On Jun 7, 2011, at 1:01 PM, Jeremy Huddleston wrote:

> 
> On Jun 7, 2011, at 12:01 PM, Jeremy Huddleston wrote:
> 
>> So what is the proper fix here?  How should libOSMesa be getting built?
>> 
>> Should libmesa.a be providing those stubs (rather than my change which put 
>> them in mesa/osmesa)?  Should the stubs be getting exported by libGL?  
>> Should GetHistogramEXT be exported by libGL?
>> 
>> Based on my understanding, it seems like we should bring these stubs into 
>> libmesa.a (and remove them from mesa/xlib).  Does that sound right?
> 
> Actually... I'm perplexed about another issue here... why is libOSMesa 
> linking against both libglapi and libGL?
> 
> It seems like the only things resolving into libGL are the stub symbols 
> missing in libglapi.  If we provide them in osmesa, why does libOSMesa need 
> to link against libGL at all?
> 
> Similarly, libOSMesa seems to build fine if I just don't include libglapi.a 
> and let it resolve those calls into libGL (which it can re-export to provide 
> the gl* entry points).  This of course requires libGL to be glapi-aware 
> (which glx/apple isn't yet).
> 
> As is, including libglapi.a in libOSMesa and in libGL will result in two 
> copies of libglapi existing in memory which seems like a recipe for disaster 
> since there will be two different dispatch tables, etc.  Perhaps this isn't a 
> problem on systems with flat namespaces, but on darwin, it is.

Ok, I see what is happening.  configs/darwin bitrotted and included the -lGL, 
but other configs have since removed that.  It looks like I should just remove 
-lGL from OSMESA_LIB_DEPS in addition to the other changes that I've already 
sent for osmesa.c

It looks like -lGL is still done in default, beos, and aix ... so someone might 
want to update those as well...


_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev

Reply via email to