[cxx-abi-dev] trivial __dynamic_cast fails?

[David Baraff]

Begin forwarded message:

> From: Daveed Vandevoorde <daveed at edg.com>
> Date: February 27, 2004 1:10:18 PM PST
> To: cxx-abi-dev at codesourcery.com
> Cc: David Baraff <deb at pixar.com>, David Abrahams 
> <dave at boost-consulting.com>
> Subject: Re: [cxx-abi-dev] trivial __dynamic_cast fails?
>
>
> On Feb 27, 2004, at 12:36 PM, Mark Mitchell wrote:
>
>> David Baraff wrote:
>>
>>> I'm playing around with the __cxxabiv1::__dynamic_cast() function.  
>>> It
>>> works as I would expect, with one exception.
>>
>> The ABI document does not require that __dynamic_cast perform a 
>> derived-to-base cast.  Those __dynamic_cast operations that can 
>> actually be performed statically by the compiler must be performed 
>> statically by the compiler -- the runtime library does not expect to 
>> be called in that situation.
>

Please note that Mark Mitchell already pointed me in the right 
direction, though it shifts a lot of implementation burden to me.  I 
can use the base_type_info to compute the address shift as I chain 
upward from derived to base, which is all I'm really after in this 
case.

So the functionality I require is in fact exposed by this very nicely 
crafted API.
But please, feel free to make __dynamic_cast do *all* the work, so I 
can write less code... :)


> Indeed.  However, it is not the first time I
> encounter the desire to be able to do this
> sort of thing.  Since the ABI specs do not
> currently say what __dynamic_cast should do
> in the cases that the compiler handles alone,
> I don't think there would be a significant
> backward compatibility issue in trying to
> support a useful design technique.  If it
> can be done within the framework of the
> existing RTTI structures, I think it would
> be worth "upgrading" __dynamic_cast to
> address this.
It obviously can -- currently it returns NULL. So instead, it can 
certainly do the computation outlined above.  (I say obviously without 
bothering to consider whether there might be tricky boundary cases --- 
I mean it seems like it obviously could, to a layman such as myself...)

>
> There has been occasional talk among WG21/
> J16 actives about adding a facility like
> what David B. is looking for, but so far
> no concrete proposals.  I'm going to CC
> David Abrahams who mentioned the desire
> for this once.  Perhaps he has some insight
> into how widespread the need for this is.

Here's a biggy.  Consider the boost::any class.  Currently, if it holds 
a Derived* and you ask it for a Base*, I think it says "no can do."  
Imagine if it could actually say "yes, I can give you back a Base*", 
because it could dynamically look up the parent chain, and do the 
address shift.

That would make boost::any a heck of a lot more useful than it is today.

>
> 	Daveed
>