[SciPy-user] Install issues with lapack

[Pearu Peterson]

On Sat, 19 Oct 2002, Stephen Boulet wrote:

> On Saturday 19 October 2002 04:29 pm, Pearu Peterson wrote:
> 
> > Second, the answer to your question can be probably found in
> >   http://www.netlib.org/lapack/lawn41/index.html
> 
> Yes. The instructions are incorrect. They say "Next, if you will be using a 
> locally available BLAS library, you will need to remove blaslib from the lib 
> definition", when actually blaslib is commented out by default, causing an 
> error during the make process. It shouldn't matter, because by that time 
> lapack_LINUX.a has already been built, but I didn't know that, and seeing it 
> bork with an error was confusing.
> 
> Your Install.txt, by specifying "make lapacklib", should avoid the issue 
> though. Thanks. Are you using a pentium 2?  ;)

Yeah, it's great for producing efficient code but poor for running it ;)

> There's another problem, though. Running "python setup.py install", I get:
> 
>    detecting Absoft Fortran compiler...
>    f77 -V -c /tmp/__dummy.f -o /tmp/__dummy.o
>    32512: sh: f77: command not found
>    detecting SGI Fortran compiler...
>    f77 -version
>    32512: sh: f77: command not found
>    detecting Sun Fortran compiler...
>    f90 -V
>    32512: sh: f90: command not found
> 
> which is obviously not right.

Actually, it is ok.

> 
> But doing
> 
>    python setup.py build build_flib --fcompiler=Gnu

<snip>

>   File "scipy_distutils/command/build_flib.py", line 839, in get_opt
>     if march_flag:
> UnboundLocalError: local variable 'march_flag' referenced before assignment

Fixed in CVS. Thanks.

Pearu