Atlas and lapack provide the same library..compiled differently... is that a problem?

[Jeff Spaleta]

On Sun, Jan 4, 2009 at 1:53 PM, Kevin Kofler <kevin.kofler at chello.at> wrote:
> But this is in no way a complete regression test.


I'm quite frankly not as competent as I would like to be with regard
to runtime linker logic.
So the issue may not be as dire as it could be...if the affected
system in the bugreport is suffering due to local linker logic
reconfiguration. Jury is out on that.

But I do think its somewhat problematic in our packaging that lapack
and atlas expose the same provides.
F10: repoquery --whatprovides liblapack.so.3
atlas-0:3.6.0-15.fc10.i386
atlas-3dnow-0:3.6.0-15.fc10.i386
atlas-sse-0:3.6.0-15.fc10.i386
atlas-sse2-0:3.6.0-15.fc10.i386
lapack-0:3.1.1-4.fc10.i386

Does yum prefer to install atlas over lapack when looking to fill
liblapack.so.3 deps?
Is that a problem for applications which are compiled against lapack
but not atlas at build time?  This is the more general problem that
worries me, Its probably a corner case in the dependency resolution
space that you can only get to if you start from a pathologically
broken starting point..like forcibly removing lapack and atlas from
your system, then attempting to install something qhich requires that
library.

-jef