[Insight-developers] [ITK + Python] Wrapping classes proposal

[Gaetan Lehmann]

On Thu, 16 Jun 2005 16:14:01 +0200, Benoit Regrain  
<benoit.regrain at creatis.insa-lyon.fr> wrote:

>
>
> ----- Original Message ----- From: "Gaetan Lehmann"  
> <gaetan.lehmann at jouy.inra.fr>
> To: "Benoit Regrain" <benoit.regrain at creatis.insa-lyon.fr>;  
> <insight-developers at itk.org>
> Sent: Wednesday, June 15, 2005 3:37 PM
> Subject: Re: [Insight-developers] [ITK + Python] Wrapping classes  
> proposal
>
>
>>> Some comments :
>>> I import import itkClassParser instead of import everything in it to be
>>> able to keep only interesting attributes (for user) in itk module
>>> I use True and False instead of 1 and 0
>>> I haven't imported ItkBase because the unique method it contains is  
>>> used
>>> only in one class. But it may be nice to have one common class for all  
>>> itk
>>> classes
>>> I have used ''.join([]) instead of string.join([], '') because  
>>> string.join
>>> should be deprecated in future python
>>> I remove ITK_ prefix of types (itk.ITK_US -> itk.US). itk.US seems  
>>> enough
>>> to me :-)
> It's a very good idea
>
>
>> ho, and I forgot : items() is now called keys() to be consistent with  
>> dict interface :-)
> Good. It's better to have the same use than a dict like you have made (by
> adding some others methods too).
>
>
>
>>>>> Also, I don't really understand the can_instanciate method : some
>>>>> classes can be instanciated without New method, just with () (Ex:
>>>>> itk.Index[2]())
>>>> But some classes found like ImageToImageFilter aren't instanciable
>>>> because
>>>> they haven't the ::New method (in the C++ code). The can_instanciate
>>>> method
>>>> is to differenciate super classes and instanciable classes
>>>
>>> ok
>>> To be consistent with VTK, I have made classes callable (it delegates  
>>> to
>>> New if New exists). Doing that unify classes with New method and  
>>> classes
>>> without New method.
>>> As there is no way to know in a class is really callable, I'm not sure
>>> this method is really useful, or it should return True (when New  
>>> method is
>>> avaible) or Perhaps if New is not avaible ;-)
> I think that can_instanciable isn't working in all cases. So I have  
> removed it :-(
>
>
>
>
> I have simplified our solution :
>  - Consider the itkVersion that is not a template class.
>     We were using a liaison class (ItkClassWithoutType) that
>     will offer all static methods accessible in itkVersion, like  
> GetITKVersion.
>     The wrapping creates the itkVersion_GetITKVersion method too.
>     So, I have removed this liaison class.
>     Thus, now, when I write itk.Version, I directly have the itkVersion  
> class
>  - Consider the itkImage that is a template class.
>     I'm using the same deduction to remove the ItkClassType and direcly
>     call the itk wrapped class.
>     For example : itk.Image["US2"] returns the class  
> 'itkImage_2D.itkImageUS2'
>  - I keep the first level for template classe (ItkClass).

Good !
Why haven't I thought to it before ? ;-)


>
> Removing the ItkClassType class removes :
>  - all addons that you have made (while the call of the New method).
>    But it's always possible to add them directly to the wrapped itk  
> class :-)

I can't live without it ;-) so I keep my modified New() method (see  
attached code).

The problem is that I'm not able to do the same with classes without New()  
method :
itk.BinaryBallStructuringElement.US2(Radius=3) don't work now :-(, and add  
a New method to classes without it to be able to that don't seems to be a  
nice idea :-/

Another problem : removing New method like in vtk is a good idea; we don't  
need to know in python that we call a static method (New) or the real  
constructor. Currently, I'm not able to customize __init__() method to do  
that :-(

That's sad, but because of lacks of this feature, I'm still using my  
module without improvement from yours.


>  - the __name__ and __type__ variables added.
>    But it's always possible to add or change them directly to the  
> wrapped itk class :-)
>    But I'm not sure that's a good idea to change the class name of  
> wrapped itk classes.
>    Moreover, not all itk classes will be changed (I think of itkXxxPtr  
> or itkXxx_Pointer
>    classes)
>
> But in an other hand, now, the ITK wrappers must generate the python  
> code for the
> only python class ItkClass. That will simplify the realisation of our  
> solution.
>
> Now, the last problem is always the mangled name of wrapped itk classes.  
> If we had
> some consistent rules concerning the mangled name, our solution may work  
> in all cases
> (by adding a variable on wrapped itk classes) :-)
> It will be nice that an ITK worker answers us on this point...
>

As I pointed out in a previous mail, it would be really easy to parse <C  
itk::SmartPointer<(itk::MedianImageFilter<(itk::Image<(unsigned  
short,3)>,itk::Image<(unsigned short,3)>)>)> instance at  
_a088d208_p_itk__SmartPointerTitk__MedianImageFilterTitk__ImageTunsigned_short_3_t_itk__ImageTunsigned_short_3_t_t_t>

or
<C itk::Image<(unsigned short,3)> instance at  
_4800c708_p_itk__ImageTunsigned_short_3_t>

but repr(itk.MedianImageFilter.US3US3) returns "<class  
'itkMedianImageFilter.itkMedianImageFilterUS3US3'>" and not "<class  
itk::MedianImageFilter<(itk::Image<(unsigned  
short,3)>,itk::Image<(unsigned short,3)>)>>"

I hope it souldn't be too difficult to modify the __repr__ method :-)

If It's possible, the last problem will be with default type in template.  
For example, itk.ImageFileReader[itk.Image[itk.US, 2]] is not
    itk::ImageFileReader<(itk::Image<(unsigned short,2)>)>
but
    itk::ImageFileReader<(itk::Image<(unsigned  
short,2)>,itk::DefaultConvertPixelTraits<(unsigned short)>)>


I done some small changes :

verify that the key is really a template in __getitem__ : before that,  
itk.MedianImageFilter['__dict__'] was returning the __dict__ attribute

remove imported modules to really have clean itk attributes. I suppose it  
was not working in last code I posted.

rename __manage_digit to __manage_digit__ to really mask it

add a __flatList__ method to be able to use nested types (as it was done  
in my module)



And some questions :

Do you really want to keep __call__ in ItkClass ? I was using it before  
using the dict interface, but now, I think there is no more interest to  
have it

is_template() methode do the same than has_key(). Do we really need it ?

-- 
Gaetan Lehmann <gaetan.lehmann at jouy.inra.fr>
Tel: +33 1 34 65 29 66
Biologie du Développement et de la Reproduction
INRA de Jouy-en-Josas (France)
Web: http://voxel.jouy.inra.fr
-------------- next part --------------
A non-text attachment was scrubbed...
Name: itk.py
Type: application/octet-stream
Size: 12497 bytes
Desc: not available
Url : http://www.itk.org/mailman/private/insight-developers/attachments/20050617/7cb6a9d7/itk.obj