Re: [boost] [type_index] Interest in dynamic_cast emulation?

Subject: Re: [boost] [type_index] Interest in dynamic_cast emulation?
From: Klaim - Joël Lamotte (mjklaim_at_[hidden])
Date: 2016-07-16 07:50:43

On 15 July 2016 at 23:53, Chris Glover <c.d.glover_at_[hidden]> wrote:

> Hi,
>
> I have placed an initial implementation here:
>
> https://github.com/cdglove/type_index/tree/runtime_cast
>
> Here's the commit:
>
>
> https://github.com/cdglove/type_index/commit/ee7b15a493d10bc1fad93a12ab42f2fdfd260810
>
> Usage:
>
> 1. #include <boost/type_index/runtime_cast.hpp>
> 2. Add a macro to your classes to opt in;
>
> struct base {
>     BOOST_TYPE_INDEX_REGISTER_CLASS_RTTI
> };
>
> 3. Derived classes need to list bases
>
> struct multiple_derived : base1, base2 {
>     BOOST_TYPE_INDEX_REGISTER_CLASS_RTTI_BASES(base1, base2)
> };
>
> 4. use boost::type_index::runtime_cast the same as dynamic_cast,
> (boost::type_index::runtime_pointer_cast like other boost::*pointer
> casts.).
>
> Example:
>
> using namespace boost::typeindex;
> multiple_derived d;
> base1* b1 = &d;
> multiple_derived* d2 = runtime_pointer_cast<multiple_derived>(b1);
> BOOST_TEST_NE(d2, (multiple_derived*)nullptr);
> base2* b2 = runtime_pointer_cast<base2>(b1);
> BOOST_TEST_NE(b2, (base2*)nullptr);
>
> Some notes about the implementation:
>
> 1. I am using a variadic template. This could be removed if backwards
> compatibility is desired.
> 2. The error messages could use some work.
> 3. I'm not sure if the tests are comprehensive, does anyone see anything
> I've missed? I believe I have considered all combinations of multiple and
> virtual inheritance that could mess this up.
> 4. It's possible this doesn't belong in the type_index library.
> 5. Initial performance tests put this about 5x slower than dynamic cast.
> But those tests are trivial so I need to construct something more complex
> to get an accurate measure of real world performance.
>
> Any feedback is welcome.
>

Interesting way to provide dynamic casting.
However, if there is a way for the macro to be somewhere else than inside
the classes, I would prefer that.

Joël Lamotte