Re: GNUstep on OpenBSD..

[Sebastian Reitenbach]

On Wednesday 15 December 2010 09:36:06 am Wolfgang Lux wrote:
> Sebastian Reitenbach wrote:
> > For me the Bundle binary is a shared object, and no a standalone
> > binary. I
> > cannot just run it. For me, a shared object usually has a .so
> > extension.
> > So I still don't know why it is not a good idea to have .so
> > extension for
> > Bundle binaries.
>
> It is not a good idea because it just adds one more platform
> dependency, making the code less maintainable. Note that not all
> systems are using the .so suffix, e.g., I'm sitting right now in front
> of an OS X machine which does not use an extension for bundle
> executables (maybe not very surprising) and which uses .dylib instead
> of .so for shared libraries. On AIX, where all objects are position
> independent by default shared libraries use the extension .a (sic!).
> Still other systems were using .shlib for shared libraries. Tracking
> these differences is just adding needless complexity given that we can
> simply load the bundle executable from a file with any extension or no
> extension at all.

Thanks for sharing your insights, and yes, good reasons to not have extensions 
on those files.

>
> > I guess there were some reasons, when it was decided how they have
> > to look
> > like, and don't want to question that, but rather understand why it
> > is the way
> > it is, since I don't see the reason, yet, and why there is an
> > exception
> > explicitly for Windows?
>
> I guess its because Windows refuses to load the shared object if it
> does not have the proper extension, but then I'm not a Windows expert.
Me neither. I'll test on OpenBSD whether it will load the file without the .so 
extension, and if it does, I'll drop the patch to gnustep-make and update the 
ports depending on it. In case its unable to load those files, I'll report 
back and propose a patch for NSBundle.m.

cheers,
Sebastian

>
> Wolfgang