Re: Just wondering after updating from CVS

[Adam Fedor]

On Aug 9, 2005, at 11:20 AM, Nicola Pero wrote:

> > For one thing, it doesn't normally cause any problems to have lobjc 
> > linked and extra time.  On some systems, however, (Solaris, Darwin, 
> > MinGW, presumably Cygwin), the linker will complain if it is not 
> > linked (even though it's implicitly linked with gnustep-base.
>
> Maybe we could have a target.make flag for those systems ?
>
> The double-linking (in the tool and in gnustep-base) on, eg, 
> GNU/Linux, is not elegant but as you say should cause no problems ... 
> unless you have multiple libobjc installed ... if you have multiple 
> libobjcs then it might be a problem if the system gets messed up with 
> gnustep-base linked to one libobjc while tools are linked to another 
> one.  Unusual, but could happen -- the system is more robust if we 
> link only once.
>
> Probably different target.make flags for the different systems is best 
> ?
>
> ... also because ... isn't this just a specific example of a wider 
> problem?  I mean the systems that need to have libobjc linked in the 
> tool, are they also the ones that need all libraries linked into 
> bundles ?
>
> Then we could have a single target.make flag (or a couple of them) to 
> decide if that platform requires linking libobjc (and other stuff) 
> into everything or not.
>
> ... and we could use that flag when building the flags to use (maybe 
> sharing that code between the various Instance makefiles).
>
> Sounds like a good plan ?

I had originally had something like that in the patch, although much 
less general:

+# On windows, this is unfortunately required.
+ifeq ($(BUILD_DLL), yes)
+  DUP_OBJC_LIBS = $(AUXILIARY_OBJC_LIBS) $(OBJC_LIBS)
+endif
+

It would be nice to handle these other systems gracefully.