[vtk-developers] Smart pointer declaration macro?

[David Doria]

On Wed, Jan 27, 2010 at 8:35 AM, Jeff Baumes <jeff.baumes at kitware.com> wrote:
> On Tue, Jan 26, 2010 at 12:44 PM, Marcus D. Hanwell
> <marcus.hanwell at kitware.com> wrote:
>> On Friday 08 January 2010 16:46:47 David Cole wrote:
>>> On Fri, Jan 8, 2010 at 4:08 PM, Marcus D. Hanwell <
>>>
>>> marcus.hanwell at kitware.com> wrote:
>>> > I would like to add my -1, I think needing to split lines in order to
>>> > declare
>>> > a new local variable is a little much. I came from a C++ background where
>>> > any
>>> > object could be declared on the stack though. For things like the
>>> > examples it
>>> > seems to hurt readability to me.
>>> >
>>> > Pointer:
>>> > vtkFloatArray *myTable = vtkFloatArray::New();
>>> > myTable->Delete();
>>> > myTable = NULL;
>>> >
>>> > Smart pointer:
>>> > vtkSmartPointer<vtkFloatArray> myTable =
>>> >     vtkSmartPointer<vtkFloatArray>::New();
>>> >
>>> > Smart pointer with macro:
>>> > VTK_CREATE(vtkFloatArray, myTable);
>>> >
>>> > Stack:
>>> > vtkFloatArray myTable;
>>> >
>>> > I would prefer to be able to use something like the first or the last. In
>>> > classes etc it is often a different story. It seems like there should be
>>> > some
>>> > macro or template function to generate variables with less repetition.
>>>
>>> Prefer the first and last as much as you want, but we simply can't use them
>>> in VTK. The first leads to memory leaks because people forget the Delete
>>> calls. The last cannot be done with vtkObject derived classes because of
>>>  the nature of vtkObject reference counting...
>>>
>> Wasn't suggesting either (just pointing out the shorter syntax that people
>> miss), although the first is still widely used in VTK.
>>
>>> So we have to pick one of the middle ones...
>>>
>>> It's unfortunate that we've had two +1's and two -1's... that leaves us at
>>>  0 for the moment. I guess that and the fact that it's Friday makes it
>>>  fairly easy to put off a decision until at least next week. ;-)
>>>
>>> *If* we do go with a macro-based approach, I think we can all agree there
>>> should be one centralized macro that does this and it should be used
>>> *everywhere* that vtkSmartPointer::New is presently used.
>>
>> What about a vtkLocalPointer<vtkClass> myLocal; where the default constructor
>> makes an instance on the VTK class? It would also be possible to have a
>> constructor take an argument (may be a little clunkier) such as
>> vtkSmartPointer<vtkClass> myLocal(true); if we do not want to introduce yet
>> another class.
>>
>> Alternate options a and b...
>>
>> a) vtkLocalPointer<vtkClass> myLocal;
>> b) vtkSmartPointer<vtkClass> myLocal(true);
>>
>> Would this be preferable to a macro? It seems like a better way to go to me,
>> and in terms of API and conciseness seems to satisfy our requirements better
>> than the current approach. It would still be possible to share the pointer
>> too, due to the reference counting in vtkObject derived classes.
>
> I think option b is a good option. It seems that it would be odd to
> have two different pointer classes, where the only difference is what
> they do on construction.
>
> The biggest grudge I have about the current option is that it almost
> always requires a split line if you constrain yourself to 80
> characters. If your type name is more than about 15 characters (which
> many VTK types are), you have to split your line:
>
> vtkSmartPointer<type> name = vtkSmartPointer<type>::New();
> 16 + t + 2 + n + 19 + t + 9 = 46 + 2*t + n
>
> Option b, even though not the most concise, would make it easy to stay
> within that limit, about halving the number of characters:
>
> vtkSmartPointer<type> name(true);
> 16 + t + 2 + n + 7 = 25 + t + n
>
> Jeff

This should be an extremely trivial change, right (just adding a
(bool) constructor)? So is there any downside?

Thanks,

David