. Register all polymorphic classes with any instance of
> basic_{i,o}archive that you create
My own serialization framework has the same requirement, unfortunately. I
haven't been able to avoid it.
> 2. Register them in the same order everywhere.
Yes, this may be considered inconvenient.
> > Not only that, but you also need to register derived to base
conversions,
> > too, as I recently realized, since Derived * -> void * -> Base * is not
> > required to work.
>
> Why do you need such a conversion?
Because when you are reading a polymorphic object via a pointer to Base, you
don't have Derived available at compile time; you only have a string that
identifies that Derived. You can dispatch on that string and get back a
void* to the object; now you must convert that void* to a Base*.
> >>Except
> >>1. It's much more terse
> >
> >
> > And much less const-correct.
>
> You mean that if 'describe' is not declared const, you can't serialize
> const object without const_cast somewhere? Seems so. Did you say this
> before?
Yes, I did. :-)
Having to write essentially the same code twice seems inconvenient, but I am
willing to pay the price. In fact currently I'm writing the same code three
times (the third variation includes variable names).