[Metakit] Test failures on Opteron Debian

[Jack Diederich]

On Fri, Nov 17, 2006 at 09:33:10PM +0100, Jean-Claude Wippler wrote:
> Jack Diederich wrote:
> 
> >I had 2.4.9.3 installed and was trying to upgrade to 2.4.9.6
> >All tests pass on 2.4.9.3
> >
> >2.4.9.6 fails on a few tests (I just commented them out to find the  
> >others)
> >b26 - Partial memo field access
> >./regress: relocation error: ./regress: undefined symbol:  
> >_ZNK11c4_BytesRef6AccessEiib
> >s37 - Change short partial fields
> >./regress: relocation error: ./regress: undefined symbol:  
> >_ZNK11c4_BytesRef6AccessEiib
> >s50 - Free space usage
> >./regress: relocation error: ./regress: undefined symbol:  
> >_ZN10c4_Storage9FreeSpaceEPi
> >
> >2.4.9.5 fails on b26 as well, I didn't try to find all failures
> >2.4.9.4 fails on s50, I didn't try to find all failures
> >./regress: relocation error: ./regress: undefined symbol:  
> >_ZN10c4_Storage9FreeSpaceEPi
> >
> >It is a dual Opteron box running Debian.
> >gcc (GCC) 3.3.5 (Debian 1:3.3.5-12)
> >Linux zster 2.6.8-11-amd64-k8-smp #1 SMP Wed May 4 19:16:40 UTC  
> >2005 x86_64 GNU/Linux
> 
> Weird.  I cannot reproduce this on a slightly different 64-bit Ubuntu  
> Linux box:
> 
> 	gcc (GCC) 4.0.3 (Ubuntu 4.0.3-1ubuntu5)
> 
> 	Linux teevie 2.6.15-27-amd64-generic #1 SMP PREEMPT \
> 		Sat Sep 16 01:50:50 UTC 2006 x86_64 GNU/Linux
> 
> I did have to run the test as follows to get it to load libmk4.so:
> 	LD_LIBRARY_PATH=. make test
> 
> Looking at what symbols are reported, I get the impression that these  
> were dead-stripped for some reason, yet still required.
> 
> Would recompiling everything with -Dq4_INLINE=0 solve it, perhaps?   
> Is there a mismatch in that setting between regress and the MK lib  
> itself?  Or maybe the test is loading the wrong shared lib?  I'm  
> running my tests with MK not installed anywhere.

Perfect, that is it exactly.  The 2.4.9.3 version passes because it
finds the existing 2.4.9.3 library.  Versions start passing less and
less because the source starts to diverge more and more from 2.4.9.3!

The LD_LIBRARY_PATH solves it when running make check.
After 'make install' all the tests pass too because the /lib version
matches.

Thanks for the quick reply and sorry for the inconvenience,

-Jack