[Insight-developers] GetInverseTransform - MSVC 6 instantiation bug
Tom Vercauteren
tom.vercauteren at m4x.org
Wed Jul 15 04:27:10 EDT 2009
Hi Luis,
I know it is a really late feedback. The thing is I just tried using
itk::Transform::GetInverseTransform() in "real life" and stumbled on a
visual c++ problem.
To recall the conversation, the initial signature
Transform< double,NOutDim,NInDim>::Pointer GetInverseTransform();
led to a compilator bug on MSVC 6. This is why we decided to have the
following signature on MSVC 6 only:
TransformBase::Pointer GetInverseTransform();
However, looking back at the patch
http://www.itk.org/cgi-bin/viewcvs.cgi/Code/Common/itkTransform.h?root=Insight&r1=1.68&r2=1.69&sortby=date
it appears that TransformBase is used for all versions of visual c++.
This is a bit frustrating since the following code is not portable anymore
InverseTransformBasePointer invtrsf = trsf->GetInverseTransform();
invtrsf->TransformPoint( pt ); // fails to compile on visual c++
Since the correct signature works just fine at least on visual studio
9, I propose to change
#if defined(_MSC_VER)
typedef TransformBase InverseTransformBaseType;
#else
typedef Transform<
TScalarType, NOutputDimensions, NInputDimensions > InverseTransformBaseType;
#endif
by
#if _MSC_VER <= 1300
// Work-around for a visual c++ 6.0 bug. For details, see
// http://www.itk.org/mailman/private/insight-developers/2009-April/012191.html
typedef TransformBase InverseTransformBaseType;
#else
typedef Transform<
TScalarType, NOutputDimensions, NInputDimensions > InverseTransformBaseType;
#endif
I tested this patch on visual 9 but not visual 7 and 8. If we still
ave problems with any of those compilers, we can of course increase
the _MSC_VER threshold.
Thoughts?
Regards,
Tom
On Fri, Apr 10, 2009 at 17:30, Luis Ibanez<luis.ibanez at kitware.com> wrote:
>
> Hi Tom,
>
> Yes, of course I tried the minimal example
> without the problematic typedef.
>
> That's indeed how I could pinpoint that
> this was the problematic typedef :-)
>
> and that's how we arrive to the option of
> replacing that typedef with a different one
> that is not carrying along the trait of the
> alternative Transform instantiation.
>
> The workaround of vnl_matrix_fixed may do the
> trick here....but at the price of ugly #ifdefs.
>
> If we are going to go that route,
> then I rather simply do:
>
>
> #if VCL_VS6
> typedef TransformBase InverseTransformBaseType;
> #else
> typedef Transform<
> TScalarType,
> NOutputDimensions,
> NInputDimensions > InverseTransformBaseType;
> #endif
>
>
> It is a lot cleaner than the VXL work around
>
>
> Regards,
>
>
> Luis
>
>
> ----------------------
> Tom Vercauteren wrote:
>>
>> Thanks Luis,
>>
>> It would be good if we could have the correct return type because as
>> you mention, once it is released, the API is there "forever". However,
>> using TransformBase instead of Transform< T, NOutputDimensions,
>> NInputDimensions > might be sufficient.
>>
>> It is really good to have the minimal example you wrote. Maybe we
>> could try posting it to some visual c++ guru forum and see if someone
>> comes up with a workaround. Maybe
>> http://social.msdn.microsoft.com/Forums/en-US/vclanguage/threads
>> or
>> comp.lang.c++
>>
>> As a wild guess, did you try the minimal example without the
>> problematic typedef?
>>
>> template <class T, unsigned int N, unsigned int M>
>> class FOO
>> {
>> public:
>> typedef T SomeTrait;
>> typedef FOO<T,M,N> BAR;
>>
>> typename BAR::SomeTrait GetInverse()
>> {
>> BAR::SomeTrait a;
>> return a;
>> }
>> };
>>
>> int main()
>> {
>> typedef FOO<float, 3, 2> FooType;
>> FooType foo;
>> FooType::BAR::SomeTrait at = foo.GetInverse();
>> return 0;
>> }
>>
>> Tom
>>
>> On Fri, Apr 10, 2009 at 00:30, Luis Ibanez <luis.ibanez at kitware.com>
>> wrote:
>>
>>> Hi Bill, Tom,
>>>
>>> I tried the workaround from vnl_matrix_fixed without success.
>>>
>>> The work around seem to be indeed addressing the same case of defining
>>> a base class of transposed NumCols and NumRows.... so it is still likely
>>> that I didn't applied the workaround properly.
>>>
>>>
>>> In the meantime, I have replaced the return inverse type:
>>>
>>> Transform< T, NOutputDimensions, NInputDimensions >
>>>
>>> with
>>>
>>> TransformBase
>>>
>>> CVS commit:
>>>
>>> http://public.kitware.com/cgi-bin/viewcvs.cgi/Code/Common/itkTransform.h?root=Insight&r1=1.66&r2=1.67&sortby=date
>>>
>>> I must admit that the original type is the correct type to use here,
>>> so it may still be worth to further investigate the VS60 work around.
>>>
>>> In the meantime, the code should work fine with the change above.
>>>
>>>
>>>
>>> BTW: Something went wrong with the style checking of the first commit,
>>> but it seems that despite the problem the file was committed.
>>>
>>> I then fixed a missing space on the comment line 117 and made a second
>>> commit:
>>>
>>> http://public.kitware.com/cgi-bin/viewcvs.cgi/Code/Common/itkTransform.h?root=Insight&r1=1.67&r2=1.68&sortby=date
>>> that curiously doesn't show any change...
>>> maybe because it was just a space ?
>>>
>>>
>>> Luis
>>>
>>>
>>> --------------------------------
>>> On Thu, Apr 9, 2009 at 11:04 AM, Bill Lorensen <bill.lorensen at gmail.com>
>>> wrote:
>>>
>>>> Luis,
>>>>
>>>> Look at vnl_matrix_fixed.h. They have some kind of a VS6 workaround.
>>>>
>>>> Bill
>>>>
>>>> On Thu, Apr 9, 2009 at 10:14 AM, Luis Ibanez <luis.ibanez at kitware.com>
>>>> wrote:
>>>>
>>>>> Hi Tom,
>>>>>
>>>>> Here is what I have found so far:
>>>>> The problem in VS6.0 can be replicated with the following minimal code:
>>>>>
>>>>>
>>>>> template <class T, unsigned int N, unsigned int M>
>>>>> class FOO
>>>>> {
>>>>> public:
>>>>> typedef T SomeTrait;
>>>>> typedef FOO<T,M,N> BAR;
>>>>> typedef typename BAR::SomeTrait BARTrait; // <<-----THIS
>>>>>
>>>>> BAR GetInverse()
>>>>> {
>>>>> BAR a;
>>>>> return a;
>>>>> }
>>>>> };
>>>>>
>>>>>
>>>>> int main()
>>>>> {
>>>>> typedef FOO<float, 3, 2> FooType;
>>>>> FooType foo;
>>>>> FooType::BAR at = foo.GetInverse();
>>>>> return 0;
>>>>> }
>>>>>
>>>>>
>>>>> This generates the now familiar-looking compilation error:
>>>>>
>>>>> G:\bin\TenplateRecurssion\main.cxx(10) : error C2027: use of undefined
>>>>> type 'FOO<float,3,2>'
>>>>> G:\bin\TenplateRecurssion\main.cxx(10) : see reference to
>>>>> class template instantiation 'FOO<float,2,3>' being compiled
>>>>> G:\bin\TenplateRecurssion\main.cxx(24) : see reference to
>>>>> class template instantiation 'FOO<float,3,2>' being compiled
>>>>> Error executing cl.exe.
>>>>>
>>>>> TenplateRecurssion.exe - 1 error(s), 0 warning(s)
>>>>>
>>>>>
>>>>> Note the transposition of 3,2 and 2,3.
>>>>>
>>>>> The line with the "THIS" label is the one that triggers the compilation
>>>>> error.
>>>>>
>>>>> As you pointed out, it seems that we are missing to use some
>>>>> mechanism for forcing VS6.0 to instantiate the InverseType,
>>>>> before we attempt to use a trait from it.
>>>>>
>>>>> ...still looking into it...
>>>>>
>>>>>
>>>>>
>>>>> Luis
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --------------------------------------------------------------------------
>>>>> On Thu, Apr 9, 2009 at 9:33 AM, Luis Ibanez <luis.ibanez at kitware.com>
>>>>> wrote:
>>>>>
>>>>>> Starting up my Visual Studio 6.0 compiler...
>>>>>>
>>>>>> It's going to be long morning...
>>>>>>
>>>>>>
>>>>>>
>>>>>> Luis
>>>>>>
>>>>>>
>>>>>> -----------------------
>>>>>> Tom Vercauteren wrote:
>>>>>>
>>>>>>> Hi all,
>>>>>>>
>>>>>>> The dashboard is giving me the following error on visual 6 compiler:
>>>>>>>
>>>>>>> c:\Dashboards\My
>>>>>>> Tests\InsightContinuous\Code\Common\itkTransform.h(120) : error
>>>>>>> C2027:
>>>>>>> use of undefined type 'Transform<double,3,2>'
>>>>>>>
>>>>>>> c:\Dashboards\My
>>>>>>> Tests\InsightContinuous\Code\Common\itkTransform.h(120) : see
>>>>>>> reference to class template instantiation
>>>>>>> 'itk::Transform<double,2,3>'
>>>>>>> being compiled
>>>>>>> c:\Dashboards\My
>>>>>>>
>>>>>>> Tests\InsightContinuous\Code\Common\itkRigid3DPerspectiveTransform.h(42)
>>>>>>> : see reference to class template instantiation
>>>>>>> 'itk::Transform<double,3,2>' being compiled
>>>>>>> c:\Dashboards\My
>>>>>>> Tests\InsightContinuous\Code\Common\itkTransformFactory.h(37) : see
>>>>>>> reference to class template instantiation
>>>>>>> 'itk::Rigid3DPerspectiveTransform<double>' being compiled
>>>>>>> c:\Dashboards\My
>>>>>>> Tests\InsightContinuous\Code\Common\itkTransformFactory.h(36) : while
>>>>>>> compiling class-template member function 'void __cdecl
>>>>>>> itk::TransformFactory<class itk::Rigid3DPerspectiveTransform<double>
>>>>>>>
>>>>>>>> ::RegisterTransform(void)'
>>>>>>>
>>>>>>>
>>>>>>> http://www.cdash.org/CDash/viewBuildError.php?buildid=309403
>>>>>>>
>>>>>>> Can anyone help me understand what this means?
>>>>>>>
>>>>>>> I initially thought that this might be related to explicit
>>>>>>> instantiation since itk::Transform<double,3,2> is not listed in:
>>>>>>>
>>>>>>>
>>>>>>> http://www.itk.org/cgi-bin/viewcvs.cgi/Wrapping/ExplicitITK/Modules/Common/wrap_itkTransform.cmake?root=Insight&sortby=date&view=markup
>>>>>>>
>>>>>>> Note that I am not even sure that wrap_itkTransform.cmake is the
>>>>>>> correct file to look at since I have never worked with explicit
>>>>>>> instantiations within ITK before and since I can't manage to compile
>>>>>>> ITK with explicit instantiations turned on on my machine.
>>>>>>>
>>>>>>> Also it seems that dash13 is not using explicit instantiations:
>>>>>>> http://www.cdash.org/CDash/testDetails.php?test=20260566&build=308626
>>>>>>>
>>>>>>> Thanks for your help.
>>>>>>>
>>>>>>> Tom
>>>>>>> _______________________________________________
>>>>>>> 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
>>>>>>>
>>>>>>
>>>>> _______________________________________________
>>>>> 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
>>>>>
>>>>
>>
>
More information about the Insight-developers
mailing list