[Octave-bug-tracker] [bug #45659] version function: options and return values supported by Matlab

[Martin K]

Follow-up Comment #42, bug #45659 (project octave):

I talked to @CH today about this issue because he thinks I know something. 

The technical reason what happens, to link a application against OpenBLAS and
ATLAS, is described in
http://www2.mpi-magdeburg.mpg.de/mpcsc/koehlerm/talks/2018_FIN_Flexiblas.pdf
or http://www.netlib.org/lapack/lawnspdf/lawn284.pdf and appears in the case
of Octave because the configure script prefers to use libopenblas with not
"--with-blas" argument is given. But several depending libraries (like arpack,
suitesparse, qrupdate) used from the system are linked against libblas. If the
OS is debian/ubuntu this is set by update-alternatives to one of the BLAS
libraries installed. In the case of @mmuetzel this is ATLAS. This causes both
libraries with the same symbols to be linked against Octave and nobody can say
which symbol is called in which situation.  In order to avoid this and make
the idea of check_blas_ver.cc properly working, the configure script should
search by default for libblas.so and liblapack.so instead of libopenblas.so
and during the startup of octave one should search like the check_blas_ver.cc
idea for different characteristic symbols from the BLAS libraries and drop a
warning if octave is linked against more than one BLAS implementation. 

@mmuetzel 
Can you check against which BLAS Octave is linked using ldd `which octave-cli`
? 

    _______________________________________________________

Reply to this item at:

  <http://savannah.gnu.org/bugs/?45659>

_______________________________________________
  Message sent via/by Savannah
  http://savannah.gnu.org/