[Insight-users] Re: NumericTraits::Zero not found during WrapITK build on Windows XP

[Gaetan Lehmann]

On Thu, 16 Feb 2006 14:22:50 +0100, Zachary Pincus <zpincus at stanford.edu>  
wrote:

> I think that this is the same problem that we looked at yesterday, in a  
> different guise.
>
> Things like the numeric trait Zero are declared in the ITK header as
> ITKCommon_Export
> which is empty on non-windows systems, but
> __declspec(dllimport)
> when you are using the header on windows (it is __declspec(dllexport)  
> when itk is building the shared lib, which makes sense.)
>
> See the following for more explanation:
> http://msdn.microsoft.com/library/default.asp?url=/library/en-us/ 
> vccore/html/_core_import_into_an_application_using___declspec. 
> 28.dllimport.29.asp
>
> The above page notes that __declspec(dllimport) is not necessary when  
> importing functions from a dll, but it is needed to import other  
> symbols. Like static constants, such as numeric traits Zero and One.
>
> As we learned before, CableSwig can't deal with these sort of  
> nonstandard call attributes, and so generates code that doesn't use  
> them. So, like in the case of the MultiThreader stuff, we have three  
> options:
> (1) Remove the NumericTraits wrapping
> (2) Remove the NumericTraits wrapping on windows

I don't think we wrap NumericTraits, and I don't think variables are  
available in any class - only methods.

> (3) Provide a hacked NumericTraits header for windows that doesn't have  
> the offending static symbols, but does have the rest.
>

IMHO, this would surely break lots of code :-/

> None of this is good, and in no case will windows users ever be able to  
> access these static symbols until I (if I ever get a chance; no  
> guarantees) look into getting a replacement for cableswig going.
>
> A temporary option would be to look into the swig source files and find  
> where all of those unresolved symbols are defined, and manually adding  
> __declspec(dllimport) to them, like the proposed fix of adding __stdcall  
> to the MultiThreader code.
>
> Note: I will push a patch to Gaetan's repository soon which fixes (I  
> hope) the MultiThreader issues. These might take a little longer.
>
> Zach
>
>
>
> On Feb 16, 2006, at 6:12 AM, Gaetan Lehmann wrote:
>
>> On Thu, 16 Feb 2006 11:03:49 +0100, Charl P. Botha  
>> <cpbotha at cpbotha.net> wrote:
>>
>>> Dear Zach and Gaetan,
>>>
>>> After manually adding __stdcall to all the right places in the
>>> generated code, my WrapITK build now breaks at the _ITKCommonAPython
>>> link phase with:
>>>
>>> _ITKCommonAPython error LNK2001: unresolved external symbol "public:
>>> static class itk::CovariantVector<float,2> const
>>> itk::NumericTraits<class itk::CovariantVector<float,2> >::Zero"
>>> (?Zero@?$NumericTraits at V?$CovariantVector at M$01 at itk@@@itk@@2V? 
>>> $CovariantVector at M$01 at 2@B)
>>>
>>> and two more times, for CovariantVector<float,3> and RGBPixel<unsigned  
>>> char>.
>>>
>>> "Zero" (the unresolved symbol) is a static const variable of
>>> NumericTraits with type == template type.  Looking at the
>>> itkNumericTraits.cxx, "Zero" is initialised (outside of the class
>>> declaration or definition, as all static member variables should be)
>>> for all basic types (bool, unsigned char, float, double, etc.) but
>>> obviously not for our template types (CovariantVector and RGBPixel).
>>>
>>
>> Unless I'm completely wrong, Zero is declared in  
>> itkNumericTraitsCovariantVectorPixel.h, and is  initialized in  
>> itkNumericTraitsCovariantVectorPixel.cxx for type  
>> CovariantVector<float,2>.
>> This code should be in ITKCommon shared library.
>> Can you give the command line of the linker ?
>>
>>> Could this be part of the build problem under VS.NET 2003?  For
>>> CoviariantVector and RGBPixel, shouldn't NumericTraits::Zero be
>>> initialized somewhere in the generated CXX code?
>>>
>>> Regards,
>>> Charl
>>
>>
>>
>> --Gaëtan Lehmann
>> Biologie du Développement et de la Reproduction
>> INRA de Jouy-en-Josas (France)
>> tel: +33 1 34 65 29 66    fax: 01 34 65 29 09
>> http://voxel.jouy.inra.fr
>> _______________________________________________
>> Insight-users mailing list
>> Insight-users at itk.org
>> http://www.itk.org/mailman/listinfo/insight-users



-- 
Gaëtan Lehmann
Biologie du Développement et de la Reproduction
INRA de Jouy-en-Josas (France)
tel: +33 1 34 65 29 66    fax: 01 34 65 29 09
http://voxel.jouy.inra.fr