[Numpy-discussion] lsame_ undefined ...

[thorsten at partita.dartmouth.edu]

On Thu, 25 Sep 2008, Charles R Harris wrote:

> On Thu, Sep 25, 2008 at 11:42 AM, <thorsten at partita.dartmouth.edu> wrote:
> 
> > On Fri, 26 Sep 2008, David Cournapeau wrote:
> >
> > > thorsten at partita.dartmouth.edu wrote:
> > > >
> > > > g77, it appears.  This is a somewhat older Red Hat Enterprise Linux
> > > > system, before they switched over to gfortran.
> > >
> > > You can't mix both compilers. You have to use either one of them to
> > > build *everything*: blas, lapack, numpy (scipy, etc...). g77 and
> > > gfortran are not ABI compatible.
> >
> > Definitely g77, then -- gfortran doesn't exist on this machine.
> >
> > > How did you build numpy (did you give any option to python setup.py build
> > ?)
> >
> > No options, it was entirely the defaults.
> >
> > > >
> > > > I used 3.1.1.
> > >
> > > Note that 3.1.1 is from February 2007, so if you literally meant two
> > > years in your former email, you could not use 3.1.1. I can't remember if
> > > lsame is one function which was missing in previous versions of lapack,
> > > though.
> >
> > The "two years" was a guess -- it must have been early 2007.
> >
> > I do see "lsame" in the lapack that's on the system.  Looking
> > inside /usr/local/lib/python2.4/site-packages/numpy/linalg/lapack_lite.so,
> > I do find the string "lsame_" in a few spots in among the binary.
> > So it looks for all the world like lsame_ should be available,
> > but for some reason it's not being found ("undefined symbol: lsame_").
> >
> 
> Is there more than one version of blas/lapack on the system?

There appears to be only one.  I had attempted an install of scipy 
at one point, which might have brought something else along, but 
running "find" from the top of the directory tree shows only a single
copy as far as I can tell.

		John