[CMake] Re: [vtkusers] OSX universal binaires for VTK, cmake , ITK

[Mike Jackson]

On Jan 30, 2007, at 3:54 PM, Bill Hoffman wrote:

> Mike Jackson wrote:
>> Sean will chime in pretty quick but basically since CMake uses  
>> compilations to test things like endian then you run the risk of  
>> creating binaries that will not work. At the moment I build each  
>> type (PPC or Intel) on their respective machines then lipo them  
>> together on one of the machines. This is very labor intensive to  
>> do it this way but seems to be the only way to get the binaries to  
>> work right.
>>
>> Sean may have more..
> VTK should be changed to support the mac universal binaries.   In  
> vtkConfigure.h.in this line:
> #cmakedefine VTK_WORDS_BIGENDIAN
>
> IT should be chanced to this:
>
> #ifndef __APPLE__
> #  cmakedefine VTK_WORDS_BIGENDIAN
> #else
> #ifdef  __BIG_ENDIAN__
> #  define VTK_WORDS_BIGENDIAN
> #endif
>
> That should fix the problem, and allow for universal binaries to be  
> created.   The other issue is the SIZEOF stuff.  Same sort of thing  
> can be done with that.
>
> Basically, vtkConfigure.h.in needs to have a large ifdef __APPLE__  
> section that hard codes, or otherwise sets defines to the right  
> values based on built-in defines from the compiler and not the  
> results of try-run tests.
>
> If someone can test this and create a patch for vtkConfigure.h.in I  
> will put it in VTK.
>
> -Bill
>

So if we #ifdef special case for Apple then what about the guy on a  
linux intel box trying to cross compile for Linux PPC? Would he have  
the same problems? Isn't this becoming a cross compile issue?

And just what is #defining "__BIG_ENDIAN__"? GCC or something else?  
Can we depend on that? What about the IBM C++ compiler? Having never  
used it..

Just some random thoughts as I read this post.
-- 
Mike Jackson   Senior Research Engineer
Innovative Management & Technology Services