Created attachment 552 [details] [review]
proposed patch

better yet, i'll just remove all the remaining calls to LoaderSymbol in the
drivers.  i knew it was a hack to begin with.  still a valid point though, we
ought to no-op the loader API when doing a loader-free build.

assigning to me.

I would like to check in Mattheiu's patch until you have time to finish the dll
loader work, so that work can proceed on other bugs.

Created attachment 555 [details] [review]
version 2

i'd rather do something like this.  applies against CVS HEAD without Mattheiu's
patch.

Created attachment 559 [details] [review]
supplements to version 2

Your patch works here (OpenBSD/amd64), except for 2 missing parts from my
original patch, and one typo.

Verified that with your version 2 patch and Matthieu's supplemental patch, the
build completes.  Could you go ahead and check in these patches?

checked in, closing.  thanks for the review.

I get now a warning with the patches checked in, which our (SuSE) build system 
(i386) treats as fatal ("Program is likely not 64bit clean. Please investigate 
and fix") and fails. 
 
mga_storm32.c: In function `Mga32AccelInit': 
mga_storm32.c:748: warning: implicit declaration of function 
`XAACachePlanarMonoStippleWeak' 
mga_storm32.c:748: warning: assignment makes pointer from integer without a 
cast 
 
Could you please have a look at this? The changes of the other drivers don't 
suffer from a similar problem. 

Looks like there was a typo in mga_storm.c in Matthieu's supplemental patch 
("Weak" added). 
 
-	infoPtr->CacheMonoStipple = LoaderSymbol("XAACachePlanarMonoStipple"); 
+	infoPtr->CacheMonoStipple = XAACachePlanarMonoStippleWeak(); 

Well there's a problem with declaration of pointers to functions in the patch.
I'm fixing these now.