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

[Kevin Kofler]

Jeff Spaleta wrote:
> scipy builds against atlas just fine, and runs on systems just
> fine...unless lapack is installed on the system. Once lapack is
> installed the link prefers the lapack package's version of the same
> library and runtime errors occur.

Weird, normally ATLAS should be preferred and it's also supposed to be
binary-compatible with LAPACK.

But I wonder if we can't just make ATLAS the default and retire the slow
reference LAPACK (and BLAS) implementation entirely, except possibly on
architectures not supported by ATLAS.

        Kevin Kofler