[Rd] OS X problem with DSTEIN in IGLS code

[Prof Brian Ripley]

On Sat, 3 Jul 2004, Jan de Leeuw wrote:

> 1. R-devel on OS X  aborts when the code below  is run after spitting 
> out the
> message
> 
> Loading required package: MASS
> Parameter 5 to routine DSTEIN was incorrect
> Mac OS BLAS parameter error in DSTEIN, parameter #0, (unavailable), is 0
> 
> 2. I compiled R-patched using the included BLAS and Lapack, although
> Brian says this is impossible on OS X, because they are in FORTRAN (??).

It depends on the compiler build.  The version some (most?) people have of
the gcc tools has a -lg2c that cannot be linked into a dylib.  You seem to
have one that does but no one else has reported success.  Only a week ago
someone reported being unable to build with the default configure settings
on MacOS X for exactly this reason.

> This version of R runs the code below just fine.
> 
> 3. So there seems to be a difference between the implementation of
> DSTEIN in vecLib.framework and in other BLAS versions. The error
> message comes from vecLib.

Correction: DSTEIN is part of LAPACK, not BLAS

In short, that is a MacOS error, for there is a reference implementation
of LAPACK. It is compounded by an error in the error message.  Have you
reported it to `MacOS-bugs' yet?  Please do so.

> 4. To build libRlapack.dylib on OS X add the flags "-flat_namespace
> -undefined suppress" to the link.

That's not what works for other people.  Why is is necessary for you?

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595