Re: [Libmesh-devel] Issue with libmesh_cast_ptr

[Derek G. <fri...@gm...>]

Aha I figured I was doing it wrong... It's fine the way it is I  
think.  If we really don't have to specify that second argument then  
I'm satisfied.

Derek

Sent from my iPhone

On Feb 4, 2009, at 6:40 PM, Roy Stogner <roy...@ic...>  
wrote:

>
> On Wed, 4 Feb 2009, Derek Gaston wrote:
>
>> So I just tried to use libmesh_cast_ptr for the first time.... and  
>> I'm
>> not  really sure about the interface.  Here's what I'm doing
>>
>> PetscMatrix<T> * pmatrix = libmesh_cast_ptr<PetscMatrix<T>*,
>> SparseMatrix<T> >(this->_matrix);
>
> Try:
>
> PetscMatrix<T> *pmatrix = libmesh_cast_ptr<PetscMatrix<T>*>(this- 
> >_matrix);
>
> Told gets automatically inferred from the argument type; you shouldn't
> have to explicitly specify it at all.
>
> And this way it looks like a regular C++ something_cast<Something*>().
>
>> What I don't like is how you have to pass Tnew as a pointer type (ie
>> with a *) but Told you pass _without_ a *.  I don't really care  
>> either
>> way... but it feels like both arguments should be the same.
>
> The status of Tnew as a pointer is to make the syntax look like a C++
> cast.
>
> The status of Told as a non-pointer was originally intended to allow
> both "libmesh_cast_ptr" and "libmesh_cast_ref" to have the same name,
> "libmesh_cast", but I never quite got the overloading working
> correctly - IIRC the compiler couldn't decide that a pointer-to-Told
> shouldn't really be interpreted as a reference-to-Told (where Told is
> now a pointer-to-T) instead.
>
> Should we change Told to be a pointer too?  That will still be API
> compatible, but it might make error messages more confusing when
> someone accidentally calls libmesh_cast_ptr(some_non_pointer).
> ---
> Roy