[Insight-developers] Integrating the New Statistics framework into ITK

Bill Lorensen bill.lorensen at gmail.com
Thu Apr 9 14:34:37 EDT 2009


My thought was that we would compile both. ITK distributed classes
would be converted over to the StatisticsV2 (as long as the API
remains the same).

If we go with Option A, I think we should keep the current name for
the old Statistics and call the new Statistics StatisticsV2 or some
thing similar. We could still use the CMAKE selection mechanism.

On Thu, Apr 9, 2009 at 2:19 PM, Karthik Krishnan
<karthik.krishnan at kitware.com> wrote:
> On Sat, Apr 4, 2009 at 3:37 PM, Bill Lorensen <bill.lorensen at gmail.com>
> wrote:
>>
>> Why not 2 directories with separate namespaces.
>> Why not leave Statistics with its current namespace and do something
>> like StatisticsV2 for the new one.
>
> A significant bit of user code written to use the statistics framework will
> stay the same. This will not be possible if we put the new classes in a new
> namespace. Besides, is there something to be gained from placing them in a
> separate namespace ? Only one of them will be compiled / installed since the
> filenames are the same.
>
> Migration to the new framework should, for a user expend the least amount of
> work.
>
> I think from developer feedback, Option A has the widest support. Is that
> the general consensus ?
> Thanks
> --
> karthik
>
>>
>>
>> On Sat, Apr 4, 2009 at 12:05 PM, Karthik Krishnan
>> <karthik.krishnan at kitware.com> wrote:
>> > I forgot to mention the reason for giving you 2 options. I've tried to
>> > weigh
>> > both the options below :
>> >
>> > Option A  vs B
>> > ---------------------
>> >
>> > 1. Code duplication:
>> >
>> >     With A, if a file remains unchanged, it has to be duplicated
>> >     (and maintained, at least as long as Bill is around) in both
>> >     directories.
>> >
>> >     With B, we have to worry only about the classes which need
>> >     refactoring. Even refactoried files, where changes are minimal,
>> >     you can even code of the form:
>> >
>> >       namespace Statistics{
>> >         class Blah {
>> >            #ifdef ITK_USE_DEPRECATED_STAT
>> >               void SetInputSample( ListSampleType * );
>> >            #else
>> >               void SetInput( ListSampleType * );
>> >            #endif
>> >          }
>> >        }
>> >
>> >
>> > 2. Code separation and readability
>> >
>> >     A separates the classes clearly in two directories. The statistics
>> >     library will be devoid of ugly ifdefs. If we decide to drop the
>> >     deprecated version at some point, we simply have to terminate
>> >     a directory.
>> >
>> >     We can do the same with Option B, in a less obvious way, by
>> >     separating the classes in two files, for instance :
>> >
>> >       File StatistiitkSample.h:
>> >          #include "StatisticsDeprecated/itkSample.h"
>> >          #include "StatisticsNew/itkSample.h"
>> >
>> >       File StatisticsDeprecated/itkSample.h
>> >          namespace StatisticsDeprecated
>> >           {
>> >           class Sample : public Object { ... }
>> >           }
>> >
>> >       File StatisticsNew/itkSample.h
>> >          namespace StatisticsNew
>> >           {
>> >           class Sample : public DataObject { ... }
>> >           }
>> >
>> >
>> > 3. Surprises
>> >
>> >   With A there are no surprises. I suspect that with B, using
>> >   anonymized namespaces etc, in several translation units,
>> >   some compiler might scream ? I tried to verify it with minimal
>> >   C++ code on gcc4.3, but I feel uneasy about it.
>> >
>> >
>> > Please let us know your thoughts.
>> > Regards
>> > --
>> > karthik
>> >
>> > On Sat, Apr 4, 2009 at 11:57 AM, Karthik Krishnan
>> > <karthik.krishnan at kitware.com> wrote:
>> >>
>> >> Hello:
>> >>
>> >> As you may know, the ITK statistics framework have been revamped. The
>> >> new
>> >> framework follows ITK pipeline mechanics, removes redundant classes and
>> >> provides a consistent API. The new classes are not backwards
>> >> compatible. For
>> >> details, please visit
>> >>
>> >>
>> >>
>> >> http://www.itk.org/Wiki/Proposals:Refactoring_Statistics_Framework_2007_Migration_Users_Guide
>> >>
>> >> We would like to integrate the new statistics framework into ITK. Given
>> >> that there are numerous API changes, to keep things backward
>> >> compatible, we
>> >> are considering taking one of the following approaches :
>> >>
>> >>
>> >> --------------
>> >> Option A
>> >>
>> >> - Rename the existing Statistics/ directory in ITK to
>> >> StatisticsDeprecated/
>> >>
>> >> - Create a parallel directory Statistics/ containing the new classes.
>> >>
>> >> - Add an ITK_USE_DEPRECATED_STATITSTICS_FRAMEWORK option at configure
>> >> time
>> >> to toggle between the usage of the deprecated and the new framework.
>> >> The
>> >> option will cause (a) the right directory to be compiled (b) headers
>> >> from
>> >> there right directory to be installed (c) the right directory to be
>> >> added to
>> >> the include dirs.
>> >>
>> >> - All existing code in the toolkit using the statistics framework will
>> >> be
>> >> ifdef'ed to work with both the new and the old framework. For instance
>> >> :
>> >>
>> >> #ifdef ITK_USE_DEPRECATED_STATISTICS_FRAMEWORK
>> >>   calculator = Statistics::CovarianceCalculator< ListSampleType
>> >> >::New();
>> >>   calculator->SetInputSample(sample);
>> >>   calculator->Update();
>> >> #else
>> >>   filter = Statistics::CovarianceFilter< ListSampleType >::New();
>> >>   filter->SetInput( sampleGenerator->GetOutput() );
>> >>   filter->Update();
>> >> #endif
>> >>
>> >>
>> >> --------------------------------------------------
>> >>
>> >> Option B
>> >>
>> >> - Keep all files in a single directory but use two different
>> >> namespaces.
>> >>
>> >> For instance the file itkSample.h would look like :
>> >>
>> >> namespace itk
>> >>   {
>> >>   namespace StatisticsDeprecated
>> >>     {
>> >>     class Sample : public Object { ... }
>> >>     }
>> >>
>> >>   namespace StatisticsNew
>> >>     {
>> >>     class Sample : public DataObject { ... }
>> >>     }
>> >>   }
>> >>
>> >> - The ITK_USE_DEPRECATED_STATITSTICS_FRAMEWORK option redefines the
>> >> unused
>> >> namespace to be an anonymous namespace. The used namespace to
>> >> "Statistics".
>> >> For instance :
>> >>
>> >>   #ifdef ITK_USE_DEPRECATED_STATITSTICS_FRAMEWORK
>> >>     #define StatisticsDeprecated Statistics
>> >>     #define StatisticsNew
>> >>   #else
>> >>     #define StatisticsNew Statistics
>> >>     #define StatisticsDeprecated
>> >>   #endif
>> >>
>> >> User code still looks the same :
>> >>
>> >> #ifdef ITK_USE_DEPRECATED_STATISTICS_FRAMEWORK
>> >>   calculator = Statistics::CovarianceCalculator< ListSampleType
>> >> >::New();
>> >>   calculator->SetInputSample(sample);
>> >>   calculator->Update();
>> >> #else
>> >>   filter = Statistics::CovarianceFilter< ListSampleType >::New();
>> >>   filter->SetInput( sampleGenerator->GetOutput() );
>> >>   filter->Update();
>> >> #endif
>> >>
>> >> -----------------------------------------
>> >>
>> >> Any thoughts the developers have on this issue.
>> >>
>> >> We have gathered a wikipage to assemble ideas:
>> >>
>> >> http://www.itk.org/Wiki/Proposals:Refactoring_Statistics_Framework_2007
>> >>
>> >> The refactored code is in the NAMIC sandbox.
>> >>
>> >>
>> >>
>> >> Thanks
>> >> Regards
>> >>
>> >> --
>> >> Karthik Krishnan
>> >> R&D Engineer,
>> >> Kitware Inc.
>> >> Ph: 518 881 4919
>> >> Fax: 518 371 4573
>> >
>> >
>> >
>> > --
>> > Karthik Krishnan
>> > R&D Engineer,
>> > Kitware Inc.
>> > Ph: 518 881 4919
>> > Fax: 518 371 4573
>> >
>> > _______________________________________________
>> > Powered by www.kitware.com
>> >
>> > Visit other Kitware open-source projects at
>> > http://www.kitware.com/opensource/opensource.html
>> >
>> > Please keep messages on-topic and check the ITK FAQ at:
>> > http://www.itk.org/Wiki/ITK_FAQ
>> >
>> > Follow this link to subscribe/unsubscribe:
>> > http://www.itk.org/mailman/listinfo/insight-developers
>> >
>> >
>
>
>
> --
> Karthik Krishnan
> R&D Engineer,
> Kitware Inc.
> Ph: 518 881 4919
> Fax: 518 371 4573
>


More information about the Insight-developers mailing list